{
 "groupbg::숲속 수풀": {
  "input_fingerprint": "a0985869354f7bda",
  "meta": {
   "model": "gpt-image-2",
   "size": "1536x864",
   "pack": "9.202607211540",
   "contract": "bgfirst_full_v1",
   "group_sig": {
    "key": "숲속 수풀",
    "tags": [
     "S1sh2",
     "S1sh4"
    ]
   }
  },
  "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\n\nTHE LOCATION — South Korea, 2026; all people are Korean unless stated: A narrow path through dense forest undergrowth at dusk, enclosed by thick bushes.\n\nTIME OF DAY (lock): dusk.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.",
  "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\n\nTHE LOCATION — South Korea, 2026; all people are Korean unless stated: A narrow path through dense forest undergrowth at dusk, enclosed by thick bushes.\n\nTIME OF DAY (lock): dusk.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_숲속_수풀_0e3390.png",
  "asset_id": "9b042fa6-f7eb-4dfd-9fb5-d060c53790ac",
  "input_asset_ids": [
   "3853eb29-5169-4f12-8c12-0716218f4a2a"
  ],
  "origin_tag": "S1sh2",
  "place_text": "A narrow path through dense forest undergrowth at dusk, enclosed by thick bushes."
 },
 "S1sh2::bgfirst_bg": {
  "input_fingerprint": "5c42370f9d86dbf4",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 우거진 수풀을 헤치며 앞으로 급하게 뛰쳐나가는 도중 상체가 앞으로 쏠린 자세로 굳은 한국인 남자의 잔뜩 겁에 질린 표정 정면.\n\nLOCATION (lock): A narrow path through dense forest undergrowth at dusk, enclosed by thick bushes.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At upper-torso height directly ahead of 숲속의 한국인 남자, the camera retreats at close range while holding his frightened face near frame center and his forward-pitched shoulders across the lower half. The eye-level frontal tracking keeps the surrounding foliage peripheral and unstable while his face remains comparatively locked in the composition.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 숲속의 한국인 남자 in the middle-center of the frame, midground, moves toward camera path ahead.\n- KEY BACKGROUND ELEMENTS: 우거진 수풀 (Being pushed aside along the escape path); used as Peripheral foliage parts around his shoulders and preserves the sense of forced passage; 허벅지의 이빨 자국과 피 (Large bite wound with blood flowing); used as A partial lower-frame injury cue beneath the face-centered composition.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambient light holds a restrained, desaturated image with tense contrast across his frightened face.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 우거진 수풀을 헤치며 앞으로 급하게 뛰쳐나가는 도중 상체가 앞으로 쏠린 자세로 굳은 한국인 남자의 잔뜩 겁에 질린 표정 정면.\n\nLOCATION (lock): A narrow path through dense forest undergrowth at dusk, enclosed by thick bushes.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At upper-torso height directly ahead of 숲속의 한국인 남자, the camera retreats at close range while holding his frightened face near frame center and his forward-pitched shoulders across the lower half. The eye-level frontal tracking keeps the surrounding foliage peripheral and unstable while his face remains comparatively locked in the composition.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 숲속의 한국인 남자 in the middle-center of the frame, midground, moves toward camera path ahead.\n- KEY BACKGROUND ELEMENTS: 우거진 수풀 (Being pushed aside along the escape path); used as Peripheral foliage parts around his shoulders and preserves the sense of forced passage; 허벅지의 이빨 자국과 피 (Large bite wound with blood flowing); used as A partial lower-frame injury cue beneath the face-centered composition.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambient light holds a restrained, desaturated image with tense contrast across his frightened face.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S1sh2__bgfirst_bg.png",
  "asset_id": "70ad47ef-1710-483e-a6a0-c48526285a1e",
  "input_asset_ids": [
   "3853eb29-5169-4f12-8c12-0716218f4a2a",
   "9b042fa6-f7eb-4dfd-9fb5-d060c53790ac"
  ]
 },
 "S1sh2": {
