{
  "status": "completed",
  "completed_count": 57,
  "applicable_count": 57,
  "failed_count": 0,
  "schema_version": 8,
  "config_hash": "f12296572b3e973d",
  "data": {
    "scenes": [
      {
        "scene_index": 1,
        "heading": "S#1. 숲속 / 해 질 무렵",
        "beat_title": "정체불명의 인영에게 덮쳐지는 공포의 순간",
        "representative_moment": "겁에 질린 표정으로 위를 올려다보는 C06의 얼굴 클로즈업.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "A tighter-than-stage CU low angle through-object frame with dim blue-green dusk light",
            "t2i_prompt": "Photorealistic cinematic still. A tight low-angle observer view, framing from just under the chin. C06O11 in a blood-stained outdoor jacket, Korean man in his 20s, young male face with dark hair and dark eyes, positioned at a three-quarter angle from the left, jaw lifted upward, eyes looking up in terror. Tight framing on his expression, shoulder visible at the edge. His muddy cheek and trembling lower eyelid are sharp in the center-left frame. Wet, rain-damp forest leaves and torn grass brush the left and right edges in the extreme foreground. The darkening dusk tree canopy dissolves into a shallow depth of field in the soft overhead negative space. Dim blue-green dusk ambient light seeps through the foliage, washing his skin, with a faint red-brown glow reflecting on his cheek.",
            "outfit_assignments": [
              {
                "character_id": "C06",
                "outlook_id": "O11"
              }
            ],
            "source_facts": [
              "S#1 action: '잔뜩 겁에 질린 남자 표정(T.S)'",
              "Shot description: 겁에 질린 표정으로 위를 올려다보는 한국인 남자의 얼굴 클로즈업."
            ],
            "visual_inferences": [
              "fixed_element dusk_forest_environment: damp crushed leaves, torn grass, and dim blue-green dusk light in a humid forest setting",
              "The muddy forest floor environment heavily soils the character's cheek from the prior chase"
            ],
            "creative_decisions": [
              "Through-object foreground framing with wet leaves at the edges to heighten the claustrophobic observer perspective",
              "Cool blue-green dusk light contrasted against faint warm red-brown tones on the skin to emphasize horror elements"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "3136eb4855a031d3",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "1d13fa52042f2e56",
              "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"
        ],
        "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 low angle through-object frame holds less than a foot from the 20대 한국인 남자’s face, crouched just under his chin as wet leaves cut across the left and right edges. The man’s face occupies the center-left 70% of frame, his muddy cheek and trembling lower eyelid sharp while the foliage and dusk canopy above dissolve into shallow DOF; this is an observer view, not his POV, watching him look upward in terror.",
            "lighting_mood": "Dim blue-green dusk light seeps through the canopy, with humid low contrast and faint red-brown blood tones warming the man’s skin.",
            "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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C06",
                "outlook_id": "O11"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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, humid forest setting at dusk. The ground and immediate surroundings are covered in damp, crushed leaves, torn grass, and patches of dark mud. The ambient lighting is dim and filtered, casting deep blue-green and damp green-brown tones across the vegetation.",
                "applies_to_shots": [
                  2,
                  4
                ],
                "source_facts": [
                  "Scene heading states: 숲속 / 해 질 무렵 (Forest / Dusk)",
                  "Shot 2 mentions wet leaves, foliage, dusk canopy, and dim blue-green dusk light.",
                  "Shot 4 mentions torn grass, mud, crushed leaves, and damp green-brown forest tones."
                ],
                "visual_inferences": [
                  "The ground remains consistently muddy and disturbed from the chase and the ensuing physical struggle."
                ],
                "creative_decisions": [
                  "Emphasized the damp, blue-green and muddy brown color palette to visually unify the background across different camera framing distances."
                ],
                "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": "C06O11",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "77c410345b7119ea"
      },
      {
        "scene_index": 1,
        "heading": "S#1. 숲속 / 해 질 무렵",
        "beat_title": "정체불명의 인영에게 덮쳐짐",
        "representative_moment": "도망치던 C06이 바닥에 쓰러진 채 C10에게 강하게 짓눌리는 찰나의 정면 구도.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "ground-level medium shot on the frontal axis, handheld impact feel, deep blue-green and damp green-brown dusk tones",
            "t2i_prompt": "Photorealistic cinematic still. A ground-level medium frame. C10O14 in a black ragged cloak, an unidentifiable humanoid silhouette showing the variant features carried from stable_traits — black outline with no facial details visible, with back to camera, lunging down hard from above as the figure exerting force. C10O14 is forcefully pinning down C06O11 on the shared muddy forest ground. C06O11 in a bloody outdoor jacket, a 20s Korean man showing the variant features carried from stable_traits — young male face, black hair, and dark eyes, facing the camera, pinned on his back as the figure receiving the force, eyes looking up. Focus on the violent physical impact and the downward-pressing weight on the trapped man. The camera is at eye-level to the fallen body on the frontal axis, placing the muddy impact surface and crushed leaves in the lower frame. An abandoned mud-smeared shoe rests near the lower frame edge, its toe angled away from the camera and heel turned toward the impact area. The scene is lit by dying dusk filtered through leaves, casting deep blue-green and damp green-brown tones across the torn grass. C10O14's body volume is preserved with subtle backlight tracing the shoulders against the dim ambient fill, appearing nearly black against the forest background.",
            "outfit_assignments": [
              {
                "character_id": "C06",
                "outlook_id": "O11"
              },
              {
                "character_id": "C10",
                "outlook_id": "O14"
              }
            ],
            "source_facts": [
              "S#1 action: 어디선가 쏜살같이 달려온 검은 인영이 남자를 덮친다.",
              "shot description: A ground-level MWS eye-level-to-the-fallen-body handheld frame... The 검은 인영 crashes into the center foreground, its dark mass partially blocking the man beneath",
              "shot description: C10 lunging down hard, back to camera. C06 pinned on back, facing camera, eyes up."
            ],
            "visual_inferences": [
              "fixed_element: dusk_forest_environment (dense humid forest, crushed leaves, torn grass, dark mud) persists as the shared physical surface",
              "the abandoned shoe is integrated into the foreground mud to anchor the depth of the fallen body",
              "the interaction requires role-asymmetric physical contact tokens (exerting force, pinning down)"
            ],
            "creative_decisions": [
              "used a ground-level eye-line matching the pinned victim to emphasize the oppressive weight of the black silhouette",
              "applied deep blue-green tones to the ambient light to contrast with the solid black outline of the attacker without flattening its volume"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "6e8057b5c3d0aa75",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "09cbdb5ddc8f21e8",
              "validator": "scene_detail_owned_objects.v1",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 1,
        "dependency_reason": "이전 샷(Shot 2)에서 확립된 숲속 바닥의 축축한 진흙, 찢긴 풀, 푸른빛이 도는 녹갈색 황혼 조명(dusk_forest_environment)과의 시각적 연속성 유지.",
        "_shot_index": 4,
        "visible_entities": [
          "C06",
          "C10",
          "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 eye-level-to-the-fallen-body handheld frame lands on the frontal axis about three feet from the 20대 한국인 남자’s upper body at the exact impact. The 검은 인영 crashes into the center foreground, its dark mass partially blocking the man beneath; the man’s shoulders and face are pinned in the lower half while torn grass, mud, and crushed leaves shake around the frame.",
            "lighting_mood": "The handheld impact is lit by dying dusk filtered through leaves, with the silhouette nearly black against damp green-brown forest tones.",
            "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",
              "C10",
              "L01"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C06",
                "outlook_id": "O11"
              },
              {
                "character_id": "C10",
                "outlook_id": "O14"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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, humid forest setting at dusk. The ground and immediate surroundings are covered in damp, crushed leaves, torn grass, and patches of dark mud. The ambient lighting is dim and filtered, casting deep blue-green and damp green-brown tones across the vegetation.",
                "applies_to_shots": [
                  2,
                  4
                ],
                "source_facts": [
                  "Scene heading states: 숲속 / 해 질 무렵 (Forest / Dusk)",
                  "Shot 2 mentions wet leaves, foliage, dusk canopy, and dim blue-green dusk light.",
                  "Shot 4 mentions torn grass, mud, crushed leaves, and damp green-brown forest tones."
                ],
                "visual_inferences": [
                  "The ground remains consistently muddy and disturbed from the chase and the ensuing physical struggle."
                ],
                "creative_decisions": [
                  "Emphasized the damp, blue-green and muddy brown color palette to visually unify the background across different camera framing distances."
                ],
                "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": "C06O11",
                "policy": "required"
              },
              {
                "kind": "character_outlook",
                "id": "C10O14",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "ee2f037e76a946fe"
      },
      {
        "scene_index": 2,
        "heading": "S#2. 도시 뒷골목 / N",
        "beat_title": "드러난 정체와 치명적 흡혈",
        "representative_moment": "붉은 눈을 빛내며 여자의 목을 무는 C06의 클로즈업.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Eye-level close framing over the shoulder, isolating the tense contact point with a humid, claustrophobic palette.",
            "t2i_prompt": "Photorealistic cinematic still. Eye-level close framing over the shoulder. Both figures occupy the same immediate space. C07O13 in an ivory silk blouse, Korean woman in her 20s with black hair, positioned over the shoulder with back to camera, her shoulder and hair forming a blurred foreground wedge on the left, her face partially visible in profile with eyes closed. Across from her, C06O12 in a black hooded jacket, Korean man in his 20s with black hair, at three-quarter angle from the right, acts as the forceful initiator. His face occupies the upper center with his upper chest visible at the edge, biting her neck rigidly, his teeth pressed forcefully into her exposed limp neck at the lower right. C06O12's glowing red eyes stare into the void. The camera emphasizes his glowing eyes and the tense contact point. Sickly red eye-glow reflects on his cheek, mouth, and her pale neck. Sodium streetlight washes the figures in a humid, claustrophobic palette, with black shadows pooling at their jawlines. [: dark out-of-focus frame edges showing the cylindrical side of a weathered utility pole catching a thin rim of amber backlight, rain-slick pavement surface below reflecting a faint red glint].",
            "outfit_assignments": [
              {
                "character_id": "C06",
                "outlook_id": "O12"
              },
              {
                "character_id": "C07",
                "outlook_id": "O13"
              }
            ],
            "source_facts": [
              "S#2 action: 남자의 붉은 눈동자, 목의 살점을 뜯는 이빨… 푹 힘없이 쓰러지는 여자.",
              "shot description: close-up of the Korean man's face biting the woman's neck, red eyes glowing."
            ],
            "visual_inferences": [
              "C07's face partially visible in profile to allow C07O13 identifiability in OTS framing",
              "man acts as force initiator, gripping the woman rigidly in a tight embrace",
              "blood/bite mark implied through teeth pressed forcefully into the neck"
            ],
            "creative_decisions": [
              "eye-level CU OTS-like framing to emphasize the supernatural eye glow and violent contact point",
              "humid, claustrophobic palette with sodium streetlight and black shadows to heighten horror tension",
              "frame-edge L02 block used to replace skipped background with atmospheric edges"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "154a0dfe9bde597b",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "25467122504194fc",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "No direct visual dependency from previous scenes required for this close-up moment.",
        "_shot_index": 2,
        "visible_entities": [
          "C06",
          "C07"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 2,
            "shot_index": 2
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "An eye-level CU OTS-like frame sits just off the 20대 한국인 여자’s shoulder line, isolating the 20대 한국인 남자’s glowing red eyes, cheek, mouth, and the exposed side of her neck. The man’s face fills the upper center, his teeth pressed into the neck at lower right, while the woman’s shoulder and hair form a blurred foreground wedge on the left; the camera is almost motionless at the peak of the approach.",
            "lighting_mood": "Sodium streetlight and sickly red eye-glow create a humid, claustrophobic palette with black shadows swallowing the alley edges.",
            "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",
              "C07"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C06",
                "outlook_id": "O12"
              },
              {
                "character_id": "C07",
                "outlook_id": "O13"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "C06O12",
                "policy": "required"
              },
              {
                "kind": "character_outlook",
                "id": "C07O13",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "77f0b3deb1bd1cde"
      },
      {
        "scene_index": 3,
        "heading": "S#3. 현장 / D",
        "beat_title": "뉴스 화면 속 폴리스 라인과 수사 개시",
        "representative_moment": "폴리스 라인 너머로 흰 천에 덮인 시신과 현장을 지휘하는 C02가 뉴스 화면 구도로 잡힌다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "static eye-level through-device wide composition, muted gray-green overcast daylight",
            "t2i_prompt": "Photorealistic cinematic still. An eye-level news-camera composition positioned at a distance behind the P09 police line tape. The taut, rain-damp tape stretches horizontally across the lower foreground, its printed warning side facing the camera. Beyond the tape, the moisture-stained P10 white body cover rests center frame on the [L01: damp forest floor, leaf-covered and marked by footprints], its long side facing the camera to form a low mound. In the midground sharing the same forest clearing, C02O05 in a charcoal gray suit jacket, an Asian woman in her early 40s with dark hair and dark eyes, stands at three-quarter angle from the left, her torso bent over a clipboard, eyes fixed on the covered body. Beside her across the damp ground, an Asian man in his 40s in detective workwear remains crouched near the tarp in right profile, eyes fixed on the covered body. Overcast daylight washes the scene in a muted news-feed gray-green, with humid haze and low contrast making the white cover feel unnaturally stark against the dim ambient fill. Subtle digital compression artifacts overlay the image. Focus on the stark white P10 body cover anchoring the space between the figures.",
            "outfit_assignments": [
              {
                "character_id": "C02",
                "outlook_id": "O05"
              }
            ],
            "source_facts": [
              "S#3 action: '씬1, 2 현장을 감식하는 경찰, 폴리스 라인을 치고 시신을 수거한다.'",
              "shot description: frozen broadcast-like tableau of a covered body on the forest floor, Hye-su and Detective Kim arranged among investigators"
            ],
            "visual_inferences": [
              "through-device effect translated as subtle digital compression artifacts without drawing a physical TV frame",
              "Hye-su's posture (bent over clipboard) and Detective Kim's posture (crouched) established for the midground figures",
              "shared damp forest clearing anchor to connect the foreground tape, central body, and midground investigators"
            ],
            "creative_decisions": [
              "placed the police line tape horizontally across the lower foreground to act as a visual barrier and enforce the observer POV",
              "desaturated the environment colors to low-contrast gray-green to make the stark white body cover the dominant focal point"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "f4e4122e6a358e5b",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "b49efe6cdb275c94",
              "validator": "scene_detail_owned_objects.v1",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 1,
        "dependency_reason": "동일한 숲속 현장이므로 이전 씬의 분위기와 로케이션 질감을 참조하여 시공간적 연속성을 유지",
        "_shot_index": 1,
        "visible_entities": [
          "C02",
          "P09",
          "P10",
          "L01"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 3,
            "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 eye-level through-device news-camera composition holds about 15 meters outside the police line, using the tape as a sharp foreground boundary across the lower third. Beyond it, the covered body lies center frame on the forest floor while 혜수 and 김형사 are small figures among investigators, arranged in a frozen broadcast-like tableau with no reporter visible.",
            "lighting_mood": "Overcast daylight is flattened into a muted news-feed gray-green, with humid haze and low contrast making the white cover feel unnaturally stark.",
            "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",
              "P09",
              "P10",
              "L01"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C02",
                "outlook_id": "O05"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": [
              {
                "scene_index": 1,
                "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": "C02O05",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "d54388c33ce013f3"
      },
      {
        "scene_index": 4,
        "heading": "S#4. 옥탑방 / D",
        "beat_title": "불안감이 감도는 인천 변두리 전경",
        "representative_moment": "맑은 하늘 아래 인천 변두리의 낡은 빌라 옥탑방과 주변 주택가가 부감으로 내려다보인다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "A high angle EWS bird's eye establishing shot, 45-degree slant, muted coastal haze with pale blues and concrete grays",
            "t2i_prompt": "Photorealistic cinematic still. A high-angle bird's eye establishing shot looking down from far above at a 45-degree slant. Focus on the building and rooftop deck from the reference, positioned slightly right of center. The neighboring roof and alley from the reference form staggered low-rise blocks beneath the camera, drawing diagonal leading lines through the quiet residential streets. Clear daylight is tempered into a muted coastal haze, bright but softened by humidity, casting pale blue and concrete gray tones across the sun-faded facades. The vast coastal sky occupies the upper frame, emphasizing the sprawling, uneasy scale of the quiet neighborhood.",
            "outfit_assignments": [],
            "source_facts": [
              "scene text: '인천 변두리 부감, 오래된 빌라 옥탑방 전경.'",
              "shot description: A high angle EWS bird's eye establishing shot looks down over the Incheon outskirts.",
              "DP instructions: clear daylight tempered into a muted coastal haze, pale blues and concrete grays dominating."
            ],
            "visual_inferences": [
              "No characters are visible at this extreme distance and angle.",
              "The weathered rooftops and empty streets reflect the quiet unease of the town."
            ],
            "creative_decisions": [
              "Anchored the building, rooftop deck, neighboring roof, and alley explicitly to the reference to comply with the background_ref_attached binding rule.",
              "Omitted exact metric values like '90 meters' from the prompt, instead relying on 'far above' and the scale of the sky to dictate the extreme high angle.",
              "Used pale blue and concrete gray tones to firmly set the humid, slightly ominous coastal mood."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "89dd766a5b235d11",
              "owned_hash": "72d8385eab7418f1",
              "camera_direction_hash": "a107c71c016f43e0",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "building",
                  "violating_phrase": "the building and rooftop deck from the reference",
                  "reason": "The phrase explicitly references the existing building as part of the reference composition rather than instructing a new building to be created.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "rooftop deck",
                  "violating_phrase": "the building and rooftop deck from the reference",
                  "reason": "The rooftop deck is mentioned together with an explicit reference cue, so it is being anchored to the existing background rather than redrawn.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "neighboring roof",
                  "violating_phrase": "The neighboring roof and alley from the reference",
                  "reason": "The roof is described as part of the reference scene, which anchors it to the existing background instead of asking for a new one.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "alley",
                  "violating_phrase": "The neighboring roof and alley from the reference",
                  "reason": "The alley appears in a reference-linked phrase, so this is a scene anchor rather than a redraw instruction.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "No dependent scene",
        "_shot_index": 1,
        "visible_entities": [
          "L03"
        ],
        "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 high angle EWS bird's eye establishing shot looks down from roughly 90 meters at a 45-degree slant over the Incheon outskirts, holding before any descent begins. The old rooftop room sits slightly right of center among low-rise villas and narrow lanes, occupying only a small portion of the frame so the surrounding neighborhood and clear sky establish scale and unease.",
            "lighting_mood": "Clear daylight is tempered into a muted coastal haze, bright but not cheerful, with pale blues and concrete grays dominating.",
            "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": [
              "L03"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L03B05",
            "reference_usage": "exact_background",
            "owned_objects": [
              "AC condenser",
              "alley",
              "antenna",
              "building",
              "circle mark",
              "door",
              "drain",
              "entrance canopy",
              "laundry line",
              "neighboring roof",
              "parapet",
              "parking strip",
              "pipe",
              "rooftop deck",
              "rooftop unit",
              "satellite dish",
              "stair 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": "L03B05",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "a90641afd8001a5b"
      },
      {
        "scene_index": 5,
        "heading": "S#5. 옥탑방 안 / 실내",
        "beat_title": "TV를 끄다 놀란 민숙",
        "representative_moment": "수리영의 기척에 놀라 급히 리모컨을 쥔 손을 뻗은 민숙(C03)의 당황한 얼굴.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "MCU eye-level through-object hold, cold TV flicker contrasting with muted daylight",
            "t2i_prompt": "Photorealistic cinematic still. C03O06 in casual domestic clothes with a kitchen apron tied around her body, an East Asian woman in her early 40s with a middle-aged female face, black hair, and dark eyes. She is positioned at three-quarter angle from the right, occupying the center-right of the frame from the chest up. Her arm extends toward the camera, her hand gripping a television remote in the lower-left near foreground, a distinct, small, circular reddish mark visible on the exposed skin of her wrist. Her face holds a startled expression, her dark eyes fixed sharply on an unseen doorway off-camera. Cold flickering light from an unseen television screen washes the left side of her face and the remote. [L04: soft out-of-focus background immediately behind her, showing the blurred edge of a dim kitchenette counter, faint steam rising from a simmering pot, and a closed hazy window filtering thin daylight into the space]. Focus on her startled expression against the shallow background.",
            "outfit_assignments": [
              {
                "character_id": "C03",
                "outlook_id": "O06"
              }
            ],
            "source_facts": [
              "Scene text action: '앞치마를 두른 채 TV 뉴스를 뚫어지게 바라보는 민숙 표정이 어둡다.'",
              "Scene text action: '딸, 수리영이 방문을 열고 나오자 후다닥 TV를 끄는 민숙.'",
              "Shot description: 놀란 눈으로 TV 리모컨을 쥔 손을 뻗은 민숙의 멈춘 자세"
            ],
            "visual_inferences": [
              "fixed_element minsuk_attire_and_mark applied: casual domestic clothes with apron and reddish mark on wrist",
              "Cold flickering light on Minsuk's face infers the illumination from the active TV screen just off-camera",
              "Out-of-focus background elements like the kitchenette and steam infer the physical setting without violating close framing rules"
            ],
            "creative_decisions": [
              "Positioned the hand gripping the remote in the lower-left foreground to emphasize the sudden mid-action freeze while keeping the primary focus on her startled facial expression in the center-right",
              "Mixed cool TV light and soft daylight to create a tense domestic atmosphere, matching the DP's lighting direction"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "ecd26725e2cff414",
              "owned_hash": "336090adaa16cf9a",
              "camera_direction_hash": "0eecdac105ac324f",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "",
        "_shot_index": 3,
        "visible_entities": [
          "C03",
          "L04"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 5,
            "shot_index": 3
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "From the television side of the room, an MCU eye-level through-object push-in resolves into a hold, looking past 민숙’s outstretched remote hand in the near foreground toward her startled face. The remote and fingers sit in the lower-left foreground at natural scale, while 민숙 occupies the center-right with the kitchenette soft behind her, her eyes snapping toward 수리영’s doorway rather than the lens.",
            "lighting_mood": "Thin daylight from the small window mixes with cold TV flicker, keeping the rooftop room muted and humid with a faint domestic warmth under the alarm.",
            "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",
              "L04"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C03",
                "outlook_id": "O06"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L04B01",
            "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_attire_and_mark",
                "element_type": "character_state",
                "character_name": "Minsuk",
                "description": "A 40-year-old East Asian woman wearing casual domestic clothes with a kitchen apron tied around her body. A distinct, small, circular reddish mark is permanently present on her wrist skin.",
                "applies_to_shots": [
                  3,
                  7
                ],
                "source_facts": [
                  "Scene text action: '앞치마를 두른 채' (wearing an apron)",
                  "Scene text action: '민숙 손목에 보이는 작은 원(이하 표식)' (small circular mark on Minsuk's wrist)",
                  "Shot 3: MCU push-in framing of Minsuk",
                  "Shot 7: CU profile framing showing Minsuk's upper body in the shallow plane"
                ],
                "visual_inferences": [
                  "The apron and casual clothes are consistently worn from the living room area to the entryway.",
                  "The circular mark on the wrist is an inherent, unmoving physical trait that belongs to her overall character model."
                ],
                "creative_decisions": [
                  "Merged the persistent attire and physical mark into a single 'full/upper body' character state for Shots 3 and 7.",
                  "Excluded Shot 9 from fixed elements because it is a solitary extreme close-up, adhering to the rule to prevent conflicting zoom descriptions."
                ],
                "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": "C03O06",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "89652ce46253a878"
      },
      {
        "scene_index": 5,
        "heading": "S#5. 옥탑방 안 / 실내",
        "beat_title": "수리영의 출근 전 스킨십",
        "representative_moment": "수리영이 눈을 감고 민숙의 옷자락을 쥐며 코를 가까이 댄다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "A slightly high angle close framing, shallow depth of field, warm gray-beige tones.",
            "t2i_prompt": "Photorealistic cinematic still. A slightly high angle head-and-shoulder composition. C01O02 in a beige hoodie, a young Korean woman, occupies the left-center in left profile with eyes closed, leaning her face gently into the soft fabric of the figure beside her. Her hand clutches the fabric near the lower center. Just beyond her in the shallow background, sharing the same tight space, C03O06 in casual clothes with a kitchen apron, a middle-aged Korean woman, stands at three-quarter angle from the left, her soft smile paused as her eyes focus on C01O02. Soft room light and faint daylight blend into warm gray-beige tones across their features, casting a tender mood. [: A dim narrow wall and the vertical edge of a closed wooden door] sit immediately behind them, enclosing the frame. Focus on C01O02's peaceful expression and the gripped fabric in the foreground.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              },
              {
                "character_id": "C03",
                "outlook_id": "O06"
              }
            ],
            "source_facts": [
              "Scene text action: '엄마의 냄새를 맡으며... 아, 좋다.'",
              "Shot description: '수리영’s closed-eyed side profile fills the left-center, her hand clutching 민숙’s clothing near the lower center'",
              "Shot description: '민숙 remains just beyond her in the same shallow plane, softened but present.'"
            ],
            "visual_inferences": [
              "Fixed element applied: C03 wears casual clothes with a kitchen apron.",
              "Minsuk's wrist mark from the fixed elements is kept out of this tight head-and-shoulder frame.",
              "The phrasing 'sharing the same tight space' acts as the shared spatial anchor for the fg/bg separated figures."
            ],
            "creative_decisions": [
              "Used a shallow depth of field to isolate C01's tactile gesture in sharp focus while maintaining C03 as a soft presence in the background.",
              "Substituted the forbidden L04 with a generic [L01] boundary description to safely build the close framing's immediate surroundings.",
              "Applied a warm gray-beige lighting tone to reflect the tender intimacy of the interaction."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "eb12a2bdda026e06",
              "owned_hash": "336090adaa16cf9a",
              "camera_direction_hash": "ad505378fee2f4bf",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 5,
        "dependency_reason": "Shot 3과 동일한 공간 및 인물의 연속선상이나, close framing 정책에 따라 배경 참조를 제외하고 인접 표면만 묘사함.",
        "_shot_index": 7,
        "visible_entities": [
          "C01",
          "C03"
        ],
        "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": "Inside the narrow entry, a slightly high angle CU profile frame holds at shoulder height about 0.8 meters from 수리영 as she leans into 민숙’s side. 수리영’s closed-eyed side profile fills the left-center, her hand clutching 민숙’s clothing near the lower center, while 민숙 remains just beyond her in the same shallow plane, softened but present.",
            "lighting_mood": "Soft room light and faint daylight flatten into warm gray-beige tones, making the intimate gesture feel tender but still shadowed by the earlier news anxiety.",
            "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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O01"
              },
              {
                "character_id": "C01",
                "outlook_id": "O02"
              },
              {
                "character_id": "C03",
                "outlook_id": "O06"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L04B01",
            "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_attire_and_mark",
                "element_type": "character_state",
                "character_name": "Minsuk",
                "description": "A 40-year-old East Asian woman wearing casual domestic clothes with a kitchen apron tied around her body. A distinct, small, circular reddish mark is permanently present on her wrist skin.",
                "applies_to_shots": [
                  3,
                  7
                ],
                "source_facts": [
                  "Scene text action: '앞치마를 두른 채' (wearing an apron)",
                  "Scene text action: '민숙 손목에 보이는 작은 원(이하 표식)' (small circular mark on Minsuk's wrist)",
                  "Shot 3: MCU push-in framing of Minsuk",
                  "Shot 7: CU profile framing showing Minsuk's upper body in the shallow plane"
                ],
                "visual_inferences": [
                  "The apron and casual clothes are consistently worn from the living room area to the entryway.",
                  "The circular mark on the wrist is an inherent, unmoving physical trait that belongs to her overall character model."
                ],
                "creative_decisions": [
                  "Merged the persistent attire and physical mark into a single 'full/upper body' character state for Shots 3 and 7.",
                  "Excluded Shot 9 from fixed elements because it is a solitary extreme close-up, adhering to the rule to prevent conflicting zoom descriptions."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 5,
                "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": "C03O06",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "ceee6d75af9675c9"
      },
      {
        "scene_index": 5,
        "heading": "S#5. 옥탑방 안 / 실내",
        "beat_title": "손목의 붉은 표식",
        "representative_moment": "수리영의 시선이 민숙(C03)의 손목에 새겨진 또렷한 붉은 원형 표식(P08)에 잠시 머무는 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Subjective POV ECU Dutch angle macro, soft indirect daylight with shallow depth of field",
            "t2i_prompt": "Photorealistic cinematic still. Extreme close-up macro, subjective POV Dutch angle. Focus on the wrist of an Asian woman in her 40s, the lens positioned extremely close and nearly parallel to the skin surface. The wrist cuts diagonally across the middle third of the frame, the skin slightly damp with natural texture clearly visible. On the marked outer side of the wrist facing the lens, a distinct P08 circular mark stands out sharply, carrying a faint ominous red-brown undertone. The wrinkled sleeve cuff of a gray homewear top borders the frame edge, partly shadowing the skin. Behind the wrist, the background drops into a heavy blur, reducing an entry wall and fabric to abstract, muted beige-gray tones. Lit by soft indirect daylight, emphasizing the focal reveal of the mark while completely excluding any faces from the frame.",
            "outfit_assignments": [],
            "source_facts": [
              "S#5 action: 힐끗, 민숙 손목에 보이는 작은 원(이하 표식)을 본다.",
              "shot description: ECU macro of the distinct circular mark on Minsuk's wrist, seen from Suri-yeong's subjective glance."
            ],
            "visual_inferences": [
              "fixed_element body_full_pose: wrist held near hip, excluded from full body view",
              "the circular mark stands out vividly against the skin due to the macro focus"
            ],
            "creative_decisions": [
              "Dutch angle to emphasize the eerie, ominous feeling of the mark within an ordinary domestic setting",
              "Shallow depth of field to completely abstract the background and isolate the wrist detail, avoiding any background reference binding"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "952010adfe5c22ff",
              "owned_hash": "336090adaa16cf9a",
              "camera_direction_hash": "df94279b5c816c7b",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "",
        "_shot_index": 9,
        "visible_entities": [
          "C03",
          "P08"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 5,
            "shot_index": 9
          },
          "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 Dutch angle macro from 수리영’s glance sits less than 20 centimeters from 민숙’s wrist, the lens nearly parallel to the skin surface. Both faces are excluded; the wrist cuts diagonally across the middle third, and the clear circular mark is held sharp while the entry wall and fabric blur into abstract muted color behind it.",
            "lighting_mood": "The macro is lit by soft indirect daylight with a faint ominous red-brown undertone in the mark, turning a domestic detail into an occult warning.",
            "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",
              "P08"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C03",
                "outlook_id": "O06"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L04B01",
            "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": 5,
                "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": [],
            "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"
            ]
          },
          "render_contracts": [],
          "_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": "8a84f7da98e33b97"
      },
      {
        "scene_index": 6,
        "heading": "S#6. 해안가 도로 / D",
        "beat_title": "바닷바람을 맞으며 달리는 수리영",
        "representative_moment": "자전거를 타며 이어폰을 낀 채 눈을 감고 바닷바람을 만끽하는 수리영의 미소 띤 얼굴.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Front-right tracking eye-level CU, humid pastel lighting",
            "t2i_prompt": "Photorealistic cinematic still. Head-and-shoulder composition. C01O02 in a beige hoodie, an East Asian woman in her 20s with dark hair, dominates the center-right of the frame at a three-quarter angle from the right. Her eyes are closed in a relaxed smile, her dark hair flowing back in the direction opposite to her forward motion. Earphone wires run from her ears downward along her neck. A sliver of the curved front bicycle handlebar angles across the lower frame from her right side. The background is a motion-blurred coastal backdrop with a hazy pale blue sea horizon and a weathered sunlit road streaking softly behind her. Humid pastel daylight casts warm skin tones across her face.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#6 action: '수리영 자전거를 타고 해안가 도로를 달린다. 이어폰을 끼고 바다냄새를 실컷 마신다.'",
              "shot description: '눈을 감고 바닷바람을 맞는 수리영의 미소 띤 얼굴 클로즈업.'"
