{
  "status": "completed",
  "completed_count": 63,
  "applicable_count": 63,
  "failed_count": 0,
  "schema_version": 7,
  "config_hash": "57553b5cf4c0f862",
  "data": {
    "scenes": [
      {
        "scene_index": 1,
        "heading": "S#1. 숲속 / 해 질 무렵",
        "beat_title": "공포에 질려 도망치는 남자",
        "representative_moment": "땀방울이 맺힌 채 공포에 질린 표정으로 숲속을 내달리는 C06의 얼굴 클로즈업.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Tighter-than-stage CU at eye-level, tracking front-side with shallow depth of field and low-key dusk light",
            "t2i_prompt": "Photorealistic cinematic still. An eye-level head-and-shoulder composition, tracking in front. The camera focuses on C06O08 in a dark hiking jacket, a Korean adult man with an ordinary facial contour and short black hair, captured at a three-quarter angle from the right in mid-stride. His sweat-slick face and upper chest occupy the frame, dark eyes fixed on a distant point. His short black hair whips backward across the left edge in the direction opposite to his forward motion. Cool red-brown traces of blood and sweat reflect the fading low-key dusk light on his cheek. [L01: Compressed dark wet tree trunks and shaking damp green foliage smear behind his shoulders in shallow depth of field, bathed in thin gray forest mist].",
            "outfit_assignments": [
              {
                "character_id": "C06",
                "outlook_id": "O08"
              }
            ],
            "source_facts": [
              "Scene text: '숲속 / 해 질 무렵', '잔뜩 겁에 질린 남자 표정(T.S)'",
              "Shot 2 description: '머리카락이 뒤로 흩날리는 가운데, 땀방울이 맺힌 채 공포에 질려 일그러진 남자의 얼굴 클로즈업.'"
            ],
            "visual_inferences": [
              "fixed_element dusk_forest_environment: wet dark green foliage and low-key fading natural light",
              "Motion frozen mid-stride preserves the dynamic forward movement by showing hair whipping backward in the opposite direction"
            ],
            "creative_decisions": [
              "Used a head-and-shoulder composition to prevent the 'floating head' artifact common in extremely tight face framings",
              "Employed shallow depth of field to smear the background trunks, isolating the character's intense expression according to the skipped close framing rule"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "b6c2fecadfc33219",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "d0186e8fc8a296f5",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "N/A",
        "_shot_index": 2,
        "visible_entities": [
          "C06",
          "L01"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 1,
            "shot_index": 2
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A tighter-than-stage CU at eye-level, held in a tracking front-side position about half a meter from 남자’s face as the camera moves backward and slightly sideways through the trees. His sweat-slick face occupies roughly 75% of the frame, with windblown hair whipping across the left edge and compressed trunks smearing behind him in shallow DOF, making the running motion tremble at the margins.",
            "lighting_mood": "Low-key dusk light filters through wet foliage, cooling the skin tones while small red-brown traces of blood and sweat catch the last light.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C06",
              "L01"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C06",
                "outlook_id": "O08"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": null,
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "dusk_forest_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dense, dark forest setting at dusk, characterized by wet, dark green foliage and muddy ground covered with crushed leaves. The atmosphere is consistently illuminated by low-key, fading natural light filtering through the dense canopy, creating deep shadows.",
                "applies_to_shots": [
                  2,
                  4
                ],
                "source_facts": [
                  "Scene text: '숲속 / 해 질 무렵'",
                  "Shot 2: 'Low-key dusk light filters through wet foliage'",
                  "Shot 4: 'muddy green-black with fading dusk highlights', 'crushed leaves'"
                ],
                "visual_inferences": [
                  "The ground remains persistently muddy and damp across the scene, matching the wet foliage and suspended dust"
                ],
                "creative_decisions": [
                  "Anchoring the visual continuity across vastly different framings by maintaining a cohesive low-key dusk lighting profile and wet forest textures in the background"
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C06O08",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "fa2ba9f5faef47ad"
      },
      {
        "scene_index": 1,
        "heading": "S#1. 숲속 / 해 질 무렵",
        "beat_title": "검은 인영의 습격과 짓눌린 남자",
        "representative_moment": "C12가 흙먼지가 이는 바닥에 C06을 강하게 짓누르고 있는 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "ground-level medium shot, slightly canted low angle, muddy green-black palette",
            "t2i_prompt": "Photorealistic cinematic still. A ground-level medium shot with a slightly canted low angle, the handheld camera settled almost on the wet muddy floor of a dark forest at dusk. Broken grass stems bend toward the lens in the lower foreground. C12O00, a dark humanoid silhouette exerting force from above, with back to camera, crouched forcefully over the torso of the constrained figure. Showing the variant features carried from stable_traits, C12O00 appears as a backlit figure, body volume preserved against the dim ambient fill, head outline visible but face fully swallowed in shadow with no discernible features. Pinned down in the same crushed leaves they both occupy is C06O08 in a dark hiking jacket and pants, a Korean adult male with short black hair, in left profile, head forced back into the mud, eyes looking down, a visible blood marker and torn fabric on his thigh. Suspended airborne dust and freshly kicked-up wet leaves drift across the midground, briefly veiling the tangled bodies. The environment is illuminated by muddy green-black fading dusk highlights filtering through dense dark green foliage, casting deep shadows across the forest. Focus on the violent tension and suspended dust between the pinning silhouette and the fallen man.",
            "outfit_assignments": [
              {
                "character_id": "C06",
                "outlook_id": "O08"
              },
              {
                "character_id": "C12",
                "outlook_id": "O00"
              }
            ],
            "source_facts": [
              "Scene text: '검은 인영이 남자를 덮친다', '허벅지에 커다란 이빨 자국. 피가 흐르고'",
              "Shot description: ground-level MWS with a low angle, dust and crushed leaves drift across the foreground, briefly veiling the tangled bodies"
            ],
            "visual_inferences": [
              "fixed_element dusk_forest_environment: wet dark green foliage, muddy ground covered with crushed leaves, low-key fading natural light",
              "power imbalance pose: silhouette pinning the man from above, exerting force to hold him in the mud",
              "C12 face obscured policy applied to convey the 'no facial details' trait"
            ],
            "creative_decisions": [
              "canted angle to emphasize the sudden impact and violence of the collapse",
              "suspended dust veiling the characters to add occult horror mood and grounded texture to the interaction",
              "muddy green-black palette to reflect the suffocating weight of the attack"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "8ada9adbd3a83769",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "d36e3787385d3448",
              "validator": "scene_detail_owned_objects.v1",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 1,
        "dependency_reason": "이전 샷에서 확립된 숲속 배경(dusk_forest_environment)과 조명, 진흙 바닥의 질감을 그대로 유지",
        "_shot_index": 4,
        "visible_entities": [
          "C06",
          "C12",
          "L01"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 1,
            "shot_index": 4
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A ground-level MWS with a low angle, the handheld camera settled almost on the forest floor less than a meter from the collapsed bodies. Dust and crushed leaves drift across the foreground, briefly veiling 남자 and 검은 인영 tangled in the midground before focus breathes back onto their forms; the frame is slightly canted from the impact but remains derived from the fall path.",
            "lighting_mood": "The palette is muddy green-black with fading dusk highlights cutting through suspended dust, giving the collapse a suffocating occult weight.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C06",
              "C12",
              "L01"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C06",
                "outlook_id": "O08"
              },
              {
                "character_id": "C12",
                "outlook_id": "O00"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "not_applicable",
            "bg_id": null,
            "reference_usage": "none",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "no chain_bg is bound to this shot — describe the location freely without referencing a background image",
              "do not use 'from the reference' / 'the existing room' / 'preserving the room' style phrasing — there is no reference image to refer to"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "dusk_forest_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dense, dark forest setting at dusk, characterized by wet, dark green foliage and muddy ground covered with crushed leaves. The atmosphere is consistently illuminated by low-key, fading natural light filtering through the dense canopy, creating deep shadows.",
                "applies_to_shots": [
                  2,
                  4
                ],
                "source_facts": [
                  "Scene text: '숲속 / 해 질 무렵'",
                  "Shot 2: 'Low-key dusk light filters through wet foliage'",
                  "Shot 4: 'muddy green-black with fading dusk highlights', 'crushed leaves'"
                ],
                "visual_inferences": [
                  "The ground remains persistently muddy and damp across the scene, matching the wet foliage and suspended dust"
                ],
                "creative_decisions": [
                  "Anchoring the visual continuity across vastly different framings by maintaining a cohesive low-key dusk lighting profile and wet forest textures in the background"
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 1,
                "shot_index": 2,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C06O08",
                "policy": "required"
              },
              {
                "kind": "character_outlook",
                "id": "C12O00",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "a2caf80dd0c54c52"
      },
      {
        "scene_index": 2,
        "heading": "S#2. 도시 뒷골목 / N",
        "beat_title": "붉은 눈동자와 쓰러지는 여자",
        "representative_moment": "희미한 가로등 아래, 붉은 눈동자의 C10이 전봇대에 기대어 무릎이 꺾인 C09를 붙잡고 있는 정지된 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "A locked-off wide shot at eye-level from the mouth of the narrow alley, sickly yellow-green lighting isolating the subjects in deep shadow.",
            "t2i_prompt": "Photorealistic cinematic still. An observer's eye-level view from the mouth of a damp, narrow urban alleyway. Two full-body figures occupy the center, sharing the physical space against an old, stained, slightly wet concrete utility pole. C10O11 in dark clothes, a Korean adult male with black hair, a pale face, and red eyes acting as the only saturated color accent, stands at three-quarter angle from the left. He is bracing against the front side of the pole, exerting force to hold the other figure's weight, his eyes fixed on her. C09O10 in a night-out dress, a Korean adult female with long dark hair, is positioned at three-quarter angle from the right, her body line slanted downward with knees buckled toward the slick, dark asphalt visible beneath their feet, eyes looking down. The dirty concrete alley walls close in on both sides, reflecting a weak, sickly yellow-green pool of light cast by the dim overhead streetlamp. Focus on the harsh contrast between the isolated pool of yellow-green light and the enveloping deep black shadows, anchoring the tense power imbalance.",
            "outfit_assignments": [
              {
                "character_id": "C10",
                "outlook_id": "O11"
              },
              {
                "character_id": "C09",
                "outlook_id": "O10"
              }
            ],
            "source_facts": [
              "S#2. 도시 뒷골목 / N",
              "도심 희미한 가로등 불빛. 전봇대에 뒤엉켜있는 두 사람.",
              "남자의 붉은 눈동자",
              "무릎이 꺾여 바닥으로 기울어지는 여자를 붙잡고 있는 정지된 전신."
            ],
            "visual_inferences": [
              "fixed_element dim_alley_environment: A damp, dark narrow urban alleyway illuminated by a single, weak, sickly yellow-green streetlamp.",
              "C10 exerting physical force to hold up C09 as she slumps downward with buckled knees."
            ],
            "creative_decisions": [
              "Eye-level framing capturing both full-body figures and the ground beneath them to satisfy spatial geometry rules.",
              "Highlighting the red eyes as the sole saturated color against the sickly green/black background to emphasize the occult tone.",
              "Describing the power imbalance explicitly through posture (bracing vs slanting/buckling) to match the physical struggle without adding unintended gore."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "f8931301e77004bd",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "7bdbd7c842fd3e31",
              "validator": "scene_detail_owned_objects.v1",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "N/A",
        "_shot_index": 1,
        "visible_entities": [
          "C10",
          "C09",
          "L02"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 2,
            "shot_index": 1
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A locked-off WS at eye-level from the mouth of the narrow alley, about seven meters from the utility pole, facing straight into the weak streetlamp pool. 붉은 눈동자 남자 and 여자 are held full-body near center, small against damp walls, with her body line slanting downward as her knees give way while he braces her against the pole.",
            "lighting_mood": "A thin yellow-green streetlamp isolates the pair in a damp black alley, with the man’s red eyes as the only saturated color accent.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C10",
              "C09",
              "L02"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C10",
                "outlook_id": "O11"
              },
              {
                "character_id": "C09",
                "outlook_id": "O10"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "not_applicable",
            "bg_id": null,
            "reference_usage": "none",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "no chain_bg is bound to this shot — describe the location freely without referencing a background image",
              "do not use 'from the reference' / 'the existing room' / 'preserving the room' style phrasing — there is no reference image to refer to"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "dim_alley_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A damp, dark narrow urban alleyway illuminated by a single, weak, sickly yellow-green streetlamp. A weathered concrete utility pole stands near the light source. The dim lighting highlights the wet, dark textures of the surrounding walls, leaving the rest of the scene in deep black shadow.",
                "applies_to_shots": [
                  1,
                  3
                ],
                "source_facts": [
                  "S#2. 도시 뒷골목 / N",
                  "도심 희미한 가로등 불빛. 전봇대에 뒤엉켜있는 두 사람.",
                  "A thin yellow-green streetlamp isolates the pair in a damp black alley",
                  "Sickly low-key streetlight"
                ],
                "visual_inferences": [
                  "The alley walls and ground are wet or damp, reflecting the sickly yellow-green light to emphasize the grim setting."
                ],
                "creative_decisions": [
                  "Focused on the harsh contrast between the isolated pool of yellow-green light and the enveloping deep shadows to anchor the eerie atmosphere across both wide and close-up framings."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C10O11",
                "policy": "required"
              },
              {
                "kind": "character_outlook",
                "id": "C09O10",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "e3b6d5de514d47a0"
      },
      {
        "scene_index": 2,
        "heading": "S#2. 도시 뒷골목 / N",
        "beat_title": "붉은 눈동자와 드러난 이빨",
        "representative_moment": "C10이 C09의 목을 물어뜯으며 붉은 눈동자를 드러내는 클로즈업 찰나.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Tilted shoulder-height framing, shallow depth of field, sickly low-key yellow-green lighting with a deep red accent.",
            "t2i_prompt": "Photorealistic cinematic still. The camera is tilted, positioned at shoulder height beside a weathered concrete utility pole, utilizing a shallow depth of field to isolate the physical interaction. C10O11 in a dark jacket, a Korean adult man with pale skin and a deep red glint in his eye, occupies the left as the figure exerting force. He is in right profile, his jaw clamped firmly forward, his teeth pressed deeply into the skin at the contact point on the neck. C09O10 in a dark coat, a Korean adult woman with long dark hair, serves as the figure receiving the force, her neck and shoulder anchored in the lower right. Her head is sagging sideways with eyes closed, looking away from the camera, her features partially softened by the focal plane. Sickly low-key yellow-green ambient light grazes his pale skin, jawbone, and the damp collar of her coat, contrasting against the deep shadows. [L02: The dark stained, damp vertical surface of the concrete utility pole anchors the left edge, while the wet, pitch-black alley texture is heavily compressed in the background with a tiny blurred streetlight bokeh].",
            "outfit_assignments": [
              {
                "character_id": "C10",
                "outlook_id": "O11"
              },
              {
                "character_id": "C09",
                "outlook_id": "O10"
              }
            ],
            "source_facts": [
              "S#2. 도시 뒷골목 / N",
              "남자의 붉은 눈동자, 목의 살점을 뜯는 이빨",
              "푹 힘없이 쓰러지는 여자",
              "ECU with a Dutch angle... teeth-to-skin contact sits just off center"
            ],
            "visual_inferences": [
              "fixed_element dim_alley_environment: damp texture, sickly yellow-green light reflecting on surfaces.",
              "C10 exerts downward forward force while C09 sags sideways, receiving the attack.",
              "Shallow depth of field softens C09's face to comply with close-framing dual-face restrictions while keeping C10's jaw and bite sharp."
            ],
            "creative_decisions": [
              "Used structured violence keywords ('figure exerting force', 'figure receiving the force', 'teeth pressed deeply... at the contact point') to prevent the AI from generating an intimate embrace.",
              "Restricted the background entirely to the frame-edge surface of the utility pole and compressed dark bokeh to strictly obey the skipped_close_framing rule without referencing the prior room layout.",
              "Avoided exact measurements like 'thirty centimeters' and translating 'ECU' directly into the prompt text, using spatial arrangement instead."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "f6165554a12de2f3",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "72faaeff049949e1",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 2,
        "dependency_reason": "Same location and interaction continuing immediately from S2_Shot1, zooming into a detailed framing of the bite.",
        "_shot_index": 3,
        "visible_entities": [
          "C10",
          "C09",
          "L02"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 2,
            "shot_index": 3
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "An ECU with a Dutch angle at shoulder height beside the utility pole, only about thirty centimeters from 여자’s neck, angled tightly across 붉은 눈동자 남자’s cheek and jawline. The teeth-to-skin contact sits just off center in the lower right, while most of the alley is excluded; shallow DOF holds the instant like a diseased still photograph rather than an action beat.",
            "lighting_mood": "Sickly low-key streetlight grazes skin and jawbone, while a deep red glint in the eye punctures the otherwise muted black-green frame.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C10",
              "C09",
              "L02"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C10",
                "outlook_id": "O11"
              },
              {
                "character_id": "C09",
                "outlook_id": "O10"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": null,
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "dim_alley_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A damp, dark narrow urban alleyway illuminated by a single, weak, sickly yellow-green streetlamp. A weathered concrete utility pole stands near the light source. The dim lighting highlights the wet, dark textures of the surrounding walls, leaving the rest of the scene in deep black shadow.",
                "applies_to_shots": [
                  1,
                  3
                ],
                "source_facts": [
                  "S#2. 도시 뒷골목 / N",
                  "도심 희미한 가로등 불빛. 전봇대에 뒤엉켜있는 두 사람.",
                  "A thin yellow-green streetlamp isolates the pair in a damp black alley",
                  "Sickly low-key streetlight"
                ],
                "visual_inferences": [
                  "The alley walls and ground are wet or damp, reflecting the sickly yellow-green light to emphasize the grim setting."
                ],
                "creative_decisions": [
                  "Focused on the harsh contrast between the isolated pool of yellow-green light and the enveloping deep shadows to anchor the eerie atmosphere across both wide and close-up framings."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 2,
                "shot_index": 1,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C10O11",
                "policy": "required"
              },
              {
                "kind": "character_outlook",
                "id": "C09O10",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "7e9f2986a14ceb10"
      },
      {
        "scene_index": 3,
        "heading": "S#3. 현장 / D",
        "beat_title": "연쇄 살인 사건 수사본부 지휘",
        "representative_moment": "혜수(C02)가 감식 현장에서 긴장된 어깨를 하고 특정 지점을 매섭게 응시한다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Broadcast-like MCU from chest-height, slight low angle, desaturated digital news-camera compression texture",
            "t2i_prompt": "Photorealistic cinematic still. A broadcast-like medium close-up from a chest-height, slight low angle. C02O03 in a dark detective coat, a Korean middle-aged woman with black hair, mature facial features, and sharp eyes, in a head-and-shoulder composition occupying the center-right. She is positioned at a three-quarter left angle, her shoulders squared and tense, eyes fixed on an unseen point just outside the frame. A yellow evidence marker partially enters the lower left frame edge. The lighting features flat daylight desaturated by digital news-camera compression, with clinical gray-blue shadows washing the side of her face and coat. [L03: Soft out-of-focus background of a daytime crime scene, taut yellow police line tape stretching diagonally across the rear left, faint blurred shapes of investigators in forensic gear].",
            "outfit_assignments": [
              {
                "character_id": "C02",
                "outlook_id": "O03"
              }
            ],
            "source_facts": [
              "S#3 action: '형사무리 속에 사건을 지휘하는 혜수의 모습이 보인다.'",
              "shot description: 감식 현장 한가운데 선 혜수가 미간을 찌푸린 채 특정 지점을 매섭게 응시하는 상체"
            ],
            "visual_inferences": [
              "chest-height camera angle placing the upper body in frame",
              "digital news-camera compression texture and flat daylight",
              "soft out-of-focus background of crime scene with yellow police line tape"
            ],
            "creative_decisions": [
              "head-and-shoulder composition placed center-right to emphasize her tense stance",
              "evidence marker placed at the lower left frame edge to visually anchor her off-camera gaze"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "a65583b1ccb56152",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "7f2e1ccdbcebdf4c",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "N/A",
        "_shot_index": 2,
        "visible_entities": [
          "C02",
          "L03"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 3,
            "shot_index": 2
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A broadcast-like MCU from a chest-height, slightly off-center position about 1.5 meters in front of 혜수, finishing a restrained dolly-in with a slight low angle. 혜수 occupies the center-right of frame, shoulders tense and face turned three-quarters toward an evidence point just outside the left edge, while police activity remains softened behind her as televised background noise.",
            "lighting_mood": "Flat daylight is desaturated by news-camera compression, with gray-blue shadows and damp ground texture keeping the scene clinical but uneasy.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C02",
              "L03"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C02",
                "outlook_id": "O03"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": null,
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [],
            "previous_shot_refs": [],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C02O03",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "436b272e0613281d"
      },
      {
        "scene_index": 4,
        "heading": "S#4. 옥탑방 / D",
        "beat_title": "불안감이 감도는 빈 옥탑방 전경",
        "representative_moment": "햇빛이 비치는 낡은 옥탑방 전경이 차갑고 정적인 부감으로 내려다보인다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Extreme wide overhead bird's eye view looking straight down, muted beige-gray tones",
            "t2i_prompt": "Photorealistic cinematic still. An extreme wide shot in an overhead bird's eye view, looking straight down from high above at the aging villa rooftop from the reference. Focus on the empty wooden platform from the reference sitting slightly off-center within the roof rectangle. The area is bounded by the parapet walls and features the water tank from the reference, with cramped neighboring rooftops filling the edges of the frame. Muted beige-gray tones and pale daytime sunlight wash the surfaces, while the rigid overhead stillness gives the ordinary empty space a cold, watched quality.",
            "outfit_assignments": [],
            "source_facts": [
              "S#4. 옥탑방 / D: 인천 변두리 부감, 오래된 빌라 옥탑방 전경.",
              "shot description: A static EWS in overhead / bird's eye view from roughly 25 meters above the aging villa rooftop, looking straight down."
            ],
            "visual_inferences": [
              "Empty platform symbolizing absence and tension",
              "Pale daylight conveying a cold, watched mood rather than warmth",
              "Overhead bird's eye view emphasizing urban isolation"
            ],
            "creative_decisions": [
              "Anchor objects explicitly using 'from the reference' due to exact_background mode",
              "Use muted beige-gray tones to reflect the damp urban realism and suspenseful atmosphere"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "40e9bc32864287f6",
              "owned_hash": "9cc93746643fe6bd",
              "camera_direction_hash": "5e57252bfd3c4bde",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "platform",
                  "violating_phrase": "the empty wooden platform from the reference",
                  "reason": "The phrase explicitly anchors the platform to the reference, indicating the existing background object rather than a new redraw.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "parapet",
                  "violating_phrase": "the parapet walls",
                  "reason": "The parapet is mentioned only as an existing boundary element in the scene, with no explicit create/add/transform instruction.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "water tank",
                  "violating_phrase": "the water tank from the reference",
                  "reason": "The phrase explicitly refers to the existing water tank from the reference, which is an anchor reference rather than a redraw command.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "N/A",
        "_shot_index": 1,
        "visible_entities": [
          "L04"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 4,
            "shot_index": 1
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A static EWS in overhead / bird's eye view from roughly 25 meters above the aging villa rooftop, looking straight down without beginning the later crane withdrawal. The wooden platform sits slightly off center within the roof rectangle, surrounded by parapet walls, water stains, and cramped neighboring rooftops that make the sunlit space feel exposed rather than safe.",
            "lighting_mood": "Pale daytime sun washes the rooftop in muted beige-gray tones, but the rigid overhead stillness gives the ordinary space a cold, watched quality.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "L04"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "cb_rooftop_day_normal",
            "reference_usage": "exact_background",
            "owned_objects": [
              "clothesline",
              "door",
              "drain",
              "light",
              "parapet",
              "pipe",
              "platform",
              "post",
              "rooftop deck",
              "stairwell bulkhead",
              "water tank",
              "window"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [],
            "previous_shot_refs": [],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "background",
                "id": "cb_rooftop_day_normal",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "62d381b4d3435252"
      },
      {
        "scene_index": 5,
        "heading": "S#5. 옥탑방 안 / 실내",
        "beat_title": "연쇄 살인 뉴스를 심각하게 보는 민숙",
        "representative_moment": "앞치마를 두른 민숙이 거실 한가운데 서서 굳은 표정으로 TV를 노려보고 있다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Eye-level waist-up composition, low-key gray lighting with cold TV spill.",
            "t2i_prompt": "Photorealistic cinematic still. C04O05 in an apron over indoor clothes, a Korean middle-aged woman with black hair, gentle facial lines, and dark eyes, occupies the central vertical axis, waist-up visible. She stands facing the camera, hands gripping the apron tightly, her solemn gaze locked slightly past the lens toward an unseen television. Thin sunlight enters from the sliding window from the background reference behind her, mixing with a cold blue TV light spill washing the front of her face and apron. The ambient lighting leaves her features in low-key gray. The wall and the sink from the exact background reference remain softly visible at the rear edges, keeping the foreground clear so her fixed expression remains the primary focus.",
            "outfit_assignments": [
              {
                "character_id": "C04",
                "outlook_id": "O05"
              }
            ],
            "source_facts": [
              "S#5 action: '앞치마를 두른 채 TV 뉴스를 뚫어지게 바라보는 민숙 표정이 어둡다.'",
              "shot description: Minsook standing in the living room, facing camera, gaze locked on TV"
            ],
            "visual_inferences": [
              "hands gripping the apron tightly to reflect tension",
              "waist-up visible to match the medium framing scope",
              "solemn expression to translate 'dark expression' safely"
            ],
            "creative_decisions": [
              "placed C04O05 exactly on the central vertical axis to isolate her stiff posture",
              "washed the front of her face with a cold blue TV light spill to create a low-key mood",
              "anchored the environment by mentioning the sliding window, wall, and sink from the attached background reference"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "65427a11d621be9f",
              "owned_hash": "660e140836e49bf2",
              "camera_direction_hash": "6bcdb7fef9361f3b",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "TV",
                  "violating_phrase": "unseen television",
                  "reason": "The prompt mentions the TV as an off-camera reference target, but it does not instruct creating or modifying a TV with a redraw verb.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "sliding window",
                  "violating_phrase": "the sliding window from the background reference",
                  "reason": "The sliding window is explicitly referenced as part of the background reference, which anchors it rather than requesting a new one.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "sink",
                  "violating_phrase": "the sink from the exact background reference",
                  "reason": "The sink is explicitly described as coming from the exact background reference, so it is a reference anchor, not a redraw instruction.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "N/A",
        "_shot_index": 2,
        "visible_entities": [
          "C04",
          "P02",
          "P10",
          "P01",
          "L05"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 5,
            "shot_index": 2
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A static MS at eye-level from about 2.5 meters in front of 민숙, framing her frontally in the living room while the television remains just off camera near the lens line. She occupies the central vertical axis in her apron, gaze locked slightly past camera toward the unseen screen, with the kitchen foreground excluded so her fixed expression becomes the only active element.",
            "lighting_mood": "Thin window sunlight mixes with cold TV spill, leaving 민숙’s face low-key and gray with a faint domestic warmth turning sour.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C04",
              "P02",
              "P10",
              "P01",
              "L05"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C04",
                "outlook_id": "O05"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "cb_living_day_herb_tea_news",
            "reference_usage": "exact_background",
            "owned_objects": [
              "AC",
              "TV",
              "TV console",
              "cooktop",
              "curtain",
              "document",
              "door",
              "floor cushion",
              "kitchen counter",
              "mug",
              "photograph",
              "refrigerator",
              "shoe rack",
              "sink",
              "sliding window",
              "sofa",
              "table"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [],
            "previous_shot_refs": [],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C04O05",
                "policy": "required"
              },
              {
                "kind": "background",
                "id": "cb_living_day_herb_tea_news",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "f51e95784ecb279b"
      },
      {
        "scene_index": 5,
        "heading": "S#5. 옥탑방 안 / 실내",
        "beat_title": "수리영이 엄마 손목의 또렷한 표식을 인지함",
        "representative_moment": "수리영의 어깨 너머로 시야를 낮춘 카메라가, 민숙이 식탁에 약초 달인 찻잔을 내려놓는 순간 드러난 손목의 붉은 표식에 집중한다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Table-height close-up with a low angle, tight focus on the wrist and teacup, muted gray background with narrow daylight catching the warm steam.",
            "t2i_prompt": "Photorealistic cinematic still. Close on a steaming teacup P10 of herbal drink, its rim visible from a low angle slightly below table height, resting on [L05: a worn dining table with its top plane receding away]. The cup occupies a small area in the lower center frame. Warm steam curls upward, catching the narrow daylight and creating a faintly medicinal, ominous mood against the muted gray background. In the right foreground, a young Korean woman's blurred shoulder is visible, perched at the same table's edge in an over-the-shoulder angle, her posture directed toward the cup. Crossing in from the left is a middle-aged Korean woman's lower forearm and inner wrist, captured mid-reach as her hand sets the cup down. A small, sharply defined red circular mark P01 is distinctly visible on the skin of her inner wrist, exposed past the edge of a casual sleeve. The focus is tight on the red mark P01 and the teacup P10. The distant background is softly blurred, showing a wet kitchen sink basin, damp herbs P02, and a simmering pot.",
            "outfit_assignments": [],
            "source_facts": [
              "Scene text: 힐끗, 민숙 손목에 보이는 작은 원(이하 표식)을 본다.",
              "Scene text: 오늘따라 유난히 또렷해 보이지만",
              "Shot 5 camera: A table-height CU with a low angle beside 수리영’s seat",
              "Shot 5 camera: steam curling upward between 수리영’s blurred shoulder at right foreground and 민숙’s forearm crossing in from the left."
            ],
            "visual_inferences": [
              "fixed_element: A close-up detail view of an East Asian woman's lower forearm and inner wrist. A small, sharply defined red circular mark is distinctly visible on the skin of the inner wrist.",
              "The red mark is exposed past the edge of an indoor sleeve as the arm reaches across the shared table.",
              "The scene captures the exact moment the cup is placed down (mid-reach freeze).",
              "The muted gray tone of the room emphasizes the ominous feeling of the newly placed cup."
            ],
            "creative_decisions": [
              "Used 'Close on' and common noun phrasing to trigger the body-part focus rule, explicitly avoiding pasting full faces onto the shoulder/arm slices.",
              "Placed the teacup in the lower center frame to anchor the foreground elements.",
              "Described the shared table surface directly within an [L05: ...] block, complying with skipped_close_framing rules to exclude existing background references."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "23bb1ba6c37a3db6",
              "owned_hash": "660e140836e49bf2",
              "camera_direction_hash": "703c4d8e0c4249cd",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 5,
        "dependency_reason": "같은 씬 내 이전 샷(S5_Shot2)에서 설정된 공간 분위기와 수리영의 위치를 공유하며, 민숙의 손목 표식을 클로즈업으로 확대 묘사함.",
        "_shot_index": 5,
        "visible_entities": [
          "C04",
          "C01",
          "P02",
          "P10",
          "P01",
          "L05"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 5,
            "shot_index": 5
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A table-height CU with a low angle beside 수리영’s seat, dollying slightly toward the cup as 민숙’s hand enters from the kitchen side. The herbal drink lands in the lower center and occupies less than a quarter of the frame, steam curling upward between 수리영’s blurred shoulder at right foreground and 민숙’s forearm crossing in from the left.",
            "lighting_mood": "Warm steam catches the narrow daylight while the rest of the room stays muted and gray, making the ordinary cup feel faintly medicinal and ominous.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C04",
              "C01",
              "P02",
              "P10",
              "P01",
              "L05"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C04",
                "outlook_id": "O05"
              },
              {
                "character_id": "C01",
                "outlook_id": "O01"
              },
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "cb_living_day_herb_tea_news",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "minsuk_wrist_mark_detail",
                "element_type": "character_state",
                "character_name": "민숙",
                "description": "A close-up detail view of an East Asian woman's lower forearm and inner wrist. A small, sharply defined red circular mark is distinctly visible on the skin of the inner wrist. The wrist area is exposed past the edge of a sleeve.",
                "applies_to_shots": [
                  5,
                  7
                ],
                "source_facts": [
                  "Scene text: 힐끗, 민숙 손목에 보이는 작은 원(이하 표식)을 본다.",
                  "Scene text: 오늘따라 유난히 또렷해 보이지만",
                  "Shot 5 camera: 민숙’s forearm crossing in from the left.",
                  "Shot 7 camera: pushed within thirty centimeters of 민숙’s inner wrist... small circular mark fill most of the frame"
                ],
                "visual_inferences": [
                  "The sleeve is either short or slightly pulled back to naturally expose the inner wrist and the mark during arm movements."
                ],
                "creative_decisions": [
                  "Framing class 'zoom/detail' is applied to anchor the physical appearance of the distinct mark on the skin without detailing the entire body, ensuring consistency across close-ups."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 5,
                "shot_index": 2,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C04O05",
                "policy": "required"
              },
              {
                "kind": "character_outlook",
                "id": "C01O01",
                "policy": "required"
              },
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "7069f6e3f6edf009"
      },
      {
        "scene_index": 5,
        "heading": "S#5. 옥탑방 안 / 실내",
        "beat_title": "수리영이 엄마 손목의 선명한 표식을 인지함",
        "representative_moment": "민숙의 소매 끝선 아래로 드러난 손목 안쪽의 붉고 선명한 원형 표식 클로즈업.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "subjective POV ECU at chest height, slight high angle, low-key interior with a thin strip of daylight",
            "t2i_prompt": "Photorealistic cinematic still. A subjective POV extreme close-up at chest height, tight on a middle-aged East Asian woman's lower forearm and inner wrist at a slight high angle. Focus on the inner wrist where a small, sharply defined red circular mark is distinctly visible on the skin. The wrist, turned outward, is exposed past the edge of a muted fabric sleeve. A thin strip of daylight grazes the wrist against a low-key interior, making the small red circle appear unnaturally crisp without overt supernatural glow. The sleeve edge and an out-of-focus apron blur provide scale, while [L05: a dim beige-gray entryway wall] is minimized into soft shadow at the frame edge.",
            "outfit_assignments": [],
            "source_facts": [
              "Scene text: 힐끗, 민숙 손목에 보이는 작은 원(이하 표식)을 본다.",
              "Scene text: 오늘따라 유난히 또렷해 보이지만",
              "Shot 7 camera: pushed within thirty centimeters of 민숙’s inner wrist... small circular mark fill most of the frame"
            ],
            "visual_inferences": [
              "fixed_element minsuk_wrist_mark_detail: close-up detail view of lower forearm and inner wrist exposed past sleeve",
              "wrist turned outward to face the camera POV",
              "muted fabric sleeve and out-of-focus apron"
            ],
            "creative_decisions": [
              "applied body part focus rule, using common noun and demographic descriptor instead of C04 ID to prevent floating face rendering on the wrist",
              "used dim beige-gray entryway wall in soft shadow for L05 to match close framing background rules without citing external references"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "91e27bc864a3a16c",
              "owned_hash": "660e140836e49bf2",
              "camera_direction_hash": "6f0f2a037fce5d43",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 5,
        "dependency_reason": "Shot 5의 민숙 자세 및 S5_Shot2의 배경 톤/조명 분위기 연속성 유지",
        "_shot_index": 7,
        "visible_entities": [
          "C04",
          "P02",
          "P10",
          "P01",
          "L05"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 5,
            "shot_index": 7
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A subjective POV ECU from 수리영’s side at chest height, pushed within thirty centimeters of 민숙’s inner wrist at a slight high angle. The wrist and small circular mark fill most of the frame but not the whole image, with sleeve edge and apron blur providing scale while the surrounding entryway is minimized into soft shadow.",
            "lighting_mood": "A thin strip of daylight grazes the wrist against a low-key interior, making the small circle appear unnaturally crisp without overt supernatural glow.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C04",
              "P02",
              "P10",
              "P01",
              "L05"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C04",
                "outlook_id": "O05"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "cb_living_day_herb_tea_news",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "minsuk_wrist_mark_detail",
                "element_type": "character_state",
                "character_name": "민숙",
                "description": "A close-up detail view of an East Asian woman's lower forearm and inner wrist. A small, sharply defined red circular mark is distinctly visible on the skin of the inner wrist. The wrist area is exposed past the edge of a sleeve.",
                "applies_to_shots": [
                  5,
                  7
                ],
                "source_facts": [
                  "Scene text: 힐끗, 민숙 손목에 보이는 작은 원(이하 표식)을 본다.",
                  "Scene text: 오늘따라 유난히 또렷해 보이지만",
                  "Shot 5 camera: 민숙’s forearm crossing in from the left.",
                  "Shot 7 camera: pushed within thirty centimeters of 민숙’s inner wrist... small circular mark fill most of the frame"
                ],
                "visual_inferences": [
                  "The sleeve is either short or slightly pulled back to naturally expose the inner wrist and the mark during arm movements."