  "input_fingerprint": "08a1df4dc6acb08e",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 우거진 수풀을 헤치며 앞으로 급하게 뛰쳐나가는 도중 상체가 앞으로 쏠린 자세로 굳은 한국인 남자의 잔뜩 겁에 질린 표정 정면.\n\nLOCATION (lock): A narrow path through dense forest undergrowth at dusk, enclosed by thick bushes. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At upper-torso height directly ahead of 숲속의 한국인 남자, the camera retreats at close range while holding his frightened face near frame center and his forward-pitched shoulders across the lower half. The eye-level frontal tracking keeps the surrounding foliage peripheral and unstable while his face remains comparatively locked in the composition.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 숲속의 한국인 남자 in the middle-center of the frame, midground, moves toward camera path ahead.\n- KEY BACKGROUND ELEMENTS: 우거진 수풀 (Being pushed aside along the escape path); used as Peripheral foliage parts around his shoulders and preserves the sense of forced passage; 허벅지의 이빨 자국과 피 (Large bite wound with blood flowing); used as A partial lower-frame injury cue beneath the face-centered composition.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambient light holds a restrained, desaturated image with tense contrast across his frightened face.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The man remains missing one shoe, with a large bite wound on his thigh still bleeding.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 숲속의 한국인 남자 (한국인 남성, 성인의 얼굴, 평범한 얼굴선, 헝클어진 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 우거진 수풀을 헤치며 앞으로 급하게 뛰쳐나가는 도중 상체가 앞으로 쏠린 자세로 굳은 한국인 남자의 잔뜩 겁에 질린 표정 정면.\n\nLOCATION (lock): A narrow path through dense forest undergrowth at dusk, enclosed by thick bushes. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At upper-torso height directly ahead of 숲속의 한국인 남자, the camera retreats at close range while holding his frightened face near frame center and his forward-pitched shoulders across the lower half. The eye-level frontal tracking keeps the surrounding foliage peripheral and unstable while his face remains comparatively locked in the composition.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 숲속의 한국인 남자 in the middle-center of the frame, midground, moves toward camera path ahead.\n- KEY BACKGROUND ELEMENTS: 우거진 수풀 (Being pushed aside along the escape path); used as Peripheral foliage parts around his shoulders and preserves the sense of forced passage; 허벅지의 이빨 자국과 피 (Large bite wound with blood flowing); used as A partial lower-frame injury cue beneath the face-centered composition.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambient light holds a restrained, desaturated image with tense contrast across his frightened face.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The man remains missing one shoe, with a large bite wound on his thigh still bleeding.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 숲속의 한국인 남자 (한국인 남성, 성인의 얼굴, 평범한 얼굴선, 헝클어진 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 우거진 수풀을 헤치며 앞으로 급하게 뛰쳐나가는 도중 상체가 앞으로 쏠린 자세로 굳은 한국인 남자의 잔뜩 겁에 질린 표정 정면.\n\nLOCATION (lock): A narrow path through dense forest undergrowth at dusk, enclosed by thick bushes. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At upper-torso height directly ahead of 숲속의 한국인 남자, the camera retreats at close range while holding his frightened face near frame center and his forward-pitched shoulders across the lower half. The eye-level frontal tracking keeps the surrounding foliage peripheral and unstable while his face remains comparatively locked in the composition.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 숲속의 한국인 남자 in the middle-center of the frame, midground, moves toward camera path ahead.\n- KEY BACKGROUND ELEMENTS: 우거진 수풀 (Being pushed aside along the escape path); used as Peripheral foliage parts around his shoulders and preserves the sense of forced passage; 허벅지의 이빨 자국과 피 (Large bite wound with blood flowing); used as A partial lower-frame injury cue beneath the face-centered composition.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambient light holds a restrained, desaturated image with tense contrast across his frightened face.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The man remains missing one shoe, with a large bite wound on his thigh still bleeding.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 숲속의 한국인 남자 (한국인 남성, 성인의 얼굴, 평범한 얼굴선, 헝클어진 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S1sh2__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S1sh2.png"
    },
    {
     "label": "CHARACTER REFERENCE — 숲속의 한국인 남자: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:1149945>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_숲속_수풀_0e3390.png"
    },
    {