            ],
            "visual_inferences": [
              "head-and-shoulder composition to avoid floating-head artifact in close framing",
              "motion blur on the background to convey bicycle speed",
              "hair flowing backward to preserve the direction of motion"
            ],
            "creative_decisions": [
              "humid pastel daylight palette with pale blue sea tones and warm skin highlights to emphasize a brief sense of freedom",
              "sliver of handlebar at the lower edge to ground the subject's posture without violating close framing rules"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "fbebcc3d76d22ce1",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "9ccd15073d6e3bb2",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "No dependent visual scene",
        "_shot_index": 2,
        "visible_entities": [
          "C01"
        ],
        "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 front-right tracking CU eye-level shot moves backward at bicycle speed, holding 수리영’s face close while the coast streaks softly behind her. Her closed eyes and relaxed smile occupy the center-right 70% of frame, with a sliver of handlebar and shoulder movement at the lower edge to preserve the sensation of motion.",
            "lighting_mood": "Bright coastal daylight is softened into a humid pastel palette, with pale blue sea light and warm skin tones giving her a brief sense of freedom.",
            "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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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"
            ]
          },
          "render_contracts": [],
          "_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": "f61d5418f68dc299"
      },
      {
        "scene_index": 7,
        "heading": "S#7. 유기견 보호원 / D",
        "beat_title": "맹견의 분노",
        "representative_moment": "C08이 철창 안에서 이빨을 드러내며 앞발을 들고 서서 사납게 짖고 있는 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Handheld medium low-angle shot, cold fluorescent lighting with a grim coastal-gray cast and damp reflections.",
            "t2i_prompt": "Photorealistic cinematic still. C08O00, a large dog showing its stable traits—short dark brown coat, broad head, strong jaw, and dark eyes—facing the camera with its raised paws hooked on the damp, vertical metal cage bars from the reference. Handheld medium shot, low angle from inside the enclosure looking outward, the bottom frame edge capturing the wet concrete floor. The cage bars cut the foreground into hard vertical strips. C08O00 is baring its teeth, eyes fixed on the camera. Cold fluorescent light mixes with damp concrete reflections, casting a grim coastal-gray tone and pale reflections on the wet metallic surfaces. Focus on the tension and aggressive stance of the dog pressing against the barrier.",
            "outfit_assignments": [
              {
                "character_id": "C08",
                "outlook_id": "O00"
              }
            ],
            "source_facts": [
              "S#7 action: '사납게 짖으며 날뛰는 로트와일러 T.S'",
              "Shot 1 description: '이빨을 드러낸 채 앞발을 들고 서 있는 맹견 로트와일러.'",
              "Camera description: dog's raised paws grip the metal at mid-frame and its bared teeth sit just above center, confronting the lens directly."
            ],
            "visual_inferences": [
              "fixed_element damp_kennel_bars: Thick, vertical metal cage bars in a damp, humid room",
              "low angle perspective to emphasize the intimidating size and threat of the animal"
            ],
            "creative_decisions": [
              "Replaced specific numerical camera height measurements with relative frame positioning (bottom frame edge capturing the floor) to avoid scaling distortion",
              "Anchored the cage bars to 'from the reference' to ensure correct background blending without duplication",
              "Carried over all specified stable traits for the dog to guarantee accurate physical representation"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "fe7436004001658e",
              "owned_hash": "456d19eb30dd6a5c",
              "camera_direction_hash": "3c07a1918b3a748c",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "cage",
                  "violating_phrase": "cage bars from the reference",
                  "reason": "The phrase refers to the existing cage as a reference anchor and does not explicitly instruct creating or adding a new cage.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "N/A",
        "_shot_index": 1,
        "visible_entities": [
          "C08"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 7,
            "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 inside the kennel, a handheld MS low angle through-object frame sits about 40 centimeters above the floor and one meter from 로트와일러, looking outward through thick cage bars. The bars cut the foreground into hard vertical strips while the dog’s raised paws grip the metal at mid-frame and its bared teeth sit just above center, confronting the lens directly.",
            "lighting_mood": "Cold fluorescent light mixes with damp concrete reflections, giving the shelter a grim coastal-gray cast while the dog’s mouth reads as a dark, violent 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": [
              "C08"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C08",
                "outlook_id": "O00"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L06B01",
            "reference_usage": "exact_background",
            "owned_objects": [
              "CCTV camera",
              "bathtub",
              "cabinet",
              "cage",
              "cart",
              "counter",
              "crate",
              "door",
              "dryer",
              "exam table",
              "floor drain",
              "grooming table",
              "hamper",
              "hose sprayer",
              "kennel",
              "light fixture",
              "partition",
              "rack",
              "refrigerator",
              "shelving",
              "sink",
              "trolley",
              "ventilation unit",
              "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": "damp_kennel_bars",
                "element_type": "environment_state",
                "character_name": "",
                "description": "Thick, vertical metal cage bars of an animal enclosure. The structure is set in a damp, humid room illuminated by cold fluorescent lighting, creating pale reflections on the wet metallic and concrete surfaces.",
                "applies_to_shots": [
                  1,
                  5
                ],
                "source_facts": [
                  "Shot 1 camera: 'thick cage bars', 'hard vertical strips'",
                  "Shot 1 lighting: 'Cold fluorescent light mixes with damp concrete reflections'",
                  "Shot 5 camera: 'blurred kennel bars remain as damp vertical streaks behind her'",
                  "Shot 5 lighting: 'muted by humidity', 'dim practical spill'"
                ],
                "visual_inferences": [
                  "The metal cages are uniform throughout the holding area, consistently reflecting the bleak, humid lighting of the shelter room."
                ],
                "creative_decisions": [
                  "Using the damp, vertical metal bars as a recurring environmental motif to ground the location's gritty, institutional, and enclosed atmosphere across different camera angles."
                ],
                "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": "L06B01",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "14042eb5936a358b"
      },
      {
        "scene_index": 7,
        "heading": "S#7. 유기견 보호원 / D",
        "beat_title": "연락 안 되는 엄마와 굳은 표정",
        "representative_moment": "C01이 전화를 받지 않는 휴대전화의 화면 불빛을 받으며 굳은 표정으로 화면을 내려다보고 있다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Eye-level close-up, shallow DOF push-in, cool kennel daylight mixed with pale green-blue phone glow",
            "t2i_prompt": "Photorealistic cinematic still. Head-and-shoulder composition. C01O02 in a beige hoodie, an early 20s Korean woman with black hair, at three-quarter angle from the right. Her neck is bent over a smartphone held near the lower frame edge, eyes fixed on the device. The smartphone's lit glass face is angled upward toward her and partly toward the lens, showing an unreadable call glow. Focus on her tense expression and hardened mouth. Cool kennel daylight muted by humidity fills the air, while the phone's pale green-blue glow reflects on her chin and lower cheeks against the dim ambient spill. In the shallow depth of field behind her, blurred thick, vertical damp metal kennel bars remain as wet streaks.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#7 action: '수리영, 갇혀있는 개들 보면서 한숨 후~ 늦을 거 같다. 민숙에게 연락을 취해보지만, 받지 않는다.'",
              "shot description: 휴대폰 액정을 바라보는 수리영의 굳은 얼굴 클로즈업."
            ],
            "visual_inferences": [
              "fixed_element damp_kennel_bars: blurred thick vertical metal bars at the frame edge",
              "phone screen emits a pale green-blue glow acting as an underlight on her face"
            ],
            "creative_decisions": [
              "head-and-shoulder composition to prevent the floating-head artifact in close framing",
              "changed the DP's 'carving' lighting verb to 'reflects' to prevent the T2I model from generating physical cuts on the face",
              "shallow DOF to isolate the subject from the omitted background reference, per the skipped_close_framing rule",
              "L06 location ID removed as requested, replacing it with descriptive frame-edge kennel bar elements"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "5c6b580797e49ac4",
              "owned_hash": "456d19eb30dd6a5c",
              "camera_direction_hash": "dde9c0538a830e86",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 7,
        "dependency_reason": "동일 씬 내 이전 샷(Shot 1)에서 설정된 철창(kennel bars)과 습한 분위기 연속성 유지",
        "_shot_index": 5,
        "visible_entities": [
          "C01"
        ],
        "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": "From the kennel aisle flow, the camera settles into a CU, eye-level shallow-DOF push-in about 50–60 cm from 수리영, with her face occupying the upper two-thirds and the phone held just below frame casting light up under her chin. This is neutral observer vision, not her phone POV; the frame compresses her lowered gaze and hardened mouth while blurred kennel bars remain as damp vertical streaks behind her.",
            "lighting_mood": "Cool kennel daylight is muted by humidity, with the phone’s pale green-blue glow carving her tense features against dim practical spill.",
            "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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L06B01",
            "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": "damp_kennel_bars",
                "element_type": "environment_state",
                "character_name": "",
                "description": "Thick, vertical metal cage bars of an animal enclosure. The structure is set in a damp, humid room illuminated by cold fluorescent lighting, creating pale reflections on the wet metallic and concrete surfaces.",
                "applies_to_shots": [
                  1,
                  5
                ],
                "source_facts": [
                  "Shot 1 camera: 'thick cage bars', 'hard vertical strips'",
                  "Shot 1 lighting: 'Cold fluorescent light mixes with damp concrete reflections'",
                  "Shot 5 camera: 'blurred kennel bars remain as damp vertical streaks behind her'",
                  "Shot 5 lighting: 'muted by humidity', 'dim practical spill'"
                ],
                "visual_inferences": [
                  "The metal cages are uniform throughout the holding area, consistently reflecting the bleak, humid lighting of the shelter room."
                ],
                "creative_decisions": [
                  "Using the damp, vertical metal bars as a recurring environmental motif to ground the location's gritty, institutional, and enclosed atmosphere across different camera angles."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 7,
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "da39d6cb0975111b"
      },
      {
        "scene_index": 7,
        "heading": "S#7. 유기견 보호원 / D",
        "beat_title": "바쁜 유기견 보호원 업무와 연락 시도",
        "representative_moment": "수리영(C01)이 강아지의 등에 거품 묻은 손을 얹고 있는 상체 컷.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "chest-height slight high angle, shallow depth of field through rising steam, flat fluorescent light with coastal blues",
            "t2i_prompt": "Photorealistic cinematic still. An observer view at chest height, angled slightly downward. C01O02 in a beige hooded sweatshirt, an East Asian woman in her early 20s with black hair and dark eyes, occupies the frame at three-quarter angle from the left, leaning over a wet stainless steel wash table. Her eyes are fixed on the dog below. Focus on her soapy hand resting gently on the dog's wet, foam-covered back, which anchors the lower foreground. [: A scratched, slick stainless wash table cuts across the lower frame edge, backed by a clouded plastic splash guard and a water-streaked wall-mounted soap dispenser behind her shoulder]. Shallow depth of field blurs the background elements through rising steam. Flat fluorescent light reflects on the wet surfaces, creating a humid atmosphere with dull steel reflections and subdued coastal blues.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#7 action: '‘네’, 하고 달려가 털깍고, 연고 바르고, 목욕시키고 등'",
              "shot description: 강아지 몸에 비누 거품을 묻힌 손을 얹고 있는 수리영의 상체 컷."
            ],
            "visual_inferences": [
              "C01's pose: leaning over wash table, eyes fixed on the dog",
              "dog's back is wet and covered in white foam",
              "humid environment with rising steam and wet stainless steel surfaces"
            ],
            "creative_decisions": [
              "slight high angle at chest height to naturally frame her upper body and the dog in the lower foreground",
              "shallow DOF to focus on the intimate, careful motion while softening the busy background",
              "cool coastal blues and flat fluorescent lighting to emphasize the practical, wet work environment",
              "L06 reference removed per request, replacing with L00 direct framing-edge description"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "c76f84c920f877e7",
              "owned_hash": "456d19eb30dd6a5c",
              "camera_direction_hash": "4a0fe92d51de4890",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 7,
        "dependency_reason": "같은 공간(유기견 보호원)의 직전 샷(Shot 5)과 조명, 분위기, 공간적 연속성을 유지",
        "_shot_index": 7,
        "visible_entities": [
          "C01"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 7,
            "shot_index": 7
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Beside the wash table, the tracking flow becomes a MCU with a slight high angle, cropped to 수리영’s upper body while her soapy hand and the dog’s wet back sit in the lower foreground. The camera stays chest-height and about one meter away, using shallow DOF through rising steam so her careful motion feels intimate but still workmanlike.",
            "lighting_mood": "Flat fluorescent wash-room light is softened by steam and humidity, with dull steel reflections and subdued coastal blues.",
            "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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L06B01",
            "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": 7,
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "62d59a5b007c756c"
      },
      {
        "scene_index": 8,
        "heading": "S#8. 번화가·대형 마트 / 해질 무렵",
        "beat_title": "엇갈린 동선에 혼란스러운 수리영",
        "representative_moment": "직원에게 엄마가 조퇴했다는 소식을 듣고 빈 박스 너머로 눈을 크게 뜨며 당황하는 C01.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Close framing, eye-level over-the-shoulder shot, muted greenish fluorescent and damp neon lighting",
            "t2i_prompt": "Photorealistic cinematic still. Close framing, eye-level over-the-shoulder composition. A middle-aged East Asian woman in worker uniform occupies the right foreground, viewed from behind with her shoulder and back visible. She holds an empty cardboard box tucked to her ribs, its open top tilted slightly toward the camera and front side facing left, cutting into the frame at chest height. Across the shared aisle stands C01O02 in a beige hoodie, an East Asian woman in her early 20s with a young face, black hair, and dark pupils. Occupying the center-left with her upper body in frame, C01O02 is positioned at a three-quarter angle from the right, weight shifted onto her back foot, her widened eyes fixed on the older woman. [: Compressed mart shelving runs diagonally behind C01O02, partly stocked with products facing the aisle, while a smudged glass entrance in the deep background reveals blurred street movement]. Greenish fluorescent light from overhead ceiling fixtures flattens the aisle, mixing with damp dusk neon leaking through the glass entrance, reflecting off damp surfaces to mute skin tones and add an uneasy mood. Focus on the bewildered expression of C01O02 and the physical space occupied by the cardboard box between them.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#8 action: 수리영이 민숙이 밖에서 누구랑 얘기하고 조퇴했다는 소식을 들음",
              "Shot 2 description: 빈 박스를 든 40대 한국인 여직원 앞에 서서 눈을 동그랗게 뜬 수리영의 표정"
            ],
            "visual_inferences": [
              "fixed_element damp_dusk_neon_lighting: The ambient damp neon lighting spills from the wet street into the mart's entrance, maintaining consistent environmental color tones on the character's face",
              "The cardboard box acts as a physical barrier separating the two figures during the dialogue"
            ],
            "creative_decisions": [
              "Close over-the-shoulder framing to emphasize C01's immediate emotional reaction while keeping the employee anchored in the foreground",
              "Muted greenish practical light mixed with colorful neon to highlight the grounded, slightly uneasy thriller tone"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "9eb5e66b67ad9934",
              "owned_hash": "7c29ab00671f32cf",
              "camera_direction_hash": "b06a4d780e6784e6",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 5,
        "dependency_reason": "Carrying over the damp dusk neon lighting condition from the exterior street environment into the mart's entrance to maintain visual continuity.",
        "_shot_index": 2,
        "visible_entities": [
          "C01"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 8,
            "shot_index": 2
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Inside the mart front aisle, the dolly-in settles into a near-static MCU, eye-level OTS from the 40대 한국인 여직원’s box-side shoulder toward 수리영’s widened eyes. 수리영 holds the center-left of frame while the empty cardboard box cuts into the right foreground at chest height, keeping the employee’s answer physically between them.",
            "lighting_mood": "Greenish fluorescent mart light mixes with damp dusk neon leaking through the entrance, muting skin tones and adding 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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L08B01",
            "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": "damp_dusk_neon_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A continuous lighting condition of early evening dusk mixed with colorful neon lights. The neon light reflects off damp surfaces, casting a moody, slightly muted glow that visually bridges the interior entrance and the exterior street environment.",
                "applies_to_shots": [
                  2,
                  5
                ],
                "source_facts": [
                  "Shot 2 lighting: 'mixes with damp dusk neon leaking through the entrance'",
                  "Shot 5 lighting: 'Dusk has fallen into blue-gray street darkness, punctured by damp neon'"
                ],
                "visual_inferences": [
                  "The ambient damp neon lighting spills from the wet street into the mart's entrance, maintaining consistent environmental color tones on the character's face across both locations."
                ],
                "creative_decisions": [
                  "Using wet surface reflections and neon color spills to visually connect the interior and exterior shots, maintaining a consistent somber and uneasy mood."
                ],
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "1eff32da3900b9c4"
      },
      {
        "scene_index": 8,
        "heading": "S#8. 번화가·대형 마트 / 해질 무렵",
        "beat_title": "엇갈린 약속과 귀가 결심",
        "representative_moment": "어두운 네온 불빛 아래 마트 밖에서 C01이 자전거 핸들을 꽉 쥐고 굳은 표정으로 먼 곳을 응시한다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Low observer at handlebar height framing hands and face, with blue-gray dusk and ominous red neon damp reflections.",
            "t2i_prompt": "Photorealistic cinematic still. C01O02 in a beige hoodie, a Korean female in her early 20s with a youthful face, black hair, and dark eyes, at three-quarter angle from the left. Low observer at handlebar height, torso and head in frame, the wet bicycle's curved front handlebar crossing the lower frame horizontally, the brake levers angled toward the camera. Her clenched hands sit sharp in the lower center, gripping the handlebar, with her hardened face visible above them, eyes fixed on the distant street. Focus on the gripped hands and her tense facial expression. [L08B02: blue-gray street darkness, rain-slick pavement reflecting red and green neon signage in broken streaks beneath the bicycle, darkened glass doors with moisture smears immediately behind the figure]. Continuous early evening dusk mixed with colorful neon lights reflects off damp surfaces, casting a moody blue-gray glow on her skin, with a faint ominous red spill catching on the pavement and her shoulder.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "Shot 5 description: 어두워진 마트 밖에서 자전거 핸들을 꽉 쥔 수리영의 굳은 표정.",
              "수리영 마트 밖을 나오면서 연락을 취해보지만 받지않는다. 일단 집으로-"
            ],
            "visual_inferences": [
              "fixed_element damp_dusk_neon_lighting: continuous lighting condition of early evening dusk mixed with colorful neon lights",
              "The ambient damp neon lighting spills from the wet street, casting muted reflections on the character's face"
            ],
            "creative_decisions": [
              "Low angle framing utilizing the bicycle handlebar as a horizontal foreground line to anchor her physical tension",
              "Softened, dark background via the moisture-streaked glass doors isolates the subject in a moment of unease",
              "Incorporated blue-gray street darkness with faint red neon spills to reflect the ominous thriller mood"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "ae0fd2a13e667419",
              "owned_hash": "82729a7331412f24",
              "camera_direction_hash": "3852248132c20b7d",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 8,
        "dependency_reason": "Shot 2의 damp_dusk_neon_lighting 환경 상태 및 네온 불빛 색감을 유지하여 동일 씬 내 연속성 확보",
        "_shot_index": 5,
        "visible_entities": [
          "C01"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 8,
            "shot_index": 5
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Outside the mart after dusk, the frame drops into a MCU, low angle at handlebar height, using the bicycle handlebar as a through-object foreground line across the lower third. 수리영’s clenched hands sit sharp in the lower center while her hardened face remains visible above them, slightly compressed against wet neon reflections and passing street blur.",
            "lighting_mood": "Dusk has fallen into blue-gray street darkness, punctured by damp neon and a faint ominous red spill on the wet pavement.",
            "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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L08B02",
            "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": "damp_dusk_neon_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A continuous lighting condition of early evening dusk mixed with colorful neon lights. The neon light reflects off damp surfaces, casting a moody, slightly muted glow that visually bridges the interior entrance and the exterior street environment.",
                "applies_to_shots": [
                  2,
                  5
                ],
                "source_facts": [
                  "Shot 2 lighting: 'mixes with damp dusk neon leaking through the entrance'",
                  "Shot 5 lighting: 'Dusk has fallen into blue-gray street darkness, punctured by damp neon'"
                ],
                "visual_inferences": [
                  "The ambient damp neon lighting spills from the wet street into the mart's entrance, maintaining consistent environmental color tones on the character's face across both locations."
                ],
                "creative_decisions": [
                  "Using wet surface reflections and neon color spills to visually connect the interior and exterior shots, maintaining a consistent somber and uneasy mood."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 8,
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "8d3418df55f7fc87"
      },
      {
        "scene_index": 9,
        "heading": "S#9. 해안가 / N",
        "beat_title": "다급한 야간 질주",
        "representative_moment": "어둠이 깔린 해안도로에서 입술을 꽉 깨문 채, 한쪽 발로 자전거 페달을 힘껏 내리밟는 mid-pedal 순간의 C01.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Medium wide tracking shot from the left side at waist height, cold coastal night tones with lateral motion blur in the background",
            "t2i_prompt": "Photorealistic cinematic still. A medium tracking shot from the left side at waist height. C01O02 in a beige hoodie, a young East Asian woman with a young face, black hair, and dark eyes, seen in left profile seated on a bicycle in a mid-pedal stroke. Her clenched mouth occupies the upper-left area of the frame, eyes fixed on the distance, her black hair flowing back in the direction opposite to her forward motion. Her foot forcefully drives the downward pedal in the lower-right frame. The bicycle's rain-specked frame, front wheel, and crank are clearly visible below her. Behind her, [L05: a wet, dark coastal road streaks laterally, blurred by speed, bordered by a dark, humid, foggy sea edge. A dim roadside warning sign with its red reflective face turned obliquely toward the camera slips past in the background]. The scene is lit by cold coastal night tones, with sodium streetlight glints and faint red roadside reflections washing across the wet metal crank and her sleeve.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#9 action: 해안도로를 자전거를 타고 바삐 집으로 달려가는 수리영.",
              "shot description: 어둠이 깔린 해안도로에서 입술을 꽉 깨문 채, 한쪽 발로 자전거 페달을 힘껏 내리밟는 mid-pedal 순간의 수리영."
            ],
            "visual_inferences": [
              "C01 leaning into the pedal stroke for maximum speed",
              "coastal road wet with sea mist, reflecting dim sodium lights"
            ],
            "creative_decisions": [
              "tracking camera parallel to the bicycle from the left side to capture lateral motion blur",
              "diagonal framing arrangement placing her face in the upper-left and her pedaling foot in the lower-right to emphasize downward kinetic force",
              "mid-pedal freeze with hair flowing backward to preserve motion direction"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "5ab987c7352484ae",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "c63788b3db5875d9",
              "validator": "scene_detail_owned_objects.v1",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 6,
        "dependency_reason": "해안가 도로의 야간 분위기와 축축한 환경(가로등 불빛, 안개 낀 바다 경계) 참조",
        "_shot_index": 1,
        "visible_entities": [
          "C01",
          "L05"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 9,
            "shot_index": 1
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Parallel to the bicycle on its left side, the camera tracks in a MWS with a slight high angle from waist height, widened enough to catch 수리영’s clenched mouth near the upper left and her foot driving the pedal down in the lower right. The road streaks laterally behind her in an unsteady tracking profile, preserving the flow’s speed while making one mid-pedal strike feel like a held breath.",
            "lighting_mood": "Cold coastal night tones flatten the road, with sodium streetlight glints and faint red roadside reflections flashing across wet metal.",
            "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",
              "L05"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": [
              {
                "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": [],
            "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"
            ]
          },
          "render_contracts": [],
          "_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": "1e93d1ff408db954"
      },
      {
        "scene_index": 10,
        "heading": "S#10. 옥탑방 / N",
        "beat_title": "문 앞까지 다가온 낯선 그림자",
        "representative_moment": "옥탑방 철문 현관문 틈에 전단지가 끼워져 있고, 바닥에는 낯선 남자의 긴 그림자가 드리워져 있다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "MWS, low angle at ankle height, voyeur perspective, dim rooftop practical lighting with strong contrast",
            "t2i_prompt": "Photorealistic cinematic still. Medium wide shot from a low angle at ankle height, looking across a damp, gritty rooftop concrete floor. An elongated human shadow stretches diagonally from off-frame across the blue-gray concrete, pointing directly toward the threshold. The figure's body and face remain entirely outside the frame. A creased advertising flyer is wedged edge-first in the narrow crack of the iron door from the reference, its printed face partly visible to the camera with blurred ad graphics showing. The weathered exterior face of the door from the reference looks down toward the low camera, creating a black vertical seam that forms a hard stop at the end of the shadow path. A weak rooftop practical light illuminates the scene, casting the dark shadow and highlighting subtle rust-red stains near the threshold against the humid coastal night ambient. Focus on the elongated shadow forming a diagonal shape that leads the eye to the flyer in the door crack.",
            "outfit_assignments": [],
            "source_facts": [
              "S#10 action: '남자의 긴 그림자. (얼굴은 보이지 않는다)'",
              "S#10 action: '신발은 옥탑방 철문 현관 앞에 멈춰 선다. 문틈에 낀 광고 전단지.'",
              "DP direction: 'low angle at ankle height, tracking to a near-stop as the stranger’s elongated shadow stretches from off-frame'",
              "DP direction: 'His body and face remain outside the image'",
              "DP direction: 'printed face is partly visible to camera, wedged edge-first with blurred ad graphics showing'"
            ],
            "visual_inferences": [
              "The scene focuses entirely on the ground level and the bottom of the door, eliminating the need to show the intruder.",
              "The shadow effectively conveys the unseen threat approaching the exact_background location established in prior shots.",
              "Rust-red stains on the concrete add subtle visual tension without explicit gore."
            ],
            "creative_decisions": [
              "Anchored the door explicitly as 'the iron door from the reference' to comply with background_binding owned_objects rules.",
              "Applied a muted blue-gray and rust-red color palette to match the gritty coastal thriller worldview.",
              "Maintained a strict ankle-height camera position to ensure the geometry of the floor and door base aligns perfectly."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "bf37008725ad135e",
              "owned_hash": "434f35347ebb15a4",
              "camera_direction_hash": "cb4ceaa85a4e9753",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "door",
                  "violating_phrase": "the iron door from the reference",
                  "reason": "The door is explicitly referenced as already present from the reference, so this is an anchor reference rather than a redraw instruction.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "shadow",
                  "violating_phrase": "An elongated human shadow stretches",
                  "reason": "The shadow is described as an existing compositional element and not directly created with a redraw verb.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 4,
        "dependency_reason": "exact_background continuity to preserve the same L03 옥탑방 외부 배경",
        "_shot_index": 3,
        "visible_entities": [
          "L03"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 10,
            "shot_index": 3
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "On the rooftop floor directly before the iron door, the camera holds a MWS, low angle at ankle height, tracking to a near-stop as the 낯선 남자’s elongated shadow stretches from off-frame across the concrete. His body and face remain outside the image; the shadow forms the dominant diagonal shape leading the eye to the threshold and the narrow door crack.",
            "lighting_mood": "A weak rooftop practical throws a long black shadow across blue-gray concrete, with rust-red stains near the threshold adding occult 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": [
              "L03"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L03B04",
            "reference_usage": "exact_background",
            "owned_objects": [
              "clothesline",
              "door",
              "drainage grate",
              "light fixture",
              "parapet",
              "platform",
              "post",
              "rooftop deck",
              "shadow",
              "stairwell headhouse",
              "utility pipe",
              "water tank"
            ],
            "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": "L03B04",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "0882b4f33539cc1c"
      },
      {
        "scene_index": 10,
        "heading": "S#10. 옥탑방 / N",
        "beat_title": "절체절명의 순간",
        "representative_moment": "C03이 보이지 않는 힘에 의해 어깨를 붙잡힌 채 비명을 지르며 공포에 질린 얼굴이 화면에 가득 찬다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "eye-level handheld close observation, subtly Dutch canted angle, dim practical amber and cold blue window spill",
            "t2i_prompt": "Photorealistic cinematic still. Head-and-shoulder composition with a subtly Dutch canted angle. Focus on the terrified expression of the subject. C03O06 in a gray homewear and apron, an East Asian woman in her 40s with black hair and dark eyes, at three-quarter angle from the left, eyes fixed on the void. She is mid-recoil from an unseen grip jerking her shoulder out of alignment, her open mouth and strained cheek muscles evident in the tight framing, while her neck and upper chest remain visible in the frame. Dim practical amber light fights cold blue window spill, washing the side of her face and creating a sickly, panicked skin tone. The background collapses into smeared darkness. [: composed of immediate surroundings; a half-shadowed scuffed interior doorframe cutting diagonally behind her head; a weak small room lamp softly glowing out of focus in the rear right; dirty yellowed stained wallpaper heavily blurred; a dark window glass catching a warped smear of lamp light and a vague moving darkness at the edge]",
            "outfit_assignments": [
              {
                "character_id": "C03",
                "outlook_id": "O06"
              }
            ],
            "source_facts": [
              "S#10 action: '민숙을 덮쳐 어깨를 물어뜯는 듯한 모습 ‘아악-’'",
              "shot description: 어깨를 붙잡힌 채 입을 크게 벌린 민숙의 일그러진 얼굴 클로즈업",
              "shot description: 창문에 비친 그림자를 통해 침입자에게 어깨를 물어뜯기며 끔찍한 비명을 지르는 상태"
            ],
            "visual_inferences": [
              "unseen grip causing her to recoil mid-action",
              "sickly skin tone derived from practical amber mixed with cold blue spill"
            ],
            "creative_decisions": [
              "head-and-shoulder composition to prevent the floating face artifact inherent to extreme close-up instructions",
              "Dutch canted angle to emphasize panic and instability",
              "frame-edge surfaces (doorframe, blurred lamp, stained wallpaper, window glass) replace the standard room background due to skipped close framing"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "9ce3680f56e81aa2",
              "owned_hash": "434f35347ebb15a4",
              "camera_direction_hash": "80edeb07725f8caa",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 4,
        "dependency_reason": "동일한 옥탑방 배경 및 조명 무드 (S4_Shot1) 참조, 단 프레임 가장자리 표면만으로 클로즈업 배경을 구성",
        "_shot_index": 8,
        "visible_entities": [
          "C03"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 10,
            "shot_index": 8
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Inside the room, the camera snaps into an ECU, eye-level handheld close observation less than half a meter from 민숙’s face, tightening until her open mouth and strained cheek muscles occupy most of the frame. The angle is subtly Dutch, with the unseen grip jerking her shoulder out of alignment while the background collapses into smeared darkness.",
            "lighting_mood": "Dim practical amber fights cold blue window spill, while the handheld Dutch instability makes the skin tones sickly and panicked.",
            "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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C03",
                "outlook_id": "O06"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L03B04",
            "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": 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": "C03O06",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "016819db9c758521"
      },
      {
        "scene_index": 11,
        "heading": "S#12. 옥탑방 안 / N",
        "beat_title": "방 안으로 이어지는 참혹한 핏자국",
        "representative_moment": "바닥에 흥건히 고인 피와 커튼 쪽으로 이어지는 핏자국 클로즈업.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Dutch angle ground-level close framing, weak TV spill with deep cool blue-gray ambient shadows",
            "t2i_prompt": "Photorealistic cinematic still. Low observer at ground level, tilted in a Dutch angle composition. Focus on a thick, wet pool of deep red blood occupying the lower center, heavily spread across [L04: old, dirty, dark wooden floorboards running from the foreground toward the room center]. Smeared, uneven bloody partial footprints trail away from the camera along the damp-glossed wood. The space is submerged in oppressive, humid shadows with a strong cool, blue-gray tone. A weak, muted off-camera spill leaves the floor in dim shadow, sharply broken by the unnaturally saturated, fresh glossy red of the blood. At the far edge of the visual field, the bottom hem of a long, blurred mundane fabric window curtain hangs vertically, touching the floorboards, serving as a soft, shadowy framing device for the distant destination of the blood path.",
            "outfit_assignments": [],
            "source_facts": [
              "Scene text: '바닥에 흥건한 혈흔. 피묻은 발자국은 커튼으로 이어져 있다.'",
              "Shot 11 camera: 'The pooled red blood occupies the lower center... floorboards and a blurred curtain line remain visible'",
              "Shot 11 lighting: 'dim room lamp leave the floor in humid gray shadow'"
            ],
            "visual_inferences": [
              "fixed_element pooled_blood_on_floor applied as primary subject",
              "fixed_element dim_blue_gray_lighting applied for atmospheric shadows",
              "fixed_element hanging_window_curtain applied as blurred background edge",
              "Dutch angle and low lens placement emphasize the unsettling psychological horror of the footprint trail"
            ],
            "creative_decisions": [
              "Ground level Dutch angle framing forces the wet blood to dominate the immediate foreground while leading the eye back",
              "Deep blue-gray ambient grading contrasts sharply with the highly saturated red to amplify the visceral shock",
              "Curtain is kept heavily out of focus to maintain close-framing isolation and avoid exposing the entire un-referenced room"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "5c441828ebe63cd0",
              "owned_hash": "930b9286d35f0313",
              "camera_direction_hash": "caceee152e91bc8a",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 5,
        "dependency_reason": "Occurs in the same rooftop room established in S5, although background reference is skipped due to extreme close framing on the floor.",
        "_shot_index": 11,
        "visible_entities": [
          "L04"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 11,
            "shot_index": 11
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Inside 수리영’s bedroom, the floor-tracking flow nearly stops in a CU, Dutch angle, with the lens only about 20 cm above the ground near the doorway. The pooled red blood occupies the lower center and stains the eye path deeper into the room, while floorboards and a blurred curtain line remain visible as scale references so the stain feels real rather than abstract.",
            "lighting_mood": "Weak TV spill and a dim room lamp leave the floor in humid gray shadow, broken by the unnaturally saturated red of the blood.",
            "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",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L04B06",
            "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_blue_gray_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior space is submerged in oppressive, humid shadows characterized by a strong cool, blue-gray tone. A weak, muted spill of light from an off-camera source barely illuminates the space, maintaining a grim, dark, and gloomy atmosphere.",
                "applies_to_shots": [
                  11,
                  14,
                  20
                ],
                "source_facts": [
                  "Shot 11 lighting: 'dim room lamp leave the floor in humid gray shadow'",
                  "Shot 14 lighting: 'Muted lamp spill... carve silhouette into a gray-blue silhouette'",
                  "Shot 20 lighting: 'The room stays dim and blue-gray'"
                ],
                "visual_inferences": [
                  "The ambient light is extremely low, relying on a single weak artificial light source that casts cool, unsettling shadows."