                ],
                "creative_decisions": [
                  "Framing class 'zoom/detail' is applied to anchor the physical appearance of the distinct mark on the skin without detailing the entire body, ensuring consistency across close-ups."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 5,
                "shot_index": 2,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C04O05",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "c3261436ddb09e4e"
      },
      {
        "scene_index": 6,
        "heading": "S#6. 해안가 도로 / D",
        "beat_title": "해안 도로 라이딩 중 낯선 개 짖는 소리에 고개를 돌리는 수리영",
        "representative_moment": "C01 snaps her head toward an off-screen bark while cycling, her expression shifting to sudden attention.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Eye-level moving close-up from a tight side-front position, panning to blur the background, desaturated cool blue-gray tones.",
            "t2i_prompt": "Photorealistic cinematic still. C01O02 in a dark activewear jacket, a Korean young woman, captured in a moving close-up at eye-level from a tight side-front position at three-quarter angle from the right. Her head is snapped sideways toward the left frame edge, her golden wave-like eyes fixed intensely on an off-screen point. Her soft facial structure is taut with sudden attention, and a white earphone is visible in her ear. Her long black hair flows backward, preserving the direction opposite to her forward cycling motion. At the lower edge, her hands grip the blurred bicycle handlebar, anchoring the mid-ride action. Cool blue-gray ambient light from the desaturated coastal daylight reflects on her cheek and the shoulder of her jacket. [L06: The frame edge is backed by a heavily blurred streak of a gray coastal road and a bright hazy blue-gray sea horizon, conveying rapid motion under soft wind.]",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#6 action: 수리영 자전거를 타고 해안가 도로를 달린다. 컹! 컹! (개 짖는 소리)",
              "shot description: head snaps toward the offscreen bark, eyes cutting toward frame left"
            ],
            "visual_inferences": [
              "mid-ride freeze with hair blowing backward to preserve motion direction",
              "tight framing requires the bicycle handlebar at the lower edge to ground the cycling action"
            ],
            "creative_decisions": [
              "applied heavy horizontal motion blur to the road and horizon to simulate the panning camera effect",
              "used cool blue-gray lighting to emphasize the sea haze and undercut the initially lively mood"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "4679089a408c856c",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "25654e585875eaf7",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "None",
        "_shot_index": 2,
        "visible_entities": [
          "C01",
          "L06"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 6,
            "shot_index": 2
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A moving CU at eye-level from a tight side-front position alongside 수리영’s bicycle, panning with her as her head snaps toward the offscreen bark. The earphone side of her face is framed slightly tighter, her eyes cutting toward frame left while a sliver of handlebar and the coastal road streak through the lower edge to keep the cycling motion alive.",
            "lighting_mood": "Bright coastal daylight is desaturated by sea haze, with soft wind and cool blue-gray tones undercutting her initially lively mood.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "L06"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": null,
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [],
            "previous_shot_refs": [],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "c899ee446dc0fba3"
      },
      {
        "scene_index": 7,
        "heading": "S#7. 유기견 보호원 / D",
        "beat_title": "맹견을 압도하는 수리영의 시선",
        "representative_moment": "수리영이 허리를 숙여 철망 너머 로트와일러의 눈을 정면으로 노려보며 기세를 꺾는 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Intense CU, low angle from rottweiler eye height, shallow DOF",
            "t2i_prompt": "Photorealistic cinematic still. At rottweiler eye height, low angle looking up, head-and-shoulder composition. C01O02 in a dark work jumper, a young Korean woman featuring long black hair, a soft facial contour, and striking golden wave-like irises, positioned at a three-quarter angle from the left, bending toward the cage. Her face occupies the upper two-thirds of the frame, slightly right of center, with neck and upper chest visible. Her gaze is fixed on the animal below. In the lower left foreground, C07O00, a large dog with a broad head, strong jaw, dark brown short coat, and dark eyes, appears as a smeared foreground slice seen over the shoulder, eyes fixed on C01O02. The blurred dark metal kennel bars cross the air between them, sharing a single line of sight. The sharp focus holds firmly on C01O02's intense expression. The immediate background consists of [L07: an out-of-focus stained concrete wall]. Low-key greenish fluorescent daylight mixes with cold damp shadows falling on her skin, while a faint dirty red warning tag, angled away from the camera, barely bleeds through the haze in the shallow depth of field.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              },
              {
                "character_id": "C07",
                "outlook_id": "O00"
              }
            ],
            "source_facts": [
              "Shot 2: 허리를 숙인 수리영이 로트와일러의 눈을 정면으로 노려보는 강렬한 클로즈업",
              "사납게 날뛰며 위협적으로 짖는 상태 -> 기세가 꺾이며 순한 양처럼 얌전해지는 상태",
              "Entity traits: C01 (golden wave-like irises, soft facial contour, long black hair), C07 (large dog, broad head, strong jaw, dark brown short coat)"
            ],
            "visual_inferences": [
              "fixed_element shelter_kennel_environment: dimly lit indoor dog shelter, heavy vertical dark metal bars, cold greenish overhead fluorescent lights",
              "The camera angle forces a power dynamic through the low viewpoint, emphasizing C01's dominant gaze over the dog"
            ],
            "creative_decisions": [
              "Placed C07 as a blurred foreground slice to maintain primary framing focus entirely on C01's face and eyes",
              "Used a dirty red tag to add a subtle contrasting color accent against the greenish-cold fluorescent tone, positioned to not distract from the primary subject",
              "Selected a head-and-shoulder composition to fulfill close framing rules while avoiding the 'face filling entire frame' pitfall"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "057d1af340dce1df",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "b62b06f686faf841",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "N/A",
        "_shot_index": 2,
        "visible_entities": [
          "C01",
          "C07",
          "P03",
          "L07"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 7,
            "shot_index": 2
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "From the kennel entrance at rottweiler eye height, use an intense CU, low angle, shallow DOF on 수리영’s eyes as she bends into the cage line; this is an observer’s direct view held on the confrontation axis inherited from the dolly-in. Her face occupies the upper two-thirds of frame, slightly right of center, while blurred kennel bars and the rottweiler’s dark eye-line sit as smeared foreground edges at the lower left, keeping the shelter environment present but suppressed.",
            "lighting_mood": "Low-key fluorescent daylight mixes with cold damp shadows, with a faint dirty red warning tag barely bleeding through the haze.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "C07",
              "P03",
              "L07"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              },
              {
                "character_id": "C07",
                "outlook_id": "O00"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": null,
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "shelter_kennel_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dimly lit indoor dog shelter characterized by a damp, wet concrete floor. Heavy vertical dark metal bars and wire kennel doors line the space, creating a repetitive structural boundary. The interior is illuminated by harsh, cold greenish overhead fluorescent lights, casting chilly and damp shadows across the corridor.",
                "applies_to_shots": [
                  2,
                  3,
                  5
                ],
                "source_facts": [
                  "Shot 2: 'Low-key fluorescent daylight mixes with cold damp shadows', 'blurred kennel bars'",
                  "Shot 3: 'wet concrete', 'cage bars remain vertical dark stripes', 'Cold overhead fluorescents'",
                  "Shot 5: 'Greenish shelter fluorescents', 'kennel doors stretching away'"
                ],
                "visual_inferences": [
                  "The wet concrete floor reflects the harsh fluorescent light, providing a consistent cold and damp texture across all shots.",
                  "The vertical metal bars act as a persistent architectural element that visually confines the space."
                ],
                "creative_decisions": [
                  "Maintaining a consistent greenish-cold color grade and wet floor texture to emphasize the sterile, damp, and somewhat bleak atmosphere of the animal shelter."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              },
              {
                "kind": "character_outlook",
                "id": "C07O00",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "38479319860ae31d"
      },
      {
        "scene_index": 7,
        "heading": "S#7. 유기견 보호원 / D",
        "beat_title": "제압된 맹견",
        "representative_moment": "바닥에 납작 엎드린 채 기세가 꺾인 로트와일러의 모습",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "ground-level perspective, medium wide framing, cold greenish fluorescent lighting",
            "t2i_prompt": "Photorealistic cinematic still. Viewed from ground level just inside the kennel threshold, C07O00, a large dog with short dark brown fur, a broad head, and a strong jaw, occupies the central foreground at a three-quarter angle from the right. The dog is flattened completely against the wet concrete floor, eyes directed downward. Focus on the dog's pinned ears and front paws near the bottom edge, with the body angled diagonally toward the rear wall. Cold greenish overhead fluorescent light skims the damp, scuffed concrete surface, leaving C07O00's dark coat in heavy shadow with subtle highlights visible on the eyes and muzzle. Heavy vertical dark metal bars frame both the left and right edges. A dented metal food bowl sits near the dimly lit rear wall of the shelter.",
            "outfit_assignments": [
              {
                "character_id": "C07",
                "outlook_id": "O00"
              }
            ],
            "source_facts": [
              "S7 Shot 3: worm's eye view of the 로트와일러 flattened near the floor",
              "S7 Shot 3: dog’s pinned ears and front paws sit in the central foreground",
              "S7 Shot 3: cage bars remain vertical dark stripes at both frame edges"
            ],
            "visual_inferences": [
              "fixed_element shelter_kennel_environment: cold greenish overhead fluorescent lights, damp wet concrete floor",
              "The dog's eyes directed downward visualizes the subdued and submissive state from the scene text"
            ],
            "creative_decisions": [
              "Ground-level perspective adopted without using camera directive keywords to emphasize the physical weight and submission of the dog against the floor",
              "Deep shadows on the dog's fur introduced to contrast with the harsh green reflections on the concrete, heightening the bleak atmosphere"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "d3eefbf9755689b1",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "128b7e3dcac81a45",
              "validator": "scene_detail_owned_objects.v1",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 7,
        "dependency_reason": "이전 샷의 유기견 보호원 실내 환경(젖은 바닥, 쇠창살 등)과 조명 톤(차가운 형광등) 연속성 유지",
        "_shot_index": 3,
        "visible_entities": [
          "C07",
          "P03",
          "L07"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 7,
            "shot_index": 3
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Just inside the kennel threshold, lower the lens almost to the wet concrete for a MWS, worm's eye view of the 로트와일러 flattened near the floor, continuing the tilt down from 수리영’s stare. The dog’s pinned ears and front paws sit in the central foreground, its body receding diagonally toward the rear of the kennel, while the cage bars remain vertical dark stripes at both frame edges.",
            "lighting_mood": "Cold overhead fluorescents skim the wet floor, leaving the dog’s black fur in heavy shadow with only small highlights on its eyes and muzzle.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C07",
              "P03",
              "L07"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C07",
                "outlook_id": "O00"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "not_applicable",
            "bg_id": null,
            "reference_usage": "none",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "no chain_bg is bound to this shot — describe the location freely without referencing a background image",
              "do not use 'from the reference' / 'the existing room' / 'preserving the room' style phrasing — there is no reference image to refer to"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "shelter_kennel_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dimly lit indoor dog shelter characterized by a damp, wet concrete floor. Heavy vertical dark metal bars and wire kennel doors line the space, creating a repetitive structural boundary. The interior is illuminated by harsh, cold greenish overhead fluorescent lights, casting chilly and damp shadows across the corridor.",
                "applies_to_shots": [
                  2,
                  3,
                  5
                ],
                "source_facts": [
                  "Shot 2: 'Low-key fluorescent daylight mixes with cold damp shadows', 'blurred kennel bars'",
                  "Shot 3: 'wet concrete', 'cage bars remain vertical dark stripes', 'Cold overhead fluorescents'",
                  "Shot 5: 'Greenish shelter fluorescents', 'kennel doors stretching away'"
                ],
                "visual_inferences": [
                  "The wet concrete floor reflects the harsh fluorescent light, providing a consistent cold and damp texture across all shots.",
                  "The vertical metal bars act as a persistent architectural element that visually confines the space."
                ],
                "creative_decisions": [
                  "Maintaining a consistent greenish-cold color grade and wet floor texture to emphasize the sterile, damp, and somewhat bleak atmosphere of the animal shelter."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 7,
                "shot_index": 2,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C07O00",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "eb97eb9421e6f878"
      },
      {
        "scene_index": 7,
        "heading": "S#7. 유기견 보호원 / D",
        "beat_title": "연락이 닿지 않는 엄마와 의아함",
        "representative_moment": "수리영이 굳은 표정으로 스마트폰의 발신 화면을 내려다보고 있다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "static MCU, high angle through-object composition, greenish fluorescent and cold screen glow",
            "t2i_prompt": "Photorealistic cinematic still. MCU high angle through-object composition from chest height. In the lower third foreground, P03 smartphone is positioned with the outgoing-call screen angled upward toward the camera. Above it in the center-right midground, C01O02 in a dark activity jacket, a Korean young woman with long black hair, soft facial structure, and golden wave-like irises, leans forward at three-quarter angle from the right, eyes fixed on the smartphone screen. Her upper chest and shoulders remain in the frame. Focus on C01O02's hardened expression hanging above the device. Greenish overhead fluorescent ambient washes her hair and shoulders, flattening the skin tone, the phone's small cold glow reflecting upward to pool under her tense jaw. Behind her left shoulder, the frame edge reveals [L07: dimly lit damp shelter aisle, heavy vertical dark metal bars and wire kennel doors receding into the background].",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#7 action: 민숙에게 연락을 취해보지만, 받지 않는다.",
              "Shot 5 description: 스마트폰 액정의 발신 화면을 묵묵히 내려다보는 수리영의 굳은 얼굴."
            ],
            "visual_inferences": [
              "fixed_element: dimly lit indoor dog shelter, heavy vertical dark metal bars, damp concrete",
              "phone's screen emits a cold glow upward onto her jaw due to the low angle foreground placement"
            ],
            "creative_decisions": [
              "Through-object composition places the phone in the extreme foreground lower third to visually anchor the failed contact attempt.",
              "Greenish ambient light contrasts with the cold screen glow to isolate the character emotionally in the damp environment."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "749eb889798fe73b",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "a75cf4c3d977f7e9",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 7,
        "dependency_reason": "이전 샷의 유기견 보호원 환경(철창, 조명 톤 등)과 일관성을 유지하되, 클로즈 프레이밍 규칙에 따라 배경 참조 이미지 합성 없이 직접 묘사함.",
        "_shot_index": 5,
        "visible_entities": [
          "C01",
          "P03",
          "L07"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 7,
            "shot_index": 5
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "In the shelter aisle at 수리영’s chest height, hold a static MCU, high angle through-object composition where the smartphone’s outgoing-call screen sits low in the foreground and her hardened face hangs above it in sharp emotional focus. The phone occupies only the lower third, angled upward toward camera, while 수리영 leans slightly over it in the center-right midground, kennel doors stretching away behind her left shoulder.",
            "lighting_mood": "Greenish shelter fluorescents flatten her skin tone, while the phone casts a small cold glow upward under her tense jaw.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P03",
              "L07"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": null,
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "shelter_kennel_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dimly lit indoor dog shelter characterized by a damp, wet concrete floor. Heavy vertical dark metal bars and wire kennel doors line the space, creating a repetitive structural boundary. The interior is illuminated by harsh, cold greenish overhead fluorescent lights, casting chilly and damp shadows across the corridor.",
                "applies_to_shots": [
                  2,
                  3,
                  5
                ],
                "source_facts": [
                  "Shot 2: 'Low-key fluorescent daylight mixes with cold damp shadows', 'blurred kennel bars'",
                  "Shot 3: 'wet concrete', 'cage bars remain vertical dark stripes', 'Cold overhead fluorescents'",
                  "Shot 5: 'Greenish shelter fluorescents', 'kennel doors stretching away'"
                ],
                "visual_inferences": [
                  "The wet concrete floor reflects the harsh fluorescent light, providing a consistent cold and damp texture across all shots.",
                  "The vertical metal bars act as a persistent architectural element that visually confines the space."
                ],
                "creative_decisions": [
                  "Maintaining a consistent greenish-cold color grade and wet floor texture to emphasize the sterile, damp, and somewhat bleak atmosphere of the animal shelter."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 7,
                "shot_index": 2,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "b6ab02316318f2b4"
      },
      {
        "scene_index": 8,
        "heading": "S#8. 번화가·대형 마트 / 해질 무렵",
        "beat_title": "사라진 엄마와 엇갈린 증언",
        "representative_moment": "A female employee points toward the entrance while C01 stands in the aisle, absorbing the news that her mother has already left.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Eye-level OTS-pan composition at shoulder height, contrasting harsh fluorescent lighting with blue dusk ambient spill.",
            "t2i_prompt": "Photorealistic cinematic still. From an eye-level position at shoulder height inside the supermarket, two figures stand across the shared aisle space they both occupy. A middle-aged Korean female employee in a worker uniform stands in the right foreground at three-quarter angle from the left, her arm extended pointing across the right half of the frame toward the sliding door from the reference in the background, her eyes fixed on the entrance. In the left-midground stands C01O02 in a dark jacket, a young Korean woman with long black hair, soft facial contour, and golden wave-like eyes, at three-quarter angle from the right, her weight shifted back and her eyes fixed on the employee. Harsh white fluorescent lights from the ceiling clash with the deep blue ambient dusk spill entering through the glass entrance doors from the reference, creating a stark, high-contrast boundary. Small worn red sale stickers on the product shelves from the reference add faint color accents in the dimming space. Focus on the space between the employee's extended hand and C01O02 to emphasize the sudden realization.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#8 action: 여직원 '아까 너 만나러 간다고 조퇴했는데... 집에 간 거 아냐? 얼른 가봐.'",
              "Shot 4 description: 마트 유니폼을 입은 여직원이 수리영을 향해 손가락으로 출입문 쪽을 가리킨 찰나."
            ],
            "visual_inferences": [
              "fixed_element supermarket_entrance_lighting applies: high-contrast boundary between harsh white fluorescent interior and deep blue dusk exterior",
              "The shared aisle space anchors the foreground employee and midground C01 in the same physical interaction zone"
            ],
            "creative_decisions": [
              "Framed from an eye-level shoulder height to naturally position the employee's pointing arm cutting across the right frame toward the background entrance, leading the viewer's eye.",
              "Highlighted the red sale stickers on the shelves to add subtle depth and realism to the referenced background."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "4cd60e6f36664873",
              "owned_hash": "fb12484777d97415",
              "camera_direction_hash": "db4cff78465a7348",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "sliding door",
                  "violating_phrase": "the sliding door from the reference",
                  "reason": "The prompt mentions the owned object as part of a reference anchor phrase, not as a direct instruction to create or alter it.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "",
        "_shot_index": 4,
        "visible_entities": [
          "C01",
          "P03",
          "L08",
          "L09",
          "L10"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 8,
            "shot_index": 4
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Inside the supermarket near the front aisle, use a MS, eye-level OTS-pan composition at shoulder height, resolving at the peak of 여직원’s pointing gesture toward the entrance in the background direction. 여직원’s extended hand cuts across the right half of the frame toward the lit doorway, while 수리영 remains left-midground in three-quarter view, absorbing the news rather than moving yet.",
            "lighting_mood": "Harsh supermarket fluorescents clash with blue dusk spill from the entrance, muting colors except for small red sale tags.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P03",
              "L08",
              "L09",
              "L10"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "bg_store_sales_floor_dusk_busy_exit_visible",
            "reference_usage": "exact_background",
            "owned_objects": [
              "CCTV camera",
              "ceiling light",
              "checkout counter",
              "checkout lane",
              "customer service counter",
              "endcap display",
              "floor tile",
              "produce bin",
              "promotional display",
              "refrigerated case",
              "security gate",
              "shelf",
              "shopping cart",
              "signage",
              "sliding door",
              "stanchion",
              "window"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "supermarket_entrance_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The glass entrance doors of the large supermarket are brightly illuminated by harsh, white fluorescent lights from the interior. This bright interior light spills outward, creating a stark, high-contrast boundary against the deep blue ambient light of the dusk exterior.",
                "applies_to_shots": [
                  4,
                  5
                ],
                "source_facts": [
                  "Shot 4 lighting mentions: 'Harsh supermarket fluorescents clash with blue dusk spill from the entrance'",
                  "Shot 5 lighting mentions: 'Blue dusk, sodium streetlamp haze, and supermarket spill light'"
                ],
                "visual_inferences": [
                  "The entrance area serves as a continuous visual anchor, meaning the lighting contrast must be identical whether viewed from the inside looking out (Shot 4) or the outside looking in (Shot 5)."
                ],
                "creative_decisions": [
                  "Emphasizing the clash between the cold/harsh fluorescent interior and the cool blue dusk to create a consistent mood across the location transition."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              },
              {
                "kind": "background",
                "id": "bg_store_sales_floor_dusk_busy_exit_visible",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "c04d1b1f78d5d4fc"
      },
      {
        "scene_index": 8,
        "heading": "S#8. 번화가·대형 마트 / 해질 무렵",
        "beat_title": "엇갈린 만남과 불안한 발걸음",
        "representative_moment": "마트 밖 가로등 아래서 불안하게 스마트폰을 귀에 댄 수리영의 실루엣.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Low angle from slightly below eye level, medium framing, high-contrast blue dusk and harsh interior spill",
            "t2i_prompt": "Photorealistic cinematic still. Focus on the tense stance of C01O02 in a dark jacket, a young Korean woman. Captured from slightly below eye level, her head, torso, and upper legs occupy the right third of the frame. She stands in left profile, one hand holding P03 pressed to her ear, looking down. She appears as a backlit figure, body volume preserved against the dim ambient fill, with thin rim light tracing her long black hair, shoulder, and soft jaw contour. The glass doors from the reference are brightly illuminated by harsh, white fluorescent lights from the interior. This bright interior light spills outward across the wet sidewalk from the reference, creating a stark, high-contrast boundary against the deep blue ambient light of the dusk exterior. The streetlamp from the reference is haloed by mist in the background, casting a sodium haze. Damp neon reflections pool on the street from the reference.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "Shot description: Suri-young tries calling her mom outside the mart but she doesn't answer.",
              "S8 action: '수리영 마트 밖을 나오면서 연락을 취해보지만 받지않는다.'"
            ],
            "visual_inferences": [
              "fixed_element supermarket_entrance_lighting: harsh white fluorescent lights spilling outward creating high-contrast boundary",
              "Suri-young appears as a dark figure silhouetted against the bright interior and streetlamp",
              "P03 represents the smartphone used to make the call"
            ],
            "creative_decisions": [
              "Framed C01 in the right third in left profile to emphasize her looking down in anxiety.",
              "Kept the low angle slightly below eye level to maintain geometric consistency with her visible torso and upper legs.",
              "Applied rim light and preserved body volume to ensure the silhouette does not look like a flat cutout."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "fcb72f34adde61dc",
              "owned_hash": "5e3a76c79931ff06",
              "camera_direction_hash": "ee4edca794787089",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "door",
                  "violating_phrase": "The glass doors from the reference",
                  "reason": "The phrase explicitly references existing doors from the reference rather than instructing a new door to be created.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "sidewalk",
                  "violating_phrase": "the wet sidewalk from the reference",
                  "reason": "The sidewalk is mentioned as part of the referenced background context, not as something to newly add or modify.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "streetlamp",
                  "violating_phrase": "The streetlamp from the reference",
                  "reason": "The streetlamp is explicitly tied to the reference background, which makes this an anchor rather than a redraw instruction.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "street",
                  "violating_phrase": "the street from the reference",
                  "reason": "The street is mentioned only as an existing reference element used for context and lighting reflection.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 8,
        "dependency_reason": "Shot 4에서 확립된 마트 입구의 조명 대비(실내 백색 형광등과 실외 푸른 해질녘) 및 배경 요소를 동일하게 유지하기 위함.",
        "_shot_index": 5,
        "visible_entities": [
          "C01",
          "P03",
          "L08",
          "L09",
          "L10"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 8,
            "shot_index": 5
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Outside across the sidewalk, hold a static WS, low angle silhouette from slightly below eye level, five to six meters from 수리영 under the streetlamp near the supermarket entrance. She occupies the right third as a dark figure with the phone pressed to her ear, while the entrance glow and wet pavement reflections stretch behind her, preserving the directional cut from the employee’s pointing hand.",
            "lighting_mood": "Blue dusk, sodium streetlamp haze, and supermarket spill light render her mostly black against damp neon reflections.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P03",
              "L08",
              "L09",
              "L10"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "bg_entrance_exterior_dusk_neon_streetlamp",
            "reference_usage": "exact_background",
            "owned_objects": [
              "CCTV camera",
              "awning",
              "bollard",
              "curb",
              "door",
              "sidewalk",
              "signboard",
              "storefront",
              "street",
              "streetlamp",
              "window"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "supermarket_entrance_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The glass entrance doors of the large supermarket are brightly illuminated by harsh, white fluorescent lights from the interior. This bright interior light spills outward, creating a stark, high-contrast boundary against the deep blue ambient light of the dusk exterior.",
                "applies_to_shots": [
                  4,
                  5
                ],
                "source_facts": [
                  "Shot 4 lighting mentions: 'Harsh supermarket fluorescents clash with blue dusk spill from the entrance'",
                  "Shot 5 lighting mentions: 'Blue dusk, sodium streetlamp haze, and supermarket spill light'"
                ],
                "visual_inferences": [
                  "The entrance area serves as a continuous visual anchor, meaning the lighting contrast must be identical whether viewed from the inside looking out (Shot 4) or the outside looking in (Shot 5)."
                ],
                "creative_decisions": [
                  "Emphasizing the clash between the cold/harsh fluorescent interior and the cool blue dusk to create a consistent mood across the location transition."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 8,
                "shot_index": 4,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              },
              {
                "kind": "background",
                "id": "bg_entrance_exterior_dusk_neon_streetlamp",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "5d6e80efbf32f247"
      },
      {
        "scene_index": 9,
        "heading": "S#9. 해안가 / N",
        "beat_title": "다급하게 자전거를 타고 달리는 수리영",
        "representative_moment": "페달에 체중을 싣고 상체를 숙인 C01의 창백한 얼굴.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Tight tracking frame at head height, tilted composition",
            "t2i_prompt": "Photorealistic cinematic still. C01O02 in dark activewear, a young Korean woman with long black hair, soft facial features, and golden wave-like irises, captured in mid-pedal on a bicycle. She is at a three-quarter angle from the right, leaning forward over the wet handlebars, her eyes fixed on the distance. A tight tracking frame at head height keeps her pale rigid face in the left-center, with her shoulders pitched forward and one knee driving up from the lower edge, the bicycle's front frame visible below. Her hair streams back in the direction opposite to motion. [L06: A blurred wet metal guardrail streaks past, set against a dark mist-heavy sea horizon.] Moonless blue-black coastal light and smeared road reflections wash her skin, while a distant blurred red harbor light punctures the mist behind her.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#9 action: '해안도로를 자전거를 타고 바삐 집으로 달려가는 수리영.'",
              "shot description: 페달에 체중을 싣고 상체를 앞으로 숙인 수리영의 창백하고 굳은 얼굴."
            ],
            "visual_inferences": [
              "mid-pedal freeze to preserve forward motion",
              "hair streaming backward opposite to motion",
              "blurred guardrail to imply high speed"
            ],
            "creative_decisions": [
              "head-and-shoulder composition keeping shoulders and knee in frame to avoid floating face syndrome",
              "blue-black coastal light washing the face with a contrasting distant red harbor light for a tense mood"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "6226435e3c0e3938",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "7b2f021221253605",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 6,
        "dependency_reason": "S6_Shot2의 해안가 도로 배경 연속성 및 분위기 유지 (근접 샷이므로 L06 환경을 프레임 주변부 질감으로만 묘사)",
        "_shot_index": 2,
        "visible_entities": [
          "C01",
          "L06"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 9,
            "shot_index": 2
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "From a tight front-quarter tracking position just ahead of the bicycle, use a CU, Dutch angle, handheld tracking frame at 수리영’s head height as she rides hard along the wet coastal road. Her pale rigid face fills the left-center of frame while her shoulders pitch forward and one knee drives up from the lower edge, making the weight on the pedals readable without losing the urgent face focus.",
            "lighting_mood": "Moonless blue-black coastal light and smeared road reflections bleach her face while a distant red harbor blink punctures the mist.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "L06"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": null,
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [],
            "previous_shot_refs": [
              {
                "scene_index": 6,
                "shot_index": 2,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "67d740849a3c9c01"
      },
      {
        "scene_index": 10,
        "heading": "S#10. 옥탑방 / N",
        "beat_title": "문틈 사이로 엿보는 낯선 남자의 실루엣",
        "representative_moment": "C11의 어두운 실루엣이 미세하게 열린 철문 틈 사이로 보인다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Dutch angle, tight through-object view at chest height, cold metallic and blue-black mood",
            "t2i_prompt": "Photorealistic cinematic still. Dutch angle, tight framing at chest height, through-object view. Focus on the narrow vertical slit occupying the central third of the frame. [L04: a slightly open metal entrance door with chipped dark paint and a rust-stained, mist-dampened doorframe forming oppressive foreground borders on the left and right]. Inside the vertical gap, C11O12 in a dark jacket, an adult Korean man, is pressed near the doorway at a three-quarter angle from the left, head directed toward the inside of the room. He exhibits the features carried from stable_traits — dark hair, face fully swallowed in shadow with no discernible features, and a blurred impression. His body volume is preserved against the cold blue-black darkness, with faint rim light tracing the shoulders and neck outline. A creased advertising flyer is wedged near the lower door seam. A dim ambient light leaves the door edges cold and metallic, while the interior of the gap remains almost pure black.",
            "outfit_assignments": [
              {
                "character_id": "C11",
                "outlook_id": "O12"
              }
            ],
            "source_facts": [
              "Scene text: 남자의 긴 그림자. (얼굴은 보이지 않는다)",
              "Scene text: 신발은 옥탑방 철문 현관 앞에 멈춰 선다. 문틈에 낀 광고 전단지.",
              "Shot description: 미세하게 열린 문틈 사이로 낯선 남자의 짙은 검은 실루엣이 드리워진 형태."