     "label": "CHARACTER REFERENCE — 숲속의 한국인 남자: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:1149945>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 클로즈업 샷 크기를 비교적 잘 따르며, 어깨를 화면 하단에 배치하고 허벅지 상처를 부분적으로만 노출해 프레이밍 지침을 훌륭히 수행했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "프레임을 넓혀 다리와 발을 모두 보여줌으로써 '클로즈업' 및 '어깨가 하단에 위치'해야 한다는 샷 크기와 구도 지침을 크게 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 클로즈업 샷 크기를 비교적 잘 따르며, 어깨를 화면 하단에 배치하고 허벅지 상처를 부분적으로만 노출해 프레이밍 지침을 훌륭히 수행했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "프레임을 넓혀 다리와 발을 모두 보여줌으로써 '클로즈업' 및 '어깨가 하단에 위치'해야 한다는 샷 크기와 구도 지침을 크게 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 클로즈업 프레이밍과 상체가 앞으로 쏠린 구도를 잘 구현했으며, 화면 하단에 허벅지 상처를 적절히 배치하여 지시사항을 충실히 따랐습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "요구된 클로즈업 대신 피사체의 전신이 가깝게 드러나는 너무 넓은 샷으로 촬영되어 프레이밍 및 구도 지시사항을 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요구된 클로즈업 프레이밍과 상체가 앞으로 쏠린 구도를 잘 구현했으며, 화면 하단에 허벅지 상처를 적절히 배치하여 지시사항을 충실히 따랐습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "요구된 클로즈업 대신 피사체의 전신이 가깝게 드러나는 너무 넓은 샷으로 촬영되어 프레이밍 및 구도 지시사항을 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 14,
     "B": 8
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 14,
   "B": 8
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "지정된 클로즈업 샷 크기를 비교적 잘 따르며, 어깨를 화면 하단에 배치하고 허벅지 상처를 부분적으로만 노출해 프레이밍 지침을 훌륭히 수행했습니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "프레임을 넓혀 다리와 발을 모두 보여줌으로써 '클로즈업' 및 '어깨가 하단에 위치'해야 한다는 샷 크기와 구도 지침을 크게 위반했습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_숲속_수풀_0e3390.png"
   },
   {
    "label": "CHARACTER REFERENCE — 숲속의 한국인 남자: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:1149945>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "프레이밍이 텍스트에 지정된 클로즈업(어깨가 화면 하단을 채움)이 아니라 복부와 허벅지 전체가 보이는 미디엄 샷으로 넓게 렌더링되었습니다.",
     "fix_en": "Enlarge the man significantly so his face is near the frame center and his forward-pitched shoulders span the entire lower half of the frame."
    },
    {
     "issue_ko": "배경을 원본 그대로 유지하라는 지시와 달리, 바닥의 흙길이 풀로 덮이고 하늘의 구름과 나무 실루엣이 새롭게 변형되었습니다.",
     "fix_en": "Restore the exact original reference background behind the man, leaving its path texture, foliage, and sky completely unaltered."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Enlarge the man significantly so his face is near the frame center and his forward-pitched shoulders span the entire lower half of the frame.\n- Restore the exact original reference background behind the man, leaving its path texture, foliage, and sky completely unaltered.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 샷 스케일과 레이아웃 스케치에 따라 하단의 허벅지 부상 부위를 정확하게 프레임에 포함시켰습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "프레이밍이 과도하게 크롭되어 프롬프트와 스케치에 명시된 필수 요소인 허벅지 부상이 화면에서 배제되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 샷 스케일과 레이아웃 스케치에 따라 하단의 허벅지 부상 부위를 정확하게 프레임에 포함시켰습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "프레이밍이 과도하게 크롭되어 프롬프트와 스케치에 명시된 필수 요소인 허벅지 부상이 화면에서 배제되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "얼굴 표정과 조명은 좋으나, 프레임이 과도하게 크롭되어 지시사항에 명시된 하단의 허벅지 상처가 누락되었습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "레이아웃 스케치의 구도를 정확히 따랐으며, 프레임 하단의 허벅지 부상 요소까지 지시대로 충실히 포함했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "얼굴 표정과 조명은 좋으나, 프레임이 과도하게 크롭되어 지시사항에 명시된 하단의 허벅지 상처가 누락되었습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "레이아웃 스케치의 구도를 정확히 따랐으며, 프레임 하단의 허벅지 부상 요소까지 지시대로 충실히 포함했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 15,
     "B": 10
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S1sh2__bgfirst_bg.png",
   "bg_asset_id": "70ad47ef-1710-483e-a6a0-c48526285a1e",
   "bg_record_key": "S1sh2::bgfirst_bg",
   "chain_winner": true,
   "authority": "groupbg",
   "group_key": "숲속 수풀",
   "groupbg_asset_id": "9b042fa6-f7eb-4dfd-9fb5-d060c53790ac"
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S1sh4::variants": {
  "author_fp": "9844233562eebd11",
  "author": {
   "variants": [
    {
     "approach_ko": "넓은 음영 공간과 불규칙한 수풀 테두리로 중심의 제압 자세를 고립시키는 냉정한 수직 조감 구도.",
     "prompt_en": "Use an austere, nearly orthographic overhead view with restrained perspective and generous forest-floor space surrounding the centered bodies. Compose the undergrowth as an irregular dark perimeter that closes around them, while subtle tonal separation keeps the fallen man's splayed limbs and the attacker's fully human posture legible. Let the subdued dusk illumination fall softly and evenly, with desaturated earth and foliage tones, crisp contact shadows, and no motion blur, creating the stillness of a grim forensic tableau."