                ],
                "creative_decisions": [
                  "Unifying the visual tone across wide and close-up shots by locking in the specific cool blue-gray color grading to enhance the horror atmosphere."
                ],
                "confidence": "high"
              },
              {
                "element_id": "pooled_blood_on_floor",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A thick, wet pool of deep red blood is heavily spread across the dark wooden floorboards. The saturated red liquid creates a striking, gruesome contrast against the muted, shadowy textures of the room's floor.",
                "applies_to_shots": [
                  11,
                  14
                ],
                "source_facts": [
                  "Scene text: '바닥에 흥건한 혈흔.'",
                  "Shot 11 action: '방바닥에 흥건하게 고인 붉은 핏자국'",
                  "Shot 11 camera: 'The pooled red blood occupies the lower center'",
                  "Shot 14 lighting: 'sick red stains pulling the eye downward'"
                ],
                "visual_inferences": [
                  "The blood is fresh enough to appear wet and highly saturated, starkly contrasting with the wooden floor."
                ],
                "creative_decisions": [
                  "Using the blood pool as a persistent environmental anchor between the close-up floor tracking and the wide reveal of the body."
                ],
                "confidence": "high"
              },
              {
                "element_id": "hanging_window_curtain",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "A long, mundane fabric window curtain hangs vertically, its bottom edge touching the wooden floorboards. It is partially pulled aside, serving as a soft, shadowy framing device at the edge of the visual field.",
                "applies_to_shots": [
                  11,
                  14
                ],
                "source_facts": [
                  "Scene text: '피묻은 발자국은 커튼으로 이어져 있다. 호흡을 가다듬고 커튼을 젖히자...'",
                  "Shot 11 camera: 'blurred curtain line remain visible as scale references'",
                  "Shot 14 camera: 'behind the pulled-aside fabric. The curtain edge frames the left foreground'"
                ],
                "visual_inferences": [
                  "The curtain is likely made of a heavy, neutral-colored fabric that absorbs the dim room light."
                ],
                "creative_decisions": [
                  "Maintaining the curtain as a constant foreground or background element to emphasize the cramped, hidden nature of the space."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 5,
                "shot_index": 7,
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "b12b548f9e10fa8b"
      },
      {
        "scene_index": 11,
        "heading": "S#12. 옥탑방 안 / N",
        "beat_title": "방 안에서 처참하게 죽어있는 엄마 민숙 발견",
        "representative_moment": "C03가 핏물이 고인 바닥 위, 커튼 뒤에 앉은 채 고개를 푹 수그린 자세로 죽어 있다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Observer POV medium shot, low angle framing slightly above the floor, muted lamp spill casting cool blue-gray shadows.",
            "t2i_prompt": "Photorealistic cinematic still. [L04B06: cramped bedroom interior submerged in oppressive cool blue-gray shadows]. Medium shot, low angle framing slightly above the floor, from the perspective of an observer looking inward. The left foreground is framed by the long fabric window curtain pulled aside like a torn veil, its folded front edge facing the camera. Center-right in the frame, C03O06 in gray homewear and an apron, a Korean woman in her early 40s with black hair, sits slumped on the dark wooden floor. Her head is dropped heavily forward, looking down, limbs heavy and motionless, body volume preserved against the dim ambient fill. The thick, wet pool of deep red blood spreads heavily across the floorboards beneath her, the saturated liquid smearing near the curtain edge. Muted lamp spill washes her shoulder, rendering her into a dim gray-blue silhouette while revealing sick red stains on her clothing. The rumpled, shadowed bed cuts into the right frame edge beside her. In the dim back plane behind the body, the closed, dirty window shows vague night shapes through grime. Focus on C03O06's slumped, lifeless posture amidst the cramped room depth.",
            "outfit_assignments": [
              {
                "character_id": "C03",
                "outlook_id": "O06"
              }
            ],
            "source_facts": [
              "Shot 14 description: 커튼 뒤에 앉은 채로 고개를 푹 수그린 민숙의 축 늘어진 전신.",
              "Scene text: 바닥에 흥건한 혈흔. 커튼을 젖히자 앉은 채로 고개를 떨군 채 죽어있는 엄마"
            ],
            "visual_inferences": [
              "fixed_element dim_blue_gray_lighting: interior space submerged in oppressive humid shadows with a cool blue-gray tone",
              "fixed_element pooled_blood_on_floor: thick wet pool of deep red blood spread across dark wooden floorboards",
              "fixed_element hanging_window_curtain: long fabric curtain pulled aside framing the visual field"
            ],
            "creative_decisions": [
              "Low angle composition just above the floor to emphasize the slumped weight of the body and the creeping blood pool.",
              "Using the curtain edge as a foreground veil to frame the gruesome reveal within the claustrophobic room depth, maintaining a stark blue-gray color grading."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "16e0d8943cd357d8",
              "owned_hash": "930b9286d35f0313",
              "camera_direction_hash": "ed2ca7ae2d27e9bc",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "curtain",
                  "violating_phrase": "the left foreground is framed by the long fabric window curtain pulled aside",
                  "reason": "The prompt mentions curtain as an existing framing element, not as something newly created or installed.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "bed",
                  "violating_phrase": "The rumpled, shadowed bed cuts into the right frame edge",
                  "reason": "The bed is only used as part of the scene composition and is not directly targeted by a redraw verb.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "window",
                  "violating_phrase": "The closed, dirty window shows vague night shapes",
                  "reason": "The window is described as an existing background object with no explicit instruction to create or add it.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 5,
        "dependency_reason": "S5_Shot9의 옥탑방 내부 배경 및 피가 번진 바닥, 조명 톤 등 물리적 공간과 환경의 연속성 유지",
        "_shot_index": 14,
        "visible_entities": [
          "C03",
          "L04"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 11,
            "shot_index": 14
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Near the window curtain, the completed tilt resolves into a WS, low angle from about one meter high, held just far enough back for 민숙’s entire slumped body to read behind the pulled-aside fabric. The curtain edge frames the left foreground like a torn veil, while 민숙 sits center-right with her head dropped and limbs heavy, the room’s cramped depth pressing in around her.",
            "lighting_mood": "Muted lamp spill and cold window darkness carve 민숙 into a gray-blue silhouette with sick red stains pulling the eye downward.",
            "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",
              "L04"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C03",
                "outlook_id": "O06"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L04B06",
            "reference_usage": "exact_background",
            "owned_objects": [
              "backpack",
              "bed",
              "bloodstain",
              "curtain",
              "doll",
              "door",
              "floor",
              "photo",
              "wall",
              "wardrobe",
              "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": "dim_blue_gray_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior space is submerged in oppressive, humid shadows characterized by a strong cool, blue-gray tone. A weak, muted spill of light from an off-camera source barely illuminates the space, maintaining a grim, dark, and gloomy atmosphere.",
                "applies_to_shots": [
                  11,
                  14,
                  20
                ],
                "source_facts": [
                  "Shot 11 lighting: 'dim room lamp leave the floor in humid gray shadow'",
                  "Shot 14 lighting: 'Muted lamp spill... carve silhouette into a gray-blue silhouette'",
                  "Shot 20 lighting: 'The room stays dim and blue-gray'"
                ],
                "visual_inferences": [
                  "The ambient light is extremely low, relying on a single weak artificial light source that casts cool, unsettling shadows."
                ],
                "creative_decisions": [
                  "Unifying the visual tone across wide and close-up shots by locking in the specific cool blue-gray color grading to enhance the horror atmosphere."
                ],
                "confidence": "high"
              },
              {
                "element_id": "pooled_blood_on_floor",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A thick, wet pool of deep red blood is heavily spread across the dark wooden floorboards. The saturated red liquid creates a striking, gruesome contrast against the muted, shadowy textures of the room's floor.",
                "applies_to_shots": [
                  11,
                  14
                ],
                "source_facts": [
                  "Scene text: '바닥에 흥건한 혈흔.'",
                  "Shot 11 action: '방바닥에 흥건하게 고인 붉은 핏자국'",
                  "Shot 11 camera: 'The pooled red blood occupies the lower center'",
                  "Shot 14 lighting: 'sick red stains pulling the eye downward'"
                ],
                "visual_inferences": [
                  "The blood is fresh enough to appear wet and highly saturated, starkly contrasting with the wooden floor."
                ],
                "creative_decisions": [
                  "Using the blood pool as a persistent environmental anchor between the close-up floor tracking and the wide reveal of the body."
                ],
                "confidence": "high"
              },
              {
                "element_id": "hanging_window_curtain",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "A long, mundane fabric window curtain hangs vertically, its bottom edge touching the wooden floorboards. It is partially pulled aside, serving as a soft, shadowy framing device at the edge of the visual field.",
                "applies_to_shots": [
                  11,
                  14
                ],
                "source_facts": [
                  "Scene text: '피묻은 발자국은 커튼으로 이어져 있다. 호흡을 가다듬고 커튼을 젖히자...'",
                  "Shot 11 camera: 'blurred curtain line remain visible as scale references'",
                  "Shot 14 camera: 'behind the pulled-aside fabric. The curtain edge frames the left foreground'"
                ],
                "visual_inferences": [
                  "The curtain is likely made of a heavy, neutral-colored fabric that absorbs the dim room light."
                ],
                "creative_decisions": [
                  "Maintaining the curtain as a constant foreground or background element to emphasize the cramped, hidden nature of the space."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 5,
                "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": "background",
                "id": "L04B06",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "65f9031162715805"
      },
      {
        "scene_index": 11,
        "heading": "S#12. 옥탑방 안 / N",
        "beat_title": "붉게 변하는 손목 표식",
        "representative_moment": "수리영이 민숙의 곁에서 구겨진 사진을 볼 때, 민숙(C03)의 손목에 있는 표식(P08)이 기이하게 붉은색으로 변한다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Overhead close-framing at hand level, rack focus onto the wrist mark, dim blue-gray lighting with a wet red accent.",
            "t2i_prompt": "Photorealistic cinematic still. Overhead angle looking directly down at a floor surface. The limp hand and exposed wrist of an East Asian middle-aged woman in a gray homewear sleeve rest across the lower third of the frame. Focus on the exposed wrist, revealing P08, a circular mark on the upward-facing skin visibly reddening, wet blood pooling subtly around its edges. The P08 mark catches a wet red emphasis against the dim, humid blue-gray ambient shadow. At frame left, the fingers lie loosely against a crumpled photograph, its creased printed face partly turned toward the camera but heavily obscured by folds and dark red blood smears. The hand and photo sit on a blood-spotted floor showing sticky dark smears. The dark horizontal block of a bed base shadow occupies the upper edge, preserving the grim blue-gray tone.",
            "outfit_assignments": [],
            "source_facts": [
              "S11 action: '엄마 손목에 있던 문신이 붉은색으로 변한다.'",
              "shot description: overhead / bird's eye angle at hand level, rack focus past her limp fingers to the circular wrist mark.",
              "DP instruction: marked skin surface faces upward toward the overhead camera, the round symbol visibly turning red."
            ],
            "visual_inferences": [
              "fixed_element dim_blue_gray_lighting: strong cool, blue-gray tone with weak muted spill.",
              "Minsuk's body part is isolated in the frame, necessitating a common noun descriptor 'East Asian middle-aged woman' instead of her C## ID due to the body-part focus rule.",
              "The background reference is skipped due to close framing, composing the shot purely from the immediate floor and shadow."
            ],
            "creative_decisions": [
              "Positioning the hand across the lower third and the photograph at frame left to visually anchor the supernatural element (the reddening mark) to physical evidence.",
              "Using a high color contrast between the dim blue-gray ambient shadows and the wet red reflection of the wrist mark to evoke a gritty, occult horror mood."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "e9d3dfdf24eed196",
              "owned_hash": "930b9286d35f0313",
              "camera_direction_hash": "954f06c972415b82",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 11,
        "dependency_reason": "Same continuous timeline and specific location (Minsuk's room) as the prior shot, requiring matching lighting and floor textures.",
        "_shot_index": 20,
        "visible_entities": [
          "C03",
          "P08"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 11,
            "shot_index": 20
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Beside 민숙 on the bedroom floor, the camera completes the evidence-detail push into an ECU, overhead / bird's eye angle at hand level, with a shallow rack focus from the crumpled photograph past her limp fingers to the circular wrist mark. The reddening mark sits dead center, her hand crossing the lower third and the photo kept partially visible at frame left so the supernatural change stays tied to physical evidence.",
            "lighting_mood": "The room stays dim and blue-gray, but the wrist symbol emits a subtle wet red emphasis that feels occult without breaking gritty realism.",
            "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",
              "P08"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C03",
                "outlook_id": "O06"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L04B06",
            "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_blue_gray_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior space is submerged in oppressive, humid shadows characterized by a strong cool, blue-gray tone. A weak, muted spill of light from an off-camera source barely illuminates the space, maintaining a grim, dark, and gloomy atmosphere.",
                "applies_to_shots": [
                  11,
                  14,
                  20
                ],
                "source_facts": [
                  "Shot 11 lighting: 'dim room lamp leave the floor in humid gray shadow'",
                  "Shot 14 lighting: 'Muted lamp spill... carve silhouette into a gray-blue silhouette'",
                  "Shot 20 lighting: 'The room stays dim and blue-gray'"
                ],
                "visual_inferences": [
                  "The ambient light is extremely low, relying on a single weak artificial light source that casts cool, unsettling shadows."
                ],
                "creative_decisions": [
                  "Unifying the visual tone across wide and close-up shots by locking in the specific cool blue-gray color grading to enhance the horror atmosphere."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 11,
                "shot_index": 14,
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "57adc052838e032d"
      },
      {
        "scene_index": 12,
        "heading": "S#13. 옥탑방 외곽 / N",
        "beat_title": "사라진 시신과 충격에 빠진 수리영",
        "representative_moment": "눈물을 머금은 C01이 고개를 들어 오프카메라의 혜수를 향해 절망적으로 상황을 알리는 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "A tight close-up, slightly low angle looking up, shallow depth of field, muted coastal night light with fog-softened contrast and faint red siren reflections.",
            "t2i_prompt": "Photorealistic cinematic still. C01O02 in a beige hoodie, young Korean woman with dark hair, positioned in the center-left. The composition is from a low angle looking up, her chest and shoulders visible at the bottom edge, extending upward into the frame. She is at three-quarter angle from the right, chin lifting slowly, eyes fixed on an off-camera figure above her, a tear-wet expression on her face. [: a damp stair rail running diagonally away, blurred dark concrete landing speckled with rain compressed in the shallow background]. Muted coastal night light with fog-softened contrast, a cold practical spill falling across her shoulder, and a faint pulsing red emergency siren light reflecting on her wet cheek and the damp railing.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#13 action: '비명을 지르며 옥탑방을 빠져나오는 수리영... (올라다보며) 엄마가, 엄마가... 죽어있어.'",
              "Shot description: A tight CU, slightly low angle, holding just below 수리영’s eyeline as she lifts her tear-wet face toward off-camera 혜수."
            ],
            "visual_inferences": [
              "fixed_element: Faint, pulsing red emergency siren lights reflecting on the wet surfaces.",
              "Rain-slicked stair railing and dark concrete landing grounding the exterior night atmosphere."
            ],
            "creative_decisions": [
              "Applied shallow depth of field to isolate C01's emotional shock while omitting the background reference per close-framing rules.",
              "Placed C01 center-left to visually weight her upward gaze toward the off-frame character.",
              "Used [L13] for frame-edge environmental elements, successfully avoiding the forbidden L04 ID."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "019eb02b050cd3de",
              "owned_hash": "13f8b28591f9882d",
              "camera_direction_hash": "f28cfac769a6e57c",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 5,
        "dependency_reason": "S5_Shot7의 옥탑방 외곽 계단 환경의 야간 조명 톤 연속성 유지 (근접 샷 특성상 배경 구조보다 색감과 분위기 위주로 공유)",
        "_shot_index": 4,
        "visible_entities": [
          "C01"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 12,
            "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 tight CU, slightly low angle, from the foot of the rooftop stairs, holding just below 수리영’s eyeline as she lifts her tear-wet face toward off-camera 혜수. The framing uses shallow DOF: 수리영 occupies the center-left 70% of frame, with the damp stair rail and a faint red siren smear compressed behind her; 혜수 is kept fully off-frame and exists only through 수리영’s upward gaze.",
            "lighting_mood": "Muted coastal night light with fog-softened contrast, a cold practical spill from the rooftop room, and a faint red emergency flicker catching her tears.",
            "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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L04B07",
            "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": "emergency_siren_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "Faint, pulsing red emergency siren lights from outside, casting subtle red accents and reflections across the space to create an uneasy and tense atmosphere.",
                "applies_to_shots": [
                  4,
                  13
                ],
                "source_facts": [
                  "S#13 action: '동시에 사이렌이 울리고'",
                  "Shot 4 lighting: 'a faint red emergency flicker catching her tears'",
                  "Shot 13 lighting: 'with the red accents now reduced to a barely perceptible unease'"
                ],
                "visual_inferences": [
                  "The pulsing red light from the arrived police car reaches both the exterior stairs and the interior living area."
                ],
                "creative_decisions": [
                  "Maintained the red siren light as a subtle, unifying atmospheric element across the active shots to emphasize the lingering tension of the crime scene."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 5,
                "shot_index": 7,
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "473e2e8498ac925b"
      },
      {
        "scene_index": 12,
        "heading": "S#14. 옥탑방 안 / N",
        "beat_title": "핏자국과 시신이 사라진 깨끗한 방",
        "representative_moment": "시신과 핏자국이 완전히 사라진 깨끗한 옥탑방 바닥에 백팩과 인형만 남아 있는 장면.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "A static eye-level composition looking into the room, cold grayish-blue ambient dawn light mixed with dim practical light.",
            "t2i_prompt": "Photorealistic cinematic still. A static eye-level composition from just outside the door from the reference, looking across [L04: a spotless, sterile rooftop bedroom interior]. The mattress from the reference sits mid-right, tightly arranged, its long side facing the camera and the pillow end angled slightly toward the right wall. The P04 backpack from the reference rests low-left on the dry, spotless floor from the reference, its front pocket side facing partly toward the camera with straps collapsed. The motionless doll from the reference is slightly separated near the lower center, its face and torso angled up toward the camera. A cold, grayish-blue ambient dawn light seeping into the room mixes with dim practical light from the light fixture from the reference, rendering the surrounding walls and the empty zone in a sterile, cool tone. The thin curtain from the reference hangs closed over the window from the reference, veiling the glass, with only pale dawn-gray exterior brightness leaking through. Focus on the unnerving empty space on the floor from the reference between the mattress and the backpack.",
            "outfit_assignments": [],
            "source_facts": [
              "S#14 text: '지나칠 정도로 깨끗하게 정돈된 방안. 핏자국도 보이지 않고, 침대 위에 붉은 표식은 물론 커튼 뒤의 시신도 감쪽 같이 사라졌다.'",
              "Shot 8 description: '핏자국과 시신이 사라진 채 인형과 백팩이 덩그러니 놓인 수리영의 방 내부 전경.'"
            ],
            "visual_inferences": [
              "fixed_element cool_interior_ambient_light: A cold, grayish-blue ambient dawn light seeping into the interior of the house, rendering the surrounding walls and empty space in a sterile, cool tone.",
              "No characters are visible; only the sterile room with the P04 prop remains."
            ],
            "creative_decisions": [
              "Eye-level framing looking into the room from the doorway to emphasize the missing body and the unnaturally clean state of the floor.",
              "Mixing coastal dawn seepage with dim practical light to create a humid but sterile atmosphere.",
              "Explicitly anchoring owned_objects like the door, mattress, floor, curtain, doll, and backpack to the exact_background reference to strictly prevent duplicate rendering."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "53a73192995e2d43",
              "owned_hash": "13f8b28591f9882d",
              "camera_direction_hash": "a647b1f4086409a6",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "door",
                  "violating_phrase": "from just outside the door from the reference",
                  "reason": "The phrase refers to an existing door as a reference anchor and does not instruct creating or modifying it.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "mattress",
                  "violating_phrase": "The mattress from the reference sits mid-right",
                  "reason": "The mattress is explicitly referenced as already present in the reference and only positioned in the composition.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "backpack",
                  "violating_phrase": "The P04 backpack from the reference rests low-left",
                  "reason": "The backpack is mentioned as an existing reference object and only described in placement, not redrawn.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "floor",
                  "violating_phrase": "the dry, spotless floor from the reference",
                  "reason": "The floor is described with reference anchoring and surface state only, with no creation or modification instruction.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "doll",
                  "violating_phrase": "The motionless doll from the reference is slightly separated",
                  "reason": "The doll is explicitly anchored to the reference and only framed in space, not newly generated.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "light fixture",
                  "violating_phrase": "the light fixture from the reference",
                  "reason": "The light fixture is directly tied to the reference and used as an existing light source, not as a redraw target.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "curtain",
                  "violating_phrase": "The thin curtain from the reference hangs closed",
                  "reason": "The curtain is referenced as an existing object from the reference and described by state, which is anchor-only.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "window",
                  "violating_phrase": "the window from the reference",
                  "reason": "The window is explicitly marked as coming from the reference and only serves as a background anchor.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 12,
        "dependency_reason": "exact_background policy applied from previous shot (scene 12) to preserve the established room's layout while highlighting the disappearance of the body and blood.",
        "_shot_index": 8,
        "visible_entities": [
          "P04",
          "L04"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 12,
            "shot_index": 8
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A static WS at eye-level from just outside 수리영’s bedroom doorway, looking across the spotless bed and floor area like a held evidentiary frame. The bed sits mid-right, the backpack rests low-left, and the doll is slightly separated near the lower center, leaving an unnerving empty zone where blood and the body should have been.",
            "lighting_mood": "Dim interior practical light mixes with bluish coastal dawn seepage, making the clean room feel unnaturally sterile and humid.",
            "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",
              "L04"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L04B07",
            "reference_usage": "exact_background",
            "owned_objects": [
              "backpack",
              "baseboard",
              "curtain",
              "doll",
              "door",
              "floor",
              "light fixture",
              "mattress",
              "photo",
              "wall mark",
              "wardrobe",
              "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": "cool_interior_ambient_light",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A cold, grayish-blue ambient dawn light seeping into the interior of the house, rendering the surrounding walls and empty space in a sterile, cool tone.",
                "applies_to_shots": [
                  8,
                  13
                ],
                "source_facts": [
                  "Shot 8 lighting: 'bluish coastal dawn seepage, making the clean room feel unnaturally sterile'",
                  "Shot 13 lighting: 'the surrounding room stays cool and gray'"
                ],
                "visual_inferences": [
                  "The early dawn light from the coastal area uniformly affects the interior color palette of the rooftop house, contrasting with localized practical lights."
                ],
                "creative_decisions": [
                  "Using a cool, sterile color temperature for the background interior to highlight the unnatural emptiness where the body used to be."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 12,
                "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": "L04B07",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "b399d62adfa1cf20"
      },
      {
        "scene_index": 12,
        "heading": "S#13. 옥탑방 외곽 / N ~ S#14. 옥탑방 안 / N",
        "beat_title": "다정한 위로와 안심",
        "representative_moment": "혜수(C02)가 혼란에 빠진 수리영(C01)의 양어깨를 붙잡고 다정하게 바라보며 진정시킨다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Eye-level frontal composition, shallow depth of field isolating the subjects, contrasting warm practical foreground light against a cool gray background.",
            "t2i_prompt": "Photorealistic cinematic still. Eye-level frontal composition gently pushing in. C02O04 in an investigator jacket, a middle-aged Korean woman, facing the camera, occupying the central upper half with head and shoulders visible, her warm expression softened by practical light. In the lower foreground, C01O02 in a beige hoodie, a young Korean woman, with back to camera, her shoulders held tense, anchoring the intimate over-the-shoulder connection. C02O04's hands are gripping both of C01O02's shoulders near the bottom third of the frame, bridging the shared space they both occupy. C02O04's eyes fixed on C01O02, while C01O02's head tilts slightly up, eyes fixed on C02O04. Soft practical light warms C02O04’s face and hands. A cold, grayish-blue ambient dawn light renders the [: soft, out-of-focus background surface] behind her in a sterile, cool tone. Faint, pulsing red emergency siren lights cast subtle red reflections across the side of C02O04's jaw and jacket collar. Focus on C02O04's comforting expression and the physical connection of her hands on the foreground shoulders.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              },
              {
                "character_id": "C02",
                "outlook_id": "O04"
              }
            ],
            "source_facts": [
              "S#14 action: '(수리영 어깨를 어루만지며) 영아.'",
              "Shot 13 description: 수리영의 어깨를 양손으로 감싸 쥔 채 부드러운 눈빛으로 바라보는 혜수의 정면 구도.",
              "Shot 13 lighting: 'with the red accents now reduced to a barely perceptible unease'"
            ],
            "visual_inferences": [
              "fixed_element cool_interior_ambient_light: A cold, grayish-blue ambient dawn light renders the space in a sterile, cool tone.",
              "fixed_element emergency_siren_lighting: subtle red accents and reflections across the space",
              "C01's back-to-camera foreground positioning creates an over-the-shoulder anchor to highlight C02's expression."
            ],
            "creative_decisions": [
              "Omitted specific room background references (skipped_close_framing) to focus purely on the immediate interaction and emotional warmth of C02.",
              "Applied a subtle red reflection on C02's jaw and collar to maintain the ambient tension from the siren without overpowering the warm practical light.",
              "Used a shared space keyword ('bridging the shared space they both occupy') to ensure accurate foreground-background interaction mapping."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "aa9398b1a90f6662",
              "owned_hash": "13f8b28591f9882d",
              "camera_direction_hash": "3921866889853abc",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 12,
        "dependency_reason": "동일한 옥탑방 배경 안에서의 연속된 씬이나, 본 샷은 close framing 룰에 따라 배경 참조를 생략하고 인물의 직접적인 감정적 교감에 집중함.",
        "_shot_index": 13,
        "visible_entities": [
          "C01",
          "C02"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 12,
            "shot_index": 13
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "An MCU at eye-level, frontal and gently pushing in, positioned in the living area near the bedroom side so 혜수 is centered squarely while 수리영’s shoulders anchor the lower foreground. The composition uses an intimate OTS-like foreground connection: 혜수’s hands grip both of 수리영’s shoulders near the bottom third, and 혜수’s face occupies the central upper half with the softened room receding behind her.",
            "lighting_mood": "Soft practical light warms 혜수’s face while the surrounding room stays cool and gray, with the red accents now reduced to a barely perceptible 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",
              "C02"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              },
              {
                "character_id": "C02",
                "outlook_id": "O04"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L04B07",
            "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": "emergency_siren_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "Faint, pulsing red emergency siren lights from outside, casting subtle red accents and reflections across the space to create an uneasy and tense atmosphere.",
                "applies_to_shots": [
                  4,
                  13
                ],
                "source_facts": [
                  "S#13 action: '동시에 사이렌이 울리고'",
                  "Shot 4 lighting: 'a faint red emergency flicker catching her tears'",
                  "Shot 13 lighting: 'with the red accents now reduced to a barely perceptible unease'"
                ],
                "visual_inferences": [
                  "The pulsing red light from the arrived police car reaches both the exterior stairs and the interior living area."