            ],
            "visual_inferences": [
              "fixed_element rooftop_exterior_environment: cold blue-black darkness with mist-dampened surfaces applied to the metal door and jamb",
              "Chest-height framing focuses strictly on the immediate door slit, omitting wider architectural elements to maximize tension"
            ],
            "creative_decisions": [
              "Dutch angle and through-object view to enhance the voyeuristic, chilling mood",
              "Silhouette body volume preserved via faint rim light on the shoulders to avoid a flat cutout effect, adhering strictly to the face-obscured trait"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "b9d90c50bf2b1f0f",
              "owned_hash": "47e0466dfeaed99f",
              "camera_direction_hash": "c60d6f3ccc0b7185",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 4,
        "dependency_reason": "이전 샷(Shot 4)의 옥탑방 옥상 외부 공간 분위기와 조명 방향성 유지",
        "_shot_index": 4,
        "visible_entities": [
          "C11",
          "P02",
          "L04"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 10,
            "shot_index": 4
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Very close to the rooftop room door at chest height, push into a CU, Dutch angle, through-object view aligned with the narrow vertical door gap. The slit occupies the central third of frame, never wider than a handspan, and the 낯선 남자’s black silhouette compresses inside it, face unreadable, with chipped door paint and the dark jamb forming oppressive foreground borders.",
            "lighting_mood": "A dim rooftop practical leaves the door edges cold and metallic while the interior remains almost pure black, swallowing the man’s features.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C11",
              "P02",
              "L04"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C11",
                "outlook_id": "O12"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "cb_rooftop_night_intrusion_door_ajar",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "rooftop_exterior_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The exterior of a small rooftop building at night. The surrounding walls and atmosphere are bathed in cold, blue-black darkness with mist-dampened surfaces. Dim, sparse ambient lighting leaves the architectural details largely in deep, oppressive shadows.",
                "applies_to_shots": [
                  4,
                  5
                ],
                "source_facts": [
                  "Scene text: N (Night), rooftop",
                  "Shot 4 lighting: A dim rooftop practical leaves the door edges cold and metallic",
                  "Shot 5 lighting: the exterior stays blue-black and mist dampened"
                ],
                "visual_inferences": [
                  "Consistent wet and damp texture on the exterior walls",
                  "Unified cold blue-black ambient night light across the entire rooftop exterior space"
                ],
                "creative_decisions": [
                  "Emphasizing the oppressive, chilling mood by keeping the exterior details mostly obscured by deep shadows to contrast with the violent action"
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 4,
                "shot_index": 1,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C11O12",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "1f2afae90b6668da"
      },
      {
        "scene_index": 10,
        "heading": "S#10. 옥탑방 / N",
        "beat_title": "민숙이 옥탑방 안에서 무방비하게 있는 상태 → 침입한 남자에게 덮쳐져 어깨를 물어뜯기며 고통스럽게 비명을 지르는 상태",
        "representative_moment": "옥탑방 창문 유리에 낯선 남자와 민숙의 형상이 뒤엉킨 거대한 그림자가 맺힌 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Low angle, tight medium shot favoring the glass surface, voyeuristic composition",
            "t2i_prompt": "Photorealistic cinematic still. Positioned outside the window from the reference at shoulder height in a low angle medium shot. The closed, dirty glass fills the middle of the frame, bordered by the mist-dampened exterior wall bathed in cold blue-black darkness, with the clothesline from the reference moving slightly in the out-of-focus edge. Inside the dimly lit room, sharing the same space, two backlit figures are illuminated by a weak yellow interior backlight. C11O12 in a dark jacket, a Korean adult man with dark hair, his face fully swallowed in shadow with no discernible features presenting a vague impression, stands at three-quarter angle from the right, the figure exerting intense force lunging over the shoulder of C04O05. C04O05 in a casual top, a Korean middle-aged woman with black hair, gentle facial lines and dark eyes, is captured in left profile, the constrained figure twisting away with eyes looking up. Their body volumes are preserved with subtle rim light tracing their shoulders against the backlight, while their enlarged, monstrous tangled silhouettes loom across the condensation-lined window. Focus on the glass pane separating the cold exterior night from the decisive physical struggle inside.",
            "outfit_assignments": [
              {
                "character_id": "C11",
                "outlook_id": "O12"
              },
              {
                "character_id": "C04",
                "outlook_id": "O05"
              }
            ],
            "source_facts": [
              "Scene text: '창문에 비친 그림자. 민숙을 덮쳐 어깨를 물어뜯는 듯한 모습'",
              "Shot description: 옥탑방 창문 유리에 낯선 남자와 민숙의 형상이 뒤엉킨 거대한 그림자가 맺힌 장면"
            ],
            "visual_inferences": [
              "fixed_element rooftop_exterior_environment: cold blue-black darkness and mist-dampened surfaces",
              "Backlit silhouettes with rim light to preserve body volume behind the window",
              "Asymmetrical posture of violence emphasizing the physical struggle"
            ],
            "creative_decisions": [
              "Low angle voyeuristic composition focusing on the glass surface rather than the room interior",
              "Contrast between the cold blue-black exterior night and the weak yellow oppressive interior glow"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "d6a5d6572cd0b7ae",
              "owned_hash": "47e0466dfeaed99f",
              "camera_direction_hash": "082cdc7a5d506891",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "window",
                  "violating_phrase": "the closed, dirty glass fills the middle of the frame",
                  "reason": "The prompt mentions the existing window as a framing surface and reference anchor, not as a newly created or modified object.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "clothesline",
                  "violating_phrase": "the clothesline from the reference",
                  "reason": "The clothesline is explicitly referenced as already present from the reference, so this is an anchor mention, not a redraw instruction.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 10,
        "dependency_reason": "이전 샷(Shot 4)과 동일한 옥탑방 옥상 배경(cb_rooftop_night_intrusion_door_ajar) 및 야간 환경 조명 톤 유지",
        "_shot_index": 5,
        "visible_entities": [
          "C11",
          "C04",
          "P02",
          "L04"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 10,
            "shot_index": 5
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Outside the rooftop window at shoulder height, track into a tight MS, low angle reflection composition favoring the glass surface rather than the room interior. The window fills the middle of the frame but not the whole image, with wet exterior wall edges visible around it, and the enlarged tangled shadows of 낯선 남자 and 민숙 loom across the glass as if pinned to the pane.",
            "lighting_mood": "Weak yellow interior backlight throws monstrous black silhouettes onto dirty glass, while the exterior stays blue-black and mist dampened.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C11",
              "C04",
              "P02",
              "L04"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C11",
                "outlook_id": "O12"
              },
              {
                "character_id": "C04",
                "outlook_id": "O05"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "cb_rooftop_night_intrusion_door_ajar",
            "reference_usage": "exact_background",
            "owned_objects": [
              "clothesline",
              "door",
              "parapet wall",
              "pipe",
              "platform",
              "post",
              "roof drain",
              "rooftop deck",
              "stairwell bulkhead",
              "wall light",
              "water tank",
              "window"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "rooftop_exterior_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The exterior of a small rooftop building at night. The surrounding walls and atmosphere are bathed in cold, blue-black darkness with mist-dampened surfaces. Dim, sparse ambient lighting leaves the architectural details largely in deep, oppressive shadows.",
                "applies_to_shots": [
                  4,
                  5
                ],
                "source_facts": [
                  "Scene text: N (Night), rooftop",
                  "Shot 4 lighting: A dim rooftop practical leaves the door edges cold and metallic",
                  "Shot 5 lighting: the exterior stays blue-black and mist dampened"
                ],
                "visual_inferences": [
                  "Consistent wet and damp texture on the exterior walls",
                  "Unified cold blue-black ambient night light across the entire rooftop exterior space"
                ],
                "creative_decisions": [
                  "Emphasizing the oppressive, chilling mood by keeping the exterior details mostly obscured by deep shadows to contrast with the violent action"
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 10,
                "shot_index": 4,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C11O12",
                "policy": "required"
              },
              {
                "kind": "character_outlook",
                "id": "C04O05",
                "policy": "required"
              },
              {
                "kind": "background",
                "id": "cb_rooftop_night_intrusion_door_ajar",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "7ea78055f4b190d9"
      },
      {
        "scene_index": 11,
        "heading": "S#11. 옥탑 방 밖 / N",
        "beat_title": "가파른 옥탑방 계단을 오르는 수리영",
        "representative_moment": "C01이 어두운 옥탑방 계단을 뛰어오르는 역동적인 측면 정지 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Side-on medium shot at waist height, high angle tracking lateral profile, hard blue-gray contrast",
            "t2i_prompt": "Photorealistic cinematic still. C01O02 in a casual jacket, an East Asian young woman with long black hair, soft facial features, and golden wave-like eyes, captured in a waist-height side-on medium shot in right profile. She is frozen mid-stride diagonally upward on the steep concrete stair from the reference. One foot lands on the next step and the other is suspended in the air behind her, her jacket and hair flowing back in the direction opposite to motion. Her eyes are fixed upward toward the faint offscreen light. The damp metal handrail from the reference and the peeling concrete wall from the reference form tight background planes. Deep, muted blue-black shadows fill the damp exterior. Cold night ambient and a weak offscreen rooftop bulb reflect off the wet surfaces, casting her body in hard blue-gray contrast. The composition focuses on her tense upward momentum against the grim shadows.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "Shot 2 action: 수리영 mid-stride, one foot landing on the next step and the other suspended behind her",
              "Shot 2 pose: right profile, eyes looking up toward the offscreen landing",
              "Lighting: Cold night ambient and a weak rooftop bulb catch the wet stair edges"
            ],
            "visual_inferences": [
              "fixed_element night_exterior_lighting: deep, muted blue-black shadows and wet concrete surfaces",
              "Hair and jacket flowing backward naturally visualizes the frozen kinetic instant of her rapid stair climb"
            ],
            "creative_decisions": [
              "Waist-height side-on framing strictly adheres to the DP directive to emphasize the diagonal stair incline and her suspended foot",
              "Hard blue-gray contrast isolates her moving figure from the peeling concrete wall and damp environment"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "630b4bf3d6fb30b4",
              "owned_hash": "dd4483bd10980271",
              "camera_direction_hash": "8a229f7b5723d3f6",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "stair",
                  "violating_phrase": "the steep concrete stair from the reference",
                  "reason": "The owned object is mentioned with a reference anchor phrase, not as the direct object of a redraw verb.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "handrail",
                  "violating_phrase": "The damp metal handrail from the reference",
                  "reason": "The owned object appears in a reference-based descriptive phrase, which anchors it to the existing background rather than requesting a new one.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "wall",
                  "violating_phrase": "the peeling concrete wall from the reference",
                  "reason": "The owned object is referenced as an existing background element from the reference, not explicitly redrawn.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "No direct dependence on a specific prior shot framing since it establishes a tracking motion",
        "_shot_index": 2,
        "visible_entities": [
          "C01",
          "L11"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 11,
            "shot_index": 2
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Side-on along the steep concrete staircase, track with 수리영 in a MWS, high angle at waist height, preserving the lateral profile from the approach stage. Freeze the kinetic instant with one foot landing on the next step and the other suspended behind her; she cuts diagonally upward across the center of frame while the stair rail and concrete wall rush past as damp, close background planes.",
            "lighting_mood": "Cold night ambient and a weak rooftop bulb catch the wet stair edges, leaving her moving body in hard blue-gray contrast.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "L11"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "cb_stairs_night_return_door_ajar",
            "reference_usage": "exact_background",
            "owned_objects": [
              "CCTV camera",
              "curb",
              "door",
              "drain grate",
              "gate",
              "handrail",
              "landing",
              "light",
              "paving",
              "planter",
              "stair",
              "storage box",
              "utility meter",
              "wall",
              "window"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "night_exterior_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dark, damp exterior environment of an old residential building at night. The scene is enveloped in deep, muted blue-black shadows, which are only slightly broken by the faint, cold glow of a weak light bulb reflecting off the wet concrete surfaces.",
                "applies_to_shots": [
                  2,
                  4
                ],
                "source_facts": [
                  "Shot 2: Cold night ambient and a weak rooftop bulb catch the wet stair edges",
                  "Shot 4: Muted blue-black rooftop darkness is broken only by a weak offscreen bulb"
                ],
                "visual_inferences": [
                  "The concrete surfaces appear slightly wet or damp, reflecting the weak light to enhance the chilling, grim atmosphere"
                ],
                "creative_decisions": [
                  "Establish visual continuity between the staircase and the rooftop through a consistent color grading of cold blue-black shadows and sparse practical lighting"
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              },
              {
                "kind": "background",
                "id": "cb_stairs_night_return_door_ajar",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "77155e8914688c9b"
      },
      {
        "scene_index": 11,
        "heading": "S#11. 옥탑 방 밖 / N",
        "beat_title": "불길하게 열린 옥탑방 철문",
        "representative_moment": "아무도 없는 어두운 옥상, 반쯤 열린 철문 너머로 새카만 어둠이 입을 벌리고 있다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Locked symmetrical wide composition, eye-level frontal frame, deep blue-black shadow color grading",
            "t2i_prompt": "Photorealistic cinematic still. Symmetrical wide shot, eye-level frontal frame with no human figure visible. Focus on the half-open metal door from the reference in the center of the frame, revealing an unlit, opaque black void beyond the threshold. The damp concrete floor from the reference leads toward the doorway. A dark, wet water tank is partly visible at the frame edge, and a slack clothesline hangs above. The scene is enveloped in deep, muted blue-black shadows, broken only by the faint, cold glow of a weak offscreen light bulb reflecting off the wet concrete surfaces.",
            "outfit_assignments": [],
            "source_facts": [
              "S#11 action: '옥상에 올라와 보니 반쯤 열린 현관 철문.'",
              "shot description: On the dark rooftop, half-open metal entrance door, no human figure visible"
            ],
            "visual_inferences": [
              "fixed_element night_exterior_lighting: deep, muted blue-black shadows, faint cold glow reflecting off wet surfaces",
              "water tank and clothesline at the frame edge to provide scale and anchor the empty rooftop environment"
            ],
            "creative_decisions": [
              "Locked symmetrical framing centered on the door to emphasize the psychological weight and threat of the opaque black interior void",
              "Utilized 'from the reference' wording to anchor the owned objects (door, floor) properly to the attached background plate"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "e2f21ca57c1c6376",
              "owned_hash": "dd4483bd10980271",
              "camera_direction_hash": "f9407fcfa5b013b4",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "door",
                  "violating_phrase": "the half-open metal door from the reference",
                  "reason": "The owned object is explicitly tied to a reference phrase, indicating it should be anchored to the existing background rather than newly redrawn.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "wall",
                  "violating_phrase": "offscreen light bulb reflecting off the wet concrete surfaces",
                  "reason": "The prompt mentions surfaces and lighting context only; it does not directly instruct creation or modification of the wall.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 11,
        "dependency_reason": "S11_Shot2의 어둡고 습한 건물 외부 환경(night_exterior_lighting)과 연속되는 조명·색감 공유",
        "_shot_index": 4,
        "visible_entities": [
          "L11"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 11,
            "shot_index": 4
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "On the dark rooftop, settle into a locked symmetrical WS, eye-level frontal frame centered on the half-open metal entrance door, with no human figure visible. The door sits in the middle third at medium distance, the black interior beyond it absorbing light, while the rooftop floor, water tank edge, and sagging clothesline keep scale and silence around the threat.",
            "lighting_mood": "Muted blue-black rooftop darkness is broken only by a weak offscreen bulb, making the open doorway read as a flat, threatening void.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "L11"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "cb_stairs_night_return_door_ajar",
            "reference_usage": "exact_background",
            "owned_objects": [
              "CCTV camera",
              "curb",
              "door",
              "drain grate",
              "gate",
              "handrail",
              "landing",
              "light",
              "paving",
              "planter",
              "stair",
              "storage box",
              "utility meter",
              "wall",
              "window"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "night_exterior_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dark, damp exterior environment of an old residential building at night. The scene is enveloped in deep, muted blue-black shadows, which are only slightly broken by the faint, cold glow of a weak light bulb reflecting off the wet concrete surfaces.",
                "applies_to_shots": [
                  2,
                  4
                ],
                "source_facts": [
                  "Shot 2: Cold night ambient and a weak rooftop bulb catch the wet stair edges",
                  "Shot 4: Muted blue-black rooftop darkness is broken only by a weak offscreen bulb"
                ],
                "visual_inferences": [
                  "The concrete surfaces appear slightly wet or damp, reflecting the weak light to enhance the chilling, grim atmosphere"
                ],
                "creative_decisions": [
                  "Establish visual continuity between the staircase and the rooftop through a consistent color grading of cold blue-black shadows and sparse practical lighting"
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 11,
                "shot_index": 2,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "background",
                "id": "cb_stairs_night_return_door_ajar",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "6c12ab5ec6024547"
      },
      {
        "scene_index": 12,
        "heading": "S#12. 옥탑방 안 / N",
        "beat_title": "핏자국이 이어진 수리영의 방",
        "representative_moment": "수리영 방 바닥에 흥건한 핏자국과 피 묻은 발자국이 커튼 쪽으로 이어진 참혹한 실내 풍경.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "low-waist-height wide shot with a slight high angle, cold gray-green lighting",
            "t2i_prompt": "Photorealistic cinematic still. A low-waist-height wide shot with a slight high angle. The open door from the reference stays as a dark rear boundary. C01O02 in an activewear outfit, a young Korean woman with long black hair, a soft facial shape, and golden wave-like irises, stands in right profile in the frame, her face identifiable as she looks down. The interior of the bedroom is immersed in a cold, low-key lighting with a gray-green tint. Deep, muted shadows dominate the space, casting a chilling, underlit darkness over the room's surroundings. Focus on the wet floor from the reference occupying the lower two-thirds of the frame. Deep red pooled blood and smeared footprints from the reference sit in the mid-foreground and form a diagonal line across the floor. The trail leads toward a long, hanging curtain from the reference positioned towards the right side of the room. It hangs straight down, remaining mostly in shadow and acting as a dark, vertical boundary within the dim space. A rumpled bed from the reference rests partly in shadow near the wall from the reference, under a light shallow depth of field.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "From just inside 수리영’s open bedroom doorway, hold a low-waist-height WS with a slight high angle",
              "the floor occupies the lower two-thirds while the opened doorway stays as a dark rear boundary",
              "pooled blood sits in the mid-foreground and the bloody footprints form a diagonal line toward the far curtain at frame center-right",
              "Low-key room light and spill from the living room flatten into a cold gray-green darkness, with the blood carrying the only deep red accent."
            ],
            "visual_inferences": [
              "fixed_element cold_dim_bedroom_lighting",
              "fixed_element vertical_hanging_curtain",
              "Character C01O02 is positioned in profile to remain identifiable while preserving the observer angle focus on the floor."
            ],
            "creative_decisions": [
              "Explicitly anchored owned objects (door, floor, footprints, curtain, bed, wall) with 'from the reference' to prevent dual rendering.",
              "Applied a light shallow depth of field to keep the curtain as the visual endpoint of the bloody trail."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "404d3503cbcbd04d",
              "owned_hash": "601128658431b6e5",
              "camera_direction_hash": "f4685163b3a30ad4",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "door",
                  "violating_phrase": "The open door from the reference",
                  "reason": "The phrase explicitly references the existing door from the reference, so it is an anchor rather than a request to redraw it.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "bed",
                  "violating_phrase": "A rumpled bed from the reference",
                  "reason": "The bed is described as already present from the reference, which anchors it instead of asking to create or add it.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "curtain",
                  "violating_phrase": "a long, hanging curtain from the reference",
                  "reason": "The curtain is explicitly marked as coming from the reference, so this is a reference anchor, not a redraw instruction.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "floor",
                  "violating_phrase": "Focus on the wet floor from the reference",
                  "reason": "The floor is referenced as an existing background element from the reference, not directly commanded to be newly generated.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "wall",
                  "violating_phrase": "near the wall from the reference",
                  "reason": "The wall is mentioned as part of the existing reference scene, so it functions as an anchor reference.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 5,
        "dependency_reason": "S5_Shot5 establishes the bedroom background and curtain layout which is matched via the attached background reference.",
        "_shot_index": 6,
        "visible_entities": [
          "C01",
          "P10",
          "P03",
          "P04",
          "P01",
          "L05"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 12,
            "shot_index": 6
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "From just inside 수리영’s open bedroom doorway, hold a low-waist-height WS with a slight high angle, the camera lowered below the stage default so the floor occupies the lower two-thirds while the opened doorway stays as a dark rear boundary. In direct observer vision, the pooled blood sits in the mid-foreground and the bloody footprints form a diagonal line toward the far curtain at frame center-right, using shallow DOF only lightly so the curtain remains the endpoint of the trail.",
            "lighting_mood": "Low-key room light and spill from the living room flatten into a cold gray-green darkness, with the blood carrying the only deep red accent.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P10",
              "P03",
              "P04",
              "P01",
              "L05"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "cb_daughter_room_night_blood_footprints",
            "reference_usage": "exact_background",
            "owned_objects": [
              "backpack",
              "bed",
              "curtain",
              "door",
              "floor",
              "footprint",
              "outline",
              "photograph",
              "pillow",
              "toy",
              "wall",
              "wall mark",
              "window"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "cold_dim_bedroom_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior of the bedroom is immersed in a cold, low-key lighting with a gray-green tint. Deep, muted shadows dominate the space, casting a chilling, underlit darkness over the room's surroundings.",
                "applies_to_shots": [
                  6,
                  9,
                  16
                ],
                "source_facts": [
                  "Low-key room light and spill from the living room flatten into a cold gray-green darkness (Shot 6)",
                  "room collapses into a dark blur, gray-pale against muted shadows (Shot 9)",
                  "The room remains cold and underlit (Shot 16)"
                ],
                "visual_inferences": [
                  "A persistent eerie, desaturated color palette unifies the grim discovery sequence regardless of camera angle."
                ],
                "creative_decisions": [
                  "Maintained consistent background illumination across different focal lengths to ensure the environment consistently feels oppressive and dark."
                ],
                "confidence": "high"
              },
              {
                "element_id": "vertical_hanging_curtain",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "A long, hanging curtain is positioned towards the right side of the room. It hangs straight down, remaining mostly in shadow and acting as a dark, vertical boundary within the dim space.",
                "applies_to_shots": [
                  6,
                  9
                ],
                "source_facts": [
                  "far curtain at frame center-right (Shot 6)",
                  "the curtain only a soft vertical smear at the right edge (Shot 9)"
                ],
                "visual_inferences": [
                  "The curtain is drawn heavily, concealing the space behind it and forming a distinct vertical line in the background geometry."
                ],
                "creative_decisions": [
                  "Kept the curtain aligned on the right edge of the frame to anchor spatial continuity between the wide floor shot and the tight facial close-up."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 5,
                "shot_index": 5,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              },
              {
                "kind": "background",
                "id": "cb_daughter_room_night_blood_footprints",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "bc6b48a90525cda5"
      },
      {
        "scene_index": 12,
        "heading": "S#12. 옥탑방 안 / N",
        "beat_title": "Frozen in Shock",
        "representative_moment": "C01 is frozen in shock, her face captured in a tight close-up as the background blurs into darkness.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "handheld close-up at eye-level, shallow depth of field, weak cold practical light from below-left",
            "t2i_prompt": "Photorealistic cinematic still. C01O02 in a dark casual jacket, a Korean young woman with long black hair and golden wave-like eyes, facing the camera in a frozen backward flinch, her widened eyes fixed downward, mouth half-open in shock. Tight framing on her face with her neck and upper shoulders visible in the lower frame, her face occupying the center. A weak, cold practical light grazes her face from the lower-left, catching glossy highlights in her pupils and rendering her skin gray-pale against deep shadows. [L05: The immediate frame-edge background collapses into an out-of-focus dark blur under cold gray-green ambient light, with a hanging curtain forming a soft, dark vertical smear along the right edge behind her shoulder line.] Focus on her widened eyes and frozen expression, shallow depth of field isolating her from the muted surroundings.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "Shot 9 description: 충격에 두 눈을 크게 뜨고 입을 반쯤 벌린 채 얼어붙은 수리영의 얼굴 클로즈업.",
              "Staging: 수리영 facing_camera, frozen backward flinch, eyes directed downward toward the space below the curtain.",
              "DP lighting note: weak, cold practical light grazes her face from below-left, leaving her pupils glossy and her skin gray-pale."
            ],
            "visual_inferences": [
              "fixed_element cold_dim_bedroom_lighting applied as a gray-green ambient tint to the blurred frame edges.",
              "fixed_element vertical_hanging_curtain translated into a soft dark vertical smear on the right edge due to shallow DOF."
            ],
            "creative_decisions": [
              "Placed neck and shoulders in the lower frame to ground the close-up and avoid a floating head effect per close-framing face phrasing rules.",
              "Used L05 block to define the immediate out-of-focus surroundings without invoking the omitted background reference, ensuring compliance with skipped_close_framing."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "c24d8d95d40c8d70",
              "owned_hash": "b0bfa88101c56b24",
              "camera_direction_hash": "0fd680f47280135b",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 12,
        "dependency_reason": "Maintains the cold, dim lighting and the relative position of the curtain from the prior wide shot, now rendered as a soft blur due to the tight close-up.",
        "_shot_index": 9,
        "visible_entities": [
          "C01",
          "P10",
          "P03",
          "P04",
          "P01",
          "L05"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 12,
            "shot_index": 9
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Near the curtain and bed area, push into a handheld CU at eye-level, tighter and more frontal than the flow default, isolating 수리영’s widened eyes and half-open mouth while the room collapses into a dark blur around her. The frame holds her face near center with the curtain only a soft vertical smear at the right edge, using shallow DOF to make the horror feel like it is pressing directly into her breath.",
            "lighting_mood": "A weak, cold practical light grazes her face from below-left, leaving her pupils glossy and her skin gray-pale against muted shadows.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P10",
              "P03",
              "P04",
              "P01",
              "L05"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "cb_daughter_room_night_body_red_mark",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "cold_dim_bedroom_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior of the bedroom is immersed in a cold, low-key lighting with a gray-green tint. Deep, muted shadows dominate the space, casting a chilling, underlit darkness over the room's surroundings.",
                "applies_to_shots": [
                  6,
                  9,
                  16
                ],
                "source_facts": [
                  "Low-key room light and spill from the living room flatten into a cold gray-green darkness (Shot 6)",
                  "room collapses into a dark blur, gray-pale against muted shadows (Shot 9)",
                  "The room remains cold and underlit (Shot 16)"
                ],
                "visual_inferences": [
                  "A persistent eerie, desaturated color palette unifies the grim discovery sequence regardless of camera angle."
                ],
                "creative_decisions": [
                  "Maintained consistent background illumination across different focal lengths to ensure the environment consistently feels oppressive and dark."
                ],
                "confidence": "high"
              },
              {
                "element_id": "vertical_hanging_curtain",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "A long, hanging curtain is positioned towards the right side of the room. It hangs straight down, remaining mostly in shadow and acting as a dark, vertical boundary within the dim space.",
                "applies_to_shots": [
                  6,
                  9
                ],
                "source_facts": [
                  "far curtain at frame center-right (Shot 6)",
                  "the curtain only a soft vertical smear at the right edge (Shot 9)"
                ],
                "visual_inferences": [
                  "The curtain is drawn heavily, concealing the space behind it and forming a distinct vertical line in the background geometry."
                ],
                "creative_decisions": [
                  "Kept the curtain aligned on the right edge of the frame to anchor spatial continuity between the wide floor shot and the tight facial close-up."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 12,
                "shot_index": 6,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "ce7525804d46cf86"
      },
      {
        "scene_index": 12,
        "heading": "S#12. 옥탑방 안 / N",
        "beat_title": "기이한 현상과 패닉",
        "representative_moment": "침대 위 벽에 거대한 붉은색 원형 표식이 피처럼 맺혀 있다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Low angle from eye height focusing on the wall, cold gray-green tint with deep red accent",
            "t2i_prompt": "Photorealistic cinematic still. Focus on P01, a fresh, wet, deep red circular mark painted on [L05: a pale, stained wall immersed in a cold gray-green tint]. C01O02 in a dark jacket, a young Korean woman with long black hair, is visible softly out of focus in left profile at the left frame edge, looking up. The camera is positioned at eye height from the side of the bed, tilted up at a low angle toward the wall. P01 is centered and occupies roughly one-third of the upper frame, catching a sickly red sheen as if illuminated from within the wet pigment. The dark, rumpled top edge of the bed anchors the lower frame horizontally below the wall. A dark curtain edge is barely visible in shadow at the far left. Deep, muted shadows dominate the space, casting a chilling, underlit darkness over the flat wall surface.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S12 action: '침대 위 벽에 붉은 원이 생겨난다'",
              "DP directions: 'From the side of the bed at eye height... low angle toward the wall above the bed'",
              "DP directions: 'bed’s dark top edge anchors the lower frame'",
              "fixed_element: 'cold, low-key lighting with a gray-green tint'"
            ],
            "visual_inferences": [
              "fixed_element interior lighting creates deep, muted shadows on the wall",
              "C01's face is identifiable in profile while remaining softly out of focus to keep P01 as the primary subject"
            ],
            "creative_decisions": [
              "Contrasted the fresh, wet red pigment of P01 against the desaturated, cold gray-green wall to maximize visual impact",
              "Anchored the composition with the dark silhouette of the bed at the bottom to emphasize the low angle"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "0b01ca4c8ef022d5",
              "owned_hash": "b0bfa88101c56b24",
              "camera_direction_hash": "cabb3eb5b9681d50",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 12,
        "dependency_reason": "Same bedroom location, maintaining the cold, dim gray-green lighting established in prior shots of the sequence.",
        "_shot_index": 16,
        "visible_entities": [
          "C01",
          "P10",
          "P03",
          "P04",
          "P01",
          "L05"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 12,
            "shot_index": 16
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "From the side of the bed at eye height, settle out of the upward tilt into a steady CU with a low angle toward the wall above the bed, keeping the red circular mark centered and occupying roughly one-third of the frame. In omniscient direct vision, the bed’s dark top edge anchors the lower frame while the paint-like circle appears freshly wet on the upper wall, clear but not oversized, as if the room itself has become evidence.",
            "lighting_mood": "The room remains cold and underlit, but the circular mark catches a sickly red sheen as if lit from within the wet pigment.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P10",
              "P03",
              "P04",
              "P01",
              "L05"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "cb_daughter_room_night_body_red_mark",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "cold_dim_bedroom_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior of the bedroom is immersed in a cold, low-key lighting with a gray-green tint. Deep, muted shadows dominate the space, casting a chilling, underlit darkness over the room's surroundings.",
                "applies_to_shots": [
                  6,
                  9,
                  16
                ],
                "source_facts": [
                  "Low-key room light and spill from the living room flatten into a cold gray-green darkness (Shot 6)",
                  "room collapses into a dark blur, gray-pale against muted shadows (Shot 9)",
                  "The room remains cold and underlit (Shot 16)"
                ],
                "visual_inferences": [
                  "A persistent eerie, desaturated color palette unifies the grim discovery sequence regardless of camera angle."
                ],
                "creative_decisions": [
                  "Maintained consistent background illumination across different focal lengths to ensure the environment consistently feels oppressive and dark."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 12,
                "shot_index": 9,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "096c297e59c09930"
      },
      {
        "scene_index": 13,
        "heading": "S#13. 옥탑방 외곽 / N",
        "beat_title": "어둠 속에서 뛰쳐나오며 비명 지르는 수리영",
        "representative_moment": "어두운 옥탑방 문턱을 벗어나며 비명을 지르는 C01의 넋 나간 상반신.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "tight low-angle composition looking up, cold wet night lighting",
            "t2i_prompt": "Photorealistic cinematic still. C01O02 in dark activewear, a Korean young woman, leaning forward through the threshold at three-quarter angle from the right, the figure's upper torso and head occupying the center frame viewed from below looking up. Her mouth is open in a scream, showing variant features carried from stable_traits — golden wave-like irises fixed on the void. Cold exterior night light skims her face and shoulders, casting wet concrete gray tones. [L11: A chipped doorframe presses along the left and right frame edges like a trap, set in a damp stained concrete exterior wall, while the unlit opaque darkness of the interior space recedes deeply behind her.] Focus on her distorted expression and the golden irises.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#13 action: 비명을 지르며 옥탑방을 빠져나오는 수리영",
              "shot description: 어두운 옥탑방 문 밖으로 상체를 들이민 채 입을 크게 벌린 수리영의 일그러진 표정"
            ],
            "visual_inferences": [
              "low angle composition placing her head against the upper doorframe and interior darkness",
              "cold exterior lighting highlighting her panicked face against the black background"
            ],
            "creative_decisions": [
              "framing the door edges tightly around her to create a claustrophobic, trap-like tension",
              "using wet concrete grays to heighten the raw thriller mood",
              "incorporating the golden wave-like irises trait directly into her panicked stare"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "4041c4ea395b8d38",
              "owned_hash": "a3c134dc147c7de6",
              "camera_direction_hash": "315504d8629740a5",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 11,
        "dependency_reason": "동일한 옥탑방 건물 외부 (close framing으로 인해 배경 reference 병합은 제외되고 프레임 가장자리만 질감으로 묘사)",
        "_shot_index": 1,
        "visible_entities": [
          "C01",
          "L11"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 13,
            "shot_index": 1
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Just outside the rooftop room threshold, use a handheld MCU with a low angle, slightly tilted up at 수리영 as she leans out of the black doorway within arm’s length of the lens. The tight frame cuts her at the upper torso, the doorframe pressing along both edges like a trap, while her open mouth sits high center and the dark room behind her remains unreadable.",
            "lighting_mood": "Cold exterior night light skims her face while the doorway behind her stays nearly black, heightening the scream with wet concrete grays.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "L11"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "cb_stairs_night_police_arrival_sirens",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [],
            "previous_shot_refs": [
              {
                "scene_index": 11,
                "shot_index": 2,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "d3d87304323640e5"
      },
      {
        "scene_index": 13,
        "heading": "S#13. 옥탑방 외곽 / N",
        "beat_title": "사건 현장 도착과 충격적 전언",
        "representative_moment": "사이렌 불빛이 번쩍이는 밤거리, 순찰차 문을 열고 내리던 C02가 굳은 얼굴로 어딘가를 응시하는 정지 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Low-to-medium Dutch angle, harsh red-blue police flicker through mist",
            "t2i_prompt": "Photorealistic cinematic still. Low-to-medium height observer camera in a Dutch angle. C02O03 in a dark detective coat, a middle-aged Korean woman with black hair, mature facial contours, and sharp eyes, positioned at three-quarter angle from the left. Her mid-step motion is frozen as one foot lands on the wet street from the reference in the lower frame, with her upper body extending toward the upper-right third. The open police car door from the reference occupies frame left, forming a dark vertical wing behind her shoulder. Her eyes are fixed on an off-screen figure. Focus on her hardened expression against the urban backdrop. Cool blue and red ambient light from the reference police car washes the side of her face and shoulder, pooling intensely on the damp asphalt surface below, glowing through the heavy blue-black mist.",
            "outfit_assignments": [
              {
                "character_id": "C02",
                "outlook_id": "O03"
              }
            ],
            "source_facts": [
              "S#13 action: '차에서 내린 혜수... 수리영의 모습을 발견하고 달려간다.'",
              "Shot 3 camera framing: 'catching 혜수 three-quarter front as one foot lands on the wet asphalt and the open car door remains on frame left.'"
            ],
            "visual_inferences": [
              "fixed_element wet_street_with_police_lights: A dark nighttime street environment enveloped in a blue-black mist with pulsing red and blue emergency light reflections.",
              "The Dutch angle accentuates the jarring, sudden nature of the situation she steps into."
            ],
            "creative_decisions": [
              "Positioned the reference police car door on the left frame edge to create a dark vertical wing, directing visual flow toward her face in the upper-right.",
              "Reflected the red and blue ambient light exclusively on the surfaces of her face and the wet asphalt to avoid physical light-cut violations."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "ce665a127aa81bc9",
              "owned_hash": "a3c134dc147c7de6",
              "camera_direction_hash": "a7f538e541ca34d9",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "door",
                  "violating_phrase": "The open police car door",
                  "reason": "The prompt references an existing door as part of the scene layout with a reference marker, not as a newly created or altered object.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "street",
                  "violating_phrase": "one foot lands on the wet street",
                  "reason": "The street is mentioned as the surface she steps onto, which is a spatial anchor rather than an instruction to redraw it.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "car",
                  "violating_phrase": "police car",
                  "reason": "This is part of the existing police car context and is not a separate redraw instruction for an owned object in the provided list.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 11,
        "dependency_reason": "Same physical location (street outside the villa) maintaining continuity of background layout and environment.",
        "_shot_index": 3,
        "visible_entities": [
          "C02",
          "L11"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 13,
            "shot_index": 3
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "At street level beside the patrol car, hold a static MWS in a Dutch angle from low-to-medium height, catching 혜수 three-quarter front as one foot lands on the wet asphalt and the open car door remains on frame left. The door forms a dark vertical wing behind her shoulder, while her hardened face is held in the upper-right third as she notices 수리영 off-screen.",
            "lighting_mood": "Blue-red police flicker cuts through sodium streetlight haze, giving 혜수’s face a hard, damp, noir edge.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C02",
              "L11"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C02",
                "outlook_id": "O03"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "cb_stairs_night_police_arrival_sirens",
            "reference_usage": "exact_background",
            "owned_objects": [
              "CCTV camera",
              "barred window",
              "building",
              "courtyard",
              "door",
              "drain grate",
              "landing",
              "lightbar",
              "pavement",
              "pedestrian gate",
              "planter",
              "police car",
              "sliding gate",
              "stair",
              "storage box",
              "street",
              "utility meter",
              "wall",
              "wall light",
              "window"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "wet_street_with_police_lights",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dark nighttime street environment enveloped in a blue-black mist. The ground consists of wet asphalt that slickly reflects surrounding light sources. The area is heavily illuminated by the harsh, flashing red and blue emergency lights of a nearby police vehicle, creating pulsing reflections across the damp surfaces.",
                "applies_to_shots": [
                  3,
                  6
                ],
                "source_facts": [
                  "Shot 3 lighting: 'Blue-red police flicker cuts through sodium streetlight haze'",
                  "Shot 3 framing: 'one foot lands on the wet asphalt'",
                  "Shot 6 lighting: 'Wet police-light reflections tremble... blue-black mist and muted red pulses'"
                ],
                "visual_inferences": [
                  "The street level remains persistently damp, with the patrol car's flashing emergency lights dominating the ambient lighting of the scene."
                ],
                "creative_decisions": [
                  "Used the combination of wet asphalt and pulsing red-blue police lights as a continuous atmospheric anchor to unify the street-level shots."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 11,
                "shot_index": 4,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C02O03",
                "policy": "required"
              },
              {
                "kind": "background",
                "id": "cb_stairs_night_police_arrival_sirens",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "194ef8e51e1b90da"
      },
      {
        "scene_index": 13,
        "heading": "S#13. 옥탑방 외곽 / N",
        "beat_title": "충격에 빠진 수리영과 사태를 파악하는 혜수",
        "representative_moment": "혜수에게 어깨를 붙잡힌 채, 충격으로 멍하게 혜수를 올려다보는 수리영의 눈물 고인 얼굴 클로즈업.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Tight face framing with shallow depth of field over-the-shoulder, high contrast night lighting",
            "t2i_prompt": "Photorealistic cinematic still. Tight framing on her face, upper body in frame, using a shallow depth of field over-the-shoulder composition. C01O02 in a dark jacket, a young Korean woman with long black hair, soft facial features, and golden wave-like irises, facing the camera, staring blankly upward with tear-filled eyes fixed on the figure in front of her, her face occupying the center-right. In the left foreground, C02O03 in a detective coat, a middle-aged Korean woman with mature facial lines, stands with back to camera, reduced to a blurred shoulder and cheek edge. The two figures occupy the same damp street area, C02O03's hands firmly gripping C01O02's shoulders. Wet red and blue police-light reflections tremble on C01O02's wet cheeks and golden irises, while cool ambient mist washes the side of her face. The camera focuses on C01O02's blank expression. Behind them, [L11: a dark street environment enveloped in thick blue-black mist, damp air faintly backlit by muted red and blue emergency pulses, completely obscuring any background structures].",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              },
              {
                "character_id": "C02",
                "outlook_id": "O03"
              }
            ],
            "source_facts": [
              "S13 Shot 6: 자신을 붙잡은 혜수의 얼굴을 멍하게 올려다보는 수리영의 눈물 맺힌 눈동자 클로즈업.",
              "DP instructions: over_shoulder, hands gripping shoulders, Wet police-light reflections tremble in 수리영’s eyes, while the rest of the frame sinks into blue-black mist and muted red pulses."
            ],
            "visual_inferences": [
              "fixed_element 'wet_street_with_police_lights': damp surfaces reflecting pulsing red and blue emergency lights",
              "Hyesu holding Suyoung by the shoulders in a dark mist-filled street space"
            ],
            "creative_decisions": [
              "Used 'tight framing on her face, upper body in frame' instead of full-frame face to prevent floating head artifact, while keeping the eyes prominent in the center-right",
              "Shallow depth of field to blur Hyesu in the foreground and isolate Suyoung's shocked expression",
              "Applied muted red and blue pulsing light reflections directly onto the cheek and iris surfaces to match the atmospheric mist"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "cb1616073fa6f55a",
              "owned_hash": "a3c134dc147c7de6",
              "camera_direction_hash": "d9ce932d0302e2ad",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 13,
        "dependency_reason": "동일한 씬(S13) 내 이전 샷(Shot 3)의 안개 낀 거리 및 경찰차 경광등 조명 분위기(wet_street_with_police_lights)를 연속적으로 유지",
        "_shot_index": 6,
        "visible_entities": [
          "C01",
          "C02",
          "L11"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 13,
            "shot_index": 6
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "At 수리영’s eye level, push into an extreme close-up (ECU) over 혜수’s shoulder, using an eye-level OTS with shallow DOF until 수리영’s tear-filled eyes occupy the central band of the frame. 혜수 is reduced to a blurred shoulder and cheek edge along the left foreground, making the shot feel like we are trapped inside the space between her grip and 수리영’s blank upward stare.",
            "lighting_mood": "Wet police-light reflections tremble in 수리영’s eyes, while the rest of the frame sinks into blue-black mist and muted red pulses.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "C02",
              "L11"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              },
              {
                "character_id": "C02",
                "outlook_id": "O03"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "cb_stairs_night_police_arrival_sirens",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "wet_street_with_police_lights",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dark nighttime street environment enveloped in a blue-black mist. The ground consists of wet asphalt that slickly reflects surrounding light sources. The area is heavily illuminated by the harsh, flashing red and blue emergency lights of a nearby police vehicle, creating pulsing reflections across the damp surfaces.",
                "applies_to_shots": [
                  3,
                  6
                ],
                "source_facts": [
                  "Shot 3 lighting: 'Blue-red police flicker cuts through sodium streetlight haze'",
                  "Shot 3 framing: 'one foot lands on the wet asphalt'",
                  "Shot 6 lighting: 'Wet police-light reflections tremble... blue-black mist and muted red pulses'"
                ],
                "visual_inferences": [
                  "The street level remains persistently damp, with the patrol car's flashing emergency lights dominating the ambient lighting of the scene."