    },
    {
     "approach_ko": "수풀의 층과 비스듬한 황혼빛을 활용해 중심의 몸싸움을 깊고 압박감 있게 내려다보는 조감 구도.",
     "prompt_en": "Give the fixed high overhead view a more immersive wide-lens feel, using overlapping layers of surrounding undergrowth to create depth and visual pressure around the middle-center struggle. Allow nearby foliage to form soft, uneven edge occlusions without covering the victim's readable splayed body, while low directional dusk light grazes leaves, clothing folds, exposed skin, the missing-shoe foot, and the wounded thigh. Keep the attacker's obscured face and dark clothing materially detailed rather than featureless, with the pinned contact points emphasized by compact, dense shadows."
    }
   ]
  },
  "reused": false
 },
 "S1sh4": {
  "input_fingerprint": "ae1afca164134803",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 검은 인영이 바닥에 쓰러진 한국인 남자를 위에서 짓누른 채 멈춘 정지 구도.\n\nLOCATION (lock): The forest floor amid dense dusk undergrowth, where the fallen man is pinned down. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From a fixed high overhead position, the camera looks nearly straight down on 검은 인영 pinning 숲속의 한국인 남자 at the center, with the victim's splayed body remaining readable beneath the attacker. The wide top-down geometry turns the surrounding foliage and ground into an enclosing boundary rather than shifting the established attack position.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 검은 인영 in the middle-center of the frame, foreground; 숲속의 한국인 남자 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 숲 바닥 (Visible beneath the fallen man); used as Defines the overhead silhouette of the pinned bodies and provides spatial scale around them; 우거진 수풀 (Surrounding the figures); used as Forms an irregular perimeter around the central attack.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk light and desaturated grading preserve the attack as a dark, ambiguous overhead shape.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the dense dusk forest, tangled undergrowth, muted natural light, and ground texture from the reference. EXCLUDE the running pose and frightened frontal expression; show only the fallen man pinned beneath the dark figure.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The fallen man remains missing one shoe and retains the large, bleeding bite wound on his thigh.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 숲속의 한국인 남자 (한국인 남성, 성인의 얼굴, 평범한 얼굴선, 헝클어진 검은 머리); 검은 인영 (한국인 남성형 인영, 성인형 머리 윤곽, 가려진 얼굴, 검게 뭉친 실루엣) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 검은 인영이 바닥에 쓰러진 한국인 남자를 위에서 짓누른 채 멈춘 정지 구도.\n\nLOCATION (lock): The forest floor amid dense dusk undergrowth, where the fallen man is pinned down. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From a fixed high overhead position, the camera looks nearly straight down on 검은 인영 pinning 숲속의 한국인 남자 at the center, with the victim's splayed body remaining readable beneath the attacker. The wide top-down geometry turns the surrounding foliage and ground into an enclosing boundary rather than shifting the established attack position.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 검은 인영 in the middle-center of the frame, foreground; 숲속의 한국인 남자 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 숲 바닥 (Visible beneath the fallen man); used as Defines the overhead silhouette of the pinned bodies and provides spatial scale around them; 우거진 수풀 (Surrounding the figures); used as Forms an irregular perimeter around the central attack.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk light and desaturated grading preserve the attack as a dark, ambiguous overhead shape.