                ],
                "creative_decisions": [
                  "Maintained the red siren light as a subtle, unifying atmospheric element across the active shots to emphasize the lingering tension of the crime scene."
                ],
                "confidence": "high"
              },
              {
                "element_id": "cool_interior_ambient_light",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A cold, grayish-blue ambient dawn light seeping into the interior of the house, rendering the surrounding walls and empty space in a sterile, cool tone.",
                "applies_to_shots": [
                  8,
                  13
                ],
                "source_facts": [
                  "Shot 8 lighting: 'bluish coastal dawn seepage, making the clean room feel unnaturally sterile'",
                  "Shot 13 lighting: 'the surrounding room stays cool and gray'"
                ],
                "visual_inferences": [
                  "The early dawn light from the coastal area uniformly affects the interior color palette of the rooftop house, contrasting with localized practical lights."
                ],
                "creative_decisions": [
                  "Using a cool, sterile color temperature for the background interior to highlight the unnatural emptiness where the body used to be."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 12,
                "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": "C02O04",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "d913805824f3db5b"
      },
      {
        "scene_index": 13,
        "heading": "S#15. 해안가. 버스정류장 부근 / N",
        "beat_title": "수리영이 헛것을 보았다고 합리화하며 직접 집에 가서 확인하기로 결심함",
        "representative_moment": "벤치에서 몸을 반쯤 일으킨 채 입술을 굳게 다문 수리영의 결연한 표정.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "A tight MCU from a slightly low angle, shallow focus, cold blue-gray coastal night color temperature.",
            "t2i_prompt": "Photorealistic cinematic still. C01O02 in a beige hoodie, an East Asian woman in her early 20s with a young face and black hair, captured in a tight head-and-shoulder MCU from a slightly low angle, occupying the center-right of the frame. She is half-risen, her torso leaning forward in left three-quarter profile, her dark eyes fixed on the distant darkness. Focus is sharply on her resolved expression and firm lips. The rain-damp front edge of a bench cuts across the lower frame, receding toward the left. In the soft out-of-focus background, the dark sea road falls into blurred negative space through a wet glass panel streaked with condensation, alongside a damp bus timetable board with its unreadable printed schedule face turned obliquely toward the camera. Cold blue-gray coastal night ambient light wraps the scene, with a weak fluorescent practical washing her features and causing moisture to glint on the bench.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "Shot 3: 'weak fluorescent practical flattening her face and moisture glinting on the bench'",
              "Shot 4: 'darkness ahead'",
              "S#15 action: '수리영 (훅 일어나며) 집에 가봐야 겠어.'"
            ],
            "visual_inferences": [
              "fixed_element: A dark, damp coastal roadside at night, featuring a bus stop shelter that casts a weak, cold fluorescent glow.",
              "The subject is caught mid-action (half-risen) leaving the bench, supporting the transition from seated to running.",
              "The coastal dampness translates to condensation on the glass and moisture glinting on nearby surfaces."
            ],
            "creative_decisions": [
              "Using a tight MCU with a slightly low angle to emphasize her decisive realization while maintaining the upper body in frame.",
              "Shallow depth of field isolates her resolved expression against the blurred, cold blue-black coastal background."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "d4cdf394b89480b3",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "ea4b5928fb6c1906",
              "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": [
          "C01"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 13,
            "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 tight MCU, slightly low angle, from within arm’s length of the bus-stop bench as 수리영 is caught half-risen, her firm lips and resolved eyes tightened in shallow focus. The bench cuts across the lower frame, her torso leans forward center-right, and the dark sea road falls into blurred negative space behind her.",
            "lighting_mood": "Cold blue-gray coastal night wraps the shelter, with a weak fluorescent practical flattening her face and moisture glinting on the bench.",
            "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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "damp_coastal_bus_stop",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dark, damp coastal roadside at night, featuring a bus stop shelter that casts a weak, cold fluorescent glow. The paved ground is wet, reflecting the faint artificial light, while the surrounding background is swallowed by deep, blue-black coastal darkness.",
                "applies_to_shots": [
                  3,
                  4,
                  5
                ],
                "source_facts": [
                  "Shot 3: 'weak fluorescent practical flattening her face and moisture glinting on the bench'",
                  "Shot 4: 'wet road', 'roadside sodium spill', 'darkness ahead'",
                  "Shot 5: 'Weak shelter light', 'damp blue-black roadside'"
                ],
                "visual_inferences": [
                  "The asphalt remains uniformly wet across the scene, creating a consistent cold reflection of the nearby artificial lights."
                ],
                "creative_decisions": [
                  "Emphasizing a cold blue-gray color palette for the night environment to amplify the sense of isolation and uncertainty."
                ],
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "38da50a348595721"
      },
      {
        "scene_index": 13,
        "heading": "S#15. 해안가. 버스정류장 부근 / N",
        "beat_title": "자신을 의심하며 집으로 달려가는 수리영",
        "representative_moment": "어둠 속 해안도로를 향해 황급히 달려가는 수리영의 미드스트라이드 뒷모습.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "A kinetic medium shot from a low tracking angle, cold and humid tone with sodium spill.",
            "t2i_prompt": "Photorealistic cinematic still. A kinetic medium shot from a low angle at waist-to-shoulder height, tracking behind and slightly to the side. C01O02 in a beige hoodie, an early 20s Korean woman, with back to camera, mid-stride running away into the darkness. Her back-left side occupies the right third of the frame. One foot is just lifted off the wet asphalt, the other pushing off, her collar and black hair flaring backward in the direction opposite to motion. Thin roadside sodium spill skims her moving silhouette, creating rim-lit edges across her shoulder and preserving her body volume. The damp coastal bus stop shelter shrinks behind her on the left edge, casting a weak cold fluorescent glow. The paved ground is uniformly wet, reflecting the faint artificial light and her lifted foot. Focus on her dynamic mid-stride back and the dark open space she is entering. The background ahead is swallowed by deep, blue-black coastal darkness.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S13 action: '수리영, 응. 나 갈게 (달려간다)'",
              "Shot 4 description: mid-stride running away from the bus stop into the dark, back to camera",
              "Shot 4 background: wet road, roadside sodium spill, darkness ahead"
            ],
            "visual_inferences": [
              "fixed_element damp_coastal_bus_stop: dark, damp coastal roadside at night, paved ground is wet reflecting faint artificial light",
              "motion freeze: one foot lifted, black hair and hoodie collar flaring backward due to running momentum"
            ],
            "creative_decisions": [
              "kinetic low angle waist-to-shoulder tracking composition putting her on the right third to emphasize the empty darkness ahead",
              "rim-lit edges on her shoulder to preserve body volume against the dark background while keeping her face hidden"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "e798a36ebe896191",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "b538df2e22f299df",
              "validator": "scene_detail_owned_objects.v1",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 13,
        "dependency_reason": "exact_background from prior shot (S13_Shot3) — same space, maintaining the damp coastal bus stop shelter context",
        "_shot_index": 4,
        "visible_entities": [
          "C01",
          "L09"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 13,
            "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 kinetic MWS, low angle, tracking behind and slightly to the side of 수리영 at waist-to-shoulder height as she runs away from the bus stop. Her back-left side occupies the right third, one foot just lifted from the wet road and her collar flaring backward, while the bus stop shrinks behind her on the left edge.",
            "lighting_mood": "The palette stays cold and humid, with thin roadside sodium spill skimming her moving silhouette and the darkness ahead swallowing detail.",
            "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",
              "L09"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "damp_coastal_bus_stop",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dark, damp coastal roadside at night, featuring a bus stop shelter that casts a weak, cold fluorescent glow. The paved ground is wet, reflecting the faint artificial light, while the surrounding background is swallowed by deep, blue-black coastal darkness.",
                "applies_to_shots": [
                  3,
                  4,
                  5
                ],
                "source_facts": [
                  "Shot 3: 'weak fluorescent practical flattening her face and moisture glinting on the bench'",
                  "Shot 4: 'wet road', 'roadside sodium spill', 'darkness ahead'",
                  "Shot 5: 'Weak shelter light', 'damp blue-black roadside'"
                ],
                "visual_inferences": [
                  "The asphalt remains uniformly wet across the scene, creating a consistent cold reflection of the nearby artificial lights."
                ],
                "creative_decisions": [
                  "Emphasizing a cold blue-gray color palette for the night environment to amplify the sense of isolation and uncertainty."
                ],
                "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"
              }
            ],
            "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"
            ]
          },
          "render_contracts": [],
          "_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": "ff0325286388775d"
      },
      {
        "scene_index": 13,
        "heading": "S#15. 해안가. 버스정류장 부근 / N",
        "beat_title": "멀어지는 수리영을 향해 소리치는 혜수",
        "representative_moment": "버스정류장에서 멀어져 가는 수리영을 향해 한 손을 입가에 댄 채 외치는 C02의 상반신 모습.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Static medium shot at eye-level, emphasizing a cold blue-gray color palette with ominously empty negative space.",
            "t2i_prompt": "Photorealistic cinematic still. A static medium shot at eye-level near the edge of the bus stop area. C02O04 in a detective coat, an East Asian middle-aged woman with dark hair and dark eyes, stands in the left-center frame from the waist up. She is at a three-quarter angle from the right, one hand cupped to her mouth, eyes fixed on a distant figure. On the right side of the frame, a young East Asian figure with back to camera is frozen mid-stride, receding into the deep blue-black coastal darkness. The damp paved road they both occupy stretches between them. A weak, cold fluorescent glow from the bus shelter roof falls on C02O04 and the ground, isolating her against the dark night. The rain-streaked glass panel of the shelter behind her catches her faint outline, while the uniformly wet asphalt reflects the cold artificial light. Focus on C02O04's tense posture calling out across the space.",
            "outfit_assignments": [
              {
                "character_id": "C02",
                "outlook_id": "O04"
              }
            ],
            "source_facts": [
              "S#15 action: 혜수(C02) shouts after 수리영 who is running toward home.",
              "Shot 5 description: observer at eye-level, 혜수 filling the left-center upper body area with one hand cupped to her mouth.",
              "DP instruction: right half of the frame is held open as dark negative space where 수리영 recedes as a small distant figure."
            ],
            "visual_inferences": [
              "fixed_element damp_coastal_bus_stop: the asphalt remains uniformly wet, creating a consistent cold reflection of the artificial light.",
              "수리영 is depicted as an unregistered figure (common noun) frozen mid-stride to capture the instantaneous moment of receding.",
              "The damp paved road is explicitly named as a shared spatial anchor between the foreground C02 and the background figure."
            ],
            "creative_decisions": [
              "Arranging the composition with C02O04 anchored on the left-center and the right side dominated by dark negative space to highlight the physical and emotional distance.",
              "Using the rain-streaked glass panel behind C02 to add textural depth and emphasize the cold, damp coastal atmosphere."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "cfd6a8712a120be7",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "6cdff6c1b1efa72e",
              "validator": "scene_detail_owned_objects.v1",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 13,
        "dependency_reason": "exact_background from previous shot (S13_Shot4) to maintain the damp coastal bus stop environment.",
        "_shot_index": 5,
        "visible_entities": [
          "C02"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 13,
            "shot_index": 5
          },
          "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 near the edge of the bus stop area, looking back across the roadside toward 혜수 as she calls after 수리영. 혜수 fills the left-center upper body area with one hand cupped to her mouth, while the right half of the frame is held open as dark negative space where 수리영 recedes as a small distant figure.",
            "lighting_mood": "Weak shelter light isolates 혜수 against a damp blue-black roadside, with the background darkness left ominously unfilled.",
            "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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C02",
                "outlook_id": "O04"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "damp_coastal_bus_stop",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dark, damp coastal roadside at night, featuring a bus stop shelter that casts a weak, cold fluorescent glow. The paved ground is wet, reflecting the faint artificial light, while the surrounding background is swallowed by deep, blue-black coastal darkness.",
                "applies_to_shots": [
                  3,
                  4,
                  5
                ],
                "source_facts": [
                  "Shot 3: 'weak fluorescent practical flattening her face and moisture glinting on the bench'",
                  "Shot 4: 'wet road', 'roadside sodium spill', 'darkness ahead'",
                  "Shot 5: 'Weak shelter light', 'damp blue-black roadside'"
                ],
                "visual_inferences": [
                  "The asphalt remains uniformly wet across the scene, creating a consistent cold reflection of the nearby artificial lights."
                ],
                "creative_decisions": [
                  "Emphasizing a cold blue-gray color palette for the night environment to amplify the sense of isolation and uncertainty."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 13,
                "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": "C02O04",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "2120cf7aaed6cf9b"
      },
      {
        "scene_index": 14,
        "heading": "S#16. 경찰서 사무실 / N",
        "beat_title": "위치추적을 강권하는 혜수",
        "representative_moment": "혜수(C02)가 김형사(C04)의 얼굴 가까이 스마트폰 액정을 내밀며 단호하게 요구하고, 김형사가 의자에 기댄 채 그 화면을 응시한다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Side-composed MCU at chest height with flat greenish fluorescent lighting and a cold phone glow",
            "t2i_prompt": "Photorealistic cinematic still. A side-composed MCU at chest height. In the left third of the frame, C02O04 in a dark jacket, a Korean woman in her early 40s with black hair, is positioned with her back mostly to the camera but her side-profile partially visible. Her tense, extended arm holds P01, a smartphone, in the foreground space across the shared duty desk. C04O08 in a dark duty jacket, a Korean man in his 30s with short black hair, occupies the mid-right, seated in left profile and leaning back in his chair, his eyes fixed on P01. The lit screen of P01 is angled toward C04O08 and partly toward the camera. Focus on the tense spatial proximity between P01 and C04O08. Flat, sickly greenish fluorescent overhead light bathes the scene, while a cold glow from P01 reflects on C04O08's cheek and jaw. [L10: The immediate surroundings include the cluttered edge of the duty desk with scattered papers in the lower frame and a blurred, lit office notice board covered in papers in the soft background, conveying a damp, exhausted night-shift tension].",
            "outfit_assignments": [
              {
                "character_id": "C02",
                "outlook_id": "O04"
              },
              {
                "character_id": "C04",
                "outlook_id": "O08"
              }
            ],
            "source_facts": [
              "Scene text: '이 번호 위치 좀 따봐.'",
              "Shot 3 description: '김형사의 얼굴 앞으로 휴대폰 액정을 불쑥 내민 혜수의 측면 구도.'",
              "Shot 3 DP instruction: 'A side-composed MCU at chest height... across 혜수’s shoulder as her arm thrusts a phone into 김형사’s personal space.'"
            ],
            "visual_inferences": [
              "fixed_element body_full_pose: C02 positioned with back mostly to camera, side-profile partially visible, extending arm",
              "fixed_element body_full_pose: C04 seated in left profile, leaning back in chair",
              "night_shift_office_environment: sickly greenish fluorescent overhead lights casting damp shadows, cold phone glow"
            ],
            "creative_decisions": [
              "Used shallow depth of field on the background notice board to keep the visual focus strictly on the interaction across the shared duty desk",
              "Positioned the phone screen to cast a clear, cool highlight on C04's cheek, contrasting with the ambient green light"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "933dddbc704635c6",
              "owned_hash": "f747b52ba12b45e1",
              "camera_direction_hash": "558b44ea92fb4eb2",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "",
        "_shot_index": 3,
        "visible_entities": [
          "C02",
          "C04",
          "P01",
          "L10"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 14,
            "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 side-composed MCU at chest height, dollying in from the duty desk side across 혜수’s shoulder as her arm thrusts a phone into 김형사’s personal space. The phone screen cuts into the foreground just below 김형사’s face without overwhelming the frame, while 김형사 sits mid-right in profile and 혜수’s tense shoulder and extended arm occupy the left third.",
            "lighting_mood": "Flat greenish fluorescent office light is broken by the cold phone glow, giving the confrontation a damp, exhausted night-shift tension.",
            "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",
              "C04",
              "P01",
              "L10"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C02",
                "outlook_id": "O04"
              },
              {
                "character_id": "C04",
                "outlook_id": "O08"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L10B02",
            "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": "night_shift_office_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior of an institutional office at night, illuminated by flat, sickly greenish fluorescent overhead lights. The harsh lighting casts cool, damp shadows across the room, establishing an exhausted and tense atmosphere. A faint, cold glow from a mobile device slightly pierces the ambient greenish light, while the background features muted, indistinct office furniture and duty desks.",
                "applies_to_shots": [
                  3,
                  5
                ],
                "source_facts": [
                  "Scene text: '경찰서 사무실 / N'",
                  "Shot 3 description: 'Flat greenish fluorescent office light is broken by the cold phone glow, giving the confrontation a damp, exhausted night-shift tension'",
                  "Shot 5 description: 'Sickly fluorescent light and the faint phone glow'"
                ],
                "visual_inferences": [
                  "Standard institutional office setting with overhead panel lights",
                  "The overall color palette is dominated by cold green and bluish tones due to the fluorescent fixtures"
                ],
                "creative_decisions": [
                  "Using the greenish tint to visually reinforce the fatigue and anxiety of the late-night shift",
                  "Keeping the background details relatively generic and muted to keep the focus on the foreground subjects"
                ],
                "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": "C02O04",
                "policy": "required"
              },
              {
                "kind": "character_outlook",
                "id": "C04O08",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "7ef81c9196e67e97"
      },
      {
        "scene_index": 14,
        "heading": "S#16. 경찰서 사무실 / N",
        "beat_title": "불길한 예감",
        "representative_moment": "C02가 휴대폰을 꽉 쥔 채 불안한 표정으로 허공을 응시한다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Eye-level CU composition, pushing into the subject with a shallow depth of field, cool sickly greenish fluorescent color grading",
            "t2i_prompt": "Photorealistic cinematic still. A close-up framing at eye-level. C02O04 in a dark investigator jacket, an early 40s Korean woman with a middle-aged face, black hair, and dark eyes, positioned slightly left of center at three-quarter angle from the left. She freezes with an anxious expression, her dark eyes staring past the camera into the void. At the lower frame edge, her hand grips P01 tightly, the smartphone's screen face angled downward and only partly visible, its faint cold glow leaking between her fingers and catching on her lower chin and neck. Flat, sickly greenish fluorescent light washes across her forehead and cheek, casting cool, damp shadows over her features. [: Soft, out-of-focus background depth behind her, an institutional office with muted horizontal blur of an unattended duty desk and a dimly lit wall calendar, illuminated by downward-spilling overhead light]. Focus on her frozen, anxious facial expression and the cold light pooling on her cheek.",
            "outfit_assignments": [
              {
                "character_id": "C02",
                "outlook_id": "O04"
              }
            ],
            "source_facts": [
              "Scene text: '수리영에게 내색은 안했지만 왠지 불길한.'",
              "Shot 5 description: '휴대폰을 꽉 쥔 채 허공을 노려보는 혜수의 불안한 표정.'"
            ],
            "visual_inferences": [
              "fixed_element night_shift_office_environment: The overall color palette is dominated by cold green and bluish tones due to the fluorescent fixtures",
              "C02's expression reflects sudden realization and severe worry about her daughter's safety"
            ],
            "creative_decisions": [
              "Translated the DP's 'carve worry into her face' to 'casting cool, damp shadows over her features' to avoid literal carving misinterpretation by the T2I model",
              "Rendered the background as a soft institutional blur via [L01] block to comply with the skipped_close_framing rule while maintaining spatial depth",
              "Positioned the phone and hand near the lower frame edge to anchor the tight framing without exposing the full body"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "47aa7be404b0d359",
              "owned_hash": "f747b52ba12b45e1",
              "camera_direction_hash": "0ed96a841501adc9",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 14,
        "dependency_reason": "같은 씬(S#16 / index 14) 앞 샷의 sickly greenish fluorescent 사무실 조명과 분위기를 배경 아웃포커스 영역과 인물 얼굴 조명에 연속적으로 적용",
        "_shot_index": 5,
        "visible_entities": [
          "C02",
          "P01"
        ],
        "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": "A CU at eye-level, pushing in to about two feet from 혜수’s face as she freezes with anxiety, staring past the camera into empty office space rather than at anyone. Her face sits slightly left of center, the clenched phone remains visible near the lower frame edge, and the duty room behind her falls soft into muted institutional blur.",
            "lighting_mood": "Sickly fluorescent light and the faint phone glow carve worry into her face, with cool shadows and no warm relief.",
            "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",
              "P01"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C02",
                "outlook_id": "O04"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L10B02",
            "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": "night_shift_office_environment",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior of an institutional office at night, illuminated by flat, sickly greenish fluorescent overhead lights. The harsh lighting casts cool, damp shadows across the room, establishing an exhausted and tense atmosphere. A faint, cold glow from a mobile device slightly pierces the ambient greenish light, while the background features muted, indistinct office furniture and duty desks.",
                "applies_to_shots": [
                  3,
                  5
                ],
                "source_facts": [
                  "Scene text: '경찰서 사무실 / N'",
                  "Shot 3 description: 'Flat greenish fluorescent office light is broken by the cold phone glow, giving the confrontation a damp, exhausted night-shift tension'",
                  "Shot 5 description: 'Sickly fluorescent light and the faint phone glow'"
                ],
                "visual_inferences": [
                  "Standard institutional office setting with overhead panel lights",
                  "The overall color palette is dominated by cold green and bluish tones due to the fluorescent fixtures"
                ],
                "creative_decisions": [
                  "Using the greenish tint to visually reinforce the fatigue and anxiety of the late-night shift",
                  "Keeping the background details relatively generic and muted to keep the focus on the foreground subjects"
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 14,
                "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": "C02O04",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "3e488b3a9b2a9365"
      },
      {
        "scene_index": 15,
        "heading": "S#17. 옥탑방 / 동틀무렵",
        "beat_title": "인적이 끊긴 고요한 새벽의 옥탑방",
        "representative_moment": "어스름한 새벽빛 아래 불이 켜진 옥탑방 창문이 멀리 보이는 텅 빈 마당 전경.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Low angle at waist height, pre-dawn blue haze with contrasting sharp yellow practical glow",
            "t2i_prompt": "Photorealistic cinematic still. Low observer at waist height from the far edge of the empty space, facing the rooftop room across the yard. Focus on the small, lit window from the reference sitting sharp in the upper third of the frame. The damp, dawn-gray courtyard from the reference occupies most of the lower frame as negative space. The rain-darkened stair from the reference leads obliquely upward on the right. The lit window from the reference glows with a lonely yellow practical light against the pre-dawn blue haze, edged by humid fog, its closed glass slightly fogged with condensation. The wet, dark wall from the reference runs along the side.",
            "outfit_assignments": [],
            "source_facts": [
              "S#17. 옥탑방 / 동틀무렵",
              "어스름한 새벽빛 아래 불이 켜진 옥탑방 창문이 보이는 텅 빈 마당 전경.",
              "shot description: observer at waist height, empty courtyard before any cat is revealed"
            ],
            "visual_inferences": [
              "Empty damp courtyard acting as heavy negative space in the lower frame to emphasize isolation",
              "Fogged condensation on the glass to add a cold, humid coastal thriller tone"
            ],
            "creative_decisions": [
              "Waist-height low angle to anchor the damp ground strongly in the foreground",
              "Pre-dawn blue haze contrasting with the warm yellow practical light to create an ominous mood"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "6a261bfb9e85e705",
              "owned_hash": "9788465baca39ead",
              "camera_direction_hash": "659983f6edbfa365",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "window",
                  "violating_phrase": "the small, lit window from the reference",
                  "reason": "The window is explicitly described as being from the reference, which anchors it as an existing background object rather than a newly generated one.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "courtyard",
                  "violating_phrase": "The damp, dawn-gray courtyard from the reference",
                  "reason": "The courtyard is explicitly tied to the reference, so this is a reference anchor description, not a redraw instruction.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "stair",
                  "violating_phrase": "The rain-darkened stair from the reference",
                  "reason": "The stair is mentioned as part of the reference background, which makes it an anchor reference rather than a new addition.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "window",
                  "violating_phrase": "The lit window from the reference glows",
                  "reason": "This is a descriptive reference mention of an existing window with lighting/state details, not an instruction to create a new window.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "wall",
                  "violating_phrase": "The wet, dark wall from the reference",
                  "reason": "The wall is explicitly identified as coming from the reference, so it is treated as an anchored existing object.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 4,
        "dependency_reason": "S4_Shot1과 동일한 배경 공간(마당 및 옥탑방 외부)을 새벽 시간대로 묘사하기 위해 exact_background 참조 유지",
        "_shot_index": 1,
        "visible_entities": [
          "L03"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 15,
            "shot_index": 1
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "An EWS, low angle, held static from the far edge of the empty courtyard at waist height, facing the rooftop room across the yard. The courtyard occupies most of the lower frame as damp negative space, while the lit rooftop window sits small but sharp in the upper third, glowing before any cat is revealed.",
            "lighting_mood": "Pre-dawn blue haze mutes the courtyard while the rooftop window burns with a lonely yellow practical glow, edged by humid fog.",
            "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": [
              "L03"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L03B01",
            "reference_usage": "exact_background",
            "owned_objects": [
              "cat",
              "courtyard",
              "drain",
              "gate",
              "landing",
              "light fixture",
              "railing",
              "roof",
              "stair",
              "utility pipe",
              "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": [],
            "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": "L03B01",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "0dd24e309fd9dff0"
      },
      {
        "scene_index": 16,
        "heading": "S#18. 안방 안 / D",
        "beat_title": "사라진 엄마를 확인하고 안도하는 수리영",
        "representative_moment": "침대에서 번쩍 눈을 뜬 수리영(C01)이 텅 빈 천장을 응시하며 불안해하는 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "ECU from bed level looking upward, cold dim dawn lighting",
            "t2i_prompt": "Photorealistic cinematic still. C01O02 in a beige hoodie, an East Asian woman in her early 20s with a young face, black hair, and dark pupils, at three-quarter angle from the right, head propped against a pillow. An extreme close-up, slight low angle from bed level aiming upward; tight framing on her face, upper chest and shoulder edge visible in the lower frame. Her eyes are fixed upward on the ceiling, her awakened eyes and brow catching the light, expression glossy and startled with oppressive stillness. [: A creased, slept-on pillow edge anchoring the bottom frame, and a sliver of pale, dim, blank bedroom ceiling occupying the upper background]. The scene is bathed in cold, dim, and muted grayish daylight characteristic of early dawn, soft cool-toned ambient light washing her face and casting faint room shadow from the side.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#18 action: Wakes up startled after dozing off, opens eyes wide ('번쩍 눈을 뜨고')",
              "Shot description: ECU on her awakened eyes, looking upward at the ceiling from bed level"
            ],
            "visual_inferences": [
              "fixed_element dim_dawn_lighting: cold, dim, muted grayish daylight characteristic of early dawn",
              "Pillow edge and dim ceiling frame the close-up to emphasize her isolated psychological dread without revealing the rest of the room"
            ],
            "creative_decisions": [
              "Slight low angle from bed level emphasizes the feeling of oppressive stillness and vulnerability",
              "Tight framing on her face with shoulder edge visible satisfies both the intimate ECU requirement and T2I body coherence constraints, preventing a floating head",
              "Muted grayish dawn palette with cool shadows matches the thriller and horror undertones"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "3394653392c5472a",
              "owned_hash": "6b550e5aab414324",
              "camera_direction_hash": "94de47cb1f3dd327",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 12,
        "dependency_reason": "Same bedroom environment established in scene 12 (Shot 8), maintaining consistent architectural tone, though close framing skips exact background references here.",
        "_shot_index": 2,
        "visible_entities": [
          "C01"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 16,
            "shot_index": 2
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "An ECU, slight low angle, from bed level beside 수리영’s face, aimed upward enough to catch her fixed gaze toward the blank ceiling. Only her awakened eyes, brow, and upper face dominate the frame, with the pillow edge and a sliver of dim ceiling giving scale; the camera is static, intimate, and breath-held.",
            "lighting_mood": "Cold dawn seepage and faint room shadow leave her eyes glossy and startled, with no hallucination yet—only oppressive stillness.",
            "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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L04B09",
            "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_dawn_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The indoor environment is bathed in a cold, dim, and muted grayish daylight characteristic of early dawn. Soft, cool-toned ambient light seeps into the spaces, casting faint shadows and establishing a consistently gloomy, low-contrast atmosphere across different rooms.",
                "applies_to_shots": [
                  2,
                  10,
                  14
                ],
                "source_facts": [
                  "Shot 2 lighting: 'Cold dawn seepage and faint room shadow'",
                  "Shot 10 lighting: 'cool humid daylight'",
                  "Shot 14 lighting: 'Muted gray daylight'"
                ],
                "visual_inferences": [
                  "The entire house shares the same overcast, early morning natural light coming from outside.",
                  "The lack of strong direct sunlight unifies the physically disconnected rooms (bedroom, bathroom, dining area) under one cohesive tone."
                ],
                "creative_decisions": [
                  "Applying a consistent cool-blue and grayish color grade to the environmental lighting to heighten the protagonist's psychological dread and the eerie occult undertones."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 12,
                "shot_index": 8,
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "178b69fb5f1436ba"
      },
      {
        "scene_index": 16,
        "heading": "S#18. 안방 안 / 욕실 안",
        "beat_title": "붉은 원의 환각과 극심한 공포",
        "representative_moment": "거울 속 평면적인 타일 벽면에 불길하고 선명한 P08이 반사되어 보이는 정지된 환각의 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Tight eye-level static CU reflection shot, subjective hallucinated perspective, cool grayish dawn light mixed with dim practical lighting.",
            "t2i_prompt": "Photorealistic cinematic still. A tight eye-level close-up composition directly facing a flat bathroom mirror plane, the frame edges bounded by the glassy surface. Focus on the aggressively vivid P08, a fresh-looking, saturated red painted circle, reflected and centered slightly above the middle third of the glass. The mirror reflects a near-graphic grid of [L04: pale, wet-looking bathroom tiles bathed in a cold, dim, and muted grayish dawn light]. Soft cool-toned ambient seepage and dim practical lighting mix across the fog-softened tile whites, making the unnaturally crisp red mark stand out starkly against the gloomy, low-contrast background reflection, capturing a purely subjective hallucination with no observer visible.",
            "outfit_assignments": [],
            "source_facts": [
              "Shot 10 description: 거울 표면에 반사된, 욕실 타일 위에 붉은 도료로 그려진 선명한 원의 정지 화면.",
              "S#18 action: 거울에 비친 욕실 벽에 갑자기 또 발생하는 붉은 원.",
              "Shot 10 camera directive: static CU, eye-level reflection shot, 수리영 is excluded from frame."
            ],
            "visual_inferences": [
              "fixed_element dim_dawn_lighting: cool, dim, grayish daylight and soft ambient light.",
              "The mirror flattens the reflected bathroom tiles into a near-graphic grid.",
              "The vivid red of P08 contrasts sharply with the cold, fog-softened pale tiles."