                ],
                "creative_decisions": [
                  "Used the combination of wet asphalt and pulsing red-blue police lights as a continuous atmospheric anchor to unify the street-level shots."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 13,
                "shot_index": 3,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              },
              {
                "kind": "character_outlook",
                "id": "C02O03",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "ff5cbd3baa8df6a5"
      },
      {
        "scene_index": 14,
        "heading": "S#14. 옥탑방 안 / N",
        "beat_title": "핏자국과 시신이 감쪽같이 사라진 수리영의 방",
        "representative_moment": "C01이 핏자국 하나 없이 지나치게 깨끗한 방 안을 혼란스러운 표정으로 바라보고, 바닥에는 P05 백팩만 놓여 있다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "symmetrical composition from the doorway, sterile gray-blue night light",
            "t2i_prompt": "Photorealistic cinematic still. Symmetrical composition from just outside the open bedroom doorway. C01O02 in a dark casual jacket, a young Korean woman, stands in the lower foreground with her back to the camera, looking straight ahead into the unnaturally clean room interior. The room is bathed in a sterile, cold gray-blue night light. The bare wooden floor from the reference stretches out smoothly across the lower frame, completely devoid of bloodstains. To one side, the neatly made bed from the reference sits quietly against the clean wall. On the far wall, the window curtain from the reference hangs perfectly still. Resting motionless on the floor is P05, the dark-colored backpack from the reference with a small plush doll dangling from its zipper. Focus on the spotless empty space of the floor in the center of the room, highlighting the chilling absence.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "지나칠 정도로 깨끗하게 정돈된 방안. 핏자국도 보이지 않고",
              "침대 위에 붉은 표식은 물론 커튼 뒤의 시신도 감쪽 같이 사라졌다.",
              "수리영 방안에 놓여진 백팩, 인형. 약간씩 위치가 달라진."
            ],
            "visual_inferences": [
              "fixed_element spotless_clean_bedroom: bare wooden floor, neatly made bed, still curtain",
              "fixed_element backpack_with_plush_doll: resting motionless on the floor",
              "C01 stands at the threshold observing the room to frame the unnaturally clean interior and fulfill the entity presence requirement"
            ],
            "creative_decisions": [
              "Using a symmetrical composition from the doorway to emphasize the artificial neatness and center the empty space.",
              "Applying a cold, sterile gray-blue night light to contrast sharply with the previous bloody scene, maximizing the eerie atmosphere."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "7f1b698ad7820a36",
              "owned_hash": "5ba936e2b10f2eb0",
              "camera_direction_hash": "675e9d76016c947d",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "door",
                  "violating_phrase": "open bedroom doorway",
                  "reason": "The phrase mentions the existing doorway as a framing reference and does not direct a redraw of a door object.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "floor",
                  "violating_phrase": "wooden floor from the reference",
                  "reason": "The floor is explicitly referenced as already present from the reference, so this is an anchor rather than a redraw instruction.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "bed",
                  "violating_phrase": "the neatly made bed from the reference",
                  "reason": "The bed is described as existing from the reference, which makes this an anchor reference.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "curtain",
                  "violating_phrase": "window curtain from the reference hangs",
                  "reason": "The curtain is referenced as already present from the reference and only described in place, not newly created.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "toy",
                  "violating_phrase": "plush doll",
                  "reason": "The prompt mentions a plush doll as an existing dangling object, but it does not explicitly instruct to create or add a toy object.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 12,
        "dependency_reason": "수리영의 방 배경(S12_Shot16) 구도를 동일하게 유지하되, 핏자국과 참혹한 흔적이 완전히 지워진 대비를 연출",
        "_shot_index": 4,
        "visible_entities": [
          "C01",
          "P05",
          "L05"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 14,
            "shot_index": 4
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "From just outside the bedroom threshold, hold a symmetrical EWS at eye-level, wider than the flow default, centered on the unnaturally clean room interior. In direct observer vision, the spotless floor stretches empty through the lower half, the made bed sits to one side, and the curtain hangs perfectly still where the body should have been, emphasizing absence rather than action.",
            "lighting_mood": "Sterile gray-blue night light replaces the earlier red horror, making the room feel too clean and airless.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P05",
              "L05"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "cb_daughter_room_night_clean_reset",
            "reference_usage": "exact_background",
            "owned_objects": [
              "backpack",
              "bed",
              "curtain",
              "door",
              "floor",
              "photograph",
              "pillow",
              "sliding window",
              "toy",
              "wall"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "spotless_clean_bedroom",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior of the bedroom is disturbingly spotless and meticulously clean, completely devoid of any bloodstains or disorder. The bare wooden floor stretches out smoothly, while a neatly made bed sits quietly alongside a perfectly still window curtain. The entire room is bathed in a cold, sterile gray-blue night light, emphasizing the unnatural emptiness.",
                "applies_to_shots": [
                  4,
                  5
                ],
                "source_facts": [
                  "지나칠 정도로 깨끗하게 정돈된 방안. 핏자국도 보이지 않고",
                  "침대 위에 붉은 표식은 물론 커튼 뒤의 시신도 감쪽 같이 사라졌다.",
                  "Shot 4: spotless floor stretches empty... made bed sits to one side, and the curtain hangs perfectly still",
                  "Shot 4 lighting: Sterile gray-blue night light",
                  "Shot 5 lighting: A cold overhead spill"
                ],
                "visual_inferences": [
                  "The stark cleanliness creates an eerie, almost artificial neatness.",
                  "A bare wooden floor maximizes the impression of an unblemished surface."
                ],
                "creative_decisions": [
                  "Emphasizing the absence of the previously seen horror by rendering the room as aggressively mundane, illuminated by a chilling blue-toned wash."
                ],
                "confidence": "high"
              },
              {
                "element_id": "backpack_with_plush_doll",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "A casual, dark-colored backpack with a small plush doll dangling from its zipper. It rests quietly on the floor, completely stationary, serving as a subtle reminder of ordinary life amidst the unsettling environment.",
                "applies_to_shots": [
                  4,
                  5,
                  9
                ],
                "source_facts": [
                  "수리영 방안에 놓여진 백팩, 인형.",
                  "Listed as an appearing element across Shots 4, 5, and 9."
                ],
                "visual_inferences": [
                  "The backpack sits motionless on the floor.",
                  "The attached plush doll hangs limply, offering a slight contrast to the otherwise sterile room."
                ],
                "creative_decisions": [
                  "Using the displaced backpack as a constant visual anchor across the different angles and rooms to unify the location's underlying tension."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 12,
                "shot_index": 16,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              },
              {
                "kind": "background",
                "id": "cb_daughter_room_night_clean_reset",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "0e25fc767619ed63"
      },
      {
        "scene_index": 14,
        "heading": "S#14. 옥탑방 안 / N",
        "beat_title": "감쪽같이 사라진 참혹한 흔적",
        "representative_moment": "핏자국이 하나도 남지 않은 깨끗한 바닥을 내려다보며 수리영이 충격에 빠져 있다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "low angle head-and-shoulder composition, cold sterile gray-blue overhead lighting",
            "t2i_prompt": "Photorealistic cinematic still. Focus on the stark contrast between the immaculate bare wooden floor in the lower foreground and the subject's bewildered expression in the upper frame. C01O02 in a casual dark track jacket, a Korean young woman with long black hair, soft facial structure, and golden wave-like eyes, is captured in a head-and-shoulder composition from a low angle, occupying the upper-right portion of the frame. She is leaning forward at a three-quarter angle from the left, her eyes fixed down on the completely empty floor below. A cold overhead spill washes her pale skin almost colorless, illuminated by a sterile gray-blue night ambient light. [L05: The immediate frame-edge surfaces show only the smooth, spotless wooden floor below and a soft, unmarked wall with a hint of a still curtain edge in the dim background].",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "지나칠 정도로 깨끗하게 정돈된 방안. 핏자국도 보이지 않고",
              "얼굴이 파리해지는 수리영. ‘어----?’ ‘분명히 봤는데...’",
              "Shot 5 lighting: A cold overhead spill"
            ],
            "visual_inferences": [
              "fixed_element spotless_clean_bedroom: The stark cleanliness creates an eerie, almost artificial neatness.",
              "A low angle composition naturally places the empty floor in the foreground while pushing her face to the upper frame, highlighting the contradiction."
            ],
            "creative_decisions": [
              "Utilized a head-and-shoulder composition to comply with close framing face constraints while keeping the upper-right placement to balance the lower floor plane.",
              "Described the background as 'soft' to ensure the primary subject and the immediate floor surface remain the sharpest focal points, respecting the skipped close framing rule."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "ec7f8c44c6909060",
              "owned_hash": "5ba936e2b10f2eb0",
              "camera_direction_hash": "288567c68a0a265f",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 14,
        "dependency_reason": "같은 씬 내 이전 샷에서 설정된 차가운 푸른빛 조명과 깨끗한 바닥 질감을 유지하되, 카메라 앵글과 프레이밍만 로우 앵글 클로즈업으로 전환",
        "_shot_index": 5,
        "visible_entities": [
          "C01",
          "P05",
          "L05"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 14,
            "shot_index": 5
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Inside the bedroom beside the clean floor, begin with a spotless floor detail and tilt up into a tight MCU at a low angle, ending on 수리영’s pale face as she looks down in disbelief. The floor remains in the lower foreground as a clean, empty plane, while her face enters the upper-right portion of frame, the tilt tightening the emotional contradiction between what she saw and what exists now.",
            "lighting_mood": "A cold overhead spill washes her face almost colorless, with no red accent left except the memory implied by the empty floor.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P05",
              "L05"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "cb_daughter_room_night_clean_reset",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "spotless_clean_bedroom",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior of the bedroom is disturbingly spotless and meticulously clean, completely devoid of any bloodstains or disorder. The bare wooden floor stretches out smoothly, while a neatly made bed sits quietly alongside a perfectly still window curtain. The entire room is bathed in a cold, sterile gray-blue night light, emphasizing the unnatural emptiness.",
                "applies_to_shots": [
                  4,
                  5
                ],
                "source_facts": [
                  "지나칠 정도로 깨끗하게 정돈된 방안. 핏자국도 보이지 않고",
                  "침대 위에 붉은 표식은 물론 커튼 뒤의 시신도 감쪽 같이 사라졌다.",
                  "Shot 4: spotless floor stretches empty... made bed sits to one side, and the curtain hangs perfectly still",
                  "Shot 4 lighting: Sterile gray-blue night light",
                  "Shot 5 lighting: A cold overhead spill"
                ],
                "visual_inferences": [
                  "The stark cleanliness creates an eerie, almost artificial neatness.",
                  "A bare wooden floor maximizes the impression of an unblemished surface."
                ],
                "creative_decisions": [
                  "Emphasizing the absence of the previously seen horror by rendering the room as aggressively mundane, illuminated by a chilling blue-toned wash."
                ],
                "confidence": "high"
              },
              {
                "element_id": "backpack_with_plush_doll",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "A casual, dark-colored backpack with a small plush doll dangling from its zipper. It rests quietly on the floor, completely stationary, serving as a subtle reminder of ordinary life amidst the unsettling environment.",
                "applies_to_shots": [
                  4,
                  5,
                  9
                ],
                "source_facts": [
                  "수리영 방안에 놓여진 백팩, 인형.",
                  "Listed as an appearing element across Shots 4, 5, and 9."
                ],
                "visual_inferences": [
                  "The backpack sits motionless on the floor.",
                  "The attached plush doll hangs limply, offering a slight contrast to the otherwise sterile room."
                ],
                "creative_decisions": [
                  "Using the displaced backpack as a constant visual anchor across the different angles and rooms to unify the location's underlying tension."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 14,
                "shot_index": 4,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "bb250c28aefb8fcf"
      },
      {
        "scene_index": 14,
        "heading": "S#14. 옥탑방 안 / N",
        "beat_title": "위로가 억압으로 느껴지는 순간",
        "representative_moment": "고개를 숙인 수리영(C01)의 양어깨를 혜수(C02)가 두 손으로 꽉 감싸 쥔 채 바라본다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "frontal medium close-up at chest height, muted low-key shadows",
            "t2i_prompt": "Photorealistic cinematic still. Frontal medium close-up at chest height. Focus on the tense grip between the two figures. C02O03 in a dark casual coat, a middle-aged Korean woman with black hair, mature facial lines, and distinct clear eyes, facing the camera. She is centered and bracing forward, her hands locked firmly around the shoulders of C01O02 in the near foreground. C01O02 in a dark jumper, a young Korean woman with long black hair, soft facial contours, and golden wavy irises visible as she looks down, positioned slightly lower at three-quarter angle from the right. C01O02's head is bowed inward, contained tightly between C02O03's arms. [L05: dim, damp-toned flat wall directly behind their torsos, an out-of-focus dining table edge on the right, the shade side of a faint living room lamp, and a blurred TV screen angled away]. Muted lamp glow washes across C02O03's side, while a faint TV spill reflects on C01O02's hair. Low-key shadows pool under the locked arms, keeping the grip visually tense and claustrophobic.",
            "outfit_assignments": [
              {
                "character_id": "C02",
                "outlook_id": "O03"
              },
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "혜수: (수리영 어깨를 어루만지며) 영아.",
              "수리영 방 안 지나칠 정도로 깨끗하게 정돈된 방안.",
              "shot description: 고개를 숙인 수리영의 양어깨를 두 손으로 꽉 감싸 쥔 혜수의 상체 정면."
            ],
            "visual_inferences": [
              "fixed_element backpack_with_plush_doll is physically on the floor but out of frame below the chest-height composition.",
              "Hye-soo's comforting gesture visually turns into a physical constraint due to the firm grip."
            ],
            "creative_decisions": [
              "Chest-height MCU isolates their upper bodies to enhance the feeling of claustrophobia and tension.",
              "Close framing skips the main room reference to focus purely on the immediate frame-edge background surfaces.",
              "Muted lamp and faint TV spill provide minimal practical lighting, allowing deep shadows around the arms."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "f46c7e3ff8ec207b",
              "owned_hash": "052e7e8396b50472",
              "camera_direction_hash": "5c788ba234ec6005",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 14,
        "dependency_reason": "Same room environment as S14 Shot 5, but close-framing omits direct background reference to focus on characters' immediate interaction.",
        "_shot_index": 9,
        "visible_entities": [
          "C02",
          "C01",
          "P05",
          "L05"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 14,
            "shot_index": 9
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "In the main room, dolly into a frontal MCU at chest height, keeping 혜수’s hands locked around 수리영’s shoulders in the near foreground of the composition. The camera holds their upper bodies on the same axis: 혜수 centered and braced forward, 수리영 slightly lower with her bowed head contained between 혜수’s arms, turning comfort into restraint.",
            "lighting_mood": "Muted lamp glow and faint TV spill soften the room, but the low-key shadows keep 혜수’s grip visually tense and claustrophobic.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C02",
              "C01",
              "P05",
              "L05"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C02",
                "outlook_id": "O03"
              },
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "cb_living_night_clean_reset",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "backpack_with_plush_doll",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "A casual, dark-colored backpack with a small plush doll dangling from its zipper. It rests quietly on the floor, completely stationary, serving as a subtle reminder of ordinary life amidst the unsettling environment.",
                "applies_to_shots": [
                  4,
                  5,
                  9
                ],
                "source_facts": [
                  "수리영 방안에 놓여진 백팩, 인형.",
                  "Listed as an appearing element across Shots 4, 5, and 9."
                ],
                "visual_inferences": [
                  "The backpack sits motionless on the floor.",
                  "The attached plush doll hangs limply, offering a slight contrast to the otherwise sterile room."
                ],
                "creative_decisions": [
                  "Using the displaced backpack as a constant visual anchor across the different angles and rooms to unify the location's underlying tension."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 14,
                "shot_index": 5,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C02O03",
                "policy": "required"
              },
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "b5883a65dde59f57"
      },
      {
        "scene_index": 15,
        "heading": "S#15. 해안가. 버스정류장 부근 / N",
        "beat_title": "집으로 돌아갈 결심",
        "representative_moment": "C01이 집으로 가야겠다며 벤치에서 벌떡 일어나 결의에 찬 표정을 짓는다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Tight MCU at eye-level, frontal composition, cold sodium light with blue-tinted shadows.",
            "t2i_prompt": "Photorealistic cinematic still. C01O02 in a dark casual jacket, a Korean young woman with long black hair, soft facial structure, and golden wave-like eyes, facing the camera, standing fully upright, shoulders squared toward the front. Tight framing on her face, upper body in frame, eyes fixed on the distant darkness with a determined expression. Cold coastal sodium light washing across her cheek, misty blue shadows falling on her neck, and faint distant red light reflecting on her damp shoulder. A wet glass panel streaked with condensation forms a faint reflective plane along the side frame edge, while a soft-focus [L12: cold misty night coastal environment] with the dark sea horizon and part of an empty damp bench recedes behind her.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#15 text: '수리영... (훅 일어나며) 집에 가봐야 겠어.'",
              "Shot 3 description: tight MCU at eye-level, determined expression, upright."
            ],
            "visual_inferences": [
              "fixed_element: damp bus-stop glass, cold sea mist, low-key sodium streetlights",
              "Tight framing blurs the background sea and bench into soft-focus elements."
            ],
            "creative_decisions": [
              "Adjusted the DP lighting instruction ('carve her face') to 'washing across her cheek' and 'falling on her neck' to prevent T2I physical mutilation artifacts.",
              "Emphasized the upper body and shoulder within the MCU to anchor the head properly and avoid floating-face generation."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "9f026373f66957b9",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "63b4a90dde95adec",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "",
        "_shot_index": 3,
        "visible_entities": [
          "C01",
          "L12"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 15,
            "shot_index": 3
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "In front of the seaside bus-stop bench, push into a tight MCU at eye-level as 수리영 is already fully upright, her determined expression filling most of the center frame without losing the bench and dark shoreline behind her. The composition is nearly frontal for the resolve beat, with her shoulders squared toward the road and the damp bus-stop glass forming a faint reflective plane along the side.",
            "lighting_mood": "Cold coastal sodium light and misty blue shadows carve her face, with only faint distant red tail-light glints cutting the darkness.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "L12"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": null,
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "coastal_bus_stop_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dark seaside bus stop at night, featuring a bench and damp glass panels. The surrounding air is thick with cold sea mist. The area is illuminated by low-key sodium streetlights, casting misty blue shadows across the wet ground and surfaces.",
                "applies_to_shots": [
                  3,
                  5
                ],
                "source_facts": [
                  "Location: L12 (해안가 버스정류장 부근)",
                  "Shot 3 mentions 'seaside bus-stop bench', 'damp bus-stop glass', 'cold coastal sodium light and misty blue shadows'.",
                  "Shot 5 mentions 'bus stop bench', 'wet coastal road', 'low-key sodium streetlight mixes with cold sea mist'."
                ],
                "visual_inferences": [
                  "The ground and glass panels remain visibly wet and damp throughout, reflecting the muted sodium streetlights."
                ],
                "creative_decisions": [
                  "Consistently apply a foggy, dark coastal atmosphere with blue-tinted shadows to emphasize isolation and coldness across both shots."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "7dc6aba6c2c6c4c6"
      },
      {
        "scene_index": 15,
        "heading": "S#15. 해안가. 버스정류장 부근 / N",
        "beat_title": "달려가는 수리영과 뒤에서 외치는 혜수",
        "representative_moment": "C02가 한 손을 입에 모으고 소리치며, 멀어지는 C01의 뒷모습을 바라본다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Eye-level MCU, shallow DOF, left foreground focus, right background softened.",
            "t2i_prompt": "Photorealistic cinematic still. At eye level, C02O03 in a dark detective coat, a middle-aged Korean woman with mature facial lines, black hair, and sharp eyes, occupies the left foreground from the chest up in right profile. One hand is cupped at her mouth as she calls out, eyes fixed on the distant figure. She stands beside an empty, damp metal bus stop bench sharing a single line of sight from the camera down the wet coastal road. In the right-background distance, C01O02 in a dark jacket, a young Korean woman with long black hair, appears as a small, out-of-focus soft background figure with her back to the camera, captured mid-stride running away along the road, her jacket flowing in the direction opposite to motion. Low-key sodium streetlight mixes with thick cold sea mist, leaving C02O03's features muted while casting misty blue shadows across the wet asphalt that fades into a blue-black haze. Focus on C02O03's profile and cupped hand, isolating her from the deep mist behind.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              },
              {
                "character_id": "C02",
                "outlook_id": "O03"
              }
            ],
            "source_facts": [
              "S15 action: 수리영, 달려간다. 혜수, (등 뒤에 대고 소리친다)",
              "Shot description: 혜수 한 손을 입가에 모으고 입을 크게 벌린 채 상체, 수리영은 거리가 벌어진 뒷모습",
              "Location: L12 해안가 버스정류장 부근 / N"
            ],
            "visual_inferences": [
              "fixed_element coastal_bus_stop_environment: damp bus stop bench, wet road, cold sea mist, misty blue shadows",
              "C01's face is fully obscured as she faces away from the camera in mid-run",
              "C02 is in sharp focus to emphasize her shouting action while C01 is soft in the distance"
            ],
            "creative_decisions": [
              "Used 'sharing a single line of sight from the camera' with the 'damp metal bus stop bench' to anchor the fg/bg separation",
              "Mid-stride pose with jacket flowing in the opposite direction applied to freeze the running motion dynamically"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "8dd80cc26bb09e8b",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "cba7d6cc05ae797d",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 15,
        "dependency_reason": "Shot 3의 해안가 버스정류장(exact_background)과 환경 상태 연속성 유지",
        "_shot_index": 5,
        "visible_entities": [
          "C01",
          "C02",
          "L12"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 15,
            "shot_index": 5
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "From the bus stop bench position, an eye-level MCU holds slightly behind and to the side of 혜수, using a shallow DOF as she occupies the sharp left foreground from chest up with one hand cupped at her mouth. The camera inherits the pan’s stopped midpoint: the wet coastal road opens diagonally behind her, and 수리영 is a small, softened figure running away in the right-background distance.",
            "lighting_mood": "Low-key sodium streetlight mixes with cold sea mist, leaving 혜수’s face muted while the road falls into blue-black haze.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "C02",
              "L12"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              },
              {
                "character_id": "C02",
                "outlook_id": "O03"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": null,
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "coastal_bus_stop_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dark seaside bus stop at night, featuring a bench and damp glass panels. The surrounding air is thick with cold sea mist. The area is illuminated by low-key sodium streetlights, casting misty blue shadows across the wet ground and surfaces.",
                "applies_to_shots": [
                  3,
                  5
                ],
                "source_facts": [
                  "Location: L12 (해안가 버스정류장 부근)",
                  "Shot 3 mentions 'seaside bus-stop bench', 'damp bus-stop glass', 'cold coastal sodium light and misty blue shadows'.",
                  "Shot 5 mentions 'bus stop bench', 'wet coastal road', 'low-key sodium streetlight mixes with cold sea mist'."
                ],
                "visual_inferences": [
                  "The ground and glass panels remain visibly wet and damp throughout, reflecting the muted sodium streetlights."
                ],
                "creative_decisions": [
                  "Consistently apply a foggy, dark coastal atmosphere with blue-tinted shadows to emphasize isolation and coldness across both shots."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 15,
                "shot_index": 3,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              },
              {
                "kind": "character_outlook",
                "id": "C02O03",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "95c1dadc7f442b81"
      },
      {
        "scene_index": 16,
        "heading": "S#16. 경찰서 사무실 / N",
        "beat_title": "위치추적 요청과 불길한 예감",
        "representative_moment": "김형사의 시점(POV)에서, 혜수가 핸드폰 화면을 강하게 들이밀며 위치추적을 요구하는 팽팽한 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "A subjective POV extreme close-up from a seated position, using the desk monitor as a blurred foreground framing device, cold fluorescent lighting undercut by a pale screen glow",
            "t2i_prompt": "Photorealistic cinematic still. A subjective POV from a seated observer behind a desk. Detail on a Korean middle-aged woman's hand firmly gripping P03, a smartphone, held out directly toward the camera. The brightly glowing screen sits centered and tight, taking the strongest visual weight. The dark-colored plastic casing of an office desk monitor occupies the lower foreground as a strongly blurred physical barrier, its top edge forming a dim horizontal border just below the lens. Cold fluorescent office light is undercut by the phone’s pale glow washing over the thumb and fingers, giving the moment a sickly urgency. [L13: dim, entirely out-of-focus soft textures of bureaucratic office partitions in the deep background].",
            "outfit_assignments": [],
            "source_facts": [
              "S#16 action: 혜수가 핸드폰에 저장된 민숙번호를 보여줌",
              "DP instruction: subjective POV from his seated eye line behind the desk, lens just above monitor height, fixes on 혜수’s hand thrusting a smartphone toward him",
              "DP instruction: The glowing contact entry for 민숙 is sharp, while the desk edge and monitor top form dim horizontal borders around the frame"
            ],
            "visual_inferences": [
              "fixed_element office_desk_monitor: positioned as a physical barrier between the seated viewer and the hand, occupying the lower foreground",
              "body_part_focus_rule applied: face is excluded to maintain the ECU framing on the smartphone and hand, using a common noun demographic descriptor instead of C02O03",
              "the hand is firmly held out in a mid-action freeze without motion blur to keep the screen sharp"
            ],
            "creative_decisions": [
              "using extremely shallow depth of field to keep only the hand and P03 sharp, rendering both the foreground monitor and background partitions as soft compositional anchors",
              "mixing cold ambient fluorescent light with the localized pale glow of the phone screen to emphasize the tense, unsettling mood"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "1132778dfbdf1ef7",
              "owned_hash": "88d6c6cc6ee9f273",
              "camera_direction_hash": "74f1ba95eccafe7a",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "N/A",
        "_shot_index": 3,
        "visible_entities": [
          "C02",
          "C05",
          "P03",
          "L13"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 16,
            "shot_index": 3
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A 김형사 subjective POV ECU from his seated eye line behind the desk, lens just above monitor height, fixes on 혜수’s hand thrusting a smartphone toward him; the phone screen sits centered and tight, taking the strongest visual weight without exceeding a natural prop scale. The glowing contact entry for 민숙 is sharp, while the desk edge and monitor top form dim horizontal borders around the frame.",
            "lighting_mood": "Cold fluorescent police-office light is undercut by the phone’s pale glow, giving the evidence a sickly urgency.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C02",
              "C05",
              "P03",
              "L13"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C02",
                "outlook_id": "O03"
              },
              {
                "character_id": "C05",
                "outlook_id": "O07"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "bg_office_night_duty_normal",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "office_desk_monitor",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "A dark-colored computer monitor sitting on an office desk. It is positioned as a physical barrier between the seated viewer behind the desk and anyone standing on the opposite side. The top and side edges of the monitor frequently occupy the lower and lateral foreground of the frame.",
                "applies_to_shots": [
                  3,
                  5
                ],
                "source_facts": [
                  "Shot 3: lens just above monitor height, desk edge and monitor top form dim horizontal borders around the frame.",
                  "Shot 5: two blurred monitor edges in the foreground, leans over the monitor symmetrically."
                ],
                "visual_inferences": [
                  "standard dark plastic casing for an office computer monitor",
                  "placed squarely in front of the seated officer"
                ],
                "creative_decisions": [
                  "using the monitor as a blurred foreground framing device (through-object composition) to anchor the subjective seated POV and emphasize the barricade-like feeling of the desk"
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C02O03",
                "policy": "required"
              },
              {
                "kind": "character_outlook",
                "id": "C05O07",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "66a5899e750211c1"
      },
      {
        "scene_index": 16,
        "heading": "경찰서 사무실 / N",
        "beat_title": "민숙의 실종 가능성에 대한 확신과 위치추적 지시",
        "representative_moment": "C02가 책상 모니터 위로 상체를 숙이며 심각한 표정으로 카메라(C05 시점)를 내려다본다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Low angle medium shot from a seated subjective POV, flat green-white fluorescent lighting with cold monitor underlight.",
            "t2i_prompt": "Photorealistic cinematic still. A subjective POV from a seated position, looking up at a low angle. C02O03 in a casual jacket, a middle-aged Asian woman, faces the camera directly, leaning symmetrically forward over the dark computer monitor from the reference. Her hands are planted on the scuffed desk from the reference near the monitor's side bezels. Her eyes are fixed intensely down into the lens. The blurred top and side edges of the monitor occupy the lower and lateral foreground, acting as a physical barricade between the viewer and the subject. Flat green-white fluorescent light washes her hair and shoulders, casting dark shadows under her eyes, while a cold underlight from the monitor spill reflects on her jaw and lower face. Behind her, the closed metal filing cabinet from the reference stands in the dim office background. Focus on the tense posture of C02O03 pressing forward over the foreground barrier.",
            "outfit_assignments": [
              {
                "character_id": "C02",
                "outlook_id": "O03"
              }
            ],
            "source_facts": [
              "S16 action: '혜수 ... 이상해, 예감이 안 좋아.'",
              "Shot 5 description: two blurred monitor edges in the foreground, leans over the monitor symmetrically.",
              "Shot 5 camera instructions: From the same seated 김형사 POV, a low angle MS frames 혜수."
            ],
            "visual_inferences": [
              "fixed_element office_desk_monitor: positioned as a physical barrier, top and side edges occupy lower and lateral foreground.",
              "C05 is the subjective viewer, thus excluded from the frame.",
              "The subjective POV naturally places the desk and monitor in the immediate lower frame."
            ],
            "creative_decisions": [
              "Through-object composition using the monitor to frame C02 and create a barricade feel.",
              "Lighting contrast between the harsh ceiling fluorescents and the cold monitor spill to emphasize the character's urgency."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "fceb860627fcc5c3",
              "owned_hash": "88d6c6cc6ee9f273",
              "camera_direction_hash": "e77f63361f3dba20",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "monitor",
                  "violating_phrase": "the dark computer monitor from the reference",
                  "reason": "The monitor is referenced as an existing object from the reference, so it is anchored rather than newly created or modified.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "desk",
                  "violating_phrase": "the scuffed desk from the reference",
                  "reason": "The desk is explicitly described as being from the reference, which marks it as an anchored existing object.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "monitor",
                  "violating_phrase": "the blurred top and side edges of the monitor",
                  "reason": "This is a framing/foreground description of an existing monitor, not a command to draw a new one.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "filing cabinet",
                  "violating_phrase": "the closed metal filing cabinet from the reference",
                  "reason": "The filing cabinet is explicitly tied to the reference, so it is treated as an existing background object.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 16,
        "dependency_reason": "exact_background reference from Shot 3 to maintain consistent office environment.",
        "_shot_index": 5,
        "visible_entities": [
          "C02",
          "C05",
          "P03",
          "L13"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 16,
            "shot_index": 5
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "From the same seated 김형사 POV, a low angle MS frames 혜수 between the two blurred monitor edges in the foreground, using a through-object composition that makes the desk feel like a barricade. She leans over the monitor symmetrically, both hands planted near its sides, staring down into the lens as if pressing 김형사 into action.",
            "lighting_mood": "Flat green-white fluorescents carve dark hollows under 혜수’s eyes, with the monitor spill adding a colder underlight to her urgent face.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C02",
              "C05",
              "P03",
              "L13"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C02",
                "outlook_id": "O03"
              },
              {
                "character_id": "C05",
                "outlook_id": "O07"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "bg_office_night_duty_normal",
            "reference_usage": "exact_background",
            "owned_objects": [
              "box",
              "case board",
              "chair",
              "clock",
              "computer",
              "console",
              "copier",
              "desk",
              "door",
              "filing cabinet",
              "folder",
              "monitor",
              "photograph",
              "printer",
              "radio",
              "telephone",
              "tray",
              "waste bin",
              "window"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "office_desk_monitor",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "A dark-colored computer monitor sitting on an office desk. It is positioned as a physical barrier between the seated viewer behind the desk and anyone standing on the opposite side. The top and side edges of the monitor frequently occupy the lower and lateral foreground of the frame.",
                "applies_to_shots": [
                  3,
                  5
                ],
                "source_facts": [
                  "Shot 3: lens just above monitor height, desk edge and monitor top form dim horizontal borders around the frame.",
                  "Shot 5: two blurred monitor edges in the foreground, leans over the monitor symmetrically."
                ],
                "visual_inferences": [
                  "standard dark plastic casing for an office computer monitor",
                  "placed squarely in front of the seated officer"
                ],
                "creative_decisions": [
                  "using the monitor as a blurred foreground framing device (through-object composition) to anchor the subjective seated POV and emphasize the barricade-like feeling of the desk"
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 16,
                "shot_index": 3,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C02O03",
                "policy": "required"
              },
              {
                "kind": "character_outlook",
                "id": "C05O07",
                "policy": "required"
              },
              {
                "kind": "background",
                "id": "bg_office_night_duty_normal",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "d82d04362325622e"
      },
      {
        "scene_index": 17,
        "heading": "S#17. 옥탑방 / 동틀무렵",
        "beat_title": "새벽을 맞이한 옥탑방",
        "representative_moment": "푸른 새벽빛 속 홀로 따뜻한 불빛을 내뿜는 옥탑방 창문의 고요한 외경",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Extreme wide shot, static low angle, blue predawn ambient light",
            "t2i_prompt": "Photorealistic cinematic still. Extreme wide shot, a static low angle observing the old rooftop room from across the roofs, positioned slightly below the lit window height. Focus on the warmly lit window from the reference sitting off-center. Blue predawn ambient light washes over the damp, weathered concrete and the parapet wall from the reference. The closed window glows intensely from inside against the cold blue surroundings. Wet low rooftop walls create depth with deep blue dawn shadows in the lower frame. Thin silhouetted TV antennas and cables stand stark against a wide, mist-heavy blue-gray dawn sky, capturing an unnerving stillness.",
            "outfit_assignments": [],
            "source_facts": [
              "scene_text: '동틀무렵... 불이 켜져 있는 옥탑 방'",
              "shot description: 동이 트는 푸르스름한 새벽빛 속에서 불이 켜진 낡은 옥탑방 창문이 보이는 고요한 외관 풍경"
            ],
            "visual_inferences": [
              "exact background reference used for rooftop architectural elements",
              "damp textures on the concrete and parapet walls reflecting the cold dawn moisture"
            ],
            "creative_decisions": [
              "wide blue dawn frame to emphasize the smallness of the architecture",
              "stark contrast between the cold blue ambient light and the warm glow of the single window"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "757353857f163a71",
              "owned_hash": "de6f1110b77a4310",
              "camera_direction_hash": "804551a6df170434",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "window",
                  "violating_phrase": "the warmly lit window from the reference",
                  "reason": "The prompt explicitly references the existing window from the reference, so this is an anchor reference rather than a request to redraw it.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "parapet wall",
                  "violating_phrase": "the parapet wall from the reference",
                  "reason": "The parapet wall is mentioned as part of the reference background, which anchors it instead of asking to create or modify it.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "window",
                  "violating_phrase": "The closed window glows intensely",
                  "reason": "This describes the existing window's state and appearance without an explicit redraw verb targeting it.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 4,
        "dependency_reason": "동일한 옥탑방 공간을 다루며(exact_background), 밤에서 새벽으로 전환된 조명 톤과 분위기만 변경하여 앞선 씬과의 연속성을 유지함",
        "_shot_index": 1,
        "visible_entities": [
          "L04"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 17,
            "shot_index": 1
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A static EWS low angle watches the old rooftop room from across the roofs, slightly below the lit window height, keeping the architecture small inside a wide blue dawn frame. The warm window sits off-center against damp concrete, antennas, and low parapets, held unnervingly still before any lateral movement begins.",
            "lighting_mood": "Blue predawn ambient light swallows the rooftop while the single warm window glows like an exposed nerve.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "L04"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "cb_rooftop_predawn_light_on",
            "reference_usage": "exact_background",
            "owned_objects": [
              "clothesline",
              "deck",
              "door",
              "parapet wall",
              "platform",
              "post",
              "roof drain",
              "stairwell bulkhead",
              "utility pipe",
              "wall light",
              "water tank",
              "window"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [],
            "previous_shot_refs": [
              {
                "scene_index": 4,
                "shot_index": 1,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "background",
                "id": "cb_rooftop_predawn_light_on",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "5ac63bad6cae792d"
      },
      {
        "scene_index": 18,
        "heading": "S#18. 안방 안 / D",
        "beat_title": "번쩍 눈을 뜬 수리영",
        "representative_moment": "깜빡 졸다 고양이 소리에 놀라 고개를 치켜들며 눈을 번쩍 뜬 C01의 굳은 얼굴.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Handheld close framing at eye-level from a front-left angle, shallow depth of field.",
            "t2i_prompt": "Photorealistic cinematic still. C01O02 in a casual top, a young Korean woman with long black hair, soft facial contours, and golden wave-like irises, at three-quarter angle from the right. Her head snaps awake with a rigid expression, eyes flashed wide open and fixed on the unseen closed bedroom door beyond camera-left. The composition stays low within arm's reach, framing her neck and upper chest slightly right of center from a front-left angle to avoid an isolated floating face. The shallow depth of field sharply isolates her tense expression against the immediate surroundings. [L05: Soft, out-of-focus edges of a rumpled bed and pillow occupy the bottom frame, backed by a dim, pale, slightly stained bedroom wall in shadow]. Cool daylight washes over the interior, draining warmth from her skin and leaving muted gray shadows, creating a quiet, low-key atmosphere filled with tremulous handheld unease.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#18 action: '고양이 울음소리에 번쩍 눈을 뜨고'",
              "Shot 2 description: '번쩍 뜬 눈으로 고개를 치켜든 수리영의 잔뜩 굳은 얼굴.'",
              "DP instruction: 'A handheld CU at eye-level stays low beside the bed... tightening on her rigid face... bed and pillow reduced to soft edges... slightly right of center from a front-left angle'"
            ],
            "visual_inferences": [
              "fixed_element cool_daylight_apartment_interior applied: cool daylight draining warmth, muted gray shadows, low-key lighting.",
              "Include neck and upper chest in the frame to ground the figure and prevent a floating head artifact."