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the dense dusk forest, tangled undergrowth, muted natural light, and ground texture from the reference. EXCLUDE the running pose and frightened frontal expression; show only the fallen man pinned beneath the dark figure.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The fallen man remains missing one shoe and retains the large, bleeding bite wound on his thigh.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 숲속의 한국인 남자 (한국인 남성, 성인의 얼굴, 평범한 얼굴선, 헝클어진 검은 머리); 검은 인영 (한국인 남성형 인영, 성인형 머리 윤곽, 가려진 얼굴, 검게 뭉친 실루엣) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an austere, nearly orthographic overhead view with restrained perspective and generous forest-floor space surrounding the centered bodies. Compose the undergrowth as an irregular dark perimeter that closes around them, while subtle tonal separation keeps the fallen man's splayed limbs and the attacker's fully human posture legible. Let the subdued dusk illumination fall softly and evenly, with desaturated earth and foliage tones, crisp contact shadows, and no motion blur, creating the stillness of a grim forensic tableau.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 검은 인영이 바닥에 쓰러진 한국인 남자를 위에서 짓누른 채 멈춘 정지 구도.\n\nLOCATION (lock): The forest floor amid dense dusk undergrowth, where the fallen man is pinned down. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From a fixed high overhead position, the camera looks nearly straight down on 검은 인영 pinning 숲속의 한국인 남자 at the center, with the victim's splayed body remaining readable beneath the attacker. The wide top-down geometry turns the surrounding foliage and ground into an enclosing boundary rather than shifting the established attack position.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 검은 인영 in the middle-center of the frame, foreground; 숲속의 한국인 남자 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 숲 바닥 (Visible beneath the fallen man); used as Defines the overhead silhouette of the pinned bodies and provides spatial scale around them; 우거진 수풀 (Surrounding the figures); used as Forms an irregular perimeter around the central attack.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk light and desaturated grading preserve the attack as a dark, ambiguous overhead shape.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the dense dusk forest, tangled undergrowth, muted natural light, and ground texture from the reference. EXCLUDE the running pose and frightened frontal expression; show only the fallen man pinned beneath the dark figure.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The fallen man remains missing one shoe and retains the large, bleeding bite wound on his thigh.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 숲속의 한국인 남자 (한국인 남성, 성인의 얼굴, 평범한 얼굴선, 헝클어진 검은 머리); 검은 인영 (한국인 남성형 인영, 성인형 머리 윤곽, 가려진 얼굴, 검게 뭉친 실루엣) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nGive the fixed high overhead view a more immersive wide-lens feel, using overlapping layers of surrounding undergrowth to create depth and visual pressure around the middle-center struggle. Allow nearby foliage to form soft, uneven edge occlusions without covering the victim's readable splayed body, while low directional dusk light grazes leaves, clothing folds, exposed skin, the missing-shoe foot, and the wounded thigh. Keep the attacker's obscured face and dark clothing materially detailed rather than featureless, with the pinned contact points emphasized by compact, dense shadows.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "B": 7,
   "A": 5
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "수직 하이앵글의 와이드 샷 구도를 정확히 구현했으며, 레퍼런스에 설정된 좌측 허벅지 부상 위치와 맨발 상태를 일관되게 유지했습니다."