            ],
            "creative_decisions": [
              "Restricting the framing purely to the mirror's surface to emphasize the subjective, uncanny nature of the hallucination without showing the character.",
              "Applying a strong color contrast between the saturated red mark and the cool-blue/grayish ambient light to heighten psychological dread."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "848b4aec53ca32b5",
              "owned_hash": "cdfa7376ed4f4a07",
              "camera_direction_hash": "1c02a1e42137e393",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 16,
        "dependency_reason": "동일한 공간(L04 옥탑방)에서 이어지는 새벽 조명(dim_dawn_lighting)과 분위기 연속성 유지",
        "_shot_index": 10,
        "visible_entities": [
          "P08",
          "L04"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 16,
            "shot_index": 10
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "From tight against the bathroom mirror plane, the camera holds a static CU, eye-level reflection shot; the reflected bathroom tiles flatten into a near-graphic grid, with the vivid red painted circle centered slightly above the middle third. This is 수리영’s subjective hallucinated perception, but she is excluded from frame, leaving only the mirror-surface image and the impossible mark.",
            "lighting_mood": "Dim bathroom practical light mixes with cool humid daylight, making the red circle feel aggressively vivid against fog-softened tile whites.",
            "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": [
              "P08",
              "L04"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L04B10",
            "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_dawn_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The indoor environment is bathed in a cold, dim, and muted grayish daylight characteristic of early dawn. Soft, cool-toned ambient light seeps into the spaces, casting faint shadows and establishing a consistently gloomy, low-contrast atmosphere across different rooms.",
                "applies_to_shots": [
                  2,
                  10,
                  14
                ],
                "source_facts": [
                  "Shot 2 lighting: 'Cold dawn seepage and faint room shadow'",
                  "Shot 10 lighting: 'cool humid daylight'",
                  "Shot 14 lighting: 'Muted gray daylight'"
                ],
                "visual_inferences": [
                  "The entire house shares the same overcast, early morning natural light coming from outside.",
                  "The lack of strong direct sunlight unifies the physically disconnected rooms (bedroom, bathroom, dining area) under one cohesive tone."
                ],
                "creative_decisions": [
                  "Applying a consistent cool-blue and grayish color grade to the environmental lighting to heighten the protagonist's psychological dread and the eerie occult undertones."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 16,
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "ab5009dec12f972d"
      },
      {
        "scene_index": 16,
        "heading": "S#18. 안방 안 등 / D",
        "beat_title": "기괴하게 일그러지는 사진 속 엄마의 얼굴",
        "representative_moment": "식탁에 놓인 액자 속 엄마의 웃는 얼굴이 기괴하게 흘러내리는 끔찍한 환각의 정지 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Dining-table height close-up with a slight Dutch angle, muted grayish dawn daylight with sickly occult undertones",
            "t2i_prompt": "Photorealistic cinematic still. Captured at dining-table height with a slight Dutch angle, a framed photograph, P03, occupies the center third of the frame. Inside P03, the printed image of a middle-aged East Asian woman facing the camera is grotesquely warped, her portrait smile and facial features melting downward in a sickening distortion as if the ink is liquefied. The dull, fingerprinted glass of P03 reflects a dim warm practical light as a faint streak across the distorted face. Focus is locked on the warped portrait inside P03. [: A dark wood dining table surface with a humid sheen fills the lower frame, holding a half-full clear water glass covered in condensation beside the photo edge]. The scene is bathed in cold, dim, muted grayish dawn daylight, establishing a gloomy, low-contrast atmosphere with a sickly occult emphasis.",
            "outfit_assignments": [],
            "source_facts": [
              "S#18 action: '식탁에 놓인 엄마와 수리영의 사진을 보는데... 환하게 웃는 엄마 얼굴이 뭉개져 흘러내린다'",
              "Shot 14 lighting: 'Muted gray daylight and weak warm practicals'",
              "Shot 14 description: 'The front photo face is visible to camera, showing the middle-aged woman’s smiling portrait warped into a grotesque shape.'"
            ],
            "visual_inferences": [
              "fixed_element dim_dawn_lighting: The indoor environment is bathed in a cold, dim, and muted grayish daylight characteristic of early dawn.",
              "The subjective POV means the observer (수리영) is completely out of frame, focusing only on the hallucinatory surface of P03.",
              "A common noun + demographic descriptor ('middle-aged East Asian woman') is used for the reproduced face inside the photograph, adhering to reproduction surface rules."
            ],
            "creative_decisions": [
              "Applied a slight Dutch angle to visually reinforce the sense of psychological dread and distorted reality.",
              "Used the [L01: ...] block to define only the immediate frame-edge surfaces (the table and water glass) since the close framing skips the larger room background."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "dcfe10c957dbc9bf",
              "owned_hash": "cdfa7376ed4f4a07",
              "camera_direction_hash": "bbd88148db5584f1",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 16,
        "dependency_reason": "동일 씬(16) 내의 연속된 조명(dim_dawn_lighting) 톤 유지 및 액자 소품(P03) 초점 유지",
        "_shot_index": 14,
        "visible_entities": [
          "P03"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 16,
            "shot_index": 14
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "At dining-table height, the completed dolly-in locks into a CU, slight Dutch angle on the framed photograph surface; the mother’s smiling face sits in the center third, grotesquely warped as if the ink is melting downward. The frame edges and a sliver of tabletop remain visible to keep it physically grounded, while the distortion is perceived through 수리영’s hallucination.",
            "lighting_mood": "Muted gray daylight and weak warm practicals leave the table dim, with the photo’s warped face carrying the shot’s sickly occult emphasis.",
            "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"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L04B10",
            "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_dawn_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The indoor environment is bathed in a cold, dim, and muted grayish daylight characteristic of early dawn. Soft, cool-toned ambient light seeps into the spaces, casting faint shadows and establishing a consistently gloomy, low-contrast atmosphere across different rooms.",
                "applies_to_shots": [
                  2,
                  10,
                  14
                ],
                "source_facts": [
                  "Shot 2 lighting: 'Cold dawn seepage and faint room shadow'",
                  "Shot 10 lighting: 'cool humid daylight'",
                  "Shot 14 lighting: 'Muted gray daylight'"
                ],
                "visual_inferences": [
                  "The entire house shares the same overcast, early morning natural light coming from outside.",
                  "The lack of strong direct sunlight unifies the physically disconnected rooms (bedroom, bathroom, dining area) under one cohesive tone."
                ],
                "creative_decisions": [
                  "Applying a consistent cool-blue and grayish color grade to the environmental lighting to heighten the protagonist's psychological dread and the eerie occult undertones."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 16,
                "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": "prop",
                "id": "P03",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "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"
            ]
          },
          "render_contracts": [
            {
              "contract_id": "rc_001",
              "scope": {
                "scene_index": 16,
                "shot_indices": [
                  14
                ],
                "duration": "single_shot"
              },
              "targets": [
                {
                  "entity_id": "P03",
                  "role": "visual_target"
                }
              ],
              "requirements": [
                {
                  "dimension": "visual_identity",
                  "operation": "preserve",
                  "strength": "required",
                  "reference_policy": "use_entity_reference"
                }
              ]
            }
          ],
          "_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": "dc80d4584b696423"
      },
      {
        "scene_index": 17,
        "heading": "S#19. 마트 / D",
        "beat_title": "마트 도착과 탐문 시작점",
        "representative_moment": "안개가 옅게 낀 이른 아침, C01이 마트 문 앞에 자전거를 세운 채 문을 바라보고 있다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "eye-level establishing frame, center foreground composition, cool fog-muted dawn light",
            "t2i_prompt": "Photorealistic cinematic still. C01O02 in a beige hoodie, Korean, early 20s woman, standing in the center foreground at a three-quarter left angle. Her full body is visible as she grips the handlebar of the bicycle from the reference, the bicycle's frame and front wheel visible below, its front brake cable visible against her hoodie. Her eyes are fixed on the closed door from the reference behind her. The scene is bathed in dim, cool-toned daylight with a visible layer of faint pale gray morning fog that softens edges and mutes surrounding colors, creating a chilly and quiet atmosphere. The quiet storefront from the reference sits symmetrically behind her, emitting a weak fluorescent spill from inside. The rain-slick sidewalk from the reference below reflects the door and bicycle tires in broken, dim patches.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "scene heading text: 이른 아침 (early morning)",
              "action: 자전거를 주차하고 안으로 들어가는 수리영 (arriving at the mart)",
              "DP instruction: full body and bicycle occupy the center foreground, handlebar readable across her waist line"
            ],
            "visual_inferences": [
              "fixed_element cold_dawn_lighting_state: dim, cool-toned daylight with a visible layer of faint morning fog",
              "pose: three-quarter left angle, gripping bicycle handlebar, eyes fixed on mart entrance",
              "weak fluorescent spill from inside the mart contrasts with the cool dawn lighting"
            ],
            "creative_decisions": [
              "Explicitly anchored owned objects ('bicycle', 'door', 'storefront', 'sidewalk') to 'from the reference' to prevent the model from generating duplicate background or prop elements",
              "Included the bicycle frame and front wheel in the lower frame description to ground the large prop according to Rule D",
              "Added reflection details on the damp sidewalk to enhance the humid coastal atmosphere requested by the visual tone"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "a537d150cd23482a",
              "owned_hash": "32ecec4c97c04d6d",
              "camera_direction_hash": "3de681063f8513d2",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "bicycle",
                  "violating_phrase": "grips the handlebar of the bicycle",
                  "reason": "The owned object is directly mentioned in a reference phrase as part of the scene description, not commanded to be newly created or altered.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "door",
                  "violating_phrase": "the closed door from the reference",
                  "reason": "The owned object appears with an explicit reference cue, indicating the existing door in the background rather than a redraw instruction.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "storefront",
                  "violating_phrase": "The quiet storefront from the reference",
                  "reason": "The owned object is explicitly tied to the reference background, which makes this an anchor reference rather than a request to add or change it.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "sidewalk",
                  "violating_phrase": "The rain-slick sidewalk from the reference",
                  "reason": "The owned object is mentioned as part of the referenced environment and only described with surface conditions, not newly created.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "N/A",
        "_shot_index": 1,
        "visible_entities": [
          "C01",
          "L07"
        ],
        "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": "Outside the mart entrance, the camera holds a WS, eye-level establishing frame about 18 feet from the glass doors before the dolly-in begins; 수리영’s full body and bicycle occupy the center foreground, with the handlebar readable across her waist line. The quiet storefront sits symmetrically behind her through faint morning fog, emphasizing the pause before she enters.",
            "lighting_mood": "Cool fog-muted dawn light dominates, with weak fluorescent spill from inside the mart and no comforting 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",
              "L07"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L07B01",
            "reference_usage": "exact_background",
            "owned_objects": [
              "CCTV camera",
              "awning",
              "bicycle",
              "bicycle rack",
              "bollard",
              "curb",
              "door",
              "entrance mat",
              "sidewalk",
              "storefront",
              "street lane",
              "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_dawn_lighting_state",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A specific lighting state of the outdoor environment, featuring cold, fog-muted dawn light. The scene is bathed in dim, cool-toned daylight with a visible layer of faint morning fog that softens edges and mutes surrounding colors, creating a chilly and quiet atmosphere.",
                "applies_to_shots": [
                  1,
                  5
                ],
                "source_facts": [
                  "scene heading text: 이른 아침 (early morning)",
                  "Shot 1 lighting: Cool fog-muted dawn light dominates",
                  "Shot 5 lighting: Cold dawn light"
                ],
                "visual_inferences": [
                  "The light casts a bluish-gray tint over the exterior, lacking any warm sunlight."
                ],
                "creative_decisions": [
                  "Unifying the color grading and atmospheric fog texture to ensure the early morning mood remains consistent across wide and tight 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"
              },
              {
                "kind": "background",
                "id": "L07B01",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "ce05300773f0a338"
      },
      {
        "scene_index": 17,
        "heading": "S#19. 마트 / D",
        "beat_title": "CCTV 위치 확인과 결심",
        "representative_moment": "C01이 CCTV 렌즈를 향해 굳은 표정으로 고개를 들고 카메라를 응시한다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "tight CU, high angle, surveillance-style frontal frame with washed-out CCTV-like contrast",
            "t2i_prompt": "Photorealistic cinematic still. Surveillance-style tight framing from a high angle, C01O02 in a beige hooded jacket, a Korean woman in her early 20s showing a young face, black hair, and dark eyes, facing the camera with her chin lifted tense, eyes fixed directly on the lens. Her intense expression occupies the center, shoulders and upper chest entering the lower frame edge. [: A compressed background plane of a damp, stained pale concrete exterior wall visible behind her at a shallow angle]. The scene is bathed in dim, cool-toned daylight with a visible layer of faint morning fog that softens edges and mutes surrounding colors, creating a chilly and quiet atmosphere. Lacking any warm sunlight, a washed-out CCTV-like contrast flattens her features, while a faint red indicator glow reflects on her cheek and the side of the concrete wall, adding menace. A subtle digital surveillance texture overlays the image.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "shot description: CCTV 렌즈를 향해 턱을 치켜든 수리영의 굳은 얼굴 정면 구도.",
              "S#19 action: 마트 밖 CCTV 위치를 확인하다.",
              "camera_direction: tight CU, high angle, surveillance-style frontal frame as 수리영 lifts her chin directly toward the lens."
            ],
            "visual_inferences": [
              "fixed_element cold_dawn_lighting_state applies dim, cool-toned daylight with faint morning fog.",
              "To avoid floating head synthesis in the tight CU, the shoulders and upper chest are kept visible at the lower frame edge.",
              "Perception mode 'through_device' is represented by a subtle digital surveillance texture and washed-out contrast without drawing explicit device frames."
            ],
            "creative_decisions": [
              "Added a faint red indicator glow reflecting on the subject's cheek to emphasize the active recording state of the CCTV and enhance the menacing tension.",
              "Compressed the background entirely to the immediate damp concrete wall to adhere to skipped_close_framing rules while maintaining the spatial logic of a wall-mounted camera."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "a52b6863807b74d3",
              "owned_hash": "9c311eb384197b50",
              "camera_direction_hash": "da6841d00ebbd309",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 17,
        "dependency_reason": "같은 마트 밖 공간의 차가운 새벽빛(cold_dawn_lighting_state) 톤과 안개 낀 분위기를 일관되게 유지하기 위함.",
        "_shot_index": 5,
        "visible_entities": [
          "C01"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 17,
            "shot_index": 5
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Mounted high beside the exterior CCTV axis, the camera holds a tight CU, high angle, surveillance-style frontal frame as 수리영 lifts her chin directly toward the lens. Her face occupies most of the center, while the wall and CCTV hardware are reduced to thin blurred edges, making the confrontation feel like she is staring into recorded evidence itself.",
            "lighting_mood": "Cold dawn light and washed-out CCTV-like contrast flatten her features, with a faint red indicator glow near the camera side adding menace.",
            "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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L07B02",
            "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_dawn_lighting_state",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A specific lighting state of the outdoor environment, featuring cold, fog-muted dawn light. The scene is bathed in dim, cool-toned daylight with a visible layer of faint morning fog that softens edges and mutes surrounding colors, creating a chilly and quiet atmosphere.",
                "applies_to_shots": [
                  1,
                  5
                ],
                "source_facts": [
                  "scene heading text: 이른 아침 (early morning)",
                  "Shot 1 lighting: Cool fog-muted dawn light dominates",
                  "Shot 5 lighting: Cold dawn light"
                ],
                "visual_inferences": [
                  "The light casts a bluish-gray tint over the exterior, lacking any warm sunlight."
                ],
                "creative_decisions": [
                  "Unifying the color grading and atmospheric fog texture to ensure the early morning mood remains consistent across wide and tight shots."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 17,
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "d27f08c6c52bb6c6"
      },
      {
        "scene_index": 18,
        "heading": "S#20. 마트 총무과 / 실내",
        "beat_title": "식별 불가능한 CCTV와 수리영의 초조함",
        "representative_moment": "혜수의 어깨 너머로 모니터를 향해 목을 빼고 초조하게 화면을 주시하는 수리영의 얼굴 클로즈업.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Over-the-shoulder close composition, eye-level, shallow depth of field, cool blue-gray monitor lighting against dim shadows",
            "t2i_prompt": "Photorealistic cinematic still. Head-and-shoulder composition at eye level. The dark shoulder and back of an adult Asian woman in a dark coat occupy the left foreground as a soft, out-of-focus silhouette with her back to camera. Just behind her, sharing a single line of sight toward an off-frame monitor, C01O02 in a beige hoodie, an Asian woman in her early 20s with a young face and dark eyes, appears at three-quarter angle from the right, occupying the center-right. Her neck is craned forward past the foreground figure's shoulder, her eyes fixed on the unseen CCTV monitor. Focus on the anxious expression on her face. Cold blue-gray monitor light washes across her cheek and forehead. The immediate frame-edge background drops into dim, stale office darkness with soft cool shadows.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "Scene text: '수리영 불안이 깊어지고.' (Su-ri-young's anxiety deepens.)",
              "Shot description: Su-ri-young's anxious face fills the center-right, neck stretched forward, eyes fixed off-frame.",
              "DP instruction: Hye-su’s shoulder becomes a soft dark foreground edge on the left."
            ],
            "visual_inferences": [
              "fixed_element 'cold_cctv_monitor_office_lighting' provides the blue-gray monitor spill on Su-ri-young's face.",
              "fixed_element 'desk_cctv_monitor_dark_playback' motivates the shared eyeline toward the off-frame monitor.",
              "Hye-su is an unregistered character in this shot, so she is rendered as a common-noun adult Asian woman in silhouette."
            ],
            "creative_decisions": [
              "Used a head-and-shoulder composition to frame the close-up without violating the 'face filling the entire frame' constraint.",
              "Omitted background room details to comply with the skipped_close_framing mode, relying entirely on dim, cool shadows at the frame edge.",
              "Applied the fg_bg_shared_anchor_rule by explicitly stating both figures share a single line of sight toward the off-frame monitor."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "0eaaaf8f6aec8e65",
              "owned_hash": "93ac7ff8cd40c3b5",
              "camera_direction_hash": "8f130c24a43289f4",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "Not strictly dependent on previous scene images for generation, relies entirely on current scene's fixed_elements for continuity.",
        "_shot_index": 2,
        "visible_entities": [
          "C01"
        ],
        "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": "Just behind and slightly beside 혜수, the camera tracks into an OTS CU, eye-level frame; 혜수’s shoulder becomes a soft dark foreground edge on the left while 수리영’s anxious face fills the center-right. 수리영 leans past 혜수 toward the unseen monitor, her neck stretched forward and eyes fixed off-frame.",
            "lighting_mood": "Blue-gray monitor light cuts across humid, stale office darkness, leaving 혜수 as a soft silhouette and 수리영’s eyes tense and wet.",
            "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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L11B02",
            "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,
                  4
                ],
                "character_name": "",
                "confidence": "high",
                "creative_decisions": [
                  "Keep the blue-gray CCTV monitor glow as the dominant visual continuity cue across both selected shots.",
                  "Maintain a cold investigative atmosphere in the office rather than introducing warm ambient light."
                ],
                "description": "The supermarket administrative security office remains dim, with cold blue-gray light from a CCTV monitor washing across the people near the desk. Any overhead fluorescent light stays dull and dirty, mixing weakly with the monitor spill and leaving the room in muted, cool shadows.",
                "element_id": "cold_cctv_monitor_office_lighting",
                "element_type": "environment_state",
                "source_facts": [
                  "Scene text: CCTV footage is being replayed in the supermarket administrative office/security room.",
                  "Shot 2 lighting: 'Blue-gray monitor light cuts across humid, stale office darkness.'",
                  "Shot 4 lighting: 'Dirty fluorescent overheads mix with blue monitor spill.'"
                ],
                "visual_inferences": [
                  "The same cold monitor spill is treated as a continuous lighting condition across the selected shots.",
                  "The office darkness is kept consistent even when the camera angle changes."
                ]
              },
              {
                "applies_to_shots": [
                  2,
                  4
                ],
                "character_name": "",
                "confidence": "high",
                "creative_decisions": [
                  "Use the CCTV monitor as the central fixed prop motivating eyelines and blue light, even when it is partly off-frame.",
                  "Keep the dark surveillance segment visually unresolved so the scene’s investigative tension remains consistent."
                ],
                "description": "A CCTV monitor at the desk is replaying surveillance footage in which one crucial section appears abnormally dark and difficult to identify. The screen faces the people gathered around the desk and acts as the source of the cold blue-gray glow in the room.",
                "element_id": "desk_cctv_monitor_dark_playback",
                "element_type": "persistent_prop",
                "source_facts": [
                  "Scene text: the security staff and others are reviewing CCTV footage.",
                  "Scene text: '그 부분만 유독 어두워 식별이 불가능하다' — that portion is unusually dark and impossible to identify.",
                  "Shot 2 staging: eyes are fixed toward the CCTV monitor, with blue-gray monitor light present.",
                  "Shot 4 lighting: blue monitor spill is present during the exchange."
                ],
                "visual_inferences": [
                  "The CCTV monitor is placed on or immediately beside the desk in front of the characters.",
                  "The dark section remains on the same playback source across both shots."
                ]
              }
            ],
            "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"
            ]
          },
          "render_contracts": [],
          "_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": "a985107b56b9e75a"
      },
      {
        "scene_index": 18,
        "heading": "S#20. 마트 총무과 / 실내",
        "beat_title": "CCTV 전체 확보 요청",
        "representative_moment": "C02가 턱에 손을 괸 채 데스크 너머의 보안직원을 주시하며 영상 복사를 요청한다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Medium close-up, slight low angle, cold blue-gray monitor spill mixing with dirty overheads",
            "t2i_prompt": "Photorealistic cinematic still. Medium close-up, slight low angle looking upward. C02O05 in a charcoal grey suit jacket, a Korean woman in her early 40s with black hair and dark eyes, occupies the left-center in right profile, upper body visible, chin resting on her hand, eyes fixed on the security staff. Across the shared security desk from her, an East Asian man in a worker uniform remains in the soft background in left profile, leaning toward the desk, his eyes fixed on C02O05. The desk's front edge occupies the lower frame, holding a CCTV playback monitor angled partly away from the camera. Cold blue-gray monitor spill washes the side of C02O05's face and the desk surface, mixing with weak yellow light from a downward-pointing desk lamp. [L11: A messy notice board surface and dirty fluorescent overhead lamps visible at the upper frame edge, casting muted cool shadows.] Focus on the investigative hardness of C02O05's profile and the tension anchored at the desk.",
            "outfit_assignments": [
              {
                "character_id": "C02",
                "outlook_id": "O05"
              }
            ],
            "source_facts": [
              "Scene text: 혜수 asks to copy the entire CCTV footage.",
              "Shot description: 혜수's upper body shot, hand braced under her chin, eyes fixed on the security staff.",
              "Staging: 혜수 in profile_right, security staff in profile_left leaning toward desk."
            ],
            "visual_inferences": [
              "fixed_element cold_cctv_monitor_office_lighting: cold blue-gray monitor spill washes the subjects' faces.",
              "fixed_element desk_cctv_monitor_dark_playback: the CCTV monitor sits on the shared desk acting as a light source and physical divider.",
              "Close framing scale makes the security staff soft background while keeping sharp focus on C02."
            ],
            "creative_decisions": [
              "Slight low angle places the overhead fluorescent lamps at the upper frame edge, reinforcing the geometrically constrained office atmosphere.",
              "Cold color temperature enhances the investigative hardness of C02's expression."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "41e7797c4571e251",
              "owned_hash": "93ac7ff8cd40c3b5",
              "camera_direction_hash": "252dad1f347820eb",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 18,
        "dependency_reason": "S18_Shot2의 CCTV 모니터 조명(blue-gray spill)과 데스크를 중심으로 한 인물 배치 연속성 유지",
        "_shot_index": 4,
        "visible_entities": [
          "C02",
          "L11"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 18,
            "shot_index": 4
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "The lateral move settles beside the desk into a steady MCU, slight low angle across 혜수 toward 보안직원’s side profile. 혜수 occupies the left-center with her hand braced under her chin, while 보안직원 remains in the right midground as her eyeline target, keeping the investigative pressure on their exchange.",
            "lighting_mood": "Dirty fluorescent overheads mix with blue monitor spill, giving 혜수’s focused profile a cold investigative hardness.",
            "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": "O05"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L11B02",
            "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,
                  4
                ],
                "character_name": "",
                "confidence": "high",
                "creative_decisions": [
                  "Keep the blue-gray CCTV monitor glow as the dominant visual continuity cue across both selected shots.",
                  "Maintain a cold investigative atmosphere in the office rather than introducing warm ambient light."
                ],
                "description": "The supermarket administrative security office remains dim, with cold blue-gray light from a CCTV monitor washing across the people near the desk. Any overhead fluorescent light stays dull and dirty, mixing weakly with the monitor spill and leaving the room in muted, cool shadows.",
                "element_id": "cold_cctv_monitor_office_lighting",
                "element_type": "environment_state",
                "source_facts": [
                  "Scene text: CCTV footage is being replayed in the supermarket administrative office/security room.",
                  "Shot 2 lighting: 'Blue-gray monitor light cuts across humid, stale office darkness.'",
                  "Shot 4 lighting: 'Dirty fluorescent overheads mix with blue monitor spill.'"
                ],
                "visual_inferences": [
                  "The same cold monitor spill is treated as a continuous lighting condition across the selected shots.",
                  "The office darkness is kept consistent even when the camera angle changes."
                ]
              },
              {
                "applies_to_shots": [
                  2,
                  4
                ],
                "character_name": "",
                "confidence": "high",
                "creative_decisions": [
                  "Use the CCTV monitor as the central fixed prop motivating eyelines and blue light, even when it is partly off-frame.",
                  "Keep the dark surveillance segment visually unresolved so the scene’s investigative tension remains consistent."
                ],
                "description": "A CCTV monitor at the desk is replaying surveillance footage in which one crucial section appears abnormally dark and difficult to identify. The screen faces the people gathered around the desk and acts as the source of the cold blue-gray glow in the room.",
                "element_id": "desk_cctv_monitor_dark_playback",
                "element_type": "persistent_prop",
                "source_facts": [
                  "Scene text: the security staff and others are reviewing CCTV footage.",
                  "Scene text: '그 부분만 유독 어두워 식별이 불가능하다' — that portion is unusually dark and impossible to identify.",
                  "Shot 2 staging: eyes are fixed toward the CCTV monitor, with blue-gray monitor light present.",
                  "Shot 4 lighting: blue monitor spill is present during the exchange."
                ],
                "visual_inferences": [
                  "The CCTV monitor is placed on or immediately beside the desk in front of the characters.",
                  "The dark section remains on the same playback source across both shots."
                ]
              }
            ],
            "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": "C02O05",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "b0d2d86932c59cdf"
      },
      {
        "scene_index": 19,
        "heading": "S#21. 경찰서 안 / D",
        "beat_title": "살인사건 사진 확인과 경악",
        "representative_moment": "책상 위 널려있는 살인사건 사진을 내려다보며 경악하는 C01의 얼굴 클로즈업.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Low angle close-up from the desk corner, shallow depth of field, cold fluorescent lighting",
            "t2i_prompt": "Photorealistic cinematic still. Low observer at desk level looking slightly up, tight framing on her expression with upper chest and shoulders visible. C01O02 in a beige hoodie, a young East Asian woman in her early 20s with a young face, black hair, and dark pupils, positioned at three-quarter angle from the left. She is hunched close over the tabletop, her head angled downward, eyes fixed on the scattered case photographs below. Focus on her wide, dilated eyes frozen in panic. In the lower foreground, a blurred, scratched desk edge cuts across the frame. Resting on the desk surface under her immediate view are several glossy crime scene photographs, scattered and overlapping, their front faces lying upward, displaying gruesome red-brown details partially visible beneath the shallow focus. [L10B01: A gray, worn office partition receding behind her shoulder and a computer monitor screen glowing at an oblique angle remain out of focus.] The space is illuminated by harsh, cold overhead fluorescent lights casting a sterile bluish-white glare across the flat surfaces, softened by the monitor spill, draining her face pale while deep, heavy shadows pool around the desk.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "scene text: '혜수 책상위에는 살인사건 확대 사진들이 어지럽게 널려있다. 목, 어깨, 얼굴등 살점이 떨어져'",
              "Shot 4: 'her gaze drops toward the evidence below'"
            ],
            "visual_inferences": [
              "fixed_element scattered_evidence_photos: glossy photographs directly resting on the desk under the character's immediate view",
              "fixed_element cold_fluorescent_lighting: cold bluish-white glare across flat surfaces, deepening the somber shadows in the background",
              "The shallow depth of field blurs both the gruesome foreground details and the impersonal office background, isolating the character's shock."
            ],
            "creative_decisions": [
              "Low angle from the desk level to emphasize her hunched, vulnerable posture against the evidence.",
              "A blurred foreground desk edge to create a claustrophobic framing that boxes her in.",
              "Cool blue-white lighting to drain warmth from the character, heightening the psychological thriller tone."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "01c9ba32a2bf6e2f",
              "owned_hash": "704a968fbe0b5a95",
              "camera_direction_hash": "ac2555c3f6e685a3",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 14,
        "dependency_reason": "S14_Shot5 provides background location context, but this shot uses close_framing_skips_background_ref to compose solely from immediate surroundings.",
        "_shot_index": 4,
        "visible_entities": [
          "C01"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 19,
            "shot_index": 4
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "From the front corner of 혜수’s desk, the camera pushes into a CU, low angle with shallow DOF; the blurred desk edge cuts across the lower foreground while 수리영’s dilated eyes dominate the upper center. She is hunched close over the tabletop, frozen between recognition and panic as her gaze drops toward the evidence below.",
            "lighting_mood": "Harsh police-station fluorescents are softened by humid air and monitor spill, draining her face pale while the desk shadows feel heavy.",
            "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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L10B01",
            "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": "scattered_evidence_photos",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "Several glossy crime scene photographs are chaotically scattered and overlapping on the surface of a standard office desk. The visible images display gruesome, torn flesh details with dark red-brown tones, contrasting sharply with the mundane desk surface.",
                "applies_to_shots": [
                  4,
                  5
                ],
                "source_facts": [
                  "scene text: '혜수 책상위에는 살인사건 확대 사진들이 어지럽게 널려있다. 목, 어깨, 얼굴등 살점이 떨어져'",
                  "Shot 4: 'her gaze drops toward the evidence below'",
                  "Shot 5: 'The images overlap chaotically across the desk, with torn neck and shoulder details isolated in the center'"
                ],
                "visual_inferences": [
                  "The photographs are directly resting on the desk under the character's immediate view, acting as a horrifying focal point among otherwise normal office items."
                ],
                "creative_decisions": [
                  "Highlighting the red-brown wound details within the photos to create a sickening, warm accent against the cold desk environment."
                ],
                "confidence": "high"
              },
              {
                "element_id": "cold_fluorescent_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The police station environment is illuminated by harsh, cold overhead fluorescent lights. This lighting casts a sterile, bluish-white glare across flat surfaces like the desk and glossy photos, while deepening the somber, heavy shadows in the background.",
                "applies_to_shots": [
                  4,
                  5,
                  7
                ],
                "source_facts": [
                  "Shot 4: 'Harsh police-station fluorescents are softened by humid air and monitor spill'",
                  "Shot 5: 'Cold fluorescent light makes the glossy photo surfaces glare in patches'",
                  "Shot 7: 'cutting through the police-office gloom'"
                ],
                "visual_inferences": [
                  "The overall ambient light in the room is distinctly cool-toned and uninviting, typical of institutional spaces at night or in tense situations."