            ],
            "creative_decisions": [
              "Replaced forbidden 'focus on' phrasing with 'shallow depth of field sharply isolates' to emphasize her expression without triggering body-part focus constraints.",
              "Skipped direct background reference due to close framing, instead constructing the frame edges (bed, pillow, wall) entirely within the L05 block."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "eb6080d10c3ac009",
              "owned_hash": "0ad90dcb6c66fe5a",
              "camera_direction_hash": "b7edabd6380f77e0",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 14,
        "dependency_reason": "동일한 아파트 내부의 톤(cool_daylight_apartment_interior)을 유지하기 위해 연속성을 참조하나, 클로즈업 구도이므로 물리적 배경 합성은 스킵됨.",
        "_shot_index": 2,
        "visible_entities": [
          "C01",
          "P05",
          "P03",
          "P01",
          "P08",
          "L05"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 18,
            "shot_index": 2
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A handheld CU at eye-level stays low beside the bed, within arm’s length of 수리영, tightening on her rigid face as her eyes snap open and her head lifts. The bed and pillow are reduced to soft edges at the bottom of frame, while her face sits slightly right of center from a front-left angle, trapped in shallow DOF.",
            "lighting_mood": "Thin daylight leaks into a low-key room, leaving her skin cool and gray with tremulous handheld unease.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P05",
              "P03",
              "P01",
              "P08",
              "L05"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "cb_master_bedroom_dawn_waiting_empty",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "applies_to_shots": [
                  2,
                  14
                ],
                "character_name": "",
                "confidence": "high",
                "creative_decisions": [
                  "Keep the apartment scenes visually unified with a cool, exposed daytime palette despite the change from bedside framing to dining-table framing."
                ],
                "description": "Cool daylight lies over the apartment interior, draining warmth from skin tones and leaving muted gray shadows. The room should feel quiet, low-key, and visually restrained, with no warm practical light dominating the frame.",
                "element_id": "cool_daylight_apartment_interior",
                "element_type": "environment_state",
                "source_facts": [
                  "Shot 2 lighting: 'Thin daylight leaks into a low-key room, leaving her skin cool and gray.'",
                  "Shot 14 lighting: 'Cool daylight and muted apartment shadows make her fear feel exposed.'"
                ],
                "visual_inferences": []
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 14,
                "shot_index": 5,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "41c137b14e96f6c6"
      },
      {
        "scene_index": 18,
        "heading": "S#18. 옥탑방 욕실 / D",
        "beat_title": "거울에 비친 붉은 환각",
        "representative_moment": "거울에 반사된 차가운 욕실 타일 위로 거대하고 붉은 원형 표식(P01)이 선명하게 나타난다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "A static Dutch angle CU pressing close to the mirror plane, cold flat lighting with an unnaturally saturated red hallucination",
            "t2i_prompt": "Photorealistic cinematic still. Close framing from a static Dutch angle pressing tightly against [L05: a slightly fogged bathroom mirror, with a wet white porcelain sink rim barely grazing the lower frame edge]. Focus on the reflection inside the glass. Within the mirrored surface, a large, unnaturally saturated red circular mark, P01, burns vividly on the reflected pale tile wall. Cold bathroom light flattens the reflected tiles, while the hallucinated red mark appears with a wet, striking contrast, entirely detached from the physical space.",
            "outfit_assignments": [],
            "source_facts": [
              "S18 action: '거울에 비친 욕실 벽에 갑자기 또 발생하는 붉은 원.'",
              "shot description: 거울 안쪽으로 반사된 욕실 타일 벽면에 맺힌 거대한 붉은 원형 표식 이미지."
            ],
            "visual_inferences": [
              "subjective POV: 수리영 is looking into the mirror, so she is out of frame",
              "hallucination reflection: the red mark only exists within the mirror's distortion",
              "sink rim provides minimal spatial scale at the bottom edge"
            ],
            "creative_decisions": [
              "Dutch angle applied to heighten the psychological tension and disorientation of the hallucination",
              "Cold, flat lighting on the tiles used to maximize the contrast of the vivid, burning red mark"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "f8058adabf7c0cf2",
              "owned_hash": "6c35426d65c6ece5",
              "camera_direction_hash": "4094be44d5b7915d",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 18,
        "dependency_reason": "같은 옥탑방 공간의 연속성 유지 (단, 이 샷은 클로즈업 프레이밍에 의한 주관적 환각이므로 배경 참조가 생략됨)",
        "_shot_index": 10,
        "visible_entities": [
          "P05",
          "P03",
          "P01",
          "P08",
          "L05"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 18,
            "shot_index": 10
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A static Dutch angle CU presses close to the bathroom mirror plane, composing only the reflected tile wall where a large red circular mark appears, with the sink edge barely grazing the lower frame for scale. This is 수리영’s subjective perception through reflection: the real room is displaced, the reflected tiles dominate, and the apparition feels detached from physical space.",
            "lighting_mood": "Cold bathroom light flattens the tiles while the red mark burns unnaturally saturated, signaling a hallucinated reflection.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "P05",
              "P03",
              "P01",
              "P08",
              "L05"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "cb_bathroom_day_mirror_red_mark_vision",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [],
            "previous_shot_refs": [
              {
                "scene_index": 18,
                "shot_index": 2,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "18901fad20ba5c0d"
      },
      {
        "scene_index": 18,
        "heading": "S#18. 안방 안 / D",
        "beat_title": "환각과 이성 사이의 공포",
        "representative_moment": "겁에 질린 C01이 양손으로 자신의 뺨을 꽉 감싸 쥔 채 식탁 위 P08을 내려다본다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Static MCU eye-level, tight frontal composition, cool daylight with muted gray shadows and a faint residual red unease.",
            "t2i_prompt": "Photorealistic cinematic still. A static MCU eye-level composition, tight on the subject with upper chest and shoulders in frame. Cool daylight washes the scene, draining warmth into muted gray shadows, with a faint residual red ambient glow catching the lower frame. C01O02 in a comfortable casual tracksuit, a young Korean woman with long black hair, a soft facial shape, and golden wave-like eyes, facing the camera. Both of her hands are firmly clamped to her own cheeks in sheer panic, her eyes fixed downward on P08. P08, a photo frame, occupies the lower foreground near the camera axis, its front glass angled slightly toward the camera. [L05: the hard edge of a clean dining table separating the camera and the figure, and a slice of a stiff, over-tidy domestic wall softly blurred just past her shoulders].",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#18 action: '(자신의 뺨을 치며) 정신 차려 수리영!'",
              "Shot 14 description: Her hands clamped to both cheeks become the emotional center",
              "DP background: photo frame near-axis trigger object in foreground"
            ],
            "visual_inferences": [
              "fixed_element cool_daylight_apartment_interior: Cool daylight washing the scene, draining warmth into muted gray shadows",
              "faint residual red ambient glow reflecting the hallucination's aftermath"
            ],
            "creative_decisions": [
              "Tight MCU framing with upper chest and shoulders visible to avoid floating face artifact while maintaining close emotional intensity",
              "[L05] background composed purely of immediate frame-edge surfaces (table edge and wall slice) to strictly comply with skipped_close_framing policy",
              "Foreground P08 placement anchors her downward gaze and establishes the physical distance across the table"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "29207e670fb2ccb8",
              "owned_hash": "d7adc39a19f9f935",
              "camera_direction_hash": "cf111baa49197294",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 18,
        "dependency_reason": "같은 공간 내 연속된 상황(background_binding.mode=skipped_close_framing 에 따라 직접 묘사로 우회)",
        "_shot_index": 14,
        "visible_entities": [
          "C01",
          "P05",
          "P03",
          "P01",
          "P08",
          "L05"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 18,
            "shot_index": 14
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A static MCU eye-level holds a tight frontal composition at the dining table, with 수리영 centered and the photo frame near the camera axis in the lower foreground. Her hands clamped to both cheeks become the emotional center, while the rest of the apartment recedes into stiff, over-clean domestic lines after the hallucination.",
            "lighting_mood": "Cool daylight and muted apartment shadows make her fear feel exposed, with only a faint residual red unease implied by the photo’s position.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P05",
              "P03",
              "P01",
              "P08",
              "L05"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "cb_living_day_photo_vision",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "applies_to_shots": [
                  2,
                  14
                ],
                "character_name": "",
                "confidence": "high",
                "creative_decisions": [
                  "Keep the apartment scenes visually unified with a cool, exposed daytime palette despite the change from bedside framing to dining-table framing."
                ],
                "description": "Cool daylight lies over the apartment interior, draining warmth from skin tones and leaving muted gray shadows. The room should feel quiet, low-key, and visually restrained, with no warm practical light dominating the frame.",
                "element_id": "cool_daylight_apartment_interior",
                "element_type": "environment_state",
                "source_facts": [
                  "Shot 2 lighting: 'Thin daylight leaks into a low-key room, leaving her skin cool and gray.'",
                  "Shot 14 lighting: 'Cool daylight and muted apartment shadows make her fear feel exposed.'"
                ],
                "visual_inferences": []
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 18,
                "shot_index": 2,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "f66214e1fad4fe40"
      },
      {
        "scene_index": 19,
        "heading": "S#19. 마트 / D",
        "beat_title": "마트 입구에서의 긴장된 멈춤",
        "representative_moment": "이른 아침 한산한 마트 입구 앞, C01이 자전거 핸들을 꽉 쥔 채 고개를 숙이고 서 있다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Static eye-level medium wide shot, muted early-morning gray light with a damp blue cast",
            "t2i_prompt": "Photorealistic cinematic still. C01O02 in a dark jacket, a young Korean woman with long black hair, soft facial shape, and golden wave-like eyes, facing the camera outside the quiet mart entrance. She stands centered directly in front of the sliding door from the reference, her eyes looking down. Both hands tightly grip the bicycle handlebars cutting across the lower foreground, with the bicycle frame and front wheel clearly visible below her grip. Muted early-morning gray light flattens the storefront from the reference, casting a cold, damp blue ambient haze over the street and concrete wall from the reference, keeping the empty space tense and uneasy.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "Scene heading: S#19. 마트 / D",
              "Scene text: 이른 아침, 아직은 한산한 마트 안. 자전거를 주차하고 안으로 들어가는 수리영.",
              "Shot 1 description: 한산한 마트 입구 앞, 자전거 핸들을 양손으로 꽉 쥔 채 멈춰 선 수리영의 정면."
            ],
            "visual_inferences": [
              "fixed_element mart_exterior_environment: cold, damp, muted gray and blue ambient haze",
              "Bicycle body needs to be visible in the lower frame to prevent floating hands",
              "Her posture is rigid, indicating the tension described in the DP directive"
            ],
            "creative_decisions": [
              "Used eye-level medium wide shot to capture both the character's tense grip in the foreground and the sliding door in the background",
              "Anchored background elements explicitly using 'from the reference' to prevent duplicate storefront generation"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "18a7c1a71ed650b9",
              "owned_hash": "aa09dcc0158b9ca2",
              "camera_direction_hash": "2e58f127d93cb26c",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "sliding door",
                  "violating_phrase": "the sliding door from the reference",
                  "reason": "The phrase explicitly marks the sliding door as coming from the reference, so it is treated as an anchored existing background object rather than a newly drawn one.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "storefront",
                  "violating_phrase": "the storefront from the reference",
                  "reason": "The storefront is referenced as existing in the reference, which indicates anchoring to the background rather than redrawing it.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "street",
                  "violating_phrase": "the street and concrete wall from the reference",
                  "reason": "The street is mentioned only as part of an existing reference-backed environment description, not as a newly created object.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 8,
        "dependency_reason": "S8_Shot5와 동일한 마트 외부 공간이므로, exact_background를 통해 배경 구조와 이른 아침의 차갑고 푸른 색감을 일관되게 유지",
        "_shot_index": 1,
        "visible_entities": [
          "C01",
          "P03",
          "L09",
          "L10"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 19,
            "shot_index": 1
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A static eye-level MWS faces 수리영 head-on outside the quiet mart entrance, placing her center frame with the glass doors directly behind her. The bicycle handlebars cut across the lower foreground as a tense horizontal line, while her white-knuckled grip is emphasized without letting the bicycle overpower the frame.",
            "lighting_mood": "Muted early-morning gray light flattens the storefront, with a damp blue cast keeping the empty mart uneasy rather than ordinary.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P03",
              "L09",
              "L10"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "bg_entrance_exterior_morning_quiet_cctv_visible",
            "reference_usage": "exact_background",
            "owned_objects": [
              "CCTV camera",
              "awning",
              "bollard",
              "curb",
              "sidewalk",
              "sliding door",
              "storefront",
              "street",
              "streetlamp",
              "window"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "mart_exterior_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The exterior surroundings of a local mart during an overcast early morning. The setting features stained concrete walls and a storefront area, enveloped in a cold, damp, muted gray and blue ambient haze.",
                "applies_to_shots": [
                  1,
                  4,
                  6
                ],
                "source_facts": [
                  "Scene heading: S#19 마트 / D",
                  "Scene text: 이른 아침, 아직은 한산한 마트 안 / 마트 밖",
                  "Shot 1: Muted early-morning gray light, damp blue cast, storefront",
                  "Shot 4: Overcast morning light, cold concrete pallor, stained exterior wall",
                  "Shot 6: cold morning haze, mart wall"
                ],
                "visual_inferences": [
                  "Consistent concrete texture and a cool, damp color palette connect the different angles of the mart's exterior structure."
                ],
                "creative_decisions": [
                  "Unified the lighting atmosphere with a tense blue/gray overcast tone to ground all exterior shots in the same early morning temporal continuity."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 8,
                "shot_index": 5,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              },
              {
                "kind": "background",
                "id": "bg_entrance_exterior_morning_quiet_cctv_visible",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "f15bd6ddbca232f2"
      },
      {
        "scene_index": 19,
        "heading": "S#19. 마트 / D",
        "beat_title": "CCTV 위치 확인",
        "representative_moment": "마트 외벽 아래에서 남자 직원이 손을 뻗어 위쪽에 달린 CCTV 카메라를 가리키고 있다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "static medium shot from a low angle, overcast morning light, cold blue-gray haze",
            "t2i_prompt": "Photorealistic cinematic still. A static medium shot from a low angle beside the mart wall. An East Asian middle-aged man in a practical worker uniform, positioned in left profile, occupies the left frame, his torso and head visible. His raised arm stretches upward, pointing toward the CCTV camera from the reference mounted high in the upper right frame. His eyes are fixed on the surveillance camera. The damp, faded concrete wall extends diagonally away from the camera, carrying an exposed, slightly rusted wall-mounted conduit. Shadowed, water-stained eaves form the upper edge of the frame. Muted early-morning overcast light casts a cold, damp blue-gray ambient haze, giving the wall a pale concrete pallor, while a small hard highlight catches on the casing of the CCTV camera from the reference. Focus on the diagonal alignment between the pointing arm and the surveillance camera.",
            "outfit_assignments": [],
            "source_facts": [
              "S#19 action: '엄마와 남자 직원CCTV 위치를 확인하다. 찍혔을 만한 위치다.'",
              "shot description: '마트 건물 외벽 위쪽에 달린 CCTV 카메라를 손가락으로 가리킨 채 멈춰 선 남자직원의 측면.'"
            ],
            "visual_inferences": [
              "fixed_element mart_exterior_environment: The exterior surroundings of a local mart during an overcast early morning. The setting features stained concrete walls and a storefront area, enveloped in a cold, damp, muted gray and blue ambient haze.",
              "Unregistered character '남자직원' is represented as an East Asian middle-aged man in a worker uniform."
            ],
            "creative_decisions": [
              "Utilized a low angle medium shot to fit both the character's torso and the high-mounted CCTV in the same frame without breaking vertical consistency.",
              "Placed the figure on the left in profile to create a strong diagonal leading line with his raised arm pointing to the upper right.",
              "Added a hard highlight on the CCTV casing to draw visual attention to it as a key plot element."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "b71ef5127c7a8ed0",
              "owned_hash": "aa09dcc0158b9ca2",
              "camera_direction_hash": "02dd29e0cb76da64",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "CCTV camera",
                  "violating_phrase": "the CCTV camera from the reference mounted high in the upper right frame",
                  "reason": "The prompt explicitly mentions the CCTV camera with a reference cue and framing context, which is an anchor/reference description rather than a directive to create or modify it.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "CCTV camera",
                  "violating_phrase": "the surveillance camera",
                  "reason": "This is a contextual mention of the existing camera in the scene, with no explicit redraw instruction targeting the object.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "CCTV camera",
                  "violating_phrase": "the CCTV camera from the reference",
                  "reason": "The phrase marks the camera as coming from the reference, which indicates anchoring to the existing background object.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 19,
        "dependency_reason": "exact_background mode applied for background consistency with prior shots of the mart exterior",
        "_shot_index": 4,
        "visible_entities": [
          "P03",
          "L09",
          "L10"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 19,
            "shot_index": 4
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A static MS low angle from beside the mart wall holds 남자직원 in side profile, his raised arm forming a diagonal line up to the CCTV camera mounted high above him. The frame keeps his torso on the left and the CCTV on the upper right, making the surveillance angle visually readable against the stained exterior wall.",
            "lighting_mood": "Overcast morning light gives the wall a cold concrete pallor, while the CCTV casing catches a small hard highlight like an unblinking eye.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "P03",
              "L09",
              "L10"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "bg_entrance_exterior_morning_quiet_cctv_visible",
            "reference_usage": "exact_background",
            "owned_objects": [
              "CCTV camera",
              "awning",
              "bollard",
              "curb",
              "sidewalk",
              "sliding door",
              "storefront",
              "street",
              "streetlamp",
              "window"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "mart_exterior_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The exterior surroundings of a local mart during an overcast early morning. The setting features stained concrete walls and a storefront area, enveloped in a cold, damp, muted gray and blue ambient haze.",
                "applies_to_shots": [
                  1,
                  4,
                  6
                ],
                "source_facts": [
                  "Scene heading: S#19 마트 / D",
                  "Scene text: 이른 아침, 아직은 한산한 마트 안 / 마트 밖",
                  "Shot 1: Muted early-morning gray light, damp blue cast, storefront",
                  "Shot 4: Overcast morning light, cold concrete pallor, stained exterior wall",
                  "Shot 6: cold morning haze, mart wall"
                ],
                "visual_inferences": [
                  "Consistent concrete texture and a cool, damp color palette connect the different angles of the mart's exterior structure."
                ],
                "creative_decisions": [
                  "Unified the lighting atmosphere with a tense blue/gray overcast tone to ground all exterior shots in the same early morning temporal continuity."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 19,
                "shot_index": 1,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "background",
                "id": "bg_entrance_exterior_morning_quiet_cctv_visible",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "a9a1cca6152339a1"
      },
      {
        "scene_index": 19,
        "heading": "S#19. 마트 / D",
        "beat_title": "CCTV 위치를 확인하고 혜수에게 상황을 전화로 알리는 수리영",
        "representative_moment": "이른 아침의 한산한 마트 밖, 가로등 아래 서서 스마트폰을 귀에 바짝 대고 통화하는 C01.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Low angle MCU with shallow depth of field, cool desaturated morning haze",
            "t2i_prompt": "Photorealistic cinematic still. A low angle observer view positioned slightly off-center. C01O02 in a dark active jacket, a Korean young woman with long black hair, soft facial features, and golden wave-like eyes, positioned at a three-quarter angle from the left. Her hand presses a smartphone tight against her ear, her eyes fixed on the distant space. Focus on her expression and the phone, her lower chest and shoulders anchoring the lower frame, her head extending toward the upper edge. A weak overhead light mixes with the cold morning air, outlining her cheek and the device in a tense, desaturated glow. [L10: the stained concrete wall and empty street behind her blur into a cold, damp, muted gray and blue ambient haze, a narrow vertical streetlight pole rising close behind one shoulder as a frame-edge surface.]",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "Scene heading: S#19 마트 / D",
              "Scene text: 수리영, 혜수에게 전화 걸어 상황을 설명한다.",
              "Shot 6 description: 마트 밖 가로등 아래서 스마트폰을 귀에 바짝 댄 수리영의 상체",
              "Camera direction: A static MCU low angle sits under the streetlight about four feet from 수리영"
            ],
            "visual_inferences": [
              "fixed_element mart_exterior_environment: cold, damp, muted gray and blue ambient haze, stained concrete walls",
              "Character pose: three-quarter angle from the left, smartphone pressed against ear",
              "Shallow depth of field blurs the background to match close framing rules"
            ],
            "creative_decisions": [
              "Anchored the lower chest and shoulders in the lower frame to geometrically match the low angle composition",
              "Used a tense, desaturated glow to outline her cheek, integrating the weak streetlight with the overcast morning atmosphere",
              "Placed a narrow vertical streetlight pole behind one shoulder to provide a frame-edge surface reference without detailing the omitted background"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "f7be629bd1a45f6b",
              "owned_hash": "aa09dcc0158b9ca2",
              "camera_direction_hash": "310c5b38758111d0",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 19,
        "dependency_reason": "이전 샷의 환경 상태(mart_exterior_environment)를 참조하여 이른 아침 마트 외부의 차갑고 흐린 색감(gray and blue ambient haze)을 동일하게 유지하기 위함",
        "_shot_index": 6,
        "visible_entities": [
          "C01",
          "P03",
          "L09",
          "L10"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 19,
            "shot_index": 6
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A static MCU low angle sits under the streetlight about four feet from 수리영, slightly off-center in front of her upper body as she presses the smartphone tight to her ear. The mart wall and street space shrink into blurred context behind her, while the streetlight pole forms a narrow vertical behind one shoulder.",
            "lighting_mood": "The weak streetlight mixes with cold morning haze, outlining 수리영’s cheek and phone in a tense, desaturated glow.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P03",
              "L09",
              "L10"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "bg_entrance_exterior_morning_quiet_cctv_visible",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "mart_exterior_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The exterior surroundings of a local mart during an overcast early morning. The setting features stained concrete walls and a storefront area, enveloped in a cold, damp, muted gray and blue ambient haze.",
                "applies_to_shots": [
                  1,
                  4,
                  6
                ],
                "source_facts": [
                  "Scene heading: S#19 마트 / D",
                  "Scene text: 이른 아침, 아직은 한산한 마트 안 / 마트 밖",
                  "Shot 1: Muted early-morning gray light, damp blue cast, storefront",
                  "Shot 4: Overcast morning light, cold concrete pallor, stained exterior wall",
                  "Shot 6: cold morning haze, mart wall"
                ],
                "visual_inferences": [
                  "Consistent concrete texture and a cool, damp color palette connect the different angles of the mart's exterior structure."
                ],
                "creative_decisions": [
                  "Unified the lighting atmosphere with a tense blue/gray overcast tone to ground all exterior shots in the same early morning temporal continuity."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 19,
                "shot_index": 1,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "b31b6c1669306714"
      },
      {
        "scene_index": 20,
        "heading": "S#20. 마트 총무과 / 실내",
        "beat_title": "식별 불가능한 CCTV와 수리영의 깊어지는 불안",
        "representative_moment": "보안직원의 어깨 너머로 모니터 불빛에 의지해 고개를 빼고 화면을 주시하는 수리영의 초조한 얼굴.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Over-the-shoulder composition from slightly above the male extra, tight framing on C01's face, cold blue-green monitor glow against underexposed background",
            "t2i_prompt": "Photorealistic cinematic still. An East Asian male figure in a dark security uniform occupies the left foreground with his back to the camera, hunched forward as a foreground block. Beyond his right shoulder near the center-right, C01O02 in a dark activewear jacket, a Korean young woman with long black hair, soft facial structure, and golden wave-like irises, appears at a three-quarter angle from the right. Her neck is craned forward, head and upper chest visible in the frame, eyes fixed on a monitor. A low security monitor sits on a cluttered desk in the middle ground angled away from the camera, emitting a prominent, cold blue-green glow that washes over her face and shoulders. Faint red interface LEDs glow on the console equipment nearby. Flat, low-key institutional fluorescent light from above casts damp-looking shadows across the scene, creating a muted green-gray palette. The focus is on C01O02's strained expression as she leans into the monitor's glow. [L14: underexposed dull gray wall surface in the immediate background].",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "Shot 2: 보안팀 직원 등 뒤에서 모니터 화면을 향해 목을 길게 뺀 수리영의 초조한 얼굴.",
              "남자 직원: back_to_camera, hunched at monitor, eyes→monitor",
              "수리영: three_quarter_right, neck craned forward, eyes→monitor"
            ],
            "visual_inferences": [
              "fixed_element security_monitor_desk: monitor sits low, angled away, cold blue-green glow, faint red LEDs",
              "fixed_element office_lighting_environment: flat low-key fluorescent, muted green-gray palette, damp-looking shadows",
              "East Asian male figure used for unregistered security extra, acting as an OTS foreground block",
              "C01 stable traits applied: Korean young woman, long black hair, soft facial structure, golden wave-like irises"
            ],
            "creative_decisions": [
              "Over-the-shoulder composition from slightly above to anchor the spatial relationship while keeping close framing on C01's face.",
              "Contrasting the cold blue-green monitor glow against the muted green-gray ambient light to highlight the anxiety in C01's golden wave-like irises.",
              "Utilized [L14: underexposed dull gray wall surface] for frame-edge boundaries as required by skipped_close_framing policy."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "9a8f49a8ed2e0b0b",
              "owned_hash": "aedb005298c78a71",
              "camera_direction_hash": "e092460da8db5e6d",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "N/A",
        "_shot_index": 2,
        "visible_entities": [
          "C01",
          "L14"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 20,
            "shot_index": 2
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "From behind and slightly above the male security employee's right shoulder, the camera holds a tight MCU with a subtle high angle, using an OTS foreground block: his dark back occupies the left foreground while 수리영's strained face is framed beyond him near center-right. The monitor sits low in the middle ground, angled away from camera, and 수리영 leans into its glow as if trying to force the darkness on the CCTV to clarify.",
            "lighting_mood": "Low-key fluorescent office light mixes with the cold monitor glow, leaving damp-looking shadows and a muted green-gray palette.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "L14"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "bg_security_office_day_cctv_dark_section",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "security_monitor_desk",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "A security monitor sits low on an office desk. The screen emits a prominent, cold blue-green glow that casts light on the surrounding area. Faint red interface LEDs are glowing on the console equipment nearby.",
                "applies_to_shots": [
                  2,
                  5
                ],
                "source_facts": [
                  "Shot 2: The monitor sits low in the middle ground, angled away from camera.",
                  "Shot 5: the monitor receding behind him... monitor adding a weak blue-green rim and no warmth except faint red interface LEDs."
                ],
                "visual_inferences": [
                  "The desk is a standard, utilitarian office desk used for security monitoring."
                ],
                "creative_decisions": [
                  "Positioning the monitor so it acts as a functional practical light source, casting a cool rim light on the characters nearby."
                ],
                "confidence": "high"
              },
              {
                "element_id": "office_lighting_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior of the administrative office is dim and underexposed. The overhead lighting consists of flat, low-key institutional fluorescent fixtures that create a muted green-gray color palette and cast damp-looking, harsh shadows across the room.",
                "applies_to_shots": [
                  2,
                  5
                ],
                "source_facts": [
                  "Shot 2: Low-key fluorescent office light mixes with the cold monitor glow, leaving damp-looking shadows and a muted green-gray palette.",
                  "Shot 5: Flat institutional fluorescents are underexposed... and no warmth."
                ],
                "visual_inferences": [
                  "The room lacks any warm natural light, contributing to a cold and slightly oppressive back-room atmosphere."
                ],
                "creative_decisions": [
                  "Keeping the overall environment underexposed to emphasize the characters' anxiety and focus the viewer's attention on the glow of the monitor."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "7461f5fb015cb88d"
      },
      {
        "scene_index": 20,
        "heading": "S#20. 마트 총무과 / 실내",
        "beat_title": "영상을 복사하려는 혜수와 동의하는 보안직원",
        "representative_moment": "C02가 영상을 복사하겠다고 요청하자 보안직원이 고개를 돌려 동의하는 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "MCU at eye-level near Hye-su's sightline, over-the-shoulder composition, underexposed flat fluorescent light with cold blue-green monitor rim light",
            "t2i_prompt": "Photorealistic cinematic still. An Asian man in his 30s in a dark worker uniform, occupying the center-left at a three-quarter angle from the left, turning from the console with his eyes fixed on the figure in the right foreground. C02O03 in a dark jacket, a middle-aged Korean woman with black hair and sharp eyes, leaning toward the desk with her eyes fixed on the man, her shoulder and cheek edge remaining soft in the right foreground as an over-the-shoulder frame. [L14: a scuffed gray plastic desk partition edge in the lower foreground, with closed dusty office blinds providing a soft background texture behind the man]. A security monitor sits low on the desk, receding behind the man, its dark paused screen emitting a prominent cold blue-green glow that casts a weak rim light on his side. Faint red interface LEDs glow on the console equipment nearby. The immediate area is dim and underexposed under flat, low-key institutional fluorescents, creating a muted green-gray color palette and damp-looking shadows. Focus on the man's expression as he turns.",
            "outfit_assignments": [
              {
                "character_id": "C02",
                "outlook_id": "O03"
              }
            ],
            "source_facts": [
              "S#20 action: 혜수가 복사를 요청하고 보안직원이 '그러시죠' 라며 수긍한다.",
              "shot description: camera holds a steady MCU at eye-level near 혜수's sightline, angled three-quarters onto the male security employee's face",
              "shot description: 혜수's shoulder and cheek edge remain soft in the right foreground as an OTS frame"
            ],
            "visual_inferences": [
              "fixed_element security_monitor_desk: A security monitor sits low on an office desk, emitting a cold blue-green glow and faint red LEDs",
              "fixed_element office_lighting_environment: dim and underexposed, muted green-gray color palette",
              "The security employee is wearing a dark worker uniform appropriate for his role"
            ],
            "creative_decisions": [
              "Utilized the closed dusty blinds as a soft background texture to conform to the close-framing skip-background constraint",
              "Contrasted the flat green-gray overhead fluorescents with the cold blue-green rim light from the monitor to anchor the subjects"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "ee292b3834f7b6df",
              "owned_hash": "aedb005298c78a71",
              "camera_direction_hash": "361547f6284dbb04",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 20,
        "dependency_reason": "Same room and continuous interaction from Shot 2, carrying over the monitor's glow and the dim office lighting setup.",
        "_shot_index": 5,
        "visible_entities": [
          "C02",
          "L14"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 20,
            "shot_index": 5
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "After the pan completes, the camera holds a steady MCU at eye-level near 혜수's sightline, angled three-quarters onto the male security employee's face as he turns toward her to agree. 혜수's shoulder and cheek edge remain soft in the right foreground as an OTS frame, while the employee occupies the center-left with the monitor now receding behind him.",
            "lighting_mood": "Flat institutional fluorescents are underexposed, with the monitor adding a weak blue-green rim and no warmth except faint red interface LEDs.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C02",
              "L14"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C02",
                "outlook_id": "O03"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "bg_security_office_day_cctv_dark_section",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "security_monitor_desk",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "A security monitor sits low on an office desk. The screen emits a prominent, cold blue-green glow that casts light on the surrounding area. Faint red interface LEDs are glowing on the console equipment nearby.",
                "applies_to_shots": [
                  2,
                  5
                ],
                "source_facts": [
                  "Shot 2: The monitor sits low in the middle ground, angled away from camera.",
                  "Shot 5: the monitor receding behind him... monitor adding a weak blue-green rim and no warmth except faint red interface LEDs."
                ],
                "visual_inferences": [
                  "The desk is a standard, utilitarian office desk used for security monitoring."
                ],
                "creative_decisions": [
                  "Positioning the monitor so it acts as a functional practical light source, casting a cool rim light on the characters nearby."
                ],
                "confidence": "high"
              },
              {
                "element_id": "office_lighting_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior of the administrative office is dim and underexposed. The overhead lighting consists of flat, low-key institutional fluorescent fixtures that create a muted green-gray color palette and cast damp-looking, harsh shadows across the room.",
                "applies_to_shots": [
                  2,
                  5
                ],
                "source_facts": [
                  "Shot 2: Low-key fluorescent office light mixes with the cold monitor glow, leaving damp-looking shadows and a muted green-gray palette.",
                  "Shot 5: Flat institutional fluorescents are underexposed... and no warmth."
                ],
                "visual_inferences": [
                  "The room lacks any warm natural light, contributing to a cold and slightly oppressive back-room atmosphere."
                ],
                "creative_decisions": [
                  "Keeping the overall environment underexposed to emphasize the characters' anxiety and focus the viewer's attention on the glow of the monitor."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 20,
                "shot_index": 2,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C02O03",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "2ec4b118280f139b"
      },
      {
        "scene_index": 21,
        "heading": "S#21. 경찰서 안 / D",
        "beat_title": "충격적인 현장 사진",
        "representative_moment": "책상 위에 놓인 살인사건 현장 사진 속 훼손된 시신의 모습",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "overhead high angle, close-up, cold and desaturated with sharp red accents",
            "t2i_prompt": "Photorealistic cinematic still. Overhead high angle from above, looking down at [L13: a stark, muted gray-beige office desk illuminated by cold, desaturated fluorescent light]. Focus on P09, a glossy forensic photograph lying flat in the center of the frame. The printed image on the photo reveals the damaged neck-and-shoulder area of an unidentified Asian figure, with deep crimson tissue details clearly visible. Cold daylight and weak fluorescent spill wash over the paper surface, while the deep red wound tones inside the photograph contrast sharply against the desaturated palette. The scratched beige laminate of the desk, along with scattered, disorganized case files, paper clips, and a pen, remain slightly out of focus, providing a softly blurred, chaotic boundary along the outer edges of the frame.",
            "outfit_assignments": [],
            "source_facts": [
              "S#21 action: '살인사건 확대 사진들이 어지럽게 널려있다', '목, 어깨, 얼굴등 살점이 떨어져 보는 것만으로도 끔찍하다'",
              "shot description: 훼손된 목과 어깨 부위의 붉은 살점이 선명하게 찍힌 살인사건 현장 사진 클로즈업",
              "Cold police-office daylight and weak fluorescent spill flatten the paper surface",
              "muted gray-beige palette",
              "high angle over the desk, looking down at the forensic photograph among blurred case papers"
            ],
            "visual_inferences": [
              "fixed_element police_office_environment: stark, muted gray-beige color palette with cold, desaturated fluorescent overhead lights",
              "fixed_element cluttered_desk_surface: desk densely covered with scattered, disorganized paperwork, providing a blurred boundary",
              "The physical photo is scaled naturally against the desk, focusing the viewer entirely on the reproduced image"
            ],
            "creative_decisions": [
              "Close framing on the photograph to emphasize the visceral shock of the wound against the sterile, institutional office environment",
              "Overhead angle to align with the observer's POV while keeping the frame confined strictly to the desk surface",
              "Used L13 block to describe the immediate desk environment as the background is skipped via close framing policy"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "918ca9cedb692b2e",
              "owned_hash": "37a8e1b41294114d",
              "camera_direction_hash": "3999ca5aa8c8c961",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 16,
        "dependency_reason": "S16_Shot5 establishes the baseline look of the forensic photos and the muted gray office environment, ensuring consistent lighting and texture for this desk-level close-up.",
        "_shot_index": 5,
        "visible_entities": [
          "P09",
          "P07",
          "L13"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 21,
            "shot_index": 5
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "At slightly above desk height, the camera completes its push into a CU high angle over 혜수's desk, looking down at the forensic photograph among blurred case papers. The damaged neck-and-shoulder area within the printed image is centered and sharp, while the physical photo stays within a natural desk scale and the surrounding clutter softens at the frame edges.",
            "lighting_mood": "Cold police-office daylight and weak fluorescent spill flatten the paper surface, while the red wound tones cut through the muted gray-beige palette.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "P09",
              "P07",
              "L13"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "bg_office_day_empty_case_photos_on_desk",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "police_office_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior of the police office features a stark, muted gray-beige color palette. It is illuminated by cold, desaturated fluorescent overhead lights that cast a weak, sterile glow across the room, creating a dim and chilling atmosphere.",
                "applies_to_shots": [
                  5,
                  6,
                  10
                ],
                "source_facts": [
                  "Cold police-office daylight and weak fluorescent spill flatten the paper surface",
                  "muted gray-beige palette",
                  "Desaturated fluorescent light drains her skin tone",
                  "dim police-office ambience"
                ],
                "visual_inferences": [
                  "The institutional walls and ambient shadows reinforce the emptiness and grim tone of the workspace."