   },
   {
    "label": "A",
    "score": 5,
    "verdict_ko": "요구된 하이앵글 정지 구도와 조명 분위기는 준수하게 표현했으나, 이전 샷에서 확립된 부상 위치가 반대쪽 다리로 뒤바뀌는 연결 오류가 있습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S1sh2_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 숲속의 한국인 남자: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:1149945>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "이전 컷(Previous Shot Still)에서 오른쪽 허벅지에 있던 큰 출혈 상처가 이 사진에서는 왼쪽 허벅지로 위치가 바뀌었습니다.",
     "fix_en": "Move the large bleeding wound from the man's left thigh to his right thigh to match the established reference."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Move the large bleeding wound from the man's left thigh to his right thigh to match the established reference.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "하이앵글 구도와 인물 배치뿐만 아니라 요구된 허벅지의 출혈 상처와 맨발 상태를 정확히 유지했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "전반적인 구도는 일치하나, 요구된 허벅지의 출혈 상처가 누락되어(갈색 얼룩으로 표현됨) 디테일 유지에 실패했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "하이앵글 구도와 인물 배치뿐만 아니라 요구된 허벅지의 출혈 상처와 맨발 상태를 정확히 유지했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "전반적인 구도는 일치하나, 요구된 허벅지의 출혈 상처가 누락되어(갈색 얼룩으로 표현됨) 디테일 유지에 실패했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지정된 하이앵글 와이드 샷 구도를 완벽하게 구현했으며, 이전 컷에서 이어진 맨발과 허벅지의 출혈 상처를 정확히 묘사했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "구도와 맨발 상태는 유지했으나, 지시된 허벅지의 출혈 상처가 피가 아닌 갈색 덩어리처럼 잘못 표현되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 하이앵글 와이드 샷 구도를 완벽하게 구현했으며, 이전 컷에서 이어진 맨발과 허벅지의 출혈 상처를 정확히 묘사했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "구도와 맨발 상태는 유지했으나, 지시된 허벅지의 출혈 상처가 피가 아닌 갈색 덩어리처럼 잘못 표현되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 15,
     "B": 9
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S1sh2"
  }
 },
 "groupbg::도시 뒷골목": {
  "input_fingerprint": "e93d2a4d55624d1c",
  "meta": {
   "model": "gpt-image-2",
   "size": "1536x864",
   "pack": "9.202607211540",
   "contract": "bgfirst_full_v1",
   "group_sig": {
    "key": "도시 뒷골목",
    "tags": [
     "S2sh1",
     "S2sh3"
    ]
   }
  },
  "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\n\nTHE LOCATION — South Korea, 2026; all people are Korean unless stated: A dark urban back alley beside a utility pole, dimly lit by a streetlamp.\n\nTIME OF DAY (lock): night.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.",
  "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\n\nTHE LOCATION — South Korea, 2026; all people are Korean unless stated: A dark urban back alley beside a utility pole, dimly lit by a streetlamp.\n\nTIME OF DAY (lock): night.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_도시_뒷골목_0238ec.png",
  "asset_id": "705dca19-894c-4b77-9bde-9cf9484ca2e6",
  "input_asset_ids": [
   "7ee16fc0-4fb9-41f1-a41f-7250ffab0708"
  ],
  "origin_tag": "S2sh1",
  "place_text": "A dark urban back alley beside a utility pole, dimly lit by a streetlamp."
 },
 "S2sh1::bgfirst_bg": {
  "input_fingerprint": "cea9795b5fc346ed",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 어두운 가로등 불빛 아래 전봇대에 몸을 밀착한 채 서 있는 한국인 남녀.\n\nLOCATION (lock): A dark urban back alley beside a utility pole, dimly lit by a streetlamp.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At standing eye level near the alley entrance, a static wide frame places 도시 뒷골목의 한국인 남자 and 도시 뒷골목의 한국인 여자 together beside the utility pole on a slight three-quarter axis. Their intertwined bodies occupy the middle distance while open alley space in the foreground creates an uneasy observational gap for the eventual approach.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 도시 뒷골목의 한국인 남자 in the middle-left of the frame, midground; 도시 뒷골목의 한국인 여자 in the middle-right of the frame, midground; 전봇대 in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 전봇대 (Both figures are pressed against it) — The alley-facing side is visible behind the pair; used as Shared vertical anchor against which both figures are compressed; 도시 뒷골목 (Dimly visible at night); used as Provides the broad foreground approach corridor from the camera to the pair; 가로등 (On, casting faint light) — Its street-facing side is angled toward the pair; used as Established illumination above the figures without revealing their action clearly.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Faint streetlamp light keeps the night alley desaturated and underexposed, withholding the nature of the embrace.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 어두운 가로등 불빛 아래 전봇대에 몸을 밀착한 채 서 있는 한국인 남녀.\n\nLOCATION (lock): A dark urban back alley beside a utility pole, dimly lit by a streetlamp.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At standing eye level near the alley entrance, a static wide frame places 도시 뒷골목의 한국인 남자 and 도시 뒷골목의 한국인 여자 together beside the utility pole on a slight three-quarter axis. Their intertwined bodies occupy the middle distance while open alley space in the foreground creates an uneasy observational gap for the eventual approach.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 도시 뒷골목의 한국인 남자 in the middle-left of the frame, midground; 도시 뒷골목의 한국인 여자 in the middle-right of the frame, midground; 전봇대 in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 전봇대 (Both figures are pressed against it) — The alley-facing side is visible behind the pair; used as Shared vertical anchor against which both figures are compressed; 도시 뒷골목 (Dimly visible at night); used as Provides the broad foreground approach corridor from the camera to the pair; 가로등 (On, casting faint light) — Its street-facing side is angled toward the pair; used as Established illumination above the figures without revealing their action clearly.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Faint streetlamp light keeps the night alley desaturated and underexposed, withholding the nature of the embrace.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S2sh1__bgfirst_bg.png",
  "asset_id": "ab96b13c-d203-43c0-9def-c06b4a627629",
  "input_asset_ids": [
   "7ee16fc0-4fb9-41f1-a41f-7250ffab0708",
   "705dca19-894c-4b77-9bde-9cf9484ca2e6"
  ]
 },
 "S2sh1": {
  "input_fingerprint": "e2431d15e9b4b831",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 가로등 불빛 아래 전봇대에 몸을 밀착한 채 서 있는 한국인 남녀.\n\nLOCATION (lock): A dark urban back alley beside a utility pole, dimly lit by a streetlamp. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At standing eye level near the alley entrance, a static wide frame places 도시 뒷골목의 한국인 남자 and 도시 뒷골목의 한국인 여자 together beside the utility pole on a slight three-quarter axis. Their intertwined bodies occupy the middle distance while open alley space in the foreground creates an uneasy observational gap for the eventual approach.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 도시 뒷골목의 한국인 남자 in the middle-left of the frame, midground; 도시 뒷골목의 한국인 여자 in the middle-right of the frame, midground; 전봇대 in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 전봇대 (Both figures are pressed against it) — The alley-facing side is visible behind the pair; used as Shared vertical anchor against which both figures are compressed; 도시 뒷골목 (Dimly visible at night); used as Provides the broad foreground approach corridor from the camera to the pair; 가로등 (On, casting faint light) — Its street-facing side is angled toward the pair; used as Established illumination above the figures without revealing their action clearly.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Faint streetlamp light keeps the night alley desaturated and underexposed, withholding the nature of the embrace.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 도시 뒷골목의 한국인 남자 (포식 본능 활성화 상태) (한국인 남성, 성인의 얼굴, 평범한 얼굴선, 짧은 검은 머리); 도시 뒷골목의 한국인 여자 (한국인 여성, 성인의 얼굴, 평범한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 가로등 불빛 아래 전봇대에 몸을 밀착한 채 서 있는 한국인 남녀.\n\nLOCATION (lock): A dark urban back alley beside a utility pole, dimly lit by a streetlamp. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At standing eye level near the alley entrance, a static wide frame places 도시 뒷골목의 한국인 남자 and 도시 뒷골목의 한국인 여자 together beside the utility pole on a slight three-quarter axis. Their intertwined bodies occupy the middle distance while open alley space in the foreground creates an uneasy observational gap for the eventual approach.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 도시 뒷골목의 한국인 남자 in the middle-left of the frame, midground; 도시 뒷골목의 한국인 여자 in the middle-right of the frame, midground; 전봇대 in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 전봇대 (Both figures are pressed against it) — The alley-facing side is visible behind the pair; used as Shared vertical anchor against which both figures are compressed; 도시 뒷골목 (Dimly visible at night); used as Provides the broad foreground approach corridor from the camera to the pair; 가로등 (On, casting faint light) — Its street-facing side is angled toward the pair; used as Established illumination above the figures without revealing their action clearly.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Faint streetlamp light keeps the night alley desaturated and underexposed, withholding the nature of the embrace.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 도시 뒷골목의 한국인 남자 (포식 본능 활성화 상태) (한국인 남성, 성인의 얼굴, 평범한 얼굴선, 짧은 검은 머리); 도시 뒷골목의 한국인 여자 (한국인 여성, 성인의 얼굴, 평범한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 가로등 불빛 아래 전봇대에 몸을 밀착한 채 서 있는 한국인 남녀.\n\nLOCATION (lock): A dark urban back alley beside a utility pole, dimly lit by a streetlamp. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At standing eye level near the alley entrance, a static wide frame places 도시 뒷골목의 한국인 남자 and 도시 뒷골목의 한국인 여자 together beside the utility pole on a slight three-quarter axis. Their