                ],
                "creative_decisions": [
                  "Using the contrast between the cold, harsh overhead glares and the deep background gloom to amplify the psychological tension of the discovery."
                ],
                "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"
              }
            ],
            "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"
            ]
          },
          "render_contracts": [],
          "_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": "02dd70b95830a941"
      },
      {
        "scene_index": 19,
        "heading": "S#21. 경찰서 안 / D",
        "beat_title": "끔찍한 살인사건 사진과의 직면",
        "representative_moment": "책상 위에 어지럽게 널린 끔찍한 살인사건 현장 사진들(P07)을 위에서 내려다보는 시점.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Steep overhead subjective POV, harsh fluorescent glare with contrasting red-brown accents",
            "t2i_prompt": "Photorealistic cinematic still. Overhead composition looking directly down at a tabletop, focusing on P07. P07, a set of glossy crime scene photographs, are chaotically scattered and overlapping. The glossy front faces of the photos are turned upward to the camera, displaying forensic images of torn neck and shoulder flesh in dark red-brown tones. Small printed case-number labels are attached to the photo borders, partly cropped and unreadable. A loose cluster of dull silver paperclips rests beside the overlapping photo edges, their bent metal loops facing upward. Harsh, cold overhead fluorescent light casts a sterile bluish-white glare in patches across the glossy surfaces, making the dark red-brown wound details stand out as a sickening, warm accent. The items rest directly on [: a mundane, scratched gray laminate police desk surface] filling the frame beneath them, with all human presence completely cropped out.",
            "outfit_assignments": [],
            "source_facts": [
              "scene text: '혜수 책상위에는 살인사건 확대 사진들이 어지럽게 널려있다. 목, 어깨, 얼굴등 살점이 떨어져'",
              "Shot 5: 'The images overlap chaotically across the desk, with torn neck and shoulder details isolated in the center'"
            ],
            "visual_inferences": [
              "fixed_element scattered_evidence_photos: photographs are directly resting on the desk under the character's immediate view",
              "fixed_element cold_fluorescent_lighting: cool-toned ambient light causes glare on the glossy paper",
              "Point-of-view composition fully crops out the observing character to simulate a direct downward gaze at the evidence"
            ],
            "creative_decisions": [
              "Using a steep overhead angle to isolate P07 against the desk surface, heightening the psychological shock",
              "Applying patches of harsh bluish-white glare on the photos to contrast sharply with the dark red-brown gore, amplifying the horrific visual impact"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "5ac8c0246008d30b",
              "owned_hash": "704a968fbe0b5a95",
              "camera_direction_hash": "839082b79a8927af",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 19,
        "dependency_reason": "직전 샷(Shot 4)의 공간적 맥락에 이어, 동일한 조명(cold fluorescent)과 소품(scattered evidence photos)의 디테일과 연속성을 유지",
        "_shot_index": 5,
        "visible_entities": [
          "P07"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 19,
            "shot_index": 5
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Following 수리영’s gaze, the camera holds a steep high angle CU, point-of-view tabletop insert over the scattered murder-scene photographs. The images overlap chaotically across the desk, with torn neck and shoulder details isolated in the center while all human presence is cropped out.",
            "lighting_mood": "Cold fluorescent light makes the glossy photo surfaces glare in patches, turning the red-brown wound details into the only warm, nauseating 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": [
              "P07"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L10B01",
            "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": "scattered_evidence_photos",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "Several glossy crime scene photographs are chaotically scattered and overlapping on the surface of a standard office desk. The visible images display gruesome, torn flesh details with dark red-brown tones, contrasting sharply with the mundane desk surface.",
                "applies_to_shots": [
                  4,
                  5
                ],
                "source_facts": [
                  "scene text: '혜수 책상위에는 살인사건 확대 사진들이 어지럽게 널려있다. 목, 어깨, 얼굴등 살점이 떨어져'",
                  "Shot 4: 'her gaze drops toward the evidence below'",
                  "Shot 5: 'The images overlap chaotically across the desk, with torn neck and shoulder details isolated in the center'"
                ],
                "visual_inferences": [
                  "The photographs are directly resting on the desk under the character's immediate view, acting as a horrifying focal point among otherwise normal office items."
                ],
                "creative_decisions": [
                  "Highlighting the red-brown wound details within the photos to create a sickening, warm accent against the cold desk environment."
                ],
                "confidence": "high"
              },
              {
                "element_id": "cold_fluorescent_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The police station environment is illuminated by harsh, cold overhead fluorescent lights. This lighting casts a sterile, bluish-white glare across flat surfaces like the desk and glossy photos, while deepening the somber, heavy shadows in the background.",
                "applies_to_shots": [
                  4,
                  5,
                  7
                ],
                "source_facts": [
                  "Shot 4: 'Harsh police-station fluorescents are softened by humid air and monitor spill'",
                  "Shot 5: 'Cold fluorescent light makes the glossy photo surfaces glare in patches'",
                  "Shot 7: 'cutting through the police-office gloom'"
                ],
                "visual_inferences": [
                  "The overall ambient light in the room is distinctly cool-toned and uninviting, typical of institutional spaces at night or in tense situations."
                ],
                "creative_decisions": [
                  "Using the contrast between the cold, harsh overhead glares and the deep background gloom to amplify the psychological tension of the discovery."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 19,
                "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": [],
            "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"
            ]
          },
          "render_contracts": [],
          "_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": "1105cbf5be5a4ab7"
      },
      {
        "scene_index": 19,
        "heading": "S#21. 경찰서 안 / D",
        "beat_title": "해상 좌표 확인",
        "representative_moment": "모니터 화면 속 해상 지도에 찍힌 붉은 핀과 영문 좌표 숫자가 선명하게 빛나는 클로즈업.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Extreme close-up screen insert, eye-level lens nearly parallel to display, cold blue-white monitor spill with deep gloom at the margins",
            "t2i_prompt": "Photorealistic cinematic still. Extreme close-up on a glowing computer monitor screen, the lens positioned nearly parallel to the display. The screen shows a digital sea map interface, with a crisp, bright red location pin graphic sitting in the upper-left third and glowing white English coordinate text in the lower-right third. The monitor's cold blue-white light dominates the frame, casting a sterile digital glare, while the sharp red pin acts as a striking focal marker. The screen bezel nearly disappears at the edges, leaving only [: a thin, out-of-focus strip of a dark stationary keyboard and desk clutter at the bottom margin] to anchor the display against the surrounding deep police-office gloom. Focus on the bright red map pin and the glowing coordinate numbers on the screen.",
            "outfit_assignments": [],
            "source_facts": [
              "Shot 7: 모니터 화면 속 해상 지도의 붉은 핀과 영문 좌표 숫자가 적힌 글씨 클로즈업.",
              "S#21 screen text: LAST LOCATION 해상 지도위에 좌표. 37.484400 N / 126. 072486 E",
              "DP instructions: Beside the monitor, the camera completes the move into an ECU, eye-level screen insert with the lens nearly parallel to the display"
            ],
            "visual_inferences": [
              "fixed_element cold_fluorescent_lighting: harsh, cold overhead fluorescent lights cast a sterile glare and deepen background shadows",
              "Close framing skips the broader room, anchoring the monitor solely with the desk and blurred keyboard edge below the screen."
            ],
            "creative_decisions": [
              "Used an extreme close-up to fill the frame with the screen interface, isolating the red pin and coordinates as the sole information targets.",
              "Applied cold blue-white digital light contrasting heavily with the crisp red pin to emphasize the ominous realization of the location."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "29653e58cd878119",
              "owned_hash": "704a968fbe0b5a95",
              "camera_direction_hash": "add66a580ceccd5e",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 19,
        "dependency_reason": "Shot 5의 경찰서 환경(cold fluorescent lighting)과 연관되나, 본 샷은 ECU 화면 클로즈업이므로 배경 렌더링을 생략하고 모니터와 키보드 엣지에만 빛의 영향을 반영함.",
        "_shot_index": 7,
        "visible_entities": [],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 19,
            "shot_index": 7
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Beside the monitor, the camera completes the move into an ECU, eye-level screen insert with the lens nearly parallel to the display; the red pin sits near the upper-left third and the English coordinate numbers sit sharply in the lower-right third. The bezel nearly disappears, but a thin edge of keyboard blur and desk clutter remain at the margins to keep the screen physically anchored.",
            "lighting_mood": "The monitor’s cold blue-white light becomes the dominant source, with the red pin cutting through the police-office gloom like a warning flare.",
            "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": [],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L10B01",
            "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_fluorescent_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The police station environment is illuminated by harsh, cold overhead fluorescent lights. This lighting casts a sterile, bluish-white glare across flat surfaces like the desk and glossy photos, while deepening the somber, heavy shadows in the background.",
                "applies_to_shots": [
                  4,
                  5,
                  7
                ],
                "source_facts": [
                  "Shot 4: 'Harsh police-station fluorescents are softened by humid air and monitor spill'",
                  "Shot 5: 'Cold fluorescent light makes the glossy photo surfaces glare in patches'",
                  "Shot 7: 'cutting through the police-office gloom'"
                ],
                "visual_inferences": [
                  "The overall ambient light in the room is distinctly cool-toned and uninviting, typical of institutional spaces at night or in tense situations."
                ],
                "creative_decisions": [
                  "Using the contrast between the cold, harsh overhead glares and the deep background gloom to amplify the psychological tension of the discovery."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 19,
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "bbd966bff3ff725c"
      },
      {
        "scene_index": 20,
        "heading": "S#22. 경찰서 밖 / D",
        "beat_title": "급박한 출발",
        "representative_moment": "수리영이 자전거 안장에서 엉덩이를 떼고 페달에 체중을 실으며 경찰서 주차장을 급히 빠져나가는 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Low three-quarter side tracking, muted overcast coastal daylight with humid reflections",
            "t2i_prompt": "Photorealistic cinematic still. C01O02 in a beige hoodie, a Korean young woman with black hair and dark eyes, in left profile, caught mid-pedal on the bicycle from the reference. Hips lifted off the saddle, one foot driving down on the pedal, her body leaning forward, black hair blowing back in the direction opposite to her forward motion. Her eyes are fixed on the distant road. Observed from a low position beside the subject, her shoes and the bicycle frame occupying the lower frame, her upper body extending toward the upper edge. The wet pavement from the reference reflects the tires and her legs below. Muted overcast daylight washes across her shoulder, a faint red glow catching the rear reflector. The damp concrete of the station building and doors from the reference recede behind her. Focus on her determined profile and the tense forward lean of her posture.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "S#22 action: 주차된 자전거를 타고 이내 달리는 수리영",
              "shot description: 자전거 안장에서 엉덩이를 뗀 채 한쪽 페달에 체중을 실은 수리영의 역동적인 전신 구도"
            ],
            "visual_inferences": [
              "C01 left profile pose, mid-action freeze capturing urgent physical force",
              "low observer position framing feet and wheels at the bottom edge",
              "wet road reflection tracking the bicycle's movement"
            ],
            "creative_decisions": [
              "hair flowing backward to emphasize motion direction against the static frame",
              "anchoring the bicycle and station building explicitly to the background reference to prevent duplicate prop synthesis",
              "muted coastal daylight with a faint red accent on the reflector for visual contrast"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "cd7b71751a0a78ea",
              "owned_hash": "8ce1510a9509e291",
              "camera_direction_hash": "691f00811483660b",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "bicycle",
                  "violating_phrase": "the bicycle from the reference",
                  "reason": "The phrase explicitly anchors the bicycle to the reference instead of asking to create a new one, so it is not a redraw instruction.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "road",
                  "violating_phrase": "the distant road",
                  "reason": "This is a spatial/scene reference to where she is looking, not an instruction to add or redraw a road.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "station building",
                  "violating_phrase": "the damp concrete of the station building",
                  "reason": "The station building is only mentioned as an existing background element being observed from the reference, with no redraw verb.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "door",
                  "violating_phrase": "doors from the reference",
                  "reason": "The doors are explicitly tied to the reference/background, which makes this an anchored mention rather than a redraw instruction.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "Not directly dependent on a specific prior visual shot.",
        "_shot_index": 2,
        "visible_entities": [
          "C01",
          "L12"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 20,
            "shot_index": 2
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Tracking parallel from a low three-quarter side position about two meters away, the camera frames a WS, low angle of 수리영’s full body and bicycle as she pushes off from the parking spot. Her hips lift off the saddle, one foot drives down on the pedal, and the whole bike angles forward across the wet police-station exterior, giving the departure urgent physical force.",
            "lighting_mood": "Overcast coastal daylight stays muted and humid, with slick ground reflections and a faint red brake reflector adding the only sharp 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": [
              "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",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L12B01",
            "reference_usage": "exact_background",
            "owned_objects": [
              "CCTV pole",
              "barrier gate",
              "bicycle",
              "bicycle rack",
              "curb",
              "door",
              "parking lot",
              "parking space",
              "patrol car",
              "planter",
              "road",
              "sidewalk",
              "signboard",
              "station building",
              "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": "character_outlook",
                "id": "C01O02",
                "policy": "required"
              },
              {
                "kind": "background",
                "id": "L12B01",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "775543f554fe5faf"
      },
      {
        "scene_index": 21,
        "heading": "S#23. 어느 PC방 / 실내",
        "beat_title": "좌표 검색과 긴장된 얼굴",
        "representative_moment": "어두운 PC방 안, 푸른 모니터 불빛을 받으며 화면을 뚫어지게 응시하는 수리영(C01)의 굳은 얼굴.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "A static close-up (CU) at eye-level, symmetrical frontal monitor-axis view, cold blue monitor light reflecting on the face",
            "t2i_prompt": "Photorealistic cinematic still. Head-and-shoulder composition. C01O02 in a beige hoodie, an early 20s Korean woman with a young face, black hair, and dark eyes, facing the camera, hunched at the workstation. Her face occupies the center of the frame, with her neck and upper chest visible at the bottom edge. Her eyes are fixed straight ahead on the offscreen monitor. Focus on her tense expression in the practical glow. Cold blue monitor light reflects smoothly on her cheeks, nose, and the front of her hoodie, with muted coastal humidity in the shadows and almost no warm fill. [: dark PC cafe environment, blurred staggered booth partitions receding into a soft tunnel-like depth directly behind her, murky low-contrast interior shadows].",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "source_facts": [
              "shot description: 어두운 PC방 안, 푸른빛이 감도는 모니터 불빛에 비친 수리영의 굳은 얼굴 정면 구도.",
              "S#23 action: 모니터 화면에 앉아있는 수리영, 좌표를 찍고 위치를 확인하고..."
            ],
            "visual_inferences": [
              "C01 is hunched forward to emphasize her intense focus on the screen",
              "The monitor is off-camera but acts as the primary foreground light source"
            ],
            "creative_decisions": [
              "Used a head-and-shoulder composition to prevent floating-face artifacts while keeping the focus entirely on her expression",
              "Emphasized cold blue light reflecting on the face to capture the chilling realization of the coordinates"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "c762e26384cbc481",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "b6cb85ef99f8d1f1",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "",
        "_shot_index": 1,
        "visible_entities": [
          "C01"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 21,
            "shot_index": 1
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A static close-up (CU) at eye-level holds a symmetrical frontal monitor-axis view from just beyond the implied screen line, with 수리영 centered and the dark PC café receding into soft rows behind her. The framing is direct observer vision, not a device POV; her face occupies about 70% of the frame while the monitor itself stays offscreen, suggested only by the blue light flattening her features.",
            "lighting_mood": "Cold blue monitor light is the dominant practical source, with muted coastal humidity in the shadows and almost no warm fill.",
            "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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O02"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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"
            ]
          },
          "render_contracts": [],
          "_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": "e73260821f51f4f1"
      },
      {
        "scene_index": 22,
        "heading": "해안가 / 소나기",
        "beat_title": "거센 폭풍우와 번개가 치는 해안",
        "representative_moment": "어두운 먹구름이 가득한 해안가 하늘 위로 거대한 번개가 번쩍이는 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Extreme wide establishing shot, high upward-tilted angle, cold slate-blue and charcoal palette briefly overexposed by a massive lightning flash.",
            "t2i_prompt": "Photorealistic cinematic still. Extreme wide establishing shot from a high coastal position with an upward-tilted angle. Focus on a massive, harsh white lightning bolt illuminating the upper center of the thick, dark storm clouds. The sky and environment are dominated by cold slate-blue, charcoal, and muted storm-gray tones, briefly overexposed by the bright flash. Heavy diagonal rain streaks cut sharply through the dark foreground atmosphere. At the very bottom edge of the frame, a thin strip of black, wet shoreline sits silhouetted against the dark sea, compressing the space into one ominous, fog-softened plane. No human figure is present.",
            "outfit_assignments": [],
            "source_facts": [
              "Scene text: '천둥 번개와 함께 잔뜩 낀 먹구름이 세찬 비를 뿌린다'",
              "Shot 1 description: massive lightning bolt flashing white against dark clouds, thick rain-loaded storm clouds, thin dark shoreline at the bottom."
            ],
            "visual_inferences": [
              "fixed_element stormy_coastal_weather: heavy diagonal rain streaks cutting through the dark atmosphere, cold slate-blue and charcoal tones.",
              "The bright lightning flash creates a brief moment of stark contrast against the otherwise fog-softened darkness."
            ],
            "creative_decisions": [
              "High upward-tilted angle emphasizes the immense scale of the storm and the sky, reducing the shoreline to a minimal strip at the bottom.",
              "Cool slate-blue and storm-gray palette heightens the cold, hostile coastal atmosphere without human presence."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "38d32fb9f0330982",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "230013d2afee4183",
              "validator": "scene_detail_owned_objects.v1",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 9,
        "dependency_reason": "연속된 해안가 폭풍우 환경(stormy_coastal_weather)의 배경 분위기 참조 유지",
        "_shot_index": 1,
        "visible_entities": [],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 22,
            "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 near-frozen extreme wide / establishing shot (EWS) from a very high coastal position uses a high, upward-tilted angle across the storm ceiling, holding the massive lightning bolt in the upper center while the black shoreline sits as a thin lower strip. The static observer frame compresses sky and sea into one ominous plane, with rain streaks cutting diagonally through the foreground but no human figure visible.",
            "lighting_mood": "The palette is slate-blue and charcoal, briefly overexposed by the lightning flash before falling back into fog-softened 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": [],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "stormy_coastal_weather",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A severe storm environment characterized by heavy diagonal rain streaks cutting through the dark atmosphere. The sky and surroundings are dominated by cold slate-blue, charcoal, and muted storm-gray tones, periodically illuminated by bright, harsh lightning flashes from thick, dark storm clouds.",
                "applies_to_shots": [
                  1,
                  4
                ],
                "source_facts": [
                  "Scene text: '천둥 번개와 함께 잔뜩 낀 먹구름이 세찬 비를 뿌린다'",
                  "Shot 1: 'rain streaks cutting diagonally', 'slate-blue and charcoal, briefly overexposed by the lightning flash'",
                  "Shot 4: 'Storm-muted gray light', 'faint distant lightning flicker'"
                ],
                "visual_inferences": [
                  "The intensity of the rain and the gloomy, oppressive darkness remain constant regardless of whether the camera is far away or up close."
                ],
                "creative_decisions": [
                  "Using a highly cohesive cold gray and slate-blue color palette to visually link the extreme wide landscape with the intimate close-up."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 9,
                "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": [],
            "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"
            ]
          },
          "render_contracts": [],
          "_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": "f848e35d0d4937e6"
      },
      {
        "scene_index": 22,
        "heading": "S#24. 해안가 / 소나기",
        "beat_title": "거센 폭우 속 절박한 질주",
        "representative_moment": "C01이 빗방울이 맺힌 뺨과 핏발 선 눈으로 폭우 속을 뚫고 정면을 응시하며 자전거 페달을 밟고 있다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "tracking eye-level profile framing with shallow depth of field, storm-muted gray lighting with cold highlights and distant lightning",
            "t2i_prompt": "Photorealistic cinematic still. A tracking eye-level composition beside the subject, the right half of the frame occupied by C01O03 in a dark navy windbreaker, Korean woman in her early 20s with black hair. She is in left profile, leaning forward in mid-pedal motion, eyes fixed on the distant space, her neck and upper shoulder clearly visible. A single suspended raindrop clings to her cheek, and her bloodshot eye is wide open. Storm-muted gray light wraps her face, with cold highlights reflecting on the wet skin. Faint distant lightning flicker illuminates the heavy diagonal rain streaks falling through the dark atmosphere. Near the lower frame, the vibrating grip and front edge of the bicycle handlebar are visible, aligned forward. [: A blurred wet dark road smearing horizontally in shallow depth of field behind her, dominated by cold slate-blue and charcoal storm tones.]",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O03"
              }
            ],
            "source_facts": [
              "Shot description: 뺨을 타고 내리던 빗방울이 피부에 맺힌 채, 핏발 선 눈으로 정면을 고정하여 응시하는 수리영의 옆얼굴",
              "Scene text: 잔뜩 낀 먹구름이 세찬 비를 뿌린다. 자전거를 타고 빗속을 달리는 수리영."
            ],
            "visual_inferences": [
              "fixed_element stormy_coastal_weather: heavy diagonal rain streaks, cold slate-blue tones, faint lightning flicker",
              "mid-pedal motion with forward lean to convey the physical effort of riding in a storm",
              "shallow depth of field blurring the background to isolate her expression and emphasize forward movement"
            ],
            "creative_decisions": [
              "Positioned C01O03 in the right half of the frame to provide lead room for her intense forward gaze",
              "Included the vibrating bicycle grip at the lower edge to ground the riding motion without breaking the tight profile framing"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "01b3deecb6996487",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "d4e793989a73fafe",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 22,
        "dependency_reason": "동일 씬 내 앞선 샷의 환경 상태(stormy_coastal_weather) 및 젖은 질감 연속성 유지",
        "_shot_index": 4,
        "visible_entities": [
          "C01"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 22,
            "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 tracking extreme close-up (ECU) at eye-level runs shoulder-height beside 수리영, less than a meter from her rain-slick side profile, using profile framing and shallow DOF. Her bloodshot eye and one suspended raindrop sit in the right half of the frame, while the blurred wet road and bicycle motion smear horizontally behind her.",
            "lighting_mood": "Storm-muted gray light wraps her face with cold highlights on rain, punctuated by faint distant lightning flicker.",
            "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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O03"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "stormy_coastal_weather",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A severe storm environment characterized by heavy diagonal rain streaks cutting through the dark atmosphere. The sky and surroundings are dominated by cold slate-blue, charcoal, and muted storm-gray tones, periodically illuminated by bright, harsh lightning flashes from thick, dark storm clouds.",
                "applies_to_shots": [
                  1,
                  4
                ],
                "source_facts": [
                  "Scene text: '천둥 번개와 함께 잔뜩 낀 먹구름이 세찬 비를 뿌린다'",
                  "Shot 1: 'rain streaks cutting diagonally', 'slate-blue and charcoal, briefly overexposed by the lightning flash'",
                  "Shot 4: 'Storm-muted gray light', 'faint distant lightning flicker'"
                ],
                "visual_inferences": [
                  "The intensity of the rain and the gloomy, oppressive darkness remain constant regardless of whether the camera is far away or up close."
                ],
                "creative_decisions": [
                  "Using a highly cohesive cold gray and slate-blue color palette to visually link the extreme wide landscape with the intimate close-up."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 22,
                "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": "C01O03",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "1365c18139a417bb"
      },
      {
        "scene_index": 23,
        "heading": "S#25. 옥탑방 안 / 해질 무렵",
        "beat_title": "난장판이 된 집안과 낯선 침입의 흔적",
        "representative_moment": "비에 젖은 거실 바닥에 찍힌 진흙 묻은 발자국 클로즈업.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "A very low close-up at a shallow grazing angle, focusing tightly on the muddy tracks with the chaotic room blurred in the background, utilizing dark low-key dusk lighting with silver-gray reflections.",
            "t2i_prompt": "Photorealistic cinematic still. A very low close-up positioned just above the ground, lens nearly parallel to the surface. Focus tightly on smeared muddy footprints tracking across the center of [L04: a rain-wet interior floor]. The fresh tracks leave readable shoe shapes and dirty edges among thin, reflective rainwater puddles. The space is enveloped in dark, low-key dusk lighting with deep blue-black shadows. Fading evening light filters from an open, rain-damp window far beyond the floor line, turning the wet surface a cool silver-gray. In the softly out-of-focus background, the legs and undersides of overturned dining chairs angle toward the camera, while an unlit entrance doorway forms a heavily shadowed, dark rectangular void, conveying a sickly humid and unsettling atmosphere.",
            "outfit_assignments": [],
            "source_facts": [
              "비에 젖은 옥탑방 거실 바닥에 어지럽게 찍힌 진흙 묻은 낯선 발자국 클로즈업.",
              "거실에 난 창은 열려있고, 식탁 의자들을 쓰러져있다."
            ],
            "visual_inferences": [
              "fixed_element dusk_apartment_interior_lighting: dark, low-key dusk lighting, deep blue-black shadows, weak cool silver-gray evening light",
              "fixed_element shadowed_entrance_doorway: unlit architectural feature, heavily shadowed rectangular void",
              "A shallow grazing angle places the footprints in sharp foreground focus while the toppled chairs and dark doorway remain softly blurred in the background."
            ],
            "creative_decisions": [
              "Using a very low macro perspective to emphasize the invasion, making the muddy shoe prints the focal anchor.",
              "Reflective silver-gray puddles on the floor contrast with the deep blue-black shadows of the doorway to heighten the psychological tension and thriller tone."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "4fa55b189c65404a",
              "owned_hash": "4d307ad28636e606",
              "camera_direction_hash": "472430d9774ba92a",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 12,
        "dependency_reason": "동일한 옥탑방 내부 배경(이전 12씬에서 설정됨)을 참조하되, 본 샷은 close framing 룰에 따라 배경 레퍼런스를 배제하고 근접 피사체 주변 묘사만 수행함.",
        "_shot_index": 1,
        "visible_entities": [
          "L04"
        ],
        "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": "A very low close-up (CU) at a high-angle-to-surface perspective rides only about 15 cm above the wet living-room floor, lens nearly parallel as it dollies into muddy unfamiliar footprints. The muddy tracks occupy the central band of the frame, with the entrance side blurred behind and the disturbed room hinted by toppled shapes in the far background.",
            "lighting_mood": "Fading dusk leaks through the open window, turning the wet floor silver-gray while the room stays underlit and sickly humid.",
            "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",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L04B03",
            "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_apartment_interior_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior of the apartment is enveloped in dark, low-key dusk lighting. Deep blue-black shadows dominate the unlit space, while a weak, cool silver-gray evening light softly filters in from the outside, creating a gloomy, humid, and unsettling atmosphere across the rooms.",
                "applies_to_shots": [
                  1,
                  6,
                  11
                ],
                "source_facts": [
                  "Shot 1: Fading dusk leaks through the open window, turning the wet floor silver-gray while the room stays underlit.",
                  "Shot 6: Low dusk light mixes with a faint warm interior practical, gray-green shadows.",
                  "Shot 11: doorway falls into blue-black interior shadow while the exterior dusk gives her face a weak gray rim."
                ],
                "visual_inferences": [
                  "The lack of strong indoor lighting ensures the entire apartment interior consistently reads as an eerie, shadowed void heavily influenced by the cool dusk light from outside."
                ],
                "creative_decisions": [
                  "Unifying the color palette across different sections of the apartment with a blue-gray dusk tint and heavy black shadows to heighten the psychological tension and sense of abandonment."
                ],
                "confidence": "high"
              },
              {
                "element_id": "shadowed_entrance_doorway",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The main entrance doorway of the apartment stands as a stark, unlit architectural feature. The frame of the open door reveals a deep, heavily shadowed rectangular void, completely lacking any welcoming interior light.",
                "applies_to_shots": [
                  1,
                  11
                ],
                "source_facts": [
                  "Shot 1: the entrance side blurred behind.",
                  "Shot 11: faces back into the dark doorway... doorway remains a shadowed rectangle behind her."
                ],
                "visual_inferences": [
                  "Since the apartment is unlit, the entrance area receives no practical illumination, rendering it as a dark silhouette or a black rectangle in the background."
                ],
                "creative_decisions": [
                  "Using the unlit, shadowy doorway as a consistent framing device to emphasize the emptiness and unsettling nature of the ransacked apartment."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 12,
                "shot_index": 8,
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "7139f82293c1377a"
      },
      {
        "scene_index": 23,
        "heading": "S#25. 옥탑방 안 / 해질 무렵",
        "beat_title": "벽에 꽂힌 피 묻은 옛날 사진",
        "representative_moment": "A close-up of P02, the blood-stained photograph, pierced by a pin at the center of P08 on the wall.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Chest-height eye-level close-up, shallow DOF, central focus on the pin, low dusk lighting mixed with warm practical",
            "t2i_prompt": "Photorealistic cinematic still. A chest-height eye-level close-up with shallow depth of field, focused sharply on a small metal pin piercing paper. P02, an old, yellowed, blood-stained photograph, occupies the central area of the frame. The photograph's front surface faces the camera, revealing a faded image of a rural restaurant and two young East Asian girls, marred by dark blood smears across the cracked paper texture. Surrounding the photograph is P08, an uneven, rough red circular mark painted on the wall, completely visible to the camera like an ominous target. The composition is tightly anchored to [L04: a dim, damp-stained wall surface filling the background edges]. The scene is lit by low dusk ambient light mixing with a faint warm practical glow, making the red paint feel unnaturally saturated against the deep blue-black and gray-green shadow gradients across the wall.",
            "outfit_assignments": [],
            "source_facts": [
              "S#25 action: '그 가운데 꽂혀있는 피묻은 사진 한 장!', '7-80년대 쯤 보이는 시골 식당과 ... 어린 여자애 둘(12살)이 앉아있다.'",
              "shot description: A slow dolly-in close-up pushes toward the bedroom wall, focusing on the old photograph inside the red circle."
            ],
            "visual_inferences": [
              "The background is restricted entirely to the immediate wall surface to comply with the close framing rule.",
              "The girls in the photograph are described using common nouns and demographic descriptors as they are part of a reproduction surface.",
              "fixed_element dusk_apartment_interior_lighting creates the gray-green and blue-black shadow gradients."
            ],
            "creative_decisions": [
              "Using shallow depth of field to make the metal pin and blood flecks the sharp central focus.",
              "Describing the red paint as unnaturally saturated to heighten the occult tension against the cool dusk tones."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "a4e1fee56a6da93c",
              "owned_hash": "db5e9f1b6e8d2520",
              "camera_direction_hash": "a180b9b46640bcbf",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 23,
        "dependency_reason": "Shares the dusk interior lighting environment with the previous shot, applied tightly to the frame-edge wall surface.",
        "_shot_index": 6,
        "visible_entities": [
          "P02",
          "P08",
          "L04"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 23,
            "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 slow dolly-in close-up (CU) at chest-height eye-level pushes toward the bedroom wall, using shallow DOF so the pin point, blood flecks, and cracked paper texture become the sharp central focus. The old photograph stays within the middle 35–40% of the frame, surrounded by the red circle on the wall like an occult target rather than letting the prop unrealistically overtake the whole image.",
            "lighting_mood": "Low dusk light mixes with a faint warm interior practical, making the red circle feel unnaturally saturated against otherwise gray-green 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": [
              "P02",
              "P08",
              "L04"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L04B02",
            "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_apartment_interior_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior of the apartment is enveloped in dark, low-key dusk lighting. Deep blue-black shadows dominate the unlit space, while a weak, cool silver-gray evening light softly filters in from the outside, creating a gloomy, humid, and unsettling atmosphere across the rooms.",
                "applies_to_shots": [
                  1,
                  6,
                  11
                ],
                "source_facts": [
                  "Shot 1: Fading dusk leaks through the open window, turning the wet floor silver-gray while the room stays underlit.",
                  "Shot 6: Low dusk light mixes with a faint warm interior practical, gray-green shadows.",
                  "Shot 11: doorway falls into blue-black interior shadow while the exterior dusk gives her face a weak gray rim."
                ],
                "visual_inferences": [
                  "The lack of strong indoor lighting ensures the entire apartment interior consistently reads as an eerie, shadowed void heavily influenced by the cool dusk light from outside."
                ],
                "creative_decisions": [
                  "Unifying the color palette across different sections of the apartment with a blue-gray dusk tint and heavy black shadows to heighten the psychological tension and sense of abandonment."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 23,
                "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": "prop",
                "id": "P02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "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"
            ]
          },
          "render_contracts": [
            {
              "contract_id": "rc_001",
              "scope": {
                "scene_index": 23,
                "shot_indices": [
                  6
                ],
                "duration": "single_shot"
              },
              "targets": [
                {
                  "entity_id": "P02",
                  "role": "visual_target"
                }
              ],
              "requirements": [
                {
                  "dimension": "visual_identity",
                  "operation": "preserve",
                  "strength": "required",
                  "reference_policy": "use_entity_reference"
                }
              ]
            }
          ],
          "_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": "38d119ce9a1182f3"
      },
      {
        "scene_index": 23,
        "heading": "S#25. 옥탑방 안 / 해질 무렵",
        "beat_title": "사진을 쥔 채 홀린 듯 집을 나서는 수리영",
        "representative_moment": "어두운 현관문을 나서며 허공을 응시한 채 한쪽 발을 공중에 멈춘 C01",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Retreating slight low angle composition, utilizing the shadowed doorway as a dark framing device, with cool silver-gray rim light against blue-black interior shadows.",
            "t2i_prompt": "Photorealistic cinematic still. Positioned just outside the entrance at a slight low angle, the camera looks back into the doorway. C01O03 in a dark navy windbreaker, a Korean woman in her early 20s, occupies the center-left at a three-quarter angle from the left, captured mid-stride as she steps outward, one foot suspended midair in the lower frame, hair and jacket hem faintly swaying in the direction opposite to motion. Her eyes are fixed on the void in a vacant trance. She tightly clutches a creased, blood-stained old photograph close to her chest; the image side is turned toward her body so only a stained edge faces the camera. A worn P04 backpack is slung over her shoulder, its rear side visible. Behind her, the open door from the reference reveals a deep, heavily shadowed rectangular void, completely lacking interior light. The interior is enveloped in dark dusk lighting with deep blue-black shadows, while a weak, cool silver-gray exterior evening light softly traces a rim on her face and shoulder. Focus on her mid-stride pause and vacant expression against the pitch-black doorway.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O03"
              }
            ],
            "source_facts": [
              "S#25 action: 수리영, 마치 이끌리듯 백팩을 메고 밖으로 나간다. 사진을 손에 꼭 쥔 채.",
              "shot description: 백팩을 멘 채 어두워진 옥탑방 현관문 밖으로 걸어 나가는 도중, 한쪽 발이 공중에 떠 있는 찰나의 자세로 멈춘 수리영의 몽롱한 표정."
            ],
            "visual_inferences": [
              "fixed_element dusk_apartment_interior_lighting: The interior of the apartment is enveloped in dark, low-key dusk lighting with deep blue-black shadows.",
              "fixed_element shadowed_entrance_doorway: The main entrance doorway stands as an unlit architectural feature, revealing a heavily shadowed rectangular void.",
              "Slight low angle positions her suspended foot in the lower frame while capturing the dark doorway behind her."
            ],
            "creative_decisions": [
              "Using the pitch-black interior void of the doorway to sharply contrast with the weak exterior dusk rim light on her face.",
              "Concealing the photograph's content by angling it toward her chest, maintaining mystery while focusing on her psychological state."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "0c4b09283146ff81",
              "owned_hash": "4d307ad28636e606",
              "camera_direction_hash": "90d2421ae6e9807f",
              "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_reference rather than a request to redraw it.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 23,
        "dependency_reason": "Ref_usage exact_background requires keeping the background elements consistent with the prior shot's established room.",
        "_shot_index": 11,
        "visible_entities": [
          "C01",
          "P04",
          "L04"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 23,
            "shot_index": 11
          },
          "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 rooftop-room entrance, a retreating medium shot (MS) at a slight low angle faces back into the dark doorway as 수리영 steps out with one foot suspended midair. The observer frame holds her in the center-left, her old photograph clutched close at chest level, while the doorway remains a shadowed rectangle behind her rather than competing for attention.",
            "lighting_mood": "The doorway falls into blue-black interior shadow while the exterior dusk gives her face a weak gray rim, emphasizing her vacant trance.",
            "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",
              "L04"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O03"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L04B03",
            "reference_usage": "exact_background",
            "owned_objects": [
              "TV",
              "chair",
              "coffee table",
              "cooktop",
              "counter",
              "curtain",
              "dining table",
              "door",
              "footprint",
              "hallway",
              "herb pot",
              "kitchenette",
              "living room",
              "refrigerator",
              "shoe cabinet",
              "sink",
              "smartphone",
              "sofa",
              "washing machine",
              "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_apartment_interior_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The interior of the apartment is enveloped in dark, low-key dusk lighting. Deep blue-black shadows dominate the unlit space, while a weak, cool silver-gray evening light softly filters in from the outside, creating a gloomy, humid, and unsettling atmosphere across the rooms.",
                "applies_to_shots": [
                  1,
                  6,
                  11
                ],
                "source_facts": [
                  "Shot 1: Fading dusk leaks through the open window, turning the wet floor silver-gray while the room stays underlit.",
                  "Shot 6: Low dusk light mixes with a faint warm interior practical, gray-green shadows.",
                  "Shot 11: doorway falls into blue-black interior shadow while the exterior dusk gives her face a weak gray rim."
                ],
                "visual_inferences": [
                  "The lack of strong indoor lighting ensures the entire apartment interior consistently reads as an eerie, shadowed void heavily influenced by the cool dusk light from outside."
                ],
                "creative_decisions": [
                  "Unifying the color palette across different sections of the apartment with a blue-gray dusk tint and heavy black shadows to heighten the psychological tension and sense of abandonment."
                ],
                "confidence": "high"
              },
              {
                "element_id": "shadowed_entrance_doorway",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The main entrance doorway of the apartment stands as a stark, unlit architectural feature. The frame of the open door reveals a deep, heavily shadowed rectangular void, completely lacking any welcoming interior light.",
                "applies_to_shots": [
                  1,
                  11
                ],
                "source_facts": [
                  "Shot 1: the entrance side blurred behind.",
                  "Shot 11: faces back into the dark doorway... doorway remains a shadowed rectangle behind her."
                ],
                "visual_inferences": [
                  "Since the apartment is unlit, the entrance area receives no practical illumination, rendering it as a dark silhouette or a black rectangle in the background."
                ],
                "creative_decisions": [
                  "Using the unlit, shadowy doorway as a consistent framing device to emphasize the emptiness and unsettling nature of the ransacked apartment."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 23,
                "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": "C01O03",
                "policy": "required"
              },
              {
                "kind": "background",
                "id": "L04B03",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "3073778d9b87a657"
      },
      {
        "scene_index": 24,
        "heading": "S#26. 포구 / 해질 무렵",
        "beat_title": "혜수이모의 부재중 전화를 확인하는 수리영의 손",
        "representative_moment": "수리영이 '혜수이모'라는 발신자가 뜬 P01(스마트폰)을 쥔 채 비에 젖은 손으로 화면을 내려다보고 있다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "A tight close-up (CU) tilting down to the hands, eye-level-to-hands angle, shallow depth of field, cold gray fog light with a bluish phone glow.",
            "t2i_prompt": "Photorealistic cinematic still. Focus on the tense, rain-damp hands of a young East Asian woman in her 20s, wearing a dark navy windbreaker sleeve, gripping P01. The lit front screen of P01 faces the camera at a slight diagonal in the lower third of the frame, showing the text ‘혜수이모’ on a glowing display. The posture of her unseen body implies she is hunched on the rocks, looking down with eyes fixed on P01, though her face is completely excluded from the tight frame. Cold gray dusk light is broken by the small bluish glow from P01, catching dim practical reflections on her wet fingers and the dark, uneven slick rocks directly beneath. A dense, pervasive milky gray-blue fog blankets the immediate background, acting as a semi-opaque veil with shallow depth of field that softens the environment into a moody, low-contrast atmosphere.",
            "outfit_assignments": [],
            "source_facts": [
              "S#26 action: '안개가 자욱하다. 텅 빈 포구. 갯바위에 앉아있는 수리영... 진동벨이 울린다. 얼핏 보니 혜수이모다. 벌써 여러통 째 부재중 전화.'",
              "Shot 3 description: '수리영의 손에 들린, 혜수이모라는 발신자 이름이 뜬 스마트폰 화면 클로즈업.'"
            ],
            "visual_inferences": [
              "fixed_element 'thick_harbor_fog' applies: dense gray-blue fog obscuring the background.",
              "Body-part focus on hands holding the phone triggers id_policy, requiring demographic descriptor substitution (young East Asian woman in her 20s) instead of C01 to prevent floating face synthesis.",
              "The hands are tense and rain-damp from the recent shower, resting over dark slick rocks."
            ],
            "creative_decisions": [
              "Shallow depth of field isolates the phone and hands, softening the wet rocks and fog in the background.",
              "Contrast is created between the cold gray fog light and the small bluish glow of the smartphone screen.",
              "No spatial reference tags (e.g., L01) are used; the background is directly described using immediate surrounding surfaces as required by close framing."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "3f845ff148c122f0",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "d45e101f45e98b8a",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "No direct visual dependency on a previous scene for this localized close-up shot.",
        "_shot_index": 3,
        "visible_entities": [
          "C01",
          "P01"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 24,
            "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 tight close-up (CU) from seated shoulder height beside 수리영 tilts down to her hands, using an eye-level-to-hands angle and shallow DOF to make the vibrating smartphone screen readable without letting it exceed the lower third of the frame. This is direct observer vision: her wet fingers and sleeve frame the display showing ‘혜수이모,’ while the slick rocks and foggy harbor remain soft scale references around it.",
            "lighting_mood": "Cold gray fog light is broken by the phone’s small bluish glow, with wet skin and stone catching dim practical 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",
              "P01"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O03"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "thick_harbor_fog",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dense, pervasive fog blankets the coastal area during dusk. It creates a thick gray-blue haze that heavily obscures distant elements and softens the ambient lighting into a moody, low-contrast atmosphere.",
                "applies_to_shots": [
                  3,
                  4
                ],
                "source_facts": [
                  "S#26 action: '안개가 자욱하다. 텅 빈 포구.'",
                  "Shot 3 description: 'foggy harbor remain soft scale references'",
                  "Shot 4 description: 'compressed wall of fog', 'Fog-diffused dusk light'"
                ],
                "visual_inferences": [
                  "The fog imparts a cold, uniform gray-blue atmospheric tint across the harbor area due to the dusk timing."
                ],
                "creative_decisions": [
                  "Use the heavy fog as a consistent backdrop to visually isolate the subjects and establish a gloomy, mysterious mood across varying 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": [],
            "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"
            ]
          },
          "render_contracts": [],
          "_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": "266f8540c3efcdde"
      },
      {
        "scene_index": 24,
        "heading": "S#26. 포구 / 해질 무렵",
        "beat_title": "유일한 가능성의 등장",
        "representative_moment": "짙은 안개벽을 뚫고 낡은 어선 한 척이 뱃머리를 드러내며 포구로 다가오는 정지 컷.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "A static medium wide telephoto shot at eye-level, fog-diffused dusk light with gray-blue low contrast and subtle rust-red accents.",
            "t2i_prompt": "Photorealistic cinematic still. A static medium wide telephoto shot from eye-level. An old, small weathered fishing boat emerges bow-first in the center of the frame, pushing straight out of a compressed wall of opaque gray-white fog. Focus on the wet hull breaking the fog barrier, its chipped rust-red and red-brown paint accents cutting faintly through the gloom. The thick gray-blue haze erases the stern and the distant horizon completely, making the boat seem to push out of a blank void. Below, dark harbor water ripples in horizontal bands, its surface reflecting the broken fog and the dark underside of the bow in wavering fragments. A wet stone pier edge runs across the lower corner of the frame, angled toward the incoming boat under the fog-diffused dusk light.",
            "outfit_assignments": [],
            "source_facts": [
              "S#26 action: '안개가 자욱하다. 텅 빈 포구.'",
              "S#26 action: '안개를 뚫고 배 한척이 다가오는 게 보인다.'",
              "Shot 4 description: 짙은 안개벽을 뚫고 뱃머리를 드러낸 낡은 소형 어선의 정면 구도."
            ],
            "visual_inferences": [
              "fixed_element thick_harbor_fog: Dense, pervasive fog creating a thick gray-blue haze during dusk.",
              "fixed_element weathered_fishing_boat: Old fishing boat with chipped rust-red paint, hull slick with moisture.",
              "The pier edge is empty and wet, anchoring the observer's POV as the boat approaches."
            ],
            "creative_decisions": [
              "Telephoto lens compression to visually emphasize the dense 'wall' of fog right behind the bow.",
              "Muted gray-blue overall palette allowing only the rust-red accent of the boat to stand out slightly, matching the grounded mystery thriller tone."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "8c5cf8c86638e2bb",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "db7a884e0b54ae8c",
              "validator": "scene_detail_owned_objects.v1",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 24,
        "dependency_reason": "같은 포구 공간이므로 앞선 샷(Shot 3)의 안개 낀 해질 무렵 분위기와 조명 톤(gray-blue haze)을 일관되게 유지.",
        "_shot_index": 4,
        "visible_entities": [
          "L14"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 24,
            "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 medium wide shot (MWS) with a telephoto eye-level angle waits at the pier edge, facing straight into a compressed wall of fog as the old fishing boat emerges bow-first in the center. The bow occupies the middle third of the frame, water ripples form dark horizontal bands below, and the fog erases the stern so the boat seems to push out of a blank void.",
            "lighting_mood": "Fog-diffused dusk light flattens contrast into gray-blue, with the boat’s rust-red accent cutting faintly through the gloom.",
            "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": [
              "L14"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "thick_harbor_fog",
                "element_type": "environment_state",
                "character_name": "",
                "description": "A dense, pervasive fog blankets the coastal area during dusk. It creates a thick gray-blue haze that heavily obscures distant elements and softens the ambient lighting into a moody, low-contrast atmosphere.",
                "applies_to_shots": [
                  3,
                  4
                ],
                "source_facts": [
                  "S#26 action: '안개가 자욱하다. 텅 빈 포구.'",
                  "Shot 3 description: 'foggy harbor remain soft scale references'",
                  "Shot 4 description: 'compressed wall of fog', 'Fog-diffused dusk light'"
                ],
                "visual_inferences": [
                  "The fog imparts a cold, uniform gray-blue atmospheric tint across the harbor area due to the dusk timing."
                ],
                "creative_decisions": [
                  "Use the heavy fog as a consistent backdrop to visually isolate the subjects and establish a gloomy, mysterious mood across varying framings."
                ],
                "confidence": "high"
              },
              {
                "element_id": "weathered_fishing_boat",
                "element_type": "environment_state",
                "character_name": "",
                "description": "An old, small fishing boat exhibits a visibly weathered exterior with chipped rust-red and red-brown paint accents. Its hull and deck surfaces are slick with moisture, reflecting the dim ambient light.",
                "applies_to_shots": [
                  4,
                  9
                ],
                "source_facts": [
                  "Shot 4 description: 'old fishing boat emerges bow-first', 'rust-red accent cutting faintly through the gloom'",
                  "Shot 9 description: 'wet boat deck', 'dim red-brown boat paint accents adding a restrained ominous warmth'"
                ],
                "visual_inferences": [
                  "The boat looks visibly aged, and its exposed surfaces are completely wet due to the recent heavy rain shower mentioned in the scene text."
                ],
                "creative_decisions": [
                  "Highlight the muted red-brown paint to provide a subtle, consistent color contrast against the otherwise monochromatic gray-blue foggy environment."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 24,
                "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": [],
            "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"
            ]
          },
          "render_contracts": [],
          "_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": "08c681a18bfff4ce"
      },
      {
        "scene_index": 24,
        "heading": "S#26. 포구 / 해질 무렵",
        "beat_title": "수리영의 승선",
        "representative_moment": "C01이 어선 갑판 위로 한 발을 내디디며 승선하는 순간.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Low angle from deck level, handheld composition, emphasizing the physical step with gray ambient light and red-brown accents.",
            "t2i_prompt": "Photorealistic cinematic still. Captured from a low angle at deck level, focus on the shoe planted on the wet deck surface. C01O03 in a dark navy windbreaker, an East Asian woman in her early 20s with black hair and dark eyes, at three-quarter angle from the right. Her shoe and shin occupy the lower center of the frame in mid-step, her foot pressing onto the deck, jacket hem trailing slightly in the direction opposite to her downward motion. Her upper body leans diagonally into the upper frame from above, eyes looking down at the stepping point. [: The immediate frame edges show wet deck planks extending toward the back, a slick dark vertical quay edge visible just behind her leg, and a wet inner gunwale with chipped rust-red and red-brown paint along the right edge.] Low gray ambient light reflects off the wet surfaces, while the dim red-brown paint accents add a restrained ominous warmth against the fog-softened obscurity behind the quay.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O03"
              }
            ],
            "source_facts": [
              "S#26 action: '배로 다가가는 수리영.'",
              "Shot 9 description: '수리영’s first foot lands on the wet boat deck.'",
              "Shot 9 description: 'Low gray ambient light reflects off the wet deck, with dim red-brown boat paint accents'"
            ],
            "visual_inferences": [
              "fixed_element weathered_fishing_boat: chipped rust-red and red-brown paint accents on wet hull and deck surfaces",
              "The background behind the quay is obscured by dense, damp fog"
            ],
            "creative_decisions": [
              "Framed from a low deck-level angle to anchor the step in the lower frame while the upper body leans in, creating a suspended mid-action feel.",
              "Used [L02] block to detail the immediate frame-edge surfaces (wet deck, quay edge, gunwale) to comply with close framing background skipped policy without using L01."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "a74a932b9515974a",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "a0e6b7a70715f5ea",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 24,
        "dependency_reason": "같은 포구 공간에서의 연속된 동작이므로 직전 샷(Shot 3)의 환경 요소(비에 젖은 갑판, 붉은 페인트)를 유지하되, 클로즈 프레이밍 규칙에 따라 배경 참조 이미지는 배제함.",
        "_shot_index": 9,
        "visible_entities": [
          "C01"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 24,
            "shot_index": 9
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A handheld medium close-up (MCU) at a low angle from deck level looks back toward the quay as 수리영’s first foot lands on the wet boat deck. Her shoe and shin occupy the lower center with realistic scale, while her upper body leans diagonally into the frame from above, turning the shore-to-boat transfer into a suspended action beat.",
            "lighting_mood": "Low gray ambient light reflects off the wet deck, with dim red-brown boat paint accents adding a restrained ominous 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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O03"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "weathered_fishing_boat",
                "element_type": "environment_state",
                "character_name": "",
                "description": "An old, small fishing boat exhibits a visibly weathered exterior with chipped rust-red and red-brown paint accents. Its hull and deck surfaces are slick with moisture, reflecting the dim ambient light.",
                "applies_to_shots": [
                  4,
                  9
                ],
                "source_facts": [
                  "Shot 4 description: 'old fishing boat emerges bow-first', 'rust-red accent cutting faintly through the gloom'",
                  "Shot 9 description: 'wet boat deck', 'dim red-brown boat paint accents adding a restrained ominous warmth'"
                ],
                "visual_inferences": [
                  "The boat looks visibly aged, and its exposed surfaces are completely wet due to the recent heavy rain shower mentioned in the scene text."
                ],
                "creative_decisions": [
                  "Highlight the muted red-brown paint to provide a subtle, consistent color contrast against the otherwise monochromatic gray-blue foggy environment."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 24,
                "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": "C01O03",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "8f9540c6cf564907"
      },
      {
        "scene_index": 25,
        "heading": "S#27. 옥탑방 안 / 해질 무렵",
        "beat_title": "불안한 의심의 시작",
        "representative_moment": "어스름한 방 안에서 전화기를 귀에 댄 채 불안한 얼굴로 김형사를 돌아보는 혜수(C02).",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Eye-level close-up with shallow depth of field, cool blue-gray color temperature",
            "t2i_prompt": "Photorealistic cinematic still. Eye-level close-up, head-and-shoulder composition. C02O05 in a charcoal gray suit jacket, a Korean woman in her early 40s with a middle-aged female face, black hair, and dark eyes, in right three-quarter profile. She holds P01 smartphone pressed to her ear, the phone's dark screen side angled away from the camera, its edge visible in the foreground. Her anxious eyes are fixed on a middle-aged Asian man in the soft, out-of-focus background. The man is at a three-quarter angle from the left, leaning near a table edge, his eyes directed at her, both sharing the same dim room. Focus on her tense expression. [: A hollow, shadowed background wall recedes behind her, catching faint blue-gray dusk haze]. Blue-gray dusk mixes with weak interior practical spill falling across her cheek and jacket shoulder.",
            "outfit_assignments": [
              {
                "character_id": "C02",
                "outlook_id": "O05"
              }
            ],
            "source_facts": [
              "S#27 action: 혜수 checking recorded message, looking at 김형사",
              "shot description: slow dolly-in close-up of 혜수, phone pressed to ear, looking toward the detective"
            ],
            "visual_inferences": [
              "김형사 (unregistered extra) as a soft background presence to emphasize her isolation",
              "dim, shadowed rooftop room ambiance",
              "blue-gray dusk lighting from the window"
            ],
            "creative_decisions": [
              "head-and-shoulder framing to keep upper body in frame and avoid the floating-face artifact",
              "shallow depth of field to keep the background figure out of focus while sharing the same spatial anchor",
              "cool blue-gray tones combined with weak practical spill to heighten the tense, anxious mood"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "0a9ab16aabc4cd42",
              "owned_hash": "47c60fddbcd37070",
              "camera_direction_hash": "40c254c2ecc75eb4",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 12,
        "dependency_reason": "동일한 옥탑방 공간의 연속선상이지만, 이 샷은 close framing 룰에 의해 배경 레퍼런스를 생략하고 현재 인물의 클로즈업과 주변부 묘사에 집중함",
        "_shot_index": 1,
        "visible_entities": [
          "C02",
          "P01"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 25,
            "shot_index": 1
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "A slow dolly-in close-up (CU) at eye-level sits just off 혜수’s phone side, catching her face in three-quarter view as her eyes turn toward 김형사. The phone edge remains barely visible along the left foreground, while 김형사 is a soft presence deeper in the dim rooftop room, making her anxious look define the screen direction.",
            "lighting_mood": "Blue-gray dusk from the window mixes with weak interior practical spill, leaving 혜수’s face damp-eyed and anxious against a hollow, shadowed room.",
            "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",
              "P01"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C02",
                "outlook_id": "O05"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L04B11",
            "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": 12,
                "shot_index": 13,
                "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": "C02O05",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "bd80e05a5e7cfa42"
      },
      {
        "scene_index": 26,
        "heading": "S#28. 바다 + 갑판 / 해질 무렵",
        "beat_title": "당황한 인우",
        "representative_moment": "인우가 오래된 사진을 보고 당황하며 입을 굳게 다문다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Tight extreme close-up (ECU) at eye-level from an observer's direct vision, compressing the subject into the center-right with foreground occlusion.",
            "t2i_prompt": "Photorealistic cinematic still. Head-and-shoulder composition from an observer's direct vision at eye-level, compressing C05O10 in a dark waterproof jumper, an East Asian man in his 20s with black hair and dark eyes, into the center-right. He is positioned at three-quarter angle from the left, leaning over the photo, his eyes fixed on the old photograph. His widened eyes and tightly sealed mouth convey sudden shock, with his neck and upper shoulder visible at the bottom edge. Slicing up the left foreground and occupying less than one-third of the frame is the blurred edge of a creased, damp old black-and-white photograph. A printed face on the silvered emulsion is angled toward the camera but remains heavily blurred and unreadable, creating a sense of crowding pressure. [: Only wet deck gloom and sunset haze are visible behind his cheek; dim amber sunset mixes with cold sea-gray fill, casting wet shadows on his face while the photograph edge catches a dull, ghostly highlight.]",
            "outfit_assignments": [
              {
                "character_id": "C05",
                "outlook_id": "O10"
              }
            ],
            "source_facts": [
              "S#28 action: '인우 (사진보더니 당황) 몰라. 모른다구.'",
              "Shot 6 description: 낡은 흑백 사진을 마주한 채 눈을 크게 뜨고 입술을 굳게 다문 인우의 당황한 얼굴 클로즈업.",
              "DP orientation: Inwoo: three_quarter_left, leaning over photo, eyes fixed on old photograph"
            ],
            "visual_inferences": [
              "fixed_element rain_slick_deck: wet deck gloom",
              "fixed_element sunset_marine_atmosphere: dim amber sunset and cold sea-gray fill",
              "The foreground photo's printed face is unreadable due to blur, focusing tension on Inwoo's expression",
              "A common noun is used for the face within the photograph to comply with the reproduction surface rule"
            ],
            "creative_decisions": [
              "Head-and-shoulder composition to prevent floating-head artifacts while maintaining a tight ECU framing",
              "Foreground occlusion by the photo edge enhances the feeling of crowding pressure and isolation",
              "Dim amber and cold gray lighting isolates the subject and highlights the wet textures"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "7cc8d94334e7c2f2",
              "owned_hash": "42cfab1203ec82d1",
              "camera_direction_hash": "a2d1544025c478c0",
              "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 scene established for this specific shot.",
        "_shot_index": 6,
        "visible_entities": [
          "C05"
        ],
        "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": "Tight at Inwoo's face from just beside the old photograph, this extreme close-up (ECU) at eye-level compresses his widened eyes and sealed mouth into the center-right of frame while the blurred photo edge slices up the left foreground at less than one-third of the image. The camera is an observer's direct vision, holding the dolly-in pressure as if the photograph itself is crowding him, with only wet deck gloom and sunset haze behind his cheek.",
            "lighting_mood": "Dim amber sunset mixes with cold sea-gray fill, leaving Inwoo's eyes wet and shadowed while the photograph edge catches a dull, ghostly highlight.",
            "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"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C05",
                "outlook_id": "O10"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L15B03",
            "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": "sunset_marine_atmosphere",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The surrounding environment is characterized by a dim, humid sunset merging with a cold sea-gray haze. Weak amber and orange-gray light softly illuminates the damp air, creating a moody and isolated marine atmosphere. The background remains enveloped in a muted grayish tone reflecting the fading light over the ocean.",
                "applies_to_shots": [
                  6,
                  8,
                  11
                ],
                "source_facts": [
                  "Scene heading: 바다 + 갑판 / 해질 무렵 (Sea + deck / around sunset)",
                  "Shot 6: Dim amber sunset mixes with cold sea-gray fill, wet deck gloom and sunset haze",
                  "Shot 8: Low, humid sunset light, shallow-focus gray",
                  "Shot 11: dim sea glow, weak orange-gray sunset backlights rain"
                ],
                "visual_inferences": [
                  "The combination of the setting sun and sea spray creates a persistently hazy, damp air quality across the entire scene."
                ],
                "creative_decisions": [
                  "Using the contrasting warm amber and cool sea-gray tones to heighten the emotional tension and sense of isolation on the small boat."
                ],
                "confidence": "high"
              },
              {
                "element_id": "rain_slick_deck",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The floor of the small boat's deck is noticeably slick with rainwater and sea spray, presenting a highly reflective, glossy surface under the faint light. The dark, weathered texture of the deck emphasizes the wet, cold, and uncomfortable conditions of the exterior environment.",
                "applies_to_shots": [
                  6,
                  11
                ],
                "source_facts": [
                  "Shot 6: wet deck gloom",
                  "Shot 11: rain-slick deck, rain on their faces"
                ],
                "visual_inferences": [
                  "The deck surface is likely made of aged, dark material that heavily reflects the ambient sunset light when coated in water."
                ],
                "creative_decisions": [
                  "Emphasizing the glossy, wet texture of the deck floor to visually anchor the characters in a harsh physical space before they move inside."
                ],
                "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": "C05O10",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "28293320b7398116"
      },
      {
        "scene_index": 26,
        "heading": "바다 + 갑판 / 해질 무렵",
        "beat_title": "사진 속 숨겨진 단서 확인",
        "representative_moment": "A macro close-up on the corner of P02 revealing the handwritten word '금월'.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Extreme close-up macro insert at a slight Dutch angle, shallow-focus gray, warm amber sunset skimming light",
            "t2i_prompt": "Photorealistic cinematic still. Extreme close-up macro insert at a slight Dutch angle, held almost parallel to the bottom corner of P02, an old black-and-white photograph. Focus on the worn, stained white margin facing the camera, revealing the tiny handwritten word '금월' next to faded photographic grain. The paper fibers, creases, and salt stains are sharply defined in the central band of the frame, while the rest of the monochrome image falls away into shallow focus. Low, humid sunset light skims across the paper texture, casting a warm amber and red-brown glow over the ink against the dim orange-gray atmospheric haze. Beneath the photograph, [: an out-of-focus wet, weathered deck surface streaked with salt merges with a cold sea-gray blur].",
            "outfit_assignments": [],
            "source_facts": [
              "Scene heading: 바다 + 갑판 / 해질 무렵 (Sea + deck / around sunset)",
              "S26 action: 수리영 shows the photograph... '여기 작게 ‘금월’이라는 글씨 보이죠'",
              "shot description: macro insert holding almost parallel to the photograph surface at a slight Dutch angle, isolating the tiny handwritten '금월' on the lower corner"
            ],
            "visual_inferences": [
              "fixed_element sunset_marine_atmosphere: dim, humid sunset merging with a cold sea-gray haze, weak amber light",
              "paper fibers and salt stains are highly textured under skimming sunset light",
              "the photograph rests over the deck, placing the sea environment in the blurred background"
            ],
            "creative_decisions": [
              "Macro focus exclusively on the handwritten text and paper texture to emphasize the clue",
              "Dutch angle to add slight unease and tension to the revelation",
              "Contrast between warm amber skimming light and cool sea-gray out-of-focus background"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "b04b2fa40054efaa",
              "owned_hash": "42cfab1203ec82d1",
              "camera_direction_hash": "c8a39e3e3625535a",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 26,
        "dependency_reason": "preserves the sunset marine atmosphere and focuses precisely on the clue from the prior interaction",
        "_shot_index": 8,
        "visible_entities": [
          "P02"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 26,
            "shot_index": 8
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "An extreme close-up (ECU) macro insert holds almost parallel to the photograph surface at a slight Dutch angle, isolating the tiny handwritten '금월' on the lower corner while the paper fibers and salt stains remain visible. The clue-bearing corner occupies the central band of frame, with the rest of the faded image falling away into shallow-focus gray.",
            "lighting_mood": "Low, humid sunset light skims across the paper texture, making the ink look almost red-brown against the monochrome photograph.",
            "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": [
              "P02"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L15B03",
            "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": "sunset_marine_atmosphere",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The surrounding environment is characterized by a dim, humid sunset merging with a cold sea-gray haze. Weak amber and orange-gray light softly illuminates the damp air, creating a moody and isolated marine atmosphere. The background remains enveloped in a muted grayish tone reflecting the fading light over the ocean.",
                "applies_to_shots": [
                  6,
                  8,
                  11
                ],
                "source_facts": [
                  "Scene heading: 바다 + 갑판 / 해질 무렵 (Sea + deck / around sunset)",
                  "Shot 6: Dim amber sunset mixes with cold sea-gray fill, wet deck gloom and sunset haze",
                  "Shot 8: Low, humid sunset light, shallow-focus gray",
                  "Shot 11: dim sea glow, weak orange-gray sunset backlights rain"
                ],
                "visual_inferences": [
                  "The combination of the setting sun and sea spray creates a persistently hazy, damp air quality across the entire scene."
                ],
                "creative_decisions": [
                  "Using the contrasting warm amber and cool sea-gray tones to heighten the emotional tension and sense of isolation on the small boat."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 26,
                "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": "prop",
                "id": "P02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "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"
            ]
          },
          "render_contracts": [
            {
              "contract_id": "rc_001",
              "scope": {
                "scene_index": 26,
                "shot_indices": [
                  8
                ],
                "duration": "single_shot"
              },
              "targets": [
                {
                  "entity_id": "P02",
                  "role": "visual_target"
                }
              ],
              "requirements": [
                {
                  "dimension": "visual_identity",
                  "operation": "preserve",
                  "strength": "required",
                  "reference_policy": "use_entity_reference"
                }
              ]
            }
          ],
          "_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": "3fc4b06858a422a6"
      },
      {
        "scene_index": 26,
        "heading": "바다 + 갑판 / 해질 무렵",
        "beat_title": "조타실 진입을 막아서는 수리영",
        "representative_moment": "C01O03 tightly holds the doorframe, physically blocking the unregistered man from entering the wheelhouse.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Medium wide shot from just inside the pilothouse doorway, eye-level framing enclosed by dark doorjambs, cool interior shadow contrasting with weak orange-gray sunset backlight.",
            "t2i_prompt": "Photorealistic cinematic still. Medium framing at eye-level from just inside the wheelhouse, upper bodies visible, framed by the dark doorjambs at the edges. C01O03 in a dark navy windbreaker, a young Korean woman in her early 20s showing a young face, black hair, and dark eyes, facing the camera, occupies the center-left of the shared doorway space. One hand is clamped to the doorway from the reference, her posture firm. Across the shared doorway, she blocks a Korean man in his 40s in a fisher work jacket, positioned at a three-quarter angle from the right, his shoulder driven forward against her. His eyes are fixed on C01O03, while her eyes are fixed on him. Cool interior shadow washes the foreground figures, catching a faint ominous red glimmer reflecting on the inner edge of the doorway from the reference. A weak amber and orange-gray sunset backlights the damp air and rain on their faces. The glossy, rain-slick deck from the reference and a cold sea-gray haze recede into the dim background. Focus on the physical tension across the shared doorway space.",
            "outfit_assignments": [
              {
                "character_id": "C01",
                "outlook_id": "O03"
              }
            ],
            "source_facts": [
              "Scene heading: 바다 + 갑판 / 해질 무렵",
              "Shot 11: 조타실 문틀을 한 손으로 꽉 쥔 채 인우의 어깨를 막아선 수리영의 정면 구도.",
              "DP direction: Cool interior shadow traps the foreground while a weak orange-gray sunset backlights rain on their faces, with a faint ominous red glimmer bleeding from inside the cabin."
            ],
            "visual_inferences": [
              "fixed_element rain_slick_deck: The floor of the deck is slick with rainwater, presenting a glossy surface.",
              "fixed_element sunset_marine_atmosphere: dim, humid sunset merging with a cold sea-gray haze.",
              "Placing the unregistered man's shoulder driven forward visually supports the physical block described in the text."
            ],
            "creative_decisions": [
              "Using the doorway from the reference as a hard vertical frame on the edges to enclose the foreground confrontation.",
              "Adding the faint ominous red glimmer to the inner edge of the doorway to emphasize the moody thriller tone without altering the background reference."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "b763389db78bc38e",
              "owned_hash": "42cfab1203ec82d1",
              "camera_direction_hash": "96866822470c2087",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "doorway",
                  "violating_phrase": "the doorway from the reference",
                  "reason": "The object is mentioned with an explicit reference anchor, not as a newly created or modified object.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "deck",
                  "violating_phrase": "The glossy, rain-slick deck from the reference",
                  "reason": "The deck is explicitly tied to the reference background, so this is an anchor reference rather than a redraw instruction.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 26,
        "dependency_reason": "Utilizes the exact background reference established in prior shots of S26 (Shot 8).",
        "_shot_index": 11,
        "visible_entities": [
          "C01",
          "L15"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 26,
            "shot_index": 11
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "From just inside the pilothouse doorway at chest height, the stopped track resolves into a medium wide shot (MWS), eye-level, framed by the dark doorjambs like a hard frontal collision. 수리영 occupies the center-left with one hand clamped to the frame, blocking 인우's shoulder on the right as the rain-slick deck and dim sea glow recede behind them.",
            "lighting_mood": "Cool interior shadow traps the foreground while a weak orange-gray sunset backlights rain on their faces, with a faint ominous red glimmer bleeding from inside the cabin.",
            "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": "O03"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": false,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L15B02",
            "reference_usage": "exact_background",
            "owned_objects": [
              "cleat",
              "crate",
              "deck",
              "doorway",
              "fish crate",
              "gate",
              "hatch",
              "photograph",
              "railing",
              "rope",
              "utility light",
              "water",
              "wheelhouse",
              "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": "sunset_marine_atmosphere",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The surrounding environment is characterized by a dim, humid sunset merging with a cold sea-gray haze. Weak amber and orange-gray light softly illuminates the damp air, creating a moody and isolated marine atmosphere. The background remains enveloped in a muted grayish tone reflecting the fading light over the ocean.",
                "applies_to_shots": [
                  6,
                  8,
                  11
                ],
                "source_facts": [
                  "Scene heading: 바다 + 갑판 / 해질 무렵 (Sea + deck / around sunset)",
                  "Shot 6: Dim amber sunset mixes with cold sea-gray fill, wet deck gloom and sunset haze",
                  "Shot 8: Low, humid sunset light, shallow-focus gray",
                  "Shot 11: dim sea glow, weak orange-gray sunset backlights rain"
                ],
                "visual_inferences": [
                  "The combination of the setting sun and sea spray creates a persistently hazy, damp air quality across the entire scene."
                ],
                "creative_decisions": [
                  "Using the contrasting warm amber and cool sea-gray tones to heighten the emotional tension and sense of isolation on the small boat."
                ],
                "confidence": "high"
              },
              {
                "element_id": "rain_slick_deck",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The floor of the small boat's deck is noticeably slick with rainwater and sea spray, presenting a highly reflective, glossy surface under the faint light. The dark, weathered texture of the deck emphasizes the wet, cold, and uncomfortable conditions of the exterior environment.",
                "applies_to_shots": [
                  6,
                  11
                ],
                "source_facts": [
                  "Shot 6: wet deck gloom",
                  "Shot 11: rain-slick deck, rain on their faces"
                ],
                "visual_inferences": [
                  "The deck surface is likely made of aged, dark material that heavily reflects the ambient sunset light when coated in water."
                ],
                "creative_decisions": [
                  "Emphasizing the glossy, wet texture of the deck floor to visually anchor the characters in a harsh physical space before they move inside."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 26,
                "shot_index": 8,
                "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": "C01O03",
                "policy": "required"
              },
              {
                "kind": "background",
                "id": "L15B02",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "fb1846ecdc475f88"
      },
      {
        "scene_index": 27,
        "heading": "S#29. 조타실 / 해질 무렵",
        "beat_title": "방향키를 둘러싼 손목의 대치",
        "representative_moment": "조타실 방향키를 돌리려는 인우의 손목을 수리영이 강하게 쥐어잡는 클로즈업.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Low angle close-up at wrist height, tight framing, dual-tone warm amber and cold blue dusk lighting",
            "t2i_prompt": "Photorealistic cinematic still. Low angle at wrist height, tight on the tense wrist of a young Korean man in his mid-20s in a worn waterproof fisherman jumper, his arm twisting the worn P17 steering wheel. His forearm is forcefully gripped by the hands of a young Korean woman in her early 20s in a dark navy windbreaker, the two sets of hands colliding in the lower center of the frame. The P17 wheel's side rim faces the camera at an oblique angle, cutting diagonally through the right side as a cramped mechanical barrier and partially blocking the control panel. Below them, a sliver of the dark, slightly worn horizontal control console surface is visible; its labeled gauge faces and switch labels face partly toward the camera but are cropped and dim. A muddy amber practical light from the console grazes their knuckles, while cold blue-gray dusk light spills onto the hands. Condensed glass is visible beyond the controls in the soft background, with blurred fog and sunset color seen through it. Torso and upper arms implied beyond the frame edges.",
            "outfit_assignments": [],
            "source_facts": [
              "Shot 2 description: Very close to the helm at wrist height, close-up (CU). 인우's tense wrist, 수리영's gripping hand, and a sliver of the control area remain visible.",
              "DP instruction: The dolly-in pressure makes the hands collide in the lower center of frame, while the wheel rim cuts diagonally through the right side.",
              "DP instruction: The labeled gauge faces and switch labels face partly toward camera but are cropped and dim."
            ],
            "visual_inferences": [
              "fixed_element dark_boat_console: A dark, slightly worn horizontal control console surface located near the helm.",
              "fixed_element wheelhouse_dusk_lighting: muddy amber practical light contrasting sharply against cold blue-gray gloom of dusk.",
              "Faces and full bodies are excluded from the frame due to the extreme close-up on the wrists, adhering to single camera view rules."
            ],
            "creative_decisions": [
              "Using common nouns with demographic descriptors instead of C01/C05 IDs to enforce body-part close-up rule without triggering face-ref compositing.",
              "Describing the frame-edge surfaces (console sliver, cropped gauges, condensed glass) explicitly to satisfy skipped_close_framing background constraints without referencing banned L16 background."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "fecdfc2bdddee260",
              "owned_hash": "7a1f375a15beb5d2",
              "camera_direction_hash": "d9887f5067e9415d",
              "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",
          "C05",
          "P17"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 27,
            "shot_index": 2
          },
          "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 helm at wrist height, this close-up (CU), low angle tightens beside the steering wheel so only 인우's tense wrist, 수리영's gripping hand, and a sliver of the control area remain visible. The dolly-in pressure makes the hands collide in the lower center of frame, while the wheel rim cuts diagonally through the right side as a cramped mechanical barrier.",
            "lighting_mood": "A muddy amber practical from the console grazes their knuckles, while cold blue-gray window light leaves the cramped cabin humid and bruised.",
            "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",
              "C05",
              "P17"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O03"
              },
              {
                "character_id": "C05",
                "outlook_id": "O10"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L16B03",
            "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": "wheelhouse_dusk_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The cramped interior of the boat cabin is defined by a striking dual-tone lighting environment. A sickly, muddy amber practical light from the control area casts a weak, yellowish glow over the nearby surfaces and subjects. This warm artificial light contrasts sharply against the cold, damp blue-gray gloom of dusk spilling in from the windows, creating a humid and oppressive atmosphere.",
                "applies_to_shots": [
                  2,
                  5,
                  8
                ],
                "source_facts": [
                  "Shot 2: A muddy amber practical from the console... cold blue-gray window light",
                  "Shot 5: lit by a weak yellow cabin practical and cold window spill",
                  "Shot 8: Sickly cabin yellow and cold dusk blue"
                ],
                "visual_inferences": [
                  "The consistent contrast between the interior yellow practical light and the exterior blue dusk gloom is the primary environmental characteristic visually linking these disparate shots."
                ],
                "creative_decisions": [
                  "Using the stark warm/cold color contrast to unify the spatial continuity and emotional tension across the different framing scales and camera angles."
                ],
                "confidence": "high"
              },
              {
                "element_id": "dark_boat_console",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "A dark, slightly worn horizontal control console surface located near the helm inside the boat cabin. It is partially illuminated by the nearby amber practical light, providing a flat, shadowy base for interactions and objects.",
                "applies_to_shots": [
                  2,
                  5
                ],
                "source_facts": [
                  "Shot 2: a sliver of the control area remain visible",
                  "Shot 5: The photograph lies across the dark console"
                ],
                "visual_inferences": [
                  "The 'control area' near the steering wheel in Shot 2 and the 'dark console' supporting the photograph in Shot 5 refer to the same physical structure anchoring the close-up actions."
                ],
                "creative_decisions": [
                  "Maintaining a consistent dark, worn texture for the console surface to ground the tight hand-level shots in the same spatial context."
                ],
                "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": [],
            "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"
            ]
          },
          "render_contracts": [],
          "_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": "55e166f83012f0fe"
      },
      {
        "scene_index": 27,
        "heading": "S#29. 조타실 / 해질 무렵",
        "beat_title": "사진 속 소녀를 가리키는 수리영의 손가락",
        "representative_moment": "어두운 콘솔 위에 놓인 P02(낡은 사진) 속 소녀의 얼굴을 C01의 손가락이 강하게 짚고 있다.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "High-angle close-up at hand level, contrasting muddy amber practical light with cold blue-gray dusk spill",
            "t2i_prompt": "Photorealistic cinematic still. Focus on an East Asian young woman's hand extending from a dark navy windbreaker sleeve, the index finger pressing firmly onto the lower-right section of P02, an old, creased black-and-white photograph lying flat across a dark, scratched horizontal console surface. The printed content of P02 shows two young girls facing the camera, with the fingertip planted directly on one girl's face. A dusty, cropped round gauge face peeks from the upper edge of the frame, its numbers barely visible. The composition is lit by a sickly, muddy amber yellow practical glow pooling on the paper and knuckles, contrasting with cold blue-gray damp dusk spill around the edges. Humid, hazy air softens the shadow outlines, isolating the forceful fingertip action against the otherwise still foreground.",
            "outfit_assignments": [],
            "source_facts": [
              "Shot description: 낡은 사진 속 두 소녀 중 한 명을 손가락으로 강하게 짚은 수리영의 손 클로즈업.",
              "S#29 action: 수리영 (사진속의 한 아이를 가르키며) 이 사진속의 얘가 우리 엄마예요.",
              "DP instruction: old black-and-white photograph (P02), pilothouse console top, cropped gauge face."
            ],
            "visual_inferences": [
              "fixed_element dark_boat_console: the interaction happens directly on the worn, horizontal console surface.",
              "fixed_element wheelhouse_dusk_lighting: the lighting is a dual-tone mix of warm amber practical light and cold dusk blue.",
              "id_policy body_part_focus_rule: framing is dominated by a hand, so C01 ID is replaced with a common noun (East Asian young woman's hand) and demographic descriptor.",
              "id_policy reproduction_surface_rule: the printed faces inside P02 use common nouns (two young girls), not character IDs."
            ],
            "creative_decisions": [
              "Used a high-angle, downward perspective strictly framing the hand, photograph, and console to adhere to the skipped_close_framing background constraints.",
              "Placed the cropped gauge face at the upper frame edge to provide minimal boat interior context without revealing the wider room.",
              "Applied hazy, humid atmospheric lighting to enhance the tactile realism of the creased photo and the windbreaker sleeve."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "5766574d2e0e138c",
              "owned_hash": "cc274d34a2a21947",
              "camera_direction_hash": "2279388094c678ad",
              "validator": "scene_detail_owned_objects.v1.close_skip",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 27,
        "dependency_reason": "previous_shot_refs(Shot 2): 배경의 어두운 콘솔 표면과 조명(황갈색 인공광과 푸른 해질녘 자연광의 대비)을 동일하게 유지하며 프레이밍만 클로즈업으로 전환.",
        "_shot_index": 5,
        "visible_entities": [
          "C01",
          "P02"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 27,
            "shot_index": 5
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "close",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Close above the old photograph at hand level, this close-up (CU), high angle holds completely still as 수리영's finger presses one of the two girls in the image near the lower-right foreground. The photograph lies across the dark console, its rectangular edge tilted slightly across the frame, while her fingertip creates the only forceful movement in an otherwise frozen evidentiary composition.",
            "lighting_mood": "The photo is lit by a weak yellow cabin practical and cold window spill, with the old monochrome image turning green-gray under the damp coastal gloom.",
            "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",
              "P02"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C01",
                "outlook_id": "O03"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L16B01",
            "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": "wheelhouse_dusk_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The cramped interior of the boat cabin is defined by a striking dual-tone lighting environment. A sickly, muddy amber practical light from the control area casts a weak, yellowish glow over the nearby surfaces and subjects. This warm artificial light contrasts sharply against the cold, damp blue-gray gloom of dusk spilling in from the windows, creating a humid and oppressive atmosphere.",
                "applies_to_shots": [
                  2,
                  5,
                  8
                ],
                "source_facts": [
                  "Shot 2: A muddy amber practical from the console... cold blue-gray window light",
                  "Shot 5: lit by a weak yellow cabin practical and cold window spill",
                  "Shot 8: Sickly cabin yellow and cold dusk blue"
                ],
                "visual_inferences": [
                  "The consistent contrast between the interior yellow practical light and the exterior blue dusk gloom is the primary environmental characteristic visually linking these disparate shots."
                ],
                "creative_decisions": [
                  "Using the stark warm/cold color contrast to unify the spatial continuity and emotional tension across the different framing scales and camera angles."
                ],
                "confidence": "high"
              },
              {
                "element_id": "dark_boat_console",
                "element_type": "persistent_prop",
                "character_name": "",
                "description": "A dark, slightly worn horizontal control console surface located near the helm inside the boat cabin. It is partially illuminated by the nearby amber practical light, providing a flat, shadowy base for interactions and objects.",
                "applies_to_shots": [
                  2,
                  5
                ],
                "source_facts": [
                  "Shot 2: a sliver of the control area remain visible",
                  "Shot 5: The photograph lies across the dark console"
                ],
                "visual_inferences": [
                  "The 'control area' near the steering wheel in Shot 2 and the 'dark console' supporting the photograph in Shot 5 refer to the same physical structure anchoring the close-up actions."
                ],
                "creative_decisions": [
                  "Maintaining a consistent dark, worn texture for the console surface to ground the tight hand-level shots in the same spatial context."
                ],
                "confidence": "medium"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 27,
                "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": "prop",
                "id": "P02",
                "policy": "required"
              }
            ],
            "forbidden_refs": [
              {
                "kind": "background",
                "reason": "close framing skips chain_bg reference"
              }
            ],
            "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"
            ]
          },
          "render_contracts": [
            {
              "contract_id": "rc_001",
              "scope": {
                "scene_index": 27,
                "shot_indices": [
                  5
                ],
                "duration": "single_shot"
              },
              "targets": [
                {
                  "entity_id": "P02",
                  "role": "visual_target"
                }
              ],
              "requirements": [
                {
                  "dimension": "visual_identity",
                  "operation": "preserve",
                  "strength": "required",
                  "reference_policy": "use_entity_reference"
                }
              ]
            }
          ],
          "_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": "bc7c20d5e7716e0e"
      },
      {
        "scene_index": 27,
        "heading": "S#29. 조타실 / 해질 무렵",
        "beat_title": "The Occult Pattern Revealed",
        "representative_moment": "The camera looks over C05's dark silhouette to reveal the dim wheelhouse wall densely covered with bizarre P08 red circular markings.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "Eye-level medium over-the-shoulder shot, strong warm-cool lighting contrast, high depth-of-field on the background",
            "t2i_prompt": "Photorealistic cinematic still. Medium over-the-shoulder composition from an observer point of view at eye-level. C05O10 in a worn waterproof jumper, East Asian young man in his mid-20s with black hair, is positioned in the left foreground with back to camera, shoulders locked rigid. His figure appears as a heavy silhouette, body volume preserved by a weak muddy amber practical glow catching faintly on his fabric edges and shoulder contour. Across the shared cramped wheelhouse space, focus on the existing rear wall panel from the reference filling the center and right background. The dark, stained surface is densely covered with the P08 red circular markings from the reference, the hand-drawn crimson shapes pulsing as the only saturated color against the humid shadows. Cold blue-gray dusk gloom spills through the fogged existing window from the reference at the frame edge, contrasting sharply with the sickly yellow cabin light washing the nearby surfaces.",
            "outfit_assignments": [
              {
                "character_id": "C05",
                "outlook_id": "O10"
              }
            ],
            "source_facts": [
              "Shot 8: 인우의 등 뒤, 어두운 조타실 벽면을 빼곡하게 채운 기괴한 붉은 원형 표식들의 정지 화면.",
              "S29 action: 인우 얼굴 너머에 보이는 조타실안 수많은 원... 붉은 원!"
            ],
            "visual_inferences": [
              "fixed_element wheelhouse_dusk_lighting: sickly muddy amber practical light contrasting with cold blue-gray dusk gloom from the window.",
              "Inwoo forms a dark human-shaped barrier in the foreground, his back fully turned, locked in rigid shock."
            ],
            "creative_decisions": [
              "Using a medium over-the-shoulder framing to anchor the viewer behind Inwoo, maximizing the dramatic reveal of the markings.",
              "Ensuring the silhouette maintains depth and body volume using rim light, avoiding a flat black cutout appearance.",
              "Explicitly anchoring owned objects like the wall panel, markings, and window to the background reference to prevent double generation."
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "fb9d59b7ca7b623d",
              "owned_hash": "cc274d34a2a21947",
              "camera_direction_hash": "72352fd2c28fe81a",
              "validator": "scene_detail_owned_objects.v1",
              "violations": [
                {
                  "owned_object": "wall panel",
                  "violating_phrase": "the existing rear wall panel from the reference",
                  "reason": "The phrase explicitly references the existing wall panel from the reference, which is an anchor reference rather than a request to redraw it.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "marking",
                  "violating_phrase": "the P08 red circular markings from the reference",
                  "reason": "The markings are described as existing from the reference and only as background detail, not as a newly created object.",
                  "verdict": "anchor_reference"
                },
                {
                  "owned_object": "window",
                  "violating_phrase": "the fogged existing window from the reference",
                  "reason": "The window is explicitly said to be existing from the reference, so this is an anchor reference, not a redraw instruction.",
                  "verdict": "anchor_reference"
                }
              ]
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": 27,
        "dependency_reason": "exact_background location reference from S27_Shot2",
        "_shot_index": 8,
        "visible_entities": [
          "C05",
          "P08",
          "L16"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 27,
            "shot_index": 8
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "After sliding behind 인우, the camera stops at shoulder height in a medium shot (MS), eye-level, using an over-the-shoulder reveal: his dark back anchors the left foreground while the rear wall fills the middle distance with dense red circular markings. The frame freezes into direct observer vision, keeping 인우 mostly silhouette and letting the occult pattern stare back from the dim wheelhouse wall.",
            "lighting_mood": "Sickly cabin yellow and cold dusk blue leave the room nearly black, while the red circles pulse as the only saturated color against the humid 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": [
              "C05",
              "P08",
              "L16"
            ],
            "allowed_outlook_pairs": [
              {
                "character_id": "C05",
                "outlook_id": "O10"
              }
            ],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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": "L16B01",
            "reference_usage": "exact_background",
            "owned_objects": [
              "GPS",
              "bench locker",
              "chair",
              "console",
              "door",
              "fuse panel",
              "helm wheel",
              "marking",
              "photo",
              "radio",
              "switch panel",
              "throttle",
              "wall panel",
              "window",
              "windshield"
            ],
            "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": "wheelhouse_dusk_lighting",
                "element_type": "environment_state",
                "character_name": "",
                "description": "The cramped interior of the boat cabin is defined by a striking dual-tone lighting environment. A sickly, muddy amber practical light from the control area casts a weak, yellowish glow over the nearby surfaces and subjects. This warm artificial light contrasts sharply against the cold, damp blue-gray gloom of dusk spilling in from the windows, creating a humid and oppressive atmosphere.",
                "applies_to_shots": [
                  2,
                  5,
                  8
                ],
                "source_facts": [
                  "Shot 2: A muddy amber practical from the console... cold blue-gray window light",
                  "Shot 5: lit by a weak yellow cabin practical and cold window spill",
                  "Shot 8: Sickly cabin yellow and cold dusk blue"
                ],
                "visual_inferences": [
                  "The consistent contrast between the interior yellow practical light and the exterior blue dusk gloom is the primary environmental characteristic visually linking these disparate shots."
                ],
                "creative_decisions": [
                  "Using the stark warm/cold color contrast to unify the spatial continuity and emotional tension across the different framing scales and camera angles."
                ],
                "confidence": "high"
              }
            ],
            "previous_shot_refs": [
              {
                "scene_index": 27,
                "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": "C05O10",
                "policy": "required"
              },
              {
                "kind": "background",
                "id": "L16B01",
                "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"
            ]
          },
          "render_contracts": [],
          "_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": "4e13f6a32b32f363"
      },
      {
        "scene_index": 28,
        "heading": "S#30. 바다 / 해질 무렵",
        "beat_title": "안개 속으로 사라지는 어선",
        "representative_moment": "짙은 안개벽 속으로 뱃머리가 반쯤 집어삼켜진 낡은 어선의 뒷모습.",
        "t2i_variations": [
          {
            "variant_label": "var_1",
            "camera_effect": "wide shot from directly behind the stern at a low angle, tracking the centerline, muted sunset flattened by heavy gray fog",
            "t2i_prompt": "Photorealistic cinematic still. Directly behind an old, weathered fishing boat at stern height, a low angle view tracks along the centerline just above the foaming gray-white wake. Focus on the battered rear rail of the dark, damp boat silhouette occupying the lower middle of the frame. The bow points away, disappearing halfway into a thick, low-visibility vertical fog wall. [L17: A heavy band of blank gray fog swallows the upper half of the image. Beneath the hull, the rain-dimpled, black-green sea surface reflects a broken smear of gold sunset and the boat's stern shadow.] The overall lighting is muted, as the last gold of sunset is flattened by the cold gray fog, leaving only faint bronze highlights on the churned wake trailing into the foreground.",
            "outfit_assignments": [],
            "source_facts": [
              "S#30 action: '세상과 경계를 지우는 것 같은 짙은 안개 띠. 서서히 안개 안으로 들어가는 인우의 배.'",
              "shot description: 짙은 안개벽 속으로 뱃머리가 반쯤 집어삼켜진 낡은 어선의 뒷모습",
              "camera_direction: Directly behind the old fishing boat at stern height... stern silhouette occupies the lower middle of frame",
              "reflection constraint: The water reflects a broken smear of gold sunset and the boat's stern shadow beneath the hull."
            ],
            "visual_inferences": [
              "the boat appears as a damp dark silhouette due to the high contrast against the fog and fading light",
              "foaming wake is visible in the immediate foreground, emphasizing the forward motion into the fog wall"
            ],
            "creative_decisions": [
              "placed the stern prominently in the lower middle to emphasize the vast, swallowing nature of the fog occupying the upper half",
              "used muted bronze and cold gray tones to reflect the chilling, threshold-crossing mood of the scene"
            ],
            "confidence": "high",
            "owned_validation": {
              "schema_version": 1,
              "t2i_prompt_hash": "b3ae392b406b64f9",
              "owned_hash": "e3b0c44298fc1c14",
              "camera_direction_hash": "c26f81ddb5927bac",
              "validator": "scene_detail_owned_objects.v1",
              "violations": []
            }
          }
        ],
        "scene_type": "normal",
        "dependent_scene_index": -1,
        "dependency_reason": "독립적인 환경 묘사 샷으로 시각적 종속성 없음",
        "_shot_index": 3,
        "visible_entities": [
          "L17"
        ],
        "render_prompt_card": {
          "schema_version": 1,
          "shot_key": {
            "scene_index": 28,
            "shot_index": 3
          },
          "render_strategy": {
            "mode": "direct",
            "primary_subject": "",
            "framing_scale": "medium",
            "moment_lock": "single still moment from the selected shot only",
            "camera_direction": "Directly behind the old fishing boat at stern height, this wide shot (WS), low angle tracks along the centerline just above the wake as the stern silhouette occupies the lower middle of frame and the bow disappears halfway into the fog wall. The composition stays tight enough to read the boat's battered rear rail and cabin outline, but leaves a heavy band of blank gray fog swallowing the upper half of the image.",
            "lighting_mood": "The last gold of sunset is flattened by cold gray fog, turning the boat into a damp silhouette with only faint bronze highlights on the wake.",
            "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": [
              "L17"
            ],
            "allowed_outlook_pairs": [],
            "must_use_composite_character_ids": true,
            "common_noun_required_when": [
              "body-part close-up",
              "a reproduced face inside any declared reproduction surface",
              "unregistered extra"
            ],
            "id_use_required_when": [
              "a physical photo/poster/document/picture-frame/map/key prop appearing in the scene — use its P## ID and preserve canonical content from entity_canon"
            ],
            "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": true,
              "id_use": "forbidden — common noun + demographic descriptor only",
              "rationale_summary": "C##O## 은 얼굴 reference 이미지를 원본 해상도로 inject 하므로 사진/화면/반사 표면 안 얼굴이 표면 밖 실물 크기로 합성됨. 이 rule 은 표면 안 얼굴 (face) 에만 적용 — 물리적 사진/액자/문서 prop 자체는 P## ID 보존 (Patch A)."
            },
            "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 any element flagged by id_policy.reproduction_surface_rule.applies (declared reproduction surface) — use a common noun + demographic descriptor instead. NOTE: the physical photo/poster/document/picture-frame/map/key prop itself MUST use its P## ID — the common-noun rule applies only to a face reproduction, not to the prop entity.",
              "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"
            ]
          },
          "render_contracts": [],
          "_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": "2b597d76cc077558"
      }
    ],
    "consumed_bg_catalog_hash": "d416ac4b47723c22",
    "consumed_shot_binding_hash": "7e96fc96a43e8f42"
  },
  "step_id": "scene_detail",
  "run_id": "2f6ce5c7-f43d-44c0-b366-e5d6c9a5bc5c",
  "resolved_model": "gemini-pro",
  "updated_at": "2026-05-13T10:39:21.445607+00:00",
  "project_config_snapshot": {}
}