                ],
                "creative_decisions": [
                  "Unified the background lighting and desaturated color grading across all close-up shots to maintain a consistent feeling of dread and isolation."
                ],
                "confidence": "high"
              },
              {
                "element_id": "cluttered_desk_surface",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "The flat surface of an office desk is densely covered with scattered, disorganized paperwork and case files. These cluttered documents remain slightly out of focus, providing a chaotic, textured boundary along the edge of the frame.",
                "applies_to_shots": [
                  5,
                  6
                ],
                "source_facts": [
                  "high angle over the desk, looking down at the forensic photograph among blurred case papers",
                  "locked CU at seated face height just beyond the desk edge"
                ],
                "visual_inferences": [
                  "The paperwork consists of typical generic office folders and investigation sheets, enhancing the realistic mess of an active precinct."
                ],
                "creative_decisions": [
                  "Defined the desk clutter as a continuous blurred element across the desk-level shots to anchor spatial continuity without distracting from the main subject."
                ],
                "confidence": "medium"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 16,
                "shot_index": 5,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "e7e312cb8c0a6004"
      },
      {
        "scene_index": 21,
        "heading": "S#21. 경찰서 안 / D",
        "beat_title": "끔찍한 살인사건 사진 확인",
        "representative_moment": "수리영이 책상 위 사진들을 내려다보며 충격에 얼어붙은 얼굴 클로즈업.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "head-and-shoulder close-up, eye-level angle, shallow depth of field, desaturated cold fluorescent lighting with faint red reflection",
            "t2i_prompt": "Photorealistic cinematic still. C01O02 in a dark activity jacket, a young Korean woman with long black hair, soft facial structure, and golden wave-like irises, in a head-and-shoulder composition. She is positioned center-left at an eye-level angle, facing three-quarter left. Her expression is frozen in shock. The shallow depth of field tightly frames her face, her widened golden wave-like irises looking down toward the bottom frame edge. Cold, desaturated fluorescent light washes across her skin, draining her natural tone. A faint red reflection catches on her lower eyelids and cheeks. The lower foreground shows [L13: a blurred beige surface boundary]. Behind her lies [L13: a stark, heavily out-of-focus gray-beige space with a sterile glow].",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "Shot 6: 사진 위로 시선을 고정한 채 두 눈을 크게 뜨고 얼어붙은 수리영의 얼굴 클로즈업.",
              "Desaturated fluorescent light drains her skin tone, with a faint red reflection from the photographs catching in the lower eye line."
            ],
            "visual_inferences": [
              "fixed_element police_office_environment: stark, muted gray-beige color palette and cold, desaturated fluorescent overhead lights",
              "shallow depth of field isolates the subject, treating the background and desk edge as blurred textures per skipped_close_framing rules"
            ],
            "creative_decisions": [
              "Used head-and-shoulder composition to fulfill the close-up requirement without violating the isolated body-part rendering constraints",
              "Applied a faint red reflection on the lower eyelids to vividly imply the horrific nature of the unseen photos below the frame without showing them"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "31eb8de1ebc59234",
              "owned_hash": "37a8e1b41294114d",
              "camera_direction_hash": "8cce80d536af8b49",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 21,
        "dependency_reason": "같은 공간(경찰서 사무실)에서의 직전 샷(Shot 5)과 연속된 상황",
        "_shot_index": 6,
        "visible_entities": [
          "C01",
          "P09",
          "P07",
          "L13"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 21,
            "shot_index": 6
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "The upward tilt resolves into a locked CU at seated face height just beyond the desk edge, using shallow DOF so the photographs remain only implied below frame. 수리영's face sits center-left in an eye-level composition, her widened eyes frozen downward while the empty police office blurs behind her.",
            "lighting_mood": "Desaturated fluorescent light drains her skin tone, with a faint red reflection from the photographs catching in the lower eye line.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P09",
              "P07",
              "L13"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "bg_office_day_empty_case_photos_on_desk",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "police_office_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior of the police office features a stark, muted gray-beige color palette. It is illuminated by cold, desaturated fluorescent overhead lights that cast a weak, sterile glow across the room, creating a dim and chilling atmosphere.",
                "applies_to_shots": [
                  5,
                  6,
                  10
                ],
                "source_facts": [
                  "Cold police-office daylight and weak fluorescent spill flatten the paper surface",
                  "muted gray-beige palette",
                  "Desaturated fluorescent light drains her skin tone",
                  "dim police-office ambience"
                ],
                "visual_inferences": [
                  "The institutional walls and ambient shadows reinforce the emptiness and grim tone of the workspace."
                ],
                "creative_decisions": [
                  "Unified the background lighting and desaturated color grading across all close-up shots to maintain a consistent feeling of dread and isolation."
                ],
                "confidence": "high"
              },
              {
                "element_id": "cluttered_desk_surface",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "The flat surface of an office desk is densely covered with scattered, disorganized paperwork and case files. These cluttered documents remain slightly out of focus, providing a chaotic, textured boundary along the edge of the frame.",
                "applies_to_shots": [
                  5,
                  6
                ],
                "source_facts": [
                  "high angle over the desk, looking down at the forensic photograph among blurred case papers",
                  "locked CU at seated face height just beyond the desk edge"
                ],
                "visual_inferences": [
                  "The paperwork consists of typical generic office folders and investigation sheets, enhancing the realistic mess of an active precinct."
                ],
                "creative_decisions": [
                  "Defined the desk clutter as a continuous blurred element across the desk-level shots to anchor spatial continuity without distracting from the main subject."
                ],
                "confidence": "medium"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 21,
                "shot_index": 5,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "98b5d3ebbdcb8640"
      },
      {
        "scene_index": 21,
        "heading": "S#21. 경찰서 안 / D",
        "beat_title": "엄마의 마지막 위치 확인",
        "representative_moment": "경찰서 사무실 책상 위, 모니터 화면에 바다 위 특정 좌표와 붉은 마커가 선명하게 떠 있는 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "close framing on the monitor, eye-level flat-on view, cold blue-black screen glow",
            "t2i_prompt": "Photorealistic cinematic still. Focus on a crisp deep red location marker and numeric coordinates on a dark sea map displayed on a computer monitor. Nearly flat-on eye-level view, the screen occupying the central frame. The monitor's black bezel, a slightly dusty keyboard facing the camera below, and partially open case folders stacked diagonally provide scale around the screen. The monitor casts a cold blue-black glow, punctured by the bright red marker. [L13: stark, muted gray-beige desk surface illuminated by weak, desaturated fluorescent spill at the frame edge].",
            "outfit_assignments": [],
            "source_facts": [
              "shot description: computer monitor displaying the dark sea map window, red location marker, coordinates, timestamp, and offline status",
              "S21 action: '모니터에서 뭔가 뜬다... 엄마 폰 위치추적 결과다. 해상 지도위에 좌표.'"
            ],
            "visual_inferences": [
              "fixed_element police_office_environment: stark, muted gray-beige color palette and cold, desaturated fluorescent lights",
              "keyboard and messy case folders provide immediate foreground scale for the desk workspace"
            ],
            "creative_decisions": [
              "Close framing strictly on the monitor and immediate desk clutter, omitting the background reference to maintain focus on the critical information reveal.",
              "Used a cold blue-black glow from the screen to contrast sharply with the ominous deep red GPS marker."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "d73d3b9145297b52",
              "owned_hash": "6c3108d54d0cd477",
              "camera_direction_hash": "bba202f2a5fe74df",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 21,
        "dependency_reason": "같은 공간(경찰서 사무실) 내부이므로 기존 설정된 조명과 데스크탑 환경의 톤을 유지하되, 클로즈업 구도에 맞춰 배경 묘사를 즉각적인 주변부(책상 표면)로 제한함.",
        "_shot_index": 10,
        "visible_entities": [
          "P09",
          "P07",
          "L13"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 21,
            "shot_index": 10
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Beside 김형사's empty workstation, the lateral track ends in a nearly flat-on CU eye-level view of the computer monitor, framed so the dark sea map window, red location marker, coordinates, timestamp, and offline status are crisp and readable. The monitor screen occupies just under the central two-fifths of the frame, with the black bezel, keyboard, and desk clutter providing scale around it.",
            "lighting_mood": "The monitor casts a cold blue-black glow, punctured by the ominous deep red GPS marker against the dim police-office ambience.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "P09",
              "P07",
              "L13"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "bg_office_day_empty_location_result_on_monitor",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "police_office_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior of the police office features a stark, muted gray-beige color palette. It is illuminated by cold, desaturated fluorescent overhead lights that cast a weak, sterile glow across the room, creating a dim and chilling atmosphere.",
                "applies_to_shots": [
                  5,
                  6,
                  10
                ],
                "source_facts": [
                  "Cold police-office daylight and weak fluorescent spill flatten the paper surface",
                  "muted gray-beige palette",
                  "Desaturated fluorescent light drains her skin tone",
                  "dim police-office ambience"
                ],
                "visual_inferences": [
                  "The institutional walls and ambient shadows reinforce the emptiness and grim tone of the workspace."
                ],
                "creative_decisions": [
                  "Unified the background lighting and desaturated color grading across all close-up shots to maintain a consistent feeling of dread and isolation."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 21,
                "shot_index": 5,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "2927513721e2aff6"
      },
      {
        "scene_index": 22,
        "heading": "S#22. 경찰서 밖 / D",
        "beat_title": "자전거를 타고 급히 출발하는 수리영",
        "representative_moment": "C01이 자전거 핸들을 꽉 쥐고 상체를 바짝 숙인 채 결연한 표정으로 앞으로 달려 나간다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "tight MCU from a slight low angle, tracking backward",
            "t2i_prompt": "Photorealistic cinematic still. A tight MCU composition from a slight low angle, positioned just beyond the handlebars. C01O02 in a dark activity jacket, an East Asian young woman showing the stable traits — long black hair, soft facial shape, and golden wave-like irises — facing the camera, neck and upper chest in frame. She is leaning intensely forward over the bicycle in mid-ride, eyes fixed on the road ahead. Her hands tightly grip the slightly worn rubber handlebars cutting across the lower foreground, with the bicycle's front frame visible below her arms. Her hair and jacket collar flow backward in the direction opposite to motion. Focus on the tense forward momentum of her upper body. Overcast daylight is underexposed into steel gray, creating a hard urban dampness with no warmth. [L15: blurred gray exterior glass doors and dark damp pavement receding into the soft background].",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#22 action: 주차된 자전거를 타고 이내 달리는 수리영",
              "shot description: 꽉 쥔 자전거 핸들 위로 상체를 바짝 숙인 수리영의 결연한 얼굴 정면 구도"
            ],
            "visual_inferences": [
              "mid-ride freeze state with hair flowing backward to convey fast forward motion",
              "bicycle frame visible below the arms to anchor the large prop"
            ],
            "creative_decisions": [
              "tight MCU with a slight low angle to emphasize determination and tension",
              "steel gray overcast lighting to enhance the damp, uninviting urban atmosphere"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "18daa8286fd193a8",
              "owned_hash": "1893da4a0af6fede",
              "camera_direction_hash": "d8f60875c9ab8c3d",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "N/A",
        "_shot_index": 2,
        "visible_entities": [
          "C01",
          "L15"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 22,
            "shot_index": 2
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Directly in front of the moving bicycle, the camera tracks backward in a tight MCU with a slight low angle, positioned just beyond the handlebars. The clenched grips cut across the lower foreground, while 수리영's determined face dominates the center as the police station entrance recedes behind her.",
            "lighting_mood": "Overcast daylight is underexposed into steel gray, with hard urban dampness and no warmth in the palette.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "L15"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "bg_exterior_entrance_day_bicycle_departure",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [],
            "previous_shot_refs": [],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "086b5dd2e4634217"
      },
      {
        "scene_index": 23,
        "heading": "S#23. 어느 PC방 / 실내",
        "beat_title": "좌표 확인",
        "representative_moment": "어두운 PC방 안, 모니터 화면의 밝은 지도 이미지를 뚫어지게 응시하는 수리영의 옆얼굴.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Close-up right profile at seated eye level, high contrast cold cyan-white monitor lighting against deep black background",
            "t2i_prompt": "Photorealistic cinematic still. Head-and-shoulder composition at seated eye level. C01O02 in a dark casual jacket, a young Korean woman with long black hair, hunched toward the monitor in right profile. Her eyes, showing the golden wave-like irises carried from stable_traits, are fixed intently on the bright screen. The glowing edge of the PC monitor and a cyan-white coordinate map occupy the right side of the frame. The cold cyan-white map glow washes her side profile and the dark, scratched plastic partition behind her. The surrounding space falls into deep blackness with muted shadows, illuminated only by tiny red UI points on the monitor surface. Focus on her expression reflecting the screen's light.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#23 action: '모니터 화면에 앉아있는 수리영, 좌표를 찍고 위치를 확인하고'",
              "shot description: observer beside her, looking at her right profile as she stares at the bright map"
            ],
            "visual_inferences": [
              "head-and-shoulder framing to preserve body connection per close-up face rules",
              "high contrast cold lighting from the monitor against a dark environment",
              "golden wave-like eye trait applied as a visible iris detail"
            ],
            "creative_decisions": [
              "placed the monitor edge on the right to anchor the light source naturally within the frame",
              "used cold cyan-white light washing the face rather than carving it, ensuring realistic light reflection",
              "added dark, scratched plastic partition texturing to emphasize the grounded, gritty thriller mood"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "677df8be6f8198c6",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "306967b340f68109",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "No direct visual dependency on a prior shot required for this specific close-up.",
        "_shot_index": 1,
        "visible_entities": [
          "C01",
          "P06",
          "P05",
          "L16"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 23,
            "shot_index": 1
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "From close beside 수리영 at seated eye level, the camera holds a CU profile shot about two feet from her face, with the monitor edge and glowing map image sharing the right side of frame. The PC room around her falls into darkness, while her side profile and the bright coordinate map form the only readable plane.",
            "lighting_mood": "The monitor's cold cyan-white map glow carves her profile out of a nearly black PC room, with muted shadows and no decorative color except tiny red UI points.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P06",
              "P05",
              "L16"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": null,
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [],
            "previous_shot_refs": [],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "6a8a6c7ceb056cdc"
      },
      {
        "scene_index": 24,
        "heading": "S#24. 해안가 / 소나기",
        "beat_title": "악천후 속의 질주",
        "representative_moment": "굵은 빗방울을 맞으며 굳센 표정으로 자전거를 타고 내달리는 C01의 클로즈업.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Tight CU at eye level, unsteady push-in from front-left, storm-dark desaturated colors with intermittent silver lightning.",
            "t2i_prompt": "Photorealistic cinematic still. C01O02 in a dark waterproof jacket, a young Korean woman, at three-quarter angle from the right, leaning forward into the heavy rain. Her long black hair flows back in the direction opposite to her forward motion. She has soft facial features and golden wave-like eyes fixed firmly on the road ahead. Tight framing on her face, with her upper chest and shoulders visible in the frame. Focus on her determined expression and the heavy raindrops beading on her cheeks. Storm-dark daylight is slashed by intermittent cool lightning, reflecting a silver-gray cast on her wet skin. Raindrops streak across the camera lens. In the lower-left foreground, the wet left grip of the bicycle handlebar is angled toward the camera, with a small slice of the bicycle frame visible below it. [L06: heavy dark storm clouds and dense wind-driven rainfall forming a cold, desaturated environmental veil right behind her].",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "scene text: '천둥 번개와 함께 잔뜩 낀 먹구름이 세찬 비를 뿌린다. 자전거를 타고 빗속을 달리는 수리영. 오로지 엄마를 찾아야한다는 생각 뿐.'",
              "shot description: '굵은 빗방울이 사정없이 뺨에 맺힌 수리영의 굳센 표정 클로즈업.'",
              "S24 action: riding bicycle through heavy rain"
            ],
            "visual_inferences": [
              "fixed_element: young Korean woman, long black hair, soft facial features, golden wave-like eyes from stable_traits",
              "mid-action freeze: hair flowing back opposite to motion",
              "wet skin and clothing due to heavy rain"
            ],
            "creative_decisions": [
              "tight eye-level framing pushing in from front-left to emphasize determination against the harsh environment",
              "cool, desaturated palette with silver-grey lightning highlights to enhance the storm's severity",
              "bicycle handlebar in lower-left foreground to anchor the motion without pulling focus from the face",
              "raindrops on the lens to heighten the immersive, tempestuous atmosphere"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "95e66deea7f53bc3",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "4888f9b2f09b1360",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 9,
        "dependency_reason": "동일한 해안가 배경(S9_Shot2)을 참조하지만 close framing 정책에 따라 배경 합성에서는 제외되고 인물의 얼굴과 주변 환경의 대기감에만 집중함",
        "_shot_index": 2,
        "visible_entities": [
          "C01",
          "L06"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 24,
            "shot_index": 2
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Matched to the bicycle's speed, the camera pushes in from 수리영's front-left side for a tight CU at eye level, slightly unsteady with the storm. Raindrops streak across the lens and bead on her cheeks, while her face stays center frame and the wet handlebar edge flickers through the lower-left foreground.",
            "lighting_mood": "Storm-dark daylight is slashed by intermittent lightning, turning her wet skin silver-gray while the scene remains heavy, cold, and desaturated.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "L06"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": null,
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [],
            "previous_shot_refs": [
              {
                "scene_index": 9,
                "shot_index": 2,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "6a25872074dd512d"
      },
      {
        "scene_index": 25,
        "heading": "S#25. 옥탑방 안 / 해질 무렵",
        "beat_title": "엉망이 된 옥탑방 발견",
        "representative_moment": "수리영(C01)이 누군가 다녀간 흔적으로 아수라장이 된 거실 한가운데 서서 충격에 굳어 있다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "A medium shot from chest height with a slight low angle, low-key blue-gray dusk lighting",
            "t2i_prompt": "Photorealistic cinematic still. A medium shot from chest height with a slight low angle. C01O02 in a casual dark jacket, a Korean young woman with long black hair, occupies the center of the frame from waist to head, facing the camera, standing rigid mid-breath, her wide golden rippling irises fixed on the door from the reference. Sharp focus on C01O02. A low-key damp blue-gray dusk light washes over her face and jacket. In the background, the wrecked living room is legible: overturned wooden dining chairs scattered on the floor, and dark muddy footprints trailing toward the interior. The open window from the reference lets in a weak light, the damp curtains from the reference hanging disturbed. Deep interior shadows drape the corners, creating an eerie, airless atmosphere.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#25 action: '집안에서 뭔가를 찾기 위해 가재도구를 온통 헤집은 것으로 보인다. 거실에 난 창은 열려있고, 식탁 의자들을 쓰러져있다.'",
              "Shot 3 description: 엉망이 된 거실 한가운데 서서 두 눈을 크게 뜬 채 굳어버린 수리영의 상체."
            ],
            "visual_inferences": [
              "fixed_element ransacked_living_room: overturned wooden dining chairs scattered, muddy footprints trailing.",
              "fixed_element dusk_blue_gray_lighting: low-key damp blue-gray dusk light, deep interior shadows.",
              "C01's golden rippling irises (stable trait) are visible due to her wide-eyed expression."
            ],
            "creative_decisions": [
              "Framing C01O02 from waist up at chest height with a slight low angle to emphasize her rigid shock.",
              "Explicitly anchoring owned objects ('door from the reference', 'open window from the reference', 'curtains from the reference') to blend seamlessly with the added overturned chairs and footprints."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "18b955cfb66a1877",
              "owned_hash": "18fa02b2ac47d2f4",
              "camera_direction_hash": "4ab1ebdba1b1488d",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "door",
                  "violating_phrase": "the door from the reference",
                  "reason": "The phrase explicitly references an existing door from the reference rather than instructing a new door to be created or modified.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "window",
                  "violating_phrase": "The open window from the reference",
                  "reason": "The window is explicitly anchored to the reference, which indicates the existing background object should be reused, not redrawn.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "curtain",
                  "violating_phrase": "the damp curtains from the reference",
                  "reason": "The curtains are mentioned with an explicit reference anchor, so this is a description of an existing object rather than a redraw instruction.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 14,
        "dependency_reason": "수리영의 옥탑방 거실 공간 연속성 유지 (exact_background mode 적용, 파손 및 변동 상태 추가)",
        "_shot_index": 3,
        "visible_entities": [
          "C01",
          "P01",
          "P04",
          "P05",
          "L05"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 25,
            "shot_index": 3
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "From chest height in the ruined living room, the camera creeps forward into an MS with a slight low angle, facing 수리영 directly while keeping the wrecked room legible behind her. She occupies the center of frame from waist to head, rigid and wide-eyed, with overturned chairs, open window, and muddy traces spread around the edges like evidence closing in.",
            "lighting_mood": "Dusk light leaks through the open window in a damp blue-gray wash, while deep interior shadows make the ruined room feel violated and airless.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P01",
              "P04",
              "P05",
              "L05"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "cb_living_dusk_ransacked_window_open",
            "reference_usage": "exact_background",
            "owned_objects": [
              "TV",
              "TV console",
              "air conditioner",
              "cooktop",
              "curtain",
              "debris",
              "document",
              "door",
              "floor cushion",
              "kitchen counter",
              "mug",
              "refrigerator",
              "shoe rack",
              "sink",
              "sofa",
              "table",
              "window"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "ransacked_living_room",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The living room is completely ransacked and chaotic. Wooden dining chairs are overturned and scattered on the floor, while dark, muddy footprints trail across the ground. A main window remains wide open, exposing the interior to the outside.",
                "applies_to_shots": [
                  3,
                  8
                ],
                "source_facts": [
                  "S#25 action: '누군가 다녀간 흔적으로 보이는 발자국들. 집안에서 뭔가를 찾기 위해 가재도구를 온통 헤집은 것으로 보인다. 거실에 난 창은 열려있고, 식탁 의자들을 쓰러져있다.'",
                  "Shot 3: 'overturned chairs, open window, and muddy traces spread around'",
                  "Shot 8: 'ransacked rooftop room recedes behind her'"
                ],
                "visual_inferences": [
                  "The dining chairs are wooden, typical for a modest rooftop apartment.",
                  "The muddy footprints are distinctly dark and visible against the floor surface."
                ],
                "creative_decisions": [
                  "Arranging the overturned chairs and mud trails irregularly to emphasize a violent, hasty intrusion rather than a simple mess."
                ],
                "confidence": "high"
              },
              {
                "element_id": "dusk_blue_gray_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A low-key, damp blue-gray dusk light fills the interior space. Weak natural light leaks slowly through the open window and doorways, leaving the deeper corners of the rooms draped in murky, heavy shadows.",
                "applies_to_shots": [
                  3,
                  4,
                  8
                ],
                "source_facts": [
                  "Scene heading: '해질 무렵'",
                  "Shot 3: 'Dusk light leaks through the open window in a damp blue-gray wash'",
                  "Shot 4: 'Fading dusk and weak spill from the living room leave the bedroom murky'",
                  "Shot 8: 'Low-key dusk light... muting the room in gray-blue dampness'"
                ],
                "visual_inferences": [
                  "The color temperature is noticeably cool, dominated by blue and gray tones to reflect the fading daylight."
                ],
                "creative_decisions": [
                  "Using deep interior shadows combined with the blue-gray wash to create an airless, eerie atmosphere that unifies the sequence visually."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 14,
                "shot_index": 5,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              },
              {
                "kind": "background",
                "id": "cb_living_dusk_ransacked_window_open",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "df9308dd768ba234"
      },
      {
        "scene_index": 25,
        "heading": "S#25. 옥탑방 안 / 해질 무렵",
        "beat_title": "벽에 나타난 붉은 원과 옛 사진 발견",
        "representative_moment": "수리영 방의 열린 문 너머로, 벽에 칠해진 거대한 붉은 원과 그 가운데 꽂힌 낡은 사진이 보인다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Frontal WS at eye-level from the bedroom doorway, damp blue-gray dusk lighting with isolated red saturation on the wall mark.",
            "t2i_prompt": "Photorealistic cinematic still. Observer POV from the bedroom doorway, holding a frontal wide shot at eye-level. C01O02 in a dark activity outfit, a young Korean woman with long black hair, appears partially visible at the dark frame edge. The open door from the reference creates a dark through-object border, framing the murky bedroom interior. On the far wall from the reference, a large dry-blood-like red wall mark from the reference sits centered. The small but unmistakable photograph from the reference is pinned securely at the middle of the red circle. The bed from the reference rests unmade beneath the marked wall. A low-key, damp blue-gray dusk light fills the space, leaking weakly through the open window from the reference, leaving the deeper corners draped in heavy shadows. The dried deep-red mark serves as the only saturated color in the cool gray-blue room.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "Scene heading: '해질 무렵'",
              "Shot 4 action: '수리영 방 벽면에 피가 마른 듯한 붉은 페인트 질감으로 크게 그려진 원형 표식과 그 중앙에 꽂힌 사진.'",
              "DP directive: observer at doorway, frontal WS at eye-level"
            ],
            "visual_inferences": [
              "fixed_element dusk_blue_gray_lighting: low-key, damp blue-gray dusk light, murky heavy shadows",
              "C01O02 is partially visible at the doorway edge, maintaining the observer POV without blocking the view of the wall"
            ],
            "creative_decisions": [
              "Used explicit 'from the reference' anchors for the door, wall, wall mark, photograph, bed, and window to enforce background binding rules.",
              "Saturated the red circular mark against the muted cool background to draw immediate focus, fulfilling the DP's color and lighting directives."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "f140e101973592d8",
              "owned_hash": "b0bfa88101c56b24",
              "camera_direction_hash": "f9e53bd73baf9a3a",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "door",
                  "violating_phrase": "The open door from the reference",
                  "reason": "The door is explicitly referenced as existing in the reference, which anchors it rather than instructing a new addition or redraw.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "wall",
                  "violating_phrase": "On the far wall from the reference",
                  "reason": "The wall is mentioned as part of the referenced background, not as a newly created object.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "wall mark",
                  "violating_phrase": "a large dry-blood-like red wall mark from the reference",
                  "reason": "The wall mark is explicitly tied to the reference background, so this is an anchor reference, not a redraw instruction.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "photograph",
                  "violating_phrase": "The small but unmistakable photograph from the reference",
                  "reason": "The photograph is described as already present in the reference, which makes it an anchored existing object.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "bed",
                  "violating_phrase": "The bed from the reference",
                  "reason": "The bed is explicitly identified as coming from the reference background, so it is not a request to create a new bed.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "window",
                  "violating_phrase": "the open window from the reference",
                  "reason": "The window is clearly marked as an existing referenced object, which is an anchor reference.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 25,
        "dependency_reason": "Shares the same location and base lighting setup as S25_Shot3, focusing on the specific wall mark and maintaining the exact background.",
        "_shot_index": 4,
        "visible_entities": [
          "C01",
          "P01",
          "P04",
          "P05",
          "L05"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 25,
            "shot_index": 4
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "At shoulder height from the bedroom doorway, the camera finishes the track into a frontal WS at eye-level, holding still so the red circular mark reads graphically on the far wall. The open doorframe creates a dark through-object border, the red circle sits centered on the wall, and the pinned photograph remains small but unmistakable at its middle.",
            "lighting_mood": "Fading dusk and weak spill from the living room leave the bedroom murky, with the dried deep-red circle becoming the only saturated color in the frame.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P01",
              "P04",
              "P05",
              "L05"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "cb_daughter_room_dusk_red_mark_photo_pinned",
            "reference_usage": "exact_background",
            "owned_objects": [
              "backpack",
              "bed",
              "curtain",
              "door",
              "floor",
              "footprint",
              "photograph",
              "pillow",
              "toy",
              "wall",
              "wall mark",
              "window"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "dusk_blue_gray_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A low-key, damp blue-gray dusk light fills the interior space. Weak natural light leaks slowly through the open window and doorways, leaving the deeper corners of the rooms draped in murky, heavy shadows.",
                "applies_to_shots": [
                  3,
                  4,
                  8
                ],
                "source_facts": [
                  "Scene heading: '해질 무렵'",
                  "Shot 3: 'Dusk light leaks through the open window in a damp blue-gray wash'",
                  "Shot 4: 'Fading dusk and weak spill from the living room leave the bedroom murky'",
                  "Shot 8: 'Low-key dusk light... muting the room in gray-blue dampness'"
                ],
                "visual_inferences": [
                  "The color temperature is noticeably cool, dominated by blue and gray tones to reflect the fading daylight."
                ],
                "creative_decisions": [
                  "Using deep interior shadows combined with the blue-gray wash to create an airless, eerie atmosphere that unifies the sequence visually."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 25,
                "shot_index": 3,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              },
              {
                "kind": "background",
                "id": "cb_daughter_room_dusk_red_mark_photo_pinned",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "2af15320fb7c54e0"
      },
      {
        "scene_index": 25,
        "heading": "S#25. 옥탑방 안 / 해질 무렵",
        "beat_title": "집을 떠나는 수리영",
        "representative_moment": "C01, clutching P04 and wearing P05, half-steps out of the dark doorway of her ransacked room into the corridor.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Medium close-up (MCU), eye-level threshold composition with shallow focus background",
            "t2i_prompt": "Photorealistic cinematic still. Medium close-up, eye-level composition from just outside the threshold, a dark doorframe forming a vertical line along the right third of the frame. C01O02 in a dark jacket, a Korean young woman, long black hair, soft facial features, golden wave-like eyes, occupies the left-center foreground waist-up in right profile, mid-step through the open doorway. She wears P05, a backpack with a small doll attached, the shoulder strap and upper bag visible over her shoulder. Her lowered hand grips P04, a creased photograph, its edge visible near her waist. Eyes fixed on the outside corridor. Low-key damp blue-gray dusk light washes her side profile and the wooden doorframe, while a faint deep-red glow reflects on the surface of the photograph. Behind her, [L05: shallow focus background showing blurred overturned wooden dining chairs and muddy traces draped in murky dusk shadows]. Focus on her determined side profile.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#25 action: '마치 이끌리듯 백팩을 메고 밖으로 나간다. 사진을 손에 꼭 쥔 채.'",
              "Shot 8 description: '백팩을 멘 채 현관문 밖으로 몸을 반쯤 내민 수리영의 결연한 옆얼굴.'"
            ],
            "visual_inferences": [
              "fixed_element: ransacked_living_room (overturned chairs, muddy traces in the background)",
              "fixed_element: dusk_blue_gray_lighting (damp blue-gray wash, heavy shadows)",
              "C01's visually-distinguishing traits carried from entity_canon stable_traits (long black hair, golden wave-like eyes)"
            ],
            "creative_decisions": [
              "Framing the door edge along the right third to create a strong visual threshold boundary.",
              "Using shallow focus on L05 to comply with the close framing policy while acknowledging the chaotic background evidence.",
              "Positioning the character mid-step to preserve the motion direction of her departure while adhering to the single freeze-frame rule."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "5e248e841c6e6436",
              "owned_hash": "18fa02b2ac47d2f4",
              "camera_direction_hash": "676df160be93d17c",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 25,
        "dependency_reason": "maintaining exact background state and lighting established in prior shots of the same sequence (S25_Shot4)",
        "_shot_index": 8,
        "visible_entities": [
          "C01",
          "P01",
          "P04",
          "P05",
          "L05"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 25,
            "shot_index": 8
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "From just outside the threshold, the camera holds a medium close-up (MCU), eye-level, through-object framing with the doorframe cutting a dark vertical line along the right third. 수리영 occupies the left-center foreground, half-stepping out with her backpack visible, her determined side profile caught against the open doorway while the ransacked rooftop room recedes behind her in shallow focus.",
            "lighting_mood": "Low-key dusk light leaks from the doorway and open window, muting the room in gray-blue dampness with a faint deep-red stain from the photo as the only symbolic accent.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P01",
              "P04",
              "P05",
              "L05"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "cb_living_dusk_ransacked_window_open",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "ransacked_living_room",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The living room is completely ransacked and chaotic. Wooden dining chairs are overturned and scattered on the floor, while dark, muddy footprints trail across the ground. A main window remains wide open, exposing the interior to the outside.",
                "applies_to_shots": [
                  3,
                  8
                ],
                "source_facts": [
                  "S#25 action: '누군가 다녀간 흔적으로 보이는 발자국들. 집안에서 뭔가를 찾기 위해 가재도구를 온통 헤집은 것으로 보인다. 거실에 난 창은 열려있고, 식탁 의자들을 쓰러져있다.'",
                  "Shot 3: 'overturned chairs, open window, and muddy traces spread around'",
                  "Shot 8: 'ransacked rooftop room recedes behind her'"
                ],
                "visual_inferences": [
                  "The dining chairs are wooden, typical for a modest rooftop apartment.",
                  "The muddy footprints are distinctly dark and visible against the floor surface."
                ],
                "creative_decisions": [
                  "Arranging the overturned chairs and mud trails irregularly to emphasize a violent, hasty intrusion rather than a simple mess."
                ],
                "confidence": "high"
              },
              {
                "element_id": "dusk_blue_gray_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A low-key, damp blue-gray dusk light fills the interior space. Weak natural light leaks slowly through the open window and doorways, leaving the deeper corners of the rooms draped in murky, heavy shadows.",
                "applies_to_shots": [
                  3,
                  4,
                  8
                ],
                "source_facts": [
                  "Scene heading: '해질 무렵'",
                  "Shot 3: 'Dusk light leaks through the open window in a damp blue-gray wash'",
                  "Shot 4: 'Fading dusk and weak spill from the living room leave the bedroom murky'",
                  "Shot 8: 'Low-key dusk light... muting the room in gray-blue dampness'"
                ],
                "visual_inferences": [
                  "The color temperature is noticeably cool, dominated by blue and gray tones to reflect the fading daylight."
                ],
                "creative_decisions": [
                  "Using deep interior shadows combined with the blue-gray wash to create an airless, eerie atmosphere that unifies the sequence visually."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 25,
                "shot_index": 4,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "a35188f2d2f0f0d2"
      },
      {
        "scene_index": 26,
        "heading": "S#26. 포구 / 해질 무렵",
        "beat_title": "비가 그친 적막한 포구의 전경",
        "representative_moment": "소나기가 그치고 짙은 안개가 텅 빈 포구를 완전히 집어삼킨 고요한 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "extreme wide establishing shot from a high angle, cold desaturated blue-gray palette",
            "t2i_prompt": "Photorealistic cinematic still. Extreme wide establishing shot from a high angle looking down at an empty harbor. Focus on the vast, heavy pale mist swallowing the landscape. [L17: A deserted concrete pier sits low across the bottom quarter of the frame, the paved surface and scattered natural rocks soaked from recent heavy rain.] Puddles on the wet ground reflect a faint, muted dusk glow. Dense fog completely fills the middle and upper frame, obscuring the seawater and reducing distant mooring posts and boat hulls to barely visible, dissolving pale shapes. The overall palette is cold and desaturated, washed in a quiet, oppressive blue-gray ambient light.",
            "outfit_assignments": [],
            "source_facts": [
              "소나기가 한차례 퍼 붓는가 싶더니, 비는 그치고 안개가 자욱하다. 텅 빈 포구.",
              "짙은 안개가 깔려 바닷물이 거의 보이지 않는 적막한 포구의 넓은 풍경.",
              "The wet pier line sits low across the bottom quarter, while fog swallows nearly the entire middle and upper frame"
            ],
            "visual_inferences": [
              "fixed_element wet_foggy_harbor_environment: heavy pale mist, cool blue-gray hue, water-soaked pier reflecting faint ambient light",
              "High angle perspective places the wet pier at the bottom, leaving the negative space above completely dominated by the fog"
            ],
            "creative_decisions": [
              "Focus on the oppressive, dense fog to establish a sense of isolation and chilling mystery for the thriller genre",
              "Desaturated blue-gray color grading to emphasize the damp, post-rain atmosphere"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "ccd3ed78a6fd5c25",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "2b0a7c9fe9d6a78d",
              "validator": "scene_detail_owned_objects.v1",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "",
        "_shot_index": 1,
        "visible_entities": [
          "P03",
          "L17"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 26,
            "shot_index": 1
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A static extreme wide / establishing shot (EWS), high angle, long take implied from the empty harbor entrance, about 25 meters from the pier. The wet pier line sits low across the bottom quarter, while fog swallows nearly the entire middle and upper frame so the seawater and boats are only suggested as pale, dissolving shapes.",
            "lighting_mood": "The palette is cold blue-gray and desaturated after rain, with the last dusk glow flattened by mist into a quiet, oppressive haze.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "P03",
              "L17"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "not_applicable",
            "bg_id": null,
            "reference_usage": "none",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "no chain_bg is bound to this shot — describe the location freely without referencing a background image",
              "do not use 'from the reference' / 'the existing room' / 'preserving the room' style phrasing — there is no reference image to refer to"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "wet_foggy_harbor_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A heavy, pale mist obscures the background, severely limiting visibility and softening the dusk light into a pervasive cool blue-gray hue. The ground, including the paved pier and natural rocks, is soaked with water, reflecting faint ambient light from the recent heavy rain.",
                "applies_to_shots": [
                  1,
                  3,
                  6
                ],
                "source_facts": [
                  "소나기가 한차례 퍼 붓는가 싶더니, 비는 그치고 안개가 자욱하다.",
                  "The wet pier line sits low across the bottom quarter, while fog swallows nearly the entire middle and upper frame",
                  "wet harbor rock and pale fog blur around the phone edges",
                  "Muted blue-gray fog light"
                ],
                "visual_inferences": [
                  "Puddles and lingering moisture on the harbor surfaces subtly reflect the muted dusk lighting, highlighting the dampness."
                ],
                "creative_decisions": [
                  "Unified desaturated blue-gray color grading across all shots to emphasize isolation, anxiety, and the chilling post-rain atmosphere."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [],
            "forbidden_refs": [],
            "readiness_policy": "not_applicable",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "180ca847f3fc5c83"
      },
      {
        "scene_index": 26,
        "heading": "S#26. 포구 / 해질 무렵",
        "beat_title": "부재중 전화를 확인하는 수리영",
        "representative_moment": "C01의 손이 여러 통의 부재중 전화 알림이 떠 있는 P03을 쥐고 있다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "High angle close-up insert, shallow depth of field, cool fog-diffused dusk lighting.",
            "t2i_prompt": "Photorealistic cinematic still. Close on the hands of a young Korean woman in a dark jacket, cradling P03. High angle insert from just above the hands, using shallow depth of field to focus sharply on P03 in the central lower third. The screen of P03 faces the camera clearly, displaying active missed-call notifications, and casting a weak, anxious bluish glow onto the fingers and palms. Cool fog-diffused dusk light wraps the back of the hands. Beneath the hands, [L17: dark, rain-slick rough harbor rock provides physical support, while soft, pale blue-gray fog blurs into a featureless haze at the extreme edges of the frame].",
            "outfit_assignments": [],
            "source_facts": [
              "진동벨이 울린다. 얼핏 보니 혜수이모다. 벌써 여러통 째 부재중 전화.",
              "Shot description: observer at doorway, looking in (wait, no. A close-up, high angle insert from above hands, shallow DOF)",
              "Muted blue-gray fog light"
            ],
            "visual_inferences": [
              "fixed_element wet_foggy_harbor_environment: Puddles and lingering moisture on the harbor rocks subtly reflect the muted dusk lighting.",
              "The bright screen of the smartphone highlights the texture of the fingers against the dark, dim background."
            ],
            "creative_decisions": [
              "Used 'Close on' phrasing with common noun and demographic descriptor to strictly follow body_part_focus_rule and prevent floating face artifacts.",
              "Replaced the specific '30 centimeters' measurement from the DP notes with 'just above the hands' to avoid T2I scale misinterpretation.",
              "Utilized shallow DOF to ensure the screen content is the sharp focal point while blurring the edge background as dictated by the skipped_close_framing rule."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "0f93d363254548ad",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "caf0d96c34f1f448",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 26,
        "dependency_reason": "동일한 포구 공간의 연속으로, 젖은 갯바위 질감과 푸른 안개 낀 석양빛(cool blue-gray dusk)의 조명 톤 유지",
        "_shot_index": 3,
        "visible_entities": [
          "C01",
          "P03",
          "L17"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 26,
            "shot_index": 3
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A close-up (CU), high angle insert from less than 30 centimeters above 수리영’s hands, using shallow DOF so the smartphone screen sits sharply in the central lower third without unnaturally enlarging it. 혜수’s name and repeated missed-call alerts face the camera clearly, while the wet harbor rock and pale fog blur around the phone edges.",
            "lighting_mood": "Cool fog-diffused dusk light wraps the hands, while the phone casts a weak bluish glow that feels anxious rather than comforting.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P03",
              "L17"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": null,
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "wet_foggy_harbor_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A heavy, pale mist obscures the background, severely limiting visibility and softening the dusk light into a pervasive cool blue-gray hue. The ground, including the paved pier and natural rocks, is soaked with water, reflecting faint ambient light from the recent heavy rain.",
                "applies_to_shots": [
                  1,
                  3,
                  6
                ],
                "source_facts": [
                  "소나기가 한차례 퍼 붓는가 싶더니, 비는 그치고 안개가 자욱하다.",
                  "The wet pier line sits low across the bottom quarter, while fog swallows nearly the entire middle and upper frame",
                  "wet harbor rock and pale fog blur around the phone edges",
                  "Muted blue-gray fog light"
                ],
                "visual_inferences": [
                  "Puddles and lingering moisture on the harbor surfaces subtly reflect the muted dusk lighting, highlighting the dampness."
                ],
                "creative_decisions": [
                  "Unified desaturated blue-gray color grading across all shots to emphasize isolation, anxiety, and the chilling post-rain atmosphere."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 26,
                "shot_index": 1,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "7e265de2792a6898"
      },
      {
        "scene_index": 26,
        "heading": "S#26. 포구 / 해질 무렵",
        "beat_title": "수리영의 절박한 부탁",
        "representative_moment": "배에 닷줄을 고정하는 인우의 등 뒤에 바짝 다가선 수리영이 간절하게 탑승을 요구한다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Medium close-up, low angle, over-the-shoulder composition with muted blue-gray fog light and telephoto compression",
            "t2i_prompt": "Photorealistic cinematic still. Medium close-up, low angle, over-the-shoulder composition. C03O04 in fisherman workwear, a Korean young man with short black hair, with back to camera occupying the left foreground, his damp back and shoulder blocking the view. He is bending over, eyes fixed on a taut, rain-darkened rope in his hands. Tight behind him at center-right, C01O02 in activity wear, a Korean young woman with long black hair, soft facial features, and golden wave-like eyes, at three-quarter angle from the left, leaning forward pleading, eyes fixed on C03O04. They share the narrow gap near a wet metal mooring ring at the lower frame edge. Focus on C01O02's urgent expression leaning into the space between them. Muted blue-gray fog light keeps both figures underexposed, while faint sodium reflections cast a sickly warmth on the wet rope and a slice of the damp boat hull at the right frame edge. [L17: Thick pale fog swallows the immediate background behind C01O02, heavily compressing the space.]",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              },
              {
                "character_id": "C03",
                "outlook_id": "O04"
              }
            ],
            "source_facts": [
              "배에서 내린 인우(24세. 남)는 닷줄을 고정시키고 있다.",
              "인우의 등 뒤에 바짝 선 채 간절한 눈빛으로 그를 올려다보는 수리영의 상체.",
              "비는 그치고 안개가 자욱하다."
            ],
            "visual_inferences": [
              "fixed_element wet_foggy_harbor_environment applied as frame-edge mist",
              "over-the-shoulder (OTS) angle places C03 in the foreground acting as a visual barrier",
              "low angle emphasizes C01's upward pleading gaze toward C03"
            ],
            "creative_decisions": [
              "implemented skipped_close_framing by omitting the wide background and rendering only the immediate frame-edge surfaces (mooring ring, slice of hull, fog layer)",
              "used telephoto compression to make the space between them feel tight and urgent",
              "applied faint sodium reflections to contrast with the cold blue-gray fog, drawing subtle attention to the interaction point (the rope)"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "85ac8486abae0ef6",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "efbacba8ed823788",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 26,
        "dependency_reason": "S26_Shot1 establishes the wet, foggy harbor environment and exact background mood, though this shot uses close framing to focus purely on the characters.",
        "_shot_index": 6,
        "visible_entities": [
          "C01",
          "C03",
          "P03",
          "L17"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 26,
            "shot_index": 6
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "At dock level behind 인우’s shoulder, the camera holds a medium close-up (MCU), low angle, over-the-shoulder composition with a compressed telephoto feel. 인우’s damp back and shoulder block the left foreground, while 수리영’s upper body appears tight behind him at center-right, leaning into the narrow gap with an urgent upward gaze.",
            "lighting_mood": "Muted blue-gray fog light keeps both faces underexposed, with faint sodium reflections on the wet rope and hull adding sickly warmth.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "C03",
              "P03",
              "L17"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              },
              {
                "character_id": "C03",
                "outlook_id": "O04"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": null,
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "wet_foggy_harbor_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A heavy, pale mist obscures the background, severely limiting visibility and softening the dusk light into a pervasive cool blue-gray hue. The ground, including the paved pier and natural rocks, is soaked with water, reflecting faint ambient light from the recent heavy rain.",
                "applies_to_shots": [
                  1,
                  3,
                  6
                ],
                "source_facts": [
                  "소나기가 한차례 퍼 붓는가 싶더니, 비는 그치고 안개가 자욱하다.",
                  "The wet pier line sits low across the bottom quarter, while fog swallows nearly the entire middle and upper frame",
                  "wet harbor rock and pale fog blur around the phone edges",
                  "Muted blue-gray fog light"
                ],
                "visual_inferences": [
                  "Puddles and lingering moisture on the harbor surfaces subtly reflect the muted dusk lighting, highlighting the dampness."
                ],
                "creative_decisions": [
                  "Unified desaturated blue-gray color grading across all shots to emphasize isolation, anxiety, and the chilling post-rain atmosphere."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 26,
                "shot_index": 1,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              },
              {
                "kind": "character_outlook",
                "id": "C03O04",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "5cd6701411d064cf"
      },
      {
        "scene_index": 27,
        "heading": "S#27. 옥탑방 안 / 해질 무렵",
        "beat_title": "음성메시지 확인과 불안감",
        "representative_moment": "엉망이 된 옥탑방에서 혜수(C02)가 굳은 얼굴로 스마트폰(P03)을 응시한다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Medium close-up at eye-level, low-key dusk light with blue-white screen glow",
            "t2i_prompt": "Photorealistic cinematic still. At eye-level, C02O03 in a detective coat, a Korean middle-aged woman, occupies the center-right of the frame in a medium close-up, neck and upper chest visible. She stands at a three-quarter angle from the left, her eyes looking down at P03, a smartphone clutched chest-high. The phone screen is angled slightly upward toward the camera. Cool blue-white light from the device reflects on her tense face and chin, while a slanted, low-key dusk wash falls across her shoulder. Shallow depth of field isolates her upper body against [L05: a soft out-of-focus background showing damp gray shadows, hints of scattered debris, and a slice of an open window frame].",
            "outfit_assignments": [
              {
                "character_id": "C02",
                "outlook_id": "O03"
              }
            ],
            "source_facts": [
              "S#27 scene text: '수리영의 녹음메시지를 확인하는 혜수'",
              "Shot 1 description: '엉망이 된 옥탑방 거실 한가운데 서서 자신의 스마트폰을 뚫어지게 응시하는 혜수의 굳은 얼굴.'"
            ],
            "visual_inferences": [
              "fixed_element 'ruined_rooftop_room': low-key dusk light, damp gray shadows, and scattered items adapted for close framing",
              "Screen light contrasts with the dusk lighting to emphasize her facial expression"
            ],
            "creative_decisions": [
              "Medium close-up focusing on her upper body and face to capture the tense mood without showing the full background",
              "Shallow depth of field to soften the chaotic background and keep the visual priority on her reaction"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "489b9708b3bbf483",
              "owned_hash": "18fa02b2ac47d2f4",
              "camera_direction_hash": "96613c9432a20095",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 18,
        "dependency_reason": "동일한 옥탑방 배경을 참조하되, 본 샷은 close framing 정책에 따라 배경 참조 이미지를 건너뛰고 L05 프레임 엣지 표면으로 묘사함",
        "_shot_index": 1,
        "visible_entities": [
          "C02",
          "P03",
          "L05"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 27,
            "shot_index": 1
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Inside the rooftop room, the camera locks into a medium close-up (MCU), eye-level, with a restrained rack focus from the smartphone in 혜수’s hands up to her rigid face. She is framed slightly off-center in the middle of the wrecked living area, her gaze pinned downward at the phone just below lens line rather than confronting the camera.",
            "lighting_mood": "Low-key dusk seeps through the open window, mixing with the phone’s blue-white light on 혜수’s tense face and leaving the ruined room in damp gray shadow.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C02",
              "P03",
              "L05"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C02",
                "outlook_id": "O03"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "cb_living_dusk_ransacked_window_open",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "ruined_rooftop_room",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior of a small rooftop room in a state of complete disarray. The living area is messy, with scattered items and debris indicating a recently wrecked space. A slanted, low-key dusk light filters in, casting cold and damp gray shadows across the room, while a slightly loose entrance door stands in the background.",
                "applies_to_shots": [
                  1,
                  2
                ],
                "source_facts": [
                  "S#27 scene text: '덜커덩 덜커덩. 바람에 현관문 소리. 황량하다.'",
                  "Shot 1 description: 'wrecked living area', 'ruined room in damp gray shadow'",
                  "Shot 2 description: 'messy room depth stretches behind him toward the rattling entrance', 'slanted, low-key dusk wash'"
                ],
                "visual_inferences": [
                  "scattered objects and debris across the floor to visually depict the 'wrecked' and 'messy' state without specific items mentioned"
                ],
                "creative_decisions": [
                  "use of strong gray shadows and slanted dusk light to emphasize the desolate, cold mood of the empty room"
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 18,
                "shot_index": 10,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C02O03",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "df2d3df54f1c81c6"
      },
      {
        "scene_index": 27,
        "heading": "S#27. 옥탑방 안 / 해질 무렵",
        "beat_title": "황량한 옥탑방에서의 냉소",
        "representative_moment": "어두워지는 옥탑방 안에서 C05가 짝다리를 짚고 선 채 입술을 비틀며 혜수 쪽을 바라보고 있다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Medium-wide Dutch angle at chest height, slanted low-key dusk wash with cold gray shadows",
            "t2i_prompt": "Photorealistic cinematic still. Dutch angle, camera positioned at chest height keeping the subject's full body in frame from three meters away, floor visible below his shoes. C05O07 in a casual jacket, a Korean adult male with short black hair and deep-set eyes, occupies the right half of the frame at a three-quarter angle from the right. His weight rests heavily on one leg with a dropped hip, his mouth twisted dismissively, eyes fixed on an off-camera figure. Focus on his crooked stance and the desolate space around him. A slanted, low-key dusk wash falls across the wall and the side of his jacket, creating cold, damp gray shadows. The depth of the messy living area stretches behind him, anchoring the existing door, open window, and floor debris from the reference within the dim ambient lighting.",
            "outfit_assignments": [
              {
                "character_id": "C05",
                "outlook_id": "O07"
              }
            ],
            "source_facts": [
              "S#27 scene text: '그럴리가요. (혀를 끌끌 차며) 아이구 쯧쯧.'",
              "Shot 2 description: '입술을 비튼 채 어이없다는 듯 짝다리를 짚고 선 김형사의 삐딱한 전신.'",
              "DP instruction: 'The lateral track settles into a wide shot (WS), Dutch angle, at chest height... holding 김형사’s full body about three meters away.'",
              "Lighting: 'slanted, low-key dusk wash cuts across the room'"
            ],
            "visual_inferences": [
              "Dutch angle and crooked stance visually amplify the psychological instability and his dismissive attitude",
              "Chest-height wide framing at 3 meters allows the full body to be captured along with the floor and ceiling edges, adhering to geometric camera rules",
              "Using 'falls across the wall and the side of his jacket' to avoid physically cutting the character with lighting terminology"
            ],
            "creative_decisions": [
              "Placed C05O07 strictly on the right half of the frame to emphasize the deep, wrecked room stretching towards the background door",
              "Anchored background elements like the door, window, and debris 'from the reference' to satisfy exact_background and Rule C constraints"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "9c20f8ae0504e020",
              "owned_hash": "18fa02b2ac47d2f4",
              "camera_direction_hash": "185a323f84b50f64",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "door",
                  "violating_phrase": "existing door",
                  "reason": "The prompt explicitly references the door as already existing from the reference, which anchors it rather than asking to redraw it.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "debris",
                  "violating_phrase": "floor debris from the reference",
                  "reason": "The debris is mentioned as part of the existing reference background, not as something to be newly created or modified.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "window",
                  "violating_phrase": "open window",
                  "reason": "The window is described as an existing background element from the reference, so this is an anchor reference rather than a redraw instruction.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 27,
        "dependency_reason": "previous_shot_refs indicates S27_Shot1 (exact_background) to maintain the identical wrecked room state and dusk lighting from the prior shot.",
        "_shot_index": 2,
        "visible_entities": [
          "C05",
          "P03",
          "L05"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 27,
            "shot_index": 2
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "The lateral track settles into a wide shot (WS), Dutch angle, at chest height along the side of the living area, holding 김형사’s full body about three meters away. He occupies the right half with one hip dropped and his mouth twisted, while the messy room depth stretches behind him toward the rattling entrance.",
            "lighting_mood": "A slanted, low-key dusk wash cuts across the room, making 김형사’s dismissive posture feel colder against the hollow wind from the door.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C05",
              "P03",
              "L05"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C05",
                "outlook_id": "O07"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "cb_living_dusk_ransacked_window_open",
            "reference_usage": "exact_background",
            "owned_objects": [
              "TV",
              "TV console",
              "air conditioner",
              "cooktop",
              "curtain",
              "debris",
              "document",
              "door",
              "floor cushion",
              "kitchen counter",
              "mug",
              "refrigerator",
              "shoe rack",
              "sink",
              "sofa",
              "table",
              "window"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "ruined_rooftop_room",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior of a small rooftop room in a state of complete disarray. The living area is messy, with scattered items and debris indicating a recently wrecked space. A slanted, low-key dusk light filters in, casting cold and damp gray shadows across the room, while a slightly loose entrance door stands in the background.",
                "applies_to_shots": [
                  1,
                  2
                ],
                "source_facts": [
                  "S#27 scene text: '덜커덩 덜커덩. 바람에 현관문 소리. 황량하다.'",
                  "Shot 1 description: 'wrecked living area', 'ruined room in damp gray shadow'",
                  "Shot 2 description: 'messy room depth stretches behind him toward the rattling entrance', 'slanted, low-key dusk wash'"
                ],
                "visual_inferences": [
                  "scattered objects and debris across the floor to visually depict the 'wrecked' and 'messy' state without specific items mentioned"
                ],
                "creative_decisions": [
                  "use of strong gray shadows and slanted dusk light to emphasize the desolate, cold mood of the empty room"
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 27,
                "shot_index": 1,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C05O07",
                "policy": "required"
              },
              {
                "kind": "background",
                "id": "cb_living_dusk_ransacked_window_open",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "28674791b43e1bc6"
      },
      {
        "scene_index": 28,
        "heading": "바다 + 갑판 / 해질 무렵",
        "beat_title": "아무것도 없는 바다 한가운데의 소형어선 전경",
        "representative_moment": "붉은 노을이 지는 바다 한가운데에 외롭게 떠 있는 소형어선의 전경.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "extreme wide establishing framing from low over the water, capturing the boat as a silhouette against the muted charcoal-blue and deep-red evening sky",
            "t2i_prompt": "Photorealistic cinematic still. Captured from a low observer point just over the open water off the boat's port bow, the dark lightly choppy water occupies the lower frame. Focus on the small fishing boat drifting alone, situated in the center-left and taking up only a modest portion of the frame, dwarfed by the environment. The damp hull is angled toward the camera, appearing as a lonely, ominous silhouette. The space is [L18: a dark, rolling open sea under a deep-red evening sky]. The setting sun casts a distinct blood-red light mixed with cool, damp charcoal-blue shadows across the waves and the wet deck from the reference. A thin haze hangs over the distance. The wheelhouse window from the reference on the side of the boat is barely visible, catching a faint reflection of the deep-red sunset glow.",
            "outfit_assignments": [],
            "source_facts": [
              "Scene heading: 바다 + 갑판 / 해질 무렵 (Sea + deck / Evening)",
              "Shot 1: 해질 무렵 붉은 노을빛이 비치는 바다 한가운데 떠 있는 소형어선의 조용한 전경."
            ],
            "visual_inferences": [
              "fixed_element environment_state: A dark, rolling open sea under a deep-red evening sky.",
              "The small fishing boat sits in the center-left, dwarfed by the dark, rolling sea.",
              "Wheelhouse window and wet deck reflect the contrasting red and blue tones of the sunset."
            ],
            "creative_decisions": [
              "Positioned the camera low over the water so the sea occupies the lower frame, adhering to spatial consistency rules without using shot type keywords.",
              "Rendered the sea in muted charcoal-blue with a restrained deep-red sunset band to emphasize an ominous, occult-horror mood.",
              "Anchored the deck and wheelhouse to 'from the reference' to satisfy the background_binding owned_objects constraint."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "a0965ee978a01f3c",
              "owned_hash": "dc0f99e9f7f00426",
              "camera_direction_hash": "7d9d9762d599c09f",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "deck",
                  "violating_phrase": "the wet deck from the reference",
                  "reason": "The phrase explicitly references the existing deck from the reference, so it is an anchor reference rather than a request to redraw it.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "window",
                  "violating_phrase": "The wheelhouse window from the reference",
                  "reason": "The window is explicitly described as being from the reference and part of the existing boat, so this is an anchor reference.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "wheelhouse",
                  "violating_phrase": "The wheelhouse window from the reference",
                  "reason": "The wheelhouse is mentioned only as part of an existing referenced structure, not as a newly created object.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "독립적인 외부 EWS(Extreme Wide Shot) 전경으로, 이전 샷의 구체적인 공간 배치(인물/내부)에 시각적으로 직접 종속되지 않음.",
        "_shot_index": 1,
        "visible_entities": [
          "P06",
          "P04",
          "L18",
          "L19"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 28,
            "shot_index": 1
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "From low over the open water off the boat’s port bow, the camera begins a slow dolly-in in an extreme wide / establishing shot (EWS), low angle, silhouette framing against the red evening horizon. The small fishing boat sits in the center-left and occupies only a modest portion of the frame, dwarfed by the dark, rolling sea and the blood-colored strip of sunset.",
            "lighting_mood": "The sea is rendered in muted charcoal-blue with a restrained deep-red sunset band, turning the boat into a lonely, ominous silhouette.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "P06",
              "P04",
              "L18",
              "L19"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "bg_deck_dusk_at_sea",
            "reference_usage": "exact_background",
            "owned_objects": [
              "GPS display",
              "anchor",
              "antenna",
              "bait tub",
              "bulwark",
              "cleat",
              "deck",
              "deck mark",
              "engine hatch",
              "fish box",
              "fish crate",
              "fuel locker",
              "helm console",
              "mast",
              "navigation light",
              "net roller",
              "radio",
              "rail",
              "rope",
              "sliding door",
              "stern gate",
              "wheelhouse",
              "winch",
              "window"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "evening_sea_and_sunset",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dark, rolling open sea under a deep-red evening sky. The setting sun casts a distinct blood-red light mixed with cool, damp charcoal-blue shadows across the wet deck of the small fishing boat, creating a lonely and ominous atmosphere.",
                "applies_to_shots": [
                  1,
                  4,
                  5
                ],
                "source_facts": [
                  "Scene heading: 바다 + 갑판 / 해질 무렵 (Sea + deck / Evening)",
                  "Shot 1: muted charcoal-blue with a restrained deep-red sunset band",
                  "Shot 4: Deep red sunset light grazes... wet deck falls into cool, damp shadow",
                  "Shot 5: red sunset rim light and cold sea shadow"
                ],
                "visual_inferences": [
                  "The sea appears deep and restless, amplifying the isolation of the small boat, while the wet deck reflects the contrasting red and blue tones."
                ],
                "creative_decisions": [
                  "Using the stark contrast between the warm, bloody red of the sunset and the cold, dark shadows of the sea to build a sense of impending tension across all framings."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "background",
                "id": "bg_deck_dusk_at_sea",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "2e4779970a1ffe83"
      },
      {
        "scene_index": 28,
        "heading": "바다 + 갑판 / 해질 무렵",
        "beat_title": "피 묻은 옛 사진",
        "representative_moment": "수리영의 손안에 들린 낡고 피 묻은 옛날 사진이 클로즈업된다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Near-macro extreme close-up (ECU), high angle, shallow depth of field",
            "t2i_prompt": "Photorealistic cinematic still. A near-macro extreme close-up, high angle from above, shallow depth of field. Close on P04, a small, worn and aged paper photograph featuring noticeable dark red bloodstains smeared across its surface and tiny handwritten text visible, its edges slightly frayed. The photo sits sharply in the central third of the frame, facing the camera. It is anchored at the lower edge by the tense, damp fingertips of a young Korean woman reaching into the frame. Deep red sunset light grazes the blood marks on the paper and her fingers, while the rest of the space falls into cool, damp shadow. Beneath the hand, [L19: the wet deck of a small fishing boat, reflecting blood-red light mixed with charcoal-blue shadows] and P06, an unfolded paper map lying flat, blur out of focus in the background. A wet side rail receding diagonally is faintly visible in the reddish dusk.",
            "outfit_assignments": [],
            "source_facts": [
              "S#28 action: 25씬의 사진을 보여준다.",
              "Shot description: 수리영의 손안에 들린 낡고 피 묻은 옛날 사진 클로즈업.",
              "DP direction: A near-macro extreme close-up (ECU), high angle, shallow DOF insert hovers above the hand at chest height."
            ],
            "visual_inferences": [
              "fixed_element blood_stained_old_photograph applied to P04.",
              "fixed_element evening_sea_and_sunset applied to the deck reflection and lighting mood.",
              "Body-part focus rule applied: C01 is represented as a common noun (a young Korean woman) due to extreme close-up framing."
            ],
            "creative_decisions": [
              "Used a high angle looking down at the deck to anchor the hand and the photo geometrically.",
              "Kept the background deeply blurred (shallow DOF) to emphasize the blood-stained photo and tense fingers.",
              "Contrasted deep red sunset light against cool, damp charcoal-blue shadows for an ominous thriller mood."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "d42af71502e5dfbb",
              "owned_hash": "dc0f99e9f7f00426",
              "camera_direction_hash": "f55f178c9332fc45",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 28,
        "dependency_reason": "S28_Shot1의 공간과 해질 무렵 조명 톤 연속성 유지",
        "_shot_index": 4,
        "visible_entities": [
          "C01",
          "P06",
          "P04",
          "L18",
          "L19"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 28,
            "shot_index": 4
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A near-macro extreme close-up (ECU), high angle, shallow DOF insert hovers 20–30 centimeters above 수리영’s hand at chest height. The old blood-stained photograph sits sharply in the central third with the tiny ‘금월’ clue legible, while her fingers anchor the lower edge and the wet deck blurs into reddish dusk behind it.",
            "lighting_mood": "Deep red sunset light grazes the photo’s blood marks while the rest of the deck falls into cool, damp shadow.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P06",
              "P04",
              "L18",
              "L19"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "bg_deck_dusk_at_sea",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "evening_sea_and_sunset",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dark, rolling open sea under a deep-red evening sky. The setting sun casts a distinct blood-red light mixed with cool, damp charcoal-blue shadows across the wet deck of the small fishing boat, creating a lonely and ominous atmosphere.",
                "applies_to_shots": [
                  1,
                  4,
                  5
                ],
                "source_facts": [
                  "Scene heading: 바다 + 갑판 / 해질 무렵 (Sea + deck / Evening)",
                  "Shot 1: muted charcoal-blue with a restrained deep-red sunset band",
                  "Shot 4: Deep red sunset light grazes... wet deck falls into cool, damp shadow",
                  "Shot 5: red sunset rim light and cold sea shadow"
                ],
                "visual_inferences": [
                  "The sea appears deep and restless, amplifying the isolation of the small boat, while the wet deck reflects the contrasting red and blue tones."
                ],
                "creative_decisions": [
                  "Using the stark contrast between the warm, bloody red of the sunset and the cold, dark shadows of the sea to build a sense of impending tension across all framings."
                ],
                "confidence": "high"
              },
              {
                "element_id": "blood_stained_old_photograph",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "A small, worn and aged paper photograph featuring noticeable dark red bloodstains smeared across its surface. The edges of the photo look slightly frayed from time and handling.",
                "applies_to_shots": [
                  4,
                  5
                ],
                "source_facts": [
                  "S#28 action: 25씬의 사진을 보여준다",
                  "Shot 4: old blood-stained photograph sits sharply in the central third",
                  "Shot 5: blurred edge of the photograph low in the foreground"
                ],
                "visual_inferences": [
                  "The photograph is physically weathered, with the bloodstains dried into a dark, rusty crimson over the faded image."
                ],
                "creative_decisions": [
                  "Positioning the photograph as a persistent visual anchor in both the macro insert and the close-up reaction shot, maintaining its physical presence near the lens."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 28,
                "shot_index": 1,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "135103aa3478f2f5"
      },
      {
        "scene_index": 28,
        "heading": "S#28. 바다 + 갑판 / 해질 무렵",
        "beat_title": "사진을 보고 발작적으로 당황하는 인우",
        "representative_moment": "C03 stares down at the blood-stained photograph in the foreground, his expression warped by extreme panic under a red sunset rim light.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Close-up (CU) at eye level, shallow depth of field with foreground blur, sickly red rim light contrasting with cold blue shadows.",
            "t2i_prompt": "Photorealistic cinematic still. Head-and-shoulder composition. C03O04 in a dark fisher workwear jacket, a young Korean man with short black hair and a slender face, at three-quarter angle from the right. His jaw is clenched in a slight recoil, his wide, panicked dark eyes looking down. The blurred edge of a small, worn paper photograph with dark red bloodstains occupies the lower foreground, angled toward the camera. A sickly deep-red sunset rim light reflects on his shoulder and jawline, while cold charcoal-blue ambient shadow washes across his underlit cheeks and the front of his jacket. Focus on his wide eyes and distressed expression. Directly behind his shoulder, [L19: a dimly reflective, wet cabin door frame] and [L18: a soft out-of-focus band of dark rolling sea] fill the immediate background edge.",
            "outfit_assignments": [
              {
                "character_id": "C03",
                "outlook_id": "O04"
              }
            ],
            "source_facts": [
              "S#28 action: (사진보더니 당황) 몰라. 모른다구.",
              "shot description: 사진에서 시선을 떼지 못한 채 크게 당황하여 눈이 커진 인우의 일그러진 표정",
              "camera_direction: camera pushes into a close-up (CU), eye-level, with the blurred edge of the photograph low in the foreground"
            ],
            "visual_inferences": [
              "fixed_element evening_sea_and_sunset: deep-red sunset light mixed with cool charcoal-blue shadows falling on the character",
              "fixed_element blood_stained_old_photograph: dark red bloodstains on the edge of the paper"
            ],
            "creative_decisions": [
              "Used a head-and-shoulder composition to prevent the T2I model from generating a floating severed head, satisfying the close-framing face phrasing rule.",
              "Applied the skipped_close_framing background policy by describing only immediate frame-edge elements (door frame and a slice of sea) behind the silhouette, avoiding any mention of the full background deck reference."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "55c70dec689cd684",
              "owned_hash": "dc0f99e9f7f00426",
              "camera_direction_hash": "eb670b056cc2a9dc",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 28,
        "dependency_reason": "Close-up reaction shot maintaining lighting and prop continuity from the prior shot's wider frame.",
        "_shot_index": 5,
        "visible_entities": [
          "C03",
          "P06",
          "P04",
          "L18",
          "L19"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 28,
            "shot_index": 5
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "At 인우’s eye level, less than a meter from his face, the camera pushes into a close-up (CU), eye-level, with the blurred edge of the photograph low in the foreground keeping his gaze trapped below the lens. His face fills most of the center frame, three-quarter to camera, eyes widened and expression warped by recognition before he bolts toward the wheelhouse.",
            "lighting_mood": "A sickly mix of red sunset rim light and cold sea shadow sharpens the panic in 인우’s eyes while leaving his cheeks underlit.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C03",
              "P06",
              "P04",
              "L18",
              "L19"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C03",
                "outlook_id": "O04"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "bg_deck_dusk_at_sea",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "evening_sea_and_sunset",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dark, rolling open sea under a deep-red evening sky. The setting sun casts a distinct blood-red light mixed with cool, damp charcoal-blue shadows across the wet deck of the small fishing boat, creating a lonely and ominous atmosphere.",
                "applies_to_shots": [
                  1,
                  4,
                  5
                ],
                "source_facts": [
                  "Scene heading: 바다 + 갑판 / 해질 무렵 (Sea + deck / Evening)",
                  "Shot 1: muted charcoal-blue with a restrained deep-red sunset band",
                  "Shot 4: Deep red sunset light grazes... wet deck falls into cool, damp shadow",
                  "Shot 5: red sunset rim light and cold sea shadow"
                ],
                "visual_inferences": [
                  "The sea appears deep and restless, amplifying the isolation of the small boat, while the wet deck reflects the contrasting red and blue tones."
                ],
                "creative_decisions": [
                  "Using the stark contrast between the warm, bloody red of the sunset and the cold, dark shadows of the sea to build a sense of impending tension across all framings."
                ],
                "confidence": "high"
              },
              {
                "element_id": "blood_stained_old_photograph",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "A small, worn and aged paper photograph featuring noticeable dark red bloodstains smeared across its surface. The edges of the photo look slightly frayed from time and handling.",
                "applies_to_shots": [
                  4,
                  5
                ],
                "source_facts": [
                  "S#28 action: 25씬의 사진을 보여준다",
                  "Shot 4: old blood-stained photograph sits sharply in the central third",
                  "Shot 5: blurred edge of the photograph low in the foreground"
                ],
                "visual_inferences": [
                  "The photograph is physically weathered, with the bloodstains dried into a dark, rusty crimson over the faded image."
                ],
                "creative_decisions": [
                  "Positioning the photograph as a persistent visual anchor in both the macro insert and the close-up reaction shot, maintaining its physical presence near the lens."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 28,
                "shot_index": 1,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C03O04",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "344bb97d1a881512"
      },
      {
        "scene_index": 29,
        "heading": "S#29. 조타실 / 해질 무렵",
        "beat_title": "사진 속 소녀가 엄마(강민숙)임을 밝히는 수리영",
        "representative_moment": "C01의 손가락이 낡은 사진(P04) 속 어린 소녀의 형상을 꾹 누르고 있는 극단적 클로즈업.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Extreme close-up, high angle macro, low-key red dusk mixed with weak green instrument light",
            "t2i_prompt": "Photorealistic cinematic still. Extreme close-up, high angle macro, tilted shallowly downward. Focus on the pressed fingertip of a young East Asian woman, her hand holding P04, a creased, blood-stained old photograph. The fingertip presses firmly onto the reproduced image of a young East Asian girl on the photograph's front surface facing the camera. [L20: soft out-of-focus background of a cramped cabin, lit by low-key red dusk and a weak green instrument glow, showing a blurred mechanical console and a faint dark red circular mark in the shallow depth of field].",
            "outfit_assignments": [],
            "source_facts": [
              "수리영의 손에 들린 낡은 사진 속 어린 여자애 형상을 꾹 누른 손가락 클로즈업.",
              "blurred red circular mark on the bulkhead",
              "weak green instrument light"
            ],
            "visual_inferences": [
              "fixed_element red_circular_wall_markings and worn_steering_console interpreted as soft out-of-focus background elements via [L20] block due to skipped_close_framing policy",
              "young girl image inside P04 described using demographic descriptor instead of ID per reproduction surface rule"
            ],
            "creative_decisions": [
              "Using ECU macro framing with shallow depth of field to tightly isolate the fingertip and the reproduced girl's face",
              "Applying high angle downward tilt to naturally position the photo's front surface facing the camera without distortion"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "414dce0ffa0ca969",
              "owned_hash": "5dbcf0b558780b0b",
              "camera_direction_hash": "e0fc4fec5fc57378",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "",
        "_shot_index": 4,
        "visible_entities": [
          "C01",
          "P04",
          "P01",
          "L20"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 29,
            "shot_index": 4
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Inside the cramped wheelhouse, the camera drops low over 수리영’s hands into an extreme close-up (ECU), high angle macro, tilted shallowly downward so the pressed fingertip and the young girl in the photograph command the center without the prop exceeding natural scale. The steering console and a blurred red circular mark on the bulkhead sit in the background, hinting that the evidence connects to the room around them.",
            "lighting_mood": "Low-key red dusk through the wheelhouse glass mixes with weak green instrument light, making the damp cabin feel claustrophobic and occult-charged.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "P04",
              "P01",
              "L20"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "bg_wheelhouse_dusk_red_circle_marks",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "red_circular_wall_markings",
                "element_type": "environment_state",
                "character_name": "",
                "description": "Dense, ritualistic red circular markings are painted tightly across the dim walls of the boat's cabin. The crude circles spread edge-to-edge on the bulkheads, standing out as deep, ominous accents against the cold gray-green shadows of the grimy interior.",
                "applies_to_shots": [
                  4,
                  6,
                  11
                ],
                "source_facts": [
                  "조타실안 수많은 원. 빙글빙글 돌고 여기에서 또다시 발견된 붉은 원!",
                  "blurred red circular mark on the bulkhead",
                  "dense red circular paint markings spread edge to edge",
                  "muted red wall marks bleeding softly behind him"
                ],
                "visual_inferences": [
                  "The circles appear to be hastily painted or smeared, varying slightly in size but forming a dense, claustrophobic and occult pattern."
                ],
                "creative_decisions": [
                  "Using the red marks as a recurring background and midground motif across different focal lengths to steadily build tension."
                ],
                "confidence": "high"
              },
              {
                "element_id": "worn_steering_console",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "A worn, mechanical steering console of a small boat, faintly glowing with weak green instrument light. It remains fixed within the cramped cabin, serving as an architectural anchor for the space.",
                "applies_to_shots": [
                  4,
                  6
                ],
                "source_facts": [
                  "The steering console... sit in the background",
                  "weak green instrument light",
                  "blurred sliver of the helm console... at the left foreground"
                ],
                "visual_inferences": [
                  "The console is aged and weathered, matching the old, damp aesthetic of the small fishing vessel."
                ],
                "creative_decisions": [
                  "Keeping the console partially visible either in the background or as a foreground sliver to ground the spatial geometry of the extremely tight wheelhouse."
                ],
                "confidence": "medium"
              }
            ],
            "previous_shot_refs": [],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "ce04259bd74230d0"
      },
      {
        "scene_index": 29,
        "heading": "S#29. 조타실 / 해질 무렵",
        "beat_title": "조타실을 채운 붉은 원",
        "representative_moment": "어두운 조타실 벽면에 붉은색 페인트로 빽빽하게 그려진 수많은 원형 표식들이 기괴하게 드러난다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "WS eye-level through-object composition, low-key dusk lighting",
            "t2i_prompt": "Photorealistic cinematic still. Observer perspective at eye-level, medium framing focusing on the bulkhead from the reference. In the left foreground, a blurred sliver of the helm console from the reference and the side frame of the window from the reference create a tight, claustrophobic through-object composition. The cramped bulkhead from the reference fills the midground, its damp, wet-looking surface densely covered with crude, ritualistic red circular marks from the reference spreading edge-to-edge. Focus on the dense red circular markings painted tightly across the dim wall. Low-key dusk light leaks through the fogged glass, casting cold gray-green shadows across the grimy interior, making the deep red paint read as a deep, terrifying accent.",
            "outfit_assignments": [],
            "source_facts": [
              "blurred red circular mark on the bulkhead",
              "dense red circular paint markings spread edge to edge",
              "조타실안 수많은 원. 빙글빙글 돌고 여기에서 또다시 발견된 붉은 원!"
            ],
            "visual_inferences": [
              "The circles appear to be hastily painted or smeared, varying slightly in size but forming a dense, claustrophobic and occult pattern.",
              "The console is aged and weathered, matching the old, damp aesthetic of the small fishing vessel."
            ],
            "creative_decisions": [
              "Using the red marks as a recurring background and midground motif across different focal lengths to steadily build tension.",
              "Keeping the console partially visible either in the background or as a foreground sliver to ground the spatial geometry of the extremely tight wheelhouse."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "0ec172c99c0760b5",
              "owned_hash": "5dbcf0b558780b0b",
              "camera_direction_hash": "a5b994c9d1df04ba",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "bulkhead",
                  "violating_phrase": "bulkhead from the reference",
                  "reason": "The prompt explicitly references the existing bulkhead from the reference, which is an anchor/reference mention rather than a request to redraw it.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "helm console",
                  "violating_phrase": "helm console from the reference",
                  "reason": "The prompt mentions the existing helm console from the reference as part of the composition, not as a newly created object.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "window",
                  "violating_phrase": "window from the reference",
                  "reason": "The window is referenced as an existing background element in the foreground composition, not explicitly added or generated anew.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "bulkhead",
                  "violating_phrase": "bulkhead from the reference",
                  "reason": "The bulkhead is described as already present in the reference and used as the wall surface, so this is an anchor reference.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "mark",
                  "violating_phrase": "red circular marks from the reference",
                  "reason": "The marks are explicitly described as coming from the reference and as painted markings already on the wall, not as a redraw command.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 29,
        "dependency_reason": "S29_Shot4 조타실 내부 배경의 연속성 및 붉은 원형 표식 유지",
        "_shot_index": 6,
        "visible_entities": [
          "P04",
          "P01",
          "L20"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 29,
            "shot_index": 6
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "From an observer perspective, the camera holds at the end of the pan at eye-level near Inwoo’s shoulder line, about five feet from the wheelhouse wall, in a WS eye-level through-object composition. The cramped wall fills the midground with dense red circular paint markings spread edge to edge, while a blurred sliver of the helm console and window frame sit at the left foreground, keeping the boat’s claustrophobic geometry visible without breaking focus.",
            "lighting_mood": "Low-key dusk light leaks through dirty glass, flattening the room into cold gray-green shadows while the red paint reads as a deep, ritualistic accent.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "P04",
              "P01",
              "L20"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "background_ref_attached",
            "bg_id": "bg_wheelhouse_dusk_red_circle_marks",
            "reference_usage": "exact_background",
            "owned_objects": [
              "GPS screen",
              "VHF radio",
              "bench locker",
              "bulkhead",
              "chart",
              "chart table",
              "clipboard",
              "fishfinder screen",
              "floor",
              "helm console",
              "helm wheel",
              "mark",
              "sliding door",
              "swivel chair",
              "throttle",
              "window",
              "windscreen"
            ],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "owned_objects lists all environment objects already drawn in the background reference — do not create new ones",
              "when focusing on an owned object, anchor it explicitly: 'the door from the reference'"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "red_circular_wall_markings",
                "element_type": "environment_state",
                "character_name": "",
                "description": "Dense, ritualistic red circular markings are painted tightly across the dim walls of the boat's cabin. The crude circles spread edge-to-edge on the bulkheads, standing out as deep, ominous accents against the cold gray-green shadows of the grimy interior.",
                "applies_to_shots": [
                  4,
                  6,
                  11
                ],
                "source_facts": [
                  "조타실안 수많은 원. 빙글빙글 돌고 여기에서 또다시 발견된 붉은 원!",
                  "blurred red circular mark on the bulkhead",
                  "dense red circular paint markings spread edge to edge",
                  "muted red wall marks bleeding softly behind him"
                ],
                "visual_inferences": [
                  "The circles appear to be hastily painted or smeared, varying slightly in size but forming a dense, claustrophobic and occult pattern."
                ],
                "creative_decisions": [
                  "Using the red marks as a recurring background and midground motif across different focal lengths to steadily build tension."
                ],
                "confidence": "high"
              },
              {
                "element_id": "worn_steering_console",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "A worn, mechanical steering console of a small boat, faintly glowing with weak green instrument light. It remains fixed within the cramped cabin, serving as an architectural anchor for the space.",
                "applies_to_shots": [
                  4,
                  6
                ],
                "source_facts": [
                  "The steering console... sit in the background",
                  "weak green instrument light",
                  "blurred sliver of the helm console... at the left foreground"
                ],
                "visual_inferences": [
                  "The console is aged and weathered, matching the old, damp aesthetic of the small fishing vessel."
                ],
                "creative_decisions": [
                  "Keeping the console partially visible either in the background or as a foreground sliver to ground the spatial geometry of the extremely tight wheelhouse."
                ],
                "confidence": "medium"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 29,
                "shot_index": 4,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "background",
                "id": "bg_wheelhouse_dusk_red_circle_marks",
                "policy": "required"
              }
            ],
            "forbidden_refs": [],
            "readiness_policy": "block_if_missing",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "8c3bbdad701073d0"
      },
      {
        "scene_index": 29,
        "heading": "S#29. 조타실 / 해질 무렵",
        "beat_title": "의식을 잃고 인우에게 쓰러지는 수리영",
        "representative_moment": "의식이 혼미해진 C01이 무표정한 C03의 어깨에 쓰러진다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Tight MCU low angle from below eye level, shallow depth of field, cold dusk lighting",
            "t2i_prompt": "Photorealistic cinematic still. Tight MCU low angle from below eye level, shallow depth of field. Focus on C03O04's downward gaze. C03O04 in dark fisher workwear, a Korean young man with short black hair and a slender face, occupies the upper right third at a three-quarter angle from the left. His chin is lowered, eyes fixed downward on the figure resting against him. C01O02 in a waterproof jumper, a Korean young woman with long black hair, occupies the lower left foreground in right profile. Her eyes are closed, her head and upper chest slumping heavily against C03O04's shoulder, using his body as the shared physical anchor. Cold dusk spill and weak ambient reflections fall on C03O04's cheek and jacket. [L20: The frame edge behind them shows a shadowed doorframe compressing the right side, while out-of-focus dense, ritualistic red circular markings are smeared softly across the dim gray bulkhead surface, bleeding into the shallow background.]",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              },
              {
                "character_id": "C03",
                "outlook_id": "O04"
              }
            ],
            "source_facts": [
              "S#29 action: '의식이 혼미해지는 수리영. 수리영을 노려보는 인우의 어깨에 그대로 쓰러진다.'",
              "shot description: Inwoo looking down at Suriyoung fallen on his shoulder."
            ],
            "visual_inferences": [
              "fixed_element red_circular_wall_markings: dense ritualistic red circular markings out of focus behind them.",
              "shallow depth of field blurs the background to emphasize the close physical contact."
            ],
            "creative_decisions": [
              "Low angle MCU to emphasize Inwoo's cold, dominant presence as Suriyoung slumps over him.",
              "Cold dusk spill falling on surfaces contrasts with the blurred red marks at the frame edge."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "08b576712b24484d",
              "owned_hash": "5dbcf0b558780b0b",
              "camera_direction_hash": "7ff155c4e3756114",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 29,
        "dependency_reason": "Same location as S29 Shot 4, maintaining the fixed red circular markings but omitting the broader room reference due to close framing.",
        "_shot_index": 11,
        "visible_entities": [
          "C01",
          "C03",
          "P04",
          "P01",
          "L20"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 29,
            "shot_index": 11
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "The camera tracks back into the collapse and settles just below Inwoo’s eye line, near Suriyoung’s fallen shoulder, in a tight MCU low angle with shallow DOF. Suriyoung occupies the lower left foreground as a soft, heavy shape against Inwoo’s shoulder, while Inwoo’s face and cold downward eyes sit in the upper right third, his expression almost still enough to feel inhuman.",
            "lighting_mood": "Cold dusk spill and weak instrument reflections carve Inwoo’s eyes from shadow, with muted red wall marks bleeding softly behind him.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "C01",
              "C03",
              "P04",
              "P01",
              "L20"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              },
              {
                "character_id": "C03",
                "outlook_id": "O04"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "skipped_close_framing",
            "bg_id": "bg_wheelhouse_dusk_red_circle_marks",
            "reference_usage": "skipped_close_framing",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": true,
            "constraints": [
              "this shot uses close framing — the background reference is intentionally omitted at composition time",
              "do not mention the existing room, the chain_bg reference image, or any item from the owned_objects list",
              "do not use any of: 'the existing X', 'from the reference', 'use the X from the reference', 'preserving the same room perspective', 'maintaining the reference's framing', 'do not generate a new X' — these phrases are invalid when the reference is absent",
              "do not use any of: 'matching the reference camera', 'deviation from reference', 'same … angle as the reference', 'match the reference framing' — Rule A close-framing invalidation: close framing has no reference camera to match or deviate from",
              "compose the close subject from its immediate surroundings only — use [L##: ...] block for frame-edge surface description"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [
              {
                "element_id": "red_circular_wall_markings",
                "element_type": "environment_state",
                "character_name": "",
                "description": "Dense, ritualistic red circular markings are painted tightly across the dim walls of the boat's cabin. The crude circles spread edge-to-edge on the bulkheads, standing out as deep, ominous accents against the cold gray-green shadows of the grimy interior.",
                "applies_to_shots": [
                  4,
                  6,
                  11
                ],
                "source_facts": [
                  "조타실안 수많은 원. 빙글빙글 돌고 여기에서 또다시 발견된 붉은 원!",
                  "blurred red circular mark on the bulkhead",
                  "dense red circular paint markings spread edge to edge",
                  "muted red wall marks bleeding softly behind him"
                ],
                "visual_inferences": [
                  "The circles appear to be hastily painted or smeared, varying slightly in size but forming a dense, claustrophobic and occult pattern."
                ],
                "creative_decisions": [
                  "Using the red marks as a recurring background and midground motif across different focal lengths to steadily build tension."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 29,
                "shot_index": 4,
                "ref_usage": "",
                "kind": "location"
              }
            ],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [
              {
                "kind": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              },
              {
                "kind": "character_outlook",
                "id": "C03O04",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "readiness_policy": "skipped_by_policy",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": true,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "883571c39a50d790"
      },
      {
        "scene_index": 30,
        "heading": "S#30. 바다 / 해질 무렵",
        "beat_title": "미지의 안개 속으로 진입하는 배",
        "representative_moment": "짙은 안개 속으로 선수를 들이밀며 나아가는 소형어선의 뒷모습과 길게 이어진 하얀 항적.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "low trailing observer perspective just above the waterline, muted sunset gold smothered by gray fog",
            "t2i_prompt": "Photorealistic cinematic still. From a low observer perspective just above the waterline, a small fishing boat occupies the upper middle of the frame, its stern facing the camera. The boat's bow is half-swallowed by [L18: a dense, opaque gray fog wall ahead over the dark, rippled sea surface]. A long, foamy white wake stretches from the stern down toward the lower foreground, tearing through the black water. Muted sunset gold glows faintly behind the haze, smothered by the thick mist. Focus on the stern of the boat and the pale white wake guiding the eye into the eerie boundary.",
            "outfit_assignments": [],
            "source_facts": [
              "S#30 action: '세상과 경계를 지우는 것 같은 짙은 안개 띠. 서서히 안개 안으로 들어가는 인우의 배.'",
              "shot description: Rear view of the small fishing boat, its bow half-obscured by the dense fog wall, leaving a long white wake stretching behind the stern."
            ],
            "visual_inferences": [
              "The boat's stern occupies the upper-middle frame while the wake dominates the lower foreground due to the low waterline camera perspective.",
              "No characters are visible from this trailing distance."
            ],
            "creative_decisions": [
              "Low angle just above the waterline to emphasize the vastness of the sea and the menacing scale of the fog wall.",
              "Muted sunset gold contrasting with cool gray fog and black water to highlight the eerie, isolated mood of entering the unknown."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "f25e1182975dce12",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "1f35ae9e3340f350",
              "validator": "scene_detail_owned_objects.v1",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "none",
        "_shot_index": 3,
        "visible_entities": [
          "L18"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 30,
            "shot_index": 3
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "From a low trailing observer perspective just above the waterline, the camera tracks several boat lengths behind the small fishing boat in a WS low angle. The stern stays centered in the upper middle of frame, the bow half-swallowed by the fog wall ahead, while the long white wake stretches from the boat toward the lower foreground like a torn seam in the dark sea.",
            "lighting_mood": "Muted sunset gold is smothered by gray fog and black water, leaving only the wake as a pale, uneasy guide into the mist.",
            "perception_mode": "direct",
            "spatial_consistency": {
              "camera_frame_rule": {
                "applies_when": "camera_direction specifies a vertical height/position (low / hip-height / overhead / ground level / quay level / floor level / from above) and the scene description includes frame-edge surfaces",
                "consistency_check_summary": "elements described inside the frame must be at vertical/depth positions consistent with the camera's height — low cameras place feet/lower-body in the lower frame, overhead cameras place head/shoulders in upper frame, hip-height cameras place torso in frame with head/feet implied beyond edges",
                "core_principles": [
                  "low cameras at ground/quay/floor level place feet and lower legs in the lower frame; chest-height frame-edge surfaces behind the subject are geometrically forbidden",
                  "overhead high angles from above place the top of head and shoulders in the upper-center frame; floor surfaces visible below feet at the lower frame edge are forbidden",
                  "hip-height side cameras place torso and arms in frame; head and feet are implied beyond frame edges, not both fully visible together",
                  "full body shots require either eye-level or high-angle framing where head and feet both fit; low cameras + full silhouette in one frame is forbidden"
                ],
                "forbidden_combinations": [
                  "low camera at ground/quay/floor level + frame-edge surface described at chest/waist height behind the subject",
                  "overhead high angle + feet visible at the lower frame edge with floor visible below them",
                  "hip-height side camera + full silhouette including head and feet visible in same frame",
                  "low angle looking up + ceiling and floor both fully visible in same frame"
                ],
                "recommended_phrasings": [
                  "low observer at ground level, shoes and lower legs occupying the lower frame, upper body extending toward the upper edge",
                  "overhead high angle from above, the top of head and shoulders fill the upper-center frame, the floor visible around them",
                  "at hip height beside the subject, torso and arms in frame, head and lower legs implied beyond frame edges",
                  "low angle looking up at the figure, ceiling lamps in the upper frame, the wall behind extending downward"
                ],
                "self_check_steps": [
                  "identify the camera position phrasing (height + angle)",
                  "list elements described inside the frame (foreground prop, body parts, background surfaces)",
                  "verify each element's vertical/depth placement is geometrically consistent with that camera position",
                  "if mismatch found, adjust frame-edge description to match camera position"
                ],
                "rationale_summary": "T2I 모델은 카메라 위치와 frame 안 요소 vertical 좌표 모순 시 절충 합성 — 인물이 가구에 박히거나 신체 일부가 잘리거나 prop 이 떠 있는 부자연스러운 결과"
              },
              "fg_bg_shared_anchor_rule": {
                "applies_when": "two characters share an interaction (dialogue / handing object / facing / co-watching) AND are described as foreground/background separated AND a shared surface or spatial anchor exists between them (bench / table / door / room)",
                "shared_anchor_required": "describe the shared surface or spatial anchor explicitly so both figures occupy the same physical space",
                "forbidden_phrasings": [
                  "fg/bg separation without naming the shared surface/space",
                  "isolated foreground hand reaching with another figure 'in the rear background' (no shared anchor named)",
                  "one figure 'in the foreground' and another 'in the back' without naming the room/surface they share"
                ],
                "recommended_keywords": [
                  "shared bench / shared table / shared surface",
                  "the same X they both occupy (X = bench, sofa, table, doorway, room)",
                  "across the X from each other",
                  "<prop> crosses the air between them (interaction prop visually links the two)",
                  "the <surface> sharing a single line of sight from the camera"
                ],
                "self_check_steps": [
                  "do the two characters share a single interaction (dialogue, handing, facing, co-watching)?",
                  "is one foreground and the other background-separated?",
                  "is the shared surface/space anchor explicitly named?",
                  "if not, add one of the recommended keywords"
                ],
                "rationale_summary": "fg/bg 분리만 묘사하면 모델이 두 인물을 다른 공간에 배치하거나 인터랙션이 끊긴 어색한 합성 — 공유 anchor 명시가 같은 물리적 공간 강제"
              },
              "primary_framing_rule": {
                "applies_when": "framing_scale ∈ {close, medium, insert} from `_derive_framing_scale()` (render_strategy.framing_scale, production enum). wide-class shots are mapped to medium and use `wide_medium_rules`. The LLM may *additionally* infer wide from prose keywords (`wide shot` / `전경` / `establishing` / `aerial`) for stricter checking — these are LLM-side reasoning hints, not the production framing_scale enum (RO-4 binding — card-only classification).",
                "framing_scale_keywords": {
                  "_comment": "LLM-facing reasoning hints — not the production framing_scale enum. card framing_scale is close/medium/insert from _derive_framing_scale(); wide is LLM hint only (RO-4 / RO-5 binding). Canary close-framing shot scope filter uses production _CLOSE_FRAMING_RE (9-entry alignment) — separate from this 11-entry LLM list.",
                  "close": [
                    "close-up",
                    "CU",
                    "MCU",
                    "ECU",
                    "XCU",
                    "extreme close-up",
                    "클로즈업",
                    "손가락이",
                    "손이",
                    "눈이",
                    "얼굴이"
                  ],
                  "wide": [
                    "wide shot",
                    "establishing",
                    "aerial",
                    "전경",
                    "전신"
                  ],
                  "medium_default": "medium / mid / two-shot — anything not matching close/wide keywords"
                },
                "close_framing_rules": {
                  "primary_subject_only_fully_visible": "the named primary subject (hand / face / object) is the only fully visible entity in the frame",
                  "other_entities_appearance_options": [
                    "partial visibility — shoulder slice / arm slice / sleeve / fingertips entering the frame edge",
                    "soft background — out-of-focus silhouette via shallow depth of field",
                    "absent — not rendered at all even if listed in visible_entities"
                  ],
                  "forbidden_combinations": [
                    "same character's full body and body-part close-up in the same frame (view mixing)",
                    "two characters both rendered as sharp close-up faces in the same frame"
                  ],
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "wide_medium_rules": {
                  "primary_and_secondary_both_visible": "primary subject + secondary entities all renderable",
                  "body_part_close_up_forbidden": "body-part close-up of any character is forbidden in wide/medium framing — close-up belongs in a separate shot",
                  "fg_bg_separation_requires_shared_anchor": "if two characters are rendered fg/bg separated, also apply fg_bg_shared_anchor_rule",
                  "body_part_focus_cross_ref": "see id_policy.body_part_focus_rule (G4.3 sub-field — body-part Focus 시 C##O## 금지)"
                },
                "self_check_steps": [
                  "identify framing_scale from camera_direction",
                  "if close: verify only primary subject is fully visible; other entities are partial / soft bg / absent",
                  "verify same character's full body + body-part close-up are not combined",
                  "if close: verify two faces are not both rendered sharp"
                ],
                "rationale_summary": "한 shot 안 framing scale 가 일관되어야 모델이 단일 카메라 셔터로 해석. close/wide/medium 혼합은 multi-camera 합성으로 인식되어 인물 분열 / 부위 분리 / 시점 혼합 합성"
              },
              "rationale_summary": "Rule F/G/J 의 spatial consistency 룰 통합 — camera 위치 vs frame edge 모순, fg/bg 분리 시 공유 anchor 누락, framing scale 혼합. 3 sub-rule 모두 single t2i_prompt = single camera shutter 원칙의 spatial 측면."
            },
            "constraints": [
              "one prompt captures one still moment",
              "do not combine full-body framing and body-part close-up in one frame",
              "camera position and frame-edge elements must be geometrically consistent — see render_strategy.spatial_consistency.camera_frame_rule",
              "two characters in fg/bg separation must share a named surface/space anchor — see render_strategy.spatial_consistency.fg_bg_shared_anchor_rule",
              "framing_scale binds entity rendering — close = primary subject only fully visible, wide/medium = no body-part close-up — see render_strategy.spatial_consistency.primary_framing_rule"
            ]
          },
          "id_policy": {
            "allowed_base_entity_ids": [
              "L18"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "photo/poster/screen/mirror/reflection/projection depiction",
              "unregistered extra"
            ],
            "demographic_fallback_required": true,
            "face_identifiability_rule": {
              "use_entity_id_when": [
                "face identifiable: front, profile, three-quarter, or eyes closed",
                "OTS framing with any visible face"
              ],
              "common_noun_required_when": [
                "back to camera with no visible face",
                "silhouette or blurred outline only",
                "OTS with only back of head/shoulder, face entirely hidden"
              ],
              "id_use_summary": "C##/C##O## allowed only when face is identifiable AND framing is not dominated by an isolated body part",
              "rationale_summary": "C## 사용 시 얼굴 ref 가 자동 결합 — 얼굴이 안 보이는 인물에 C## 부착하면 ref 와 본문 묘사가 충돌"
            },
            "body_part_focus_rule": {
              "trigger_phrases": [
                "focus on",
                "close on",
                "tight on",
                "detail on"
              ],
              "applies_to": "any body part (face or non-face) when the framing is dominated by that part",
              "id_use": "forbidden — use a common noun + demographic descriptor instead",
              "common_noun_examples_generic": [
                "a trembling hand",
                "an exposed forearm",
                "the furrowed brow of a figure in shadow",
                "the clenched jaw of a figure in profile"
              ]
            },
            "close_framing_face_phrasing": {
              "forbidden_phrases": [
                "face filling the entire frame",
                "face filling nearly the entire frame",
                "her face fills the frame",
                "shoulders only entering the frame",
                "head snapped sideways",
                "his face fills the frame"
              ],
              "recommended_phrasings": [
                "tight framing on her face, upper body in frame",
                "close on his expression, shoulder visible at the edge",
                "head-and-shoulder composition, face occupying the center-right",
                "face in profile at the left, neck and upper chest visible"
              ],
              "rationale_summary": "T2I 모델은 얼굴이 프레임 전체를 가득 채우는 묘사를 받으면 몸과 분리된 얼굴만 공중에 떠 있는 합성을 낸다 — 어깨/상반신 일부를 프레임 안에 유지"
            },
            "reproduction_surface_rule": {
              "applies_to_surfaces": [
                "photograph",
                "poster",
                "painting",
                "portrait",
                "monitor",
                "TV",
                "mirror",
                "window reflection",
                "projection"
              ],
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면 속 얼굴이 사진 밖 실물 크기로 합성됨"
            },
            "demographic_descriptor_policy": {
              "required_on_first_appearance": true,
              "applies_to_id_forms": [
                "C##",
                "C##O##"
              ],
              "token_count_range": [
                1,
                2
              ],
              "components": {
                "ethnicity": [
                  "Asian",
                  "East Asian",
                  "South Asian",
                  "Southeast Asian",
                  "Black",
                  "Middle Eastern",
                  "Hispanic",
                  "Latina",
                  "Latino",
                  "Caucasian"
                ],
                "gender": [
                  "man",
                  "woman",
                  "figure"
                ],
                "age_band": [
                  "in his/her 20s",
                  "in his/her 30s",
                  "in his/her 40s",
                  "young",
                  "middle-aged",
                  "elderly"
                ],
                "role_hint_from_outfit": [
                  "in worker uniform",
                  "in fisher workwear",
                  "in detective coat",
                  "in business suit"
                ]
              },
              "format_template_a": "C##O## in <옷 1-3 단어>, <demographic descriptor>",
              "format_template_b": "C##O##, <demographic descriptor>, <자세 표현>",
              "source_priority": [
                "visual_world_rules.region",
                "visual_world_rules.era",
                "shot context outfit hint"
              ],
              "scenario_dependency_ban": "no work-specific proper nouns (character names, place names) — common nouns only",
              "regional_consistency": "한 작품 내 region 표기는 일관 — 선택된 세분 ethnicity (East Asian / Southeast Asian / South Asian / Caucasian / Black / Hispanic / Latina|o / Middle Eastern) 를 작품 단위 고정"
            },
            "constraints": [
              "visible real people with outfit binding use C##O## composite IDs",
              "do not use C##O## for a reproduced face inside a photograph, poster, painting, portrait, monitor, TV, mirror, window reflection, or projection — use a common noun + demographic descriptor instead",
              "do not use C##O## for isolated body-part focus (face or non-face) — when the framing is dominated by a single body part triggered by 'focus on / close on / tight on / detail on' phrasing, use a common noun (hand / eye / brow / jaw / etc.)",
              "for close-up of a face, do not use 'face filling the entire frame' / 'face filling nearly the entire frame' / 'her face fills the frame' / 'shoulders only entering the frame' / 'head snapped sideways' — keep upper body or shoulder edge visible in the frame ('tight framing on her face, upper body in frame')",
              "when introducing a C##O## (or C##) for the first time in this t2i_prompt, append a 1-2 token demographic descriptor (ethnicity + age band, or ethnicity + role hint) consistent with visual_world_rules.region — never use scenario-specific proper nouns",
              "use the format `C##O## in <옷 1-3 단어>, <demographic descriptor>` or `C##O##, <demographic descriptor>, <자세 표현>` — main characters with face refs may include this descriptor too (image inject takes precedence)"
            ]
          },
          "background_binding": {
            "mode": "not_applicable",
            "bg_id": null,
            "reference_usage": "none",
            "owned_objects": [],
            "camera_reference": null,
            "close_framing_skips_background_ref": false,
            "constraints": [
              "no chain_bg is bound to this shot — describe the location freely without referencing a background image",
              "do not use 'from the reference' / 'the existing room' / 'preserving the room' style phrasing — there is no reference image to refer to"
            ]
          },
          "continuity_elements_used": {
            "fixed_elements": [],
            "previous_shot_refs": [],
            "forward_zoom_targets": [],
            "cross_shot_id_substitution_rule": {
              "applies_when": "fixed_elements[i].character_name maps to a C## allowed in id_policy.allowed_base_entity_ids OR a C##O## in id_policy.allowed_outlook_pairs",
              "substitution": "replace the common-noun person reference inside fixed_elements[i].description (e.g. 'An Asian man' / 'a woman' / 'a figure') with the matched C## or C##O##",
              "double_description_forbidden": "do not describe the same character with both a C##/C##O## and a separate common-noun person reference in the same t2i_prompt — produces phantom extra figures",
              "preserve_pose_unchanged": "fixed_elements[i].description 의 자세/위치/상태 keyword 는 그대로 유지 — 인물 참조 토큰만 교체",
              "no_repeat_after_pose": "이미 t2i_prompt 안에서 같은 인물의 자세를 묘사했다면, 같은 fixed_element description 을 별도 문장으로 반복하지 마라",
              "rationale_summary": "fixed_element description 이 보통명사로 작성되었고 같은 인물의 C## 묘사가 t2i_prompt 에 따로 있으면 모델이 두 인물로 인식해 인원 수가 부풀려진다"
            },
            "ref_usage_constraints": {
              "zoom_in_detail": {
                "scope_summary": "same moment, same space, same primary subject; camera moved/zoomed onto a sub-region only",
                "forbidden_additions": [
                  "new persons not in the prior shot",
                  "new props not in the prior shot",
                  "new background elements not in the prior shot",
                  "new poses, gestures, or expressions not in the prior shot",
                  "different moments of action"
                ],
                "required_phrasings": [
                  "describe only the focused sub-region in concrete terms",
                  "explicitly state the same environment (lighting, background) is preserved",
                  "use continuity tokens such as 'the same figure', 'the same room', 'the same surface'"
                ],
                "body_part_cross_ref": "id_policy.body_part_focus_rule applies — when the focused sub-region is a body part, use a common noun + demographic descriptor (no C##/C##O##)",
                "rationale_summary": "zoom_in_detail = 앞 샷의 특정 영역을 확대한 '같은 사진'. 새 정보 추가 금지 — 현재 프레임이 보여주는 부분만 정확히 묘사"
              },
              "exact_background": {
                "scope_summary": "same room, possibly different moment or angle; background carries through, persons/actions are described fresh for this shot",
                "permitted_additions": [
                  "new persons, poses, expressions, actions for this shot's moment",
                  "different camera angle within the same room"
                ],
                "background_consistency_rule": "background elements established in the prior shot must remain consistent (no contradictory walls, doors, furniture)",
                "ignore_keep_handled_elsewhere": "ignore/keep directives are auto-applied via reference image processing — do not echo them inside t2i_prompt",
                "rationale_summary": "같은 공간, 새 순간 — 인물·동작은 이 샷의 instantaneous moment 로 묘사하되 배경은 prior shot 과 일관"
              },
              "atmosphere_reference": {
                "scope_summary": "different room/floor/angle entirely — only mood/tone is shared with the prior shot",
                "forbidden_imports": [
                  "furniture layout from the prior shot",
                  "wall positions from the prior shot",
                  "specific architectural features from the prior shot"
                ],
                "required_handling": "describe the new space's background directly in this shot; carry through only color tone or lighting mood (e.g. 'cool gray ambient' / 'dim warm pool')",
                "rationale_summary": "atmosphere_reference = 분위기 reference 만, 레이아웃 reference 아님. 가구·벽 배치를 끌어오면 모델이 두 공간을 한 공간으로 합성"
              }
            },
            "view_consistency": {
              "framing_scope_options": [
                "third_person",
                "close_up"
              ],
              "single_camera_rule": "one t2i_prompt = one camera position. do not combine a third-person full-body description and a close-up of the same character's body part.",
              "third_person_handling": "describe persons with full or partial body in frame; their hands/arms remain attached to their bodies (no isolated hand close-up alongside the full body)",
              "close_up_handling": "isolated body part (hand, wrist, fingers, brow) — do not include the same character's full body in the same prompt",
              "focus_on_is_focus_area_not_view_switch": "the phrasing 'Focus on ...' specifies the area of focus while the camera position remains the same; it is NOT a view switch",
              "id_policy_cross_ref_for_body_part_focus": "id_policy.body_part_focus_rule applies — body-part Focus 시 C##O## 금지 (얼굴 ref 합성 차단)",
              "prop_orientation_rule": "props the character holds must face one direction only — toward camera OR toward character, not both. forbidden contradiction example: 'photo angled partly toward the camera' + 'eyes fixed on the photograph' (양방향 모순)",
              "mixing_forbidden": true,
              "rationale_summary": "single t2i_prompt = single camera shutter at 1/1000s. 시점 혼합은 모델이 두 카메라 합성으로 해석해 인물이 두 번 등장 또는 부위 분리"
            },
            "constraints": [
              "fixed_elements are continuity inputs derived from prior shots — treat them as binding context, not as source facts to be invented",
              "when fixed_elements[i].character_name maps to a C## in id_policy.allowed_base_entity_ids or a C##O## in id_policy.allowed_outlook_pairs, replace the common-noun person reference inside fixed_elements[i].description with that C##/C##O## and do not also write a separate common-noun reference for the same character in the same t2i_prompt",
              "previous_shot_refs with ref_usage='zoom_in_detail' must preserve the same moment, the same space, and the same lighting as the prior shot — no new persons, props, background elements, poses, or moments",
              "previous_shot_refs with ref_usage='exact_background' may freely describe this shot's persons, poses, and moment, but must keep background elements consistent with the prior shot's established room",
              "previous_shot_refs with ref_usage='atmosphere_reference' must describe a new space directly — do not import furniture layout or wall positions from the prior shot, only mood/tone",
              "one t2i_prompt = one camera position — do not combine a third-person full-body description and a close-up of the same character's body part; 'Focus on' specifies the area of focus, not a view switch (id_policy.body_part_focus_rule applies for any body-part Focus)",
              "props held by characters face one direction only — toward camera OR toward character, never both"
            ]
          },
          "asset_requirements": {
            "required_refs": [],
            "forbidden_refs": [],
            "readiness_policy": "not_applicable",
            "constraints": [
              "do not imply or describe a reference image that is not listed in required_refs (no phantom references)",
              "if a required character_outlook ref is missing, scene image generation must block before the image stage"
            ]
          },
          "_card_metadata": {
            "lift_status": {
              "rule_a_lifted": true,
              "rule_c_lifted": true,
              "rule_e_lifted": false,
              "rule_h_lifted": true,
              "id_policy_composite_lifted": true,
              "id_policy_close_framing_face_lifted": true,
              "id_policy_reproduction_surface_lifted": true,
              "continuity_cross_shot_id_substitution_lifted": true,
              "continuity_ref_usage_constraints_lifted": true,
              "continuity_view_consistency_lifted": true,
              "continuity_constraints_extended": true,
              "spatial_camera_frame_lifted": true,
              "spatial_fg_bg_shared_anchor_lifted": true,
              "spatial_primary_framing_lifted": true,
              "spatial_constraints_extended": true
            },
            "rule_source": {
              "camera_rule": "A",
              "owned_rule": "C",
              "close_skip_rule": "E",
              "demographic_rule": "H",
              "composite_id_rule": "C## 사용 규칙 — 참조 이미지 연동 (v17:32-64)",
              "close_framing_face_rule": "극단 클로즈업 표현 (v17:66-85)",
              "reproduction_surface_rule": "사진·포스터·화면·거울 속 인물 규칙 (v17:133-146)",
              "cross_shot_id_substitution_rule": "v20/scene_detail/system.md#continuity-section",
              "ref_usage_rule": "v20/scene_detail/system.md#continuity-section",
              "view_consistency_rule": "v20/scene_detail/system.md#continuity-section",
              "camera_frame_rule": "v20/scene_detail/system.md#spatial-consistency",
              "fg_bg_shared_anchor_rule": "v20/scene_detail/system.md#spatial-consistency",
              "primary_framing_rule": "v20/scene_detail/system.md#spatial-consistency"
            }
          }
        },
        "render_prompt_card_hash": "3cf59978048ff84e"
      }
    ]
  },
  "step_id": "scene_detail",
  "run_id": "c8a46fa4-5683-44bc-8570-a61788cd30e8",
  "resolved_model": "gemini-pro",
  "updated_at": "2026-05-07T11:20:09.549553+00:00",
  "project_config_snapshot": {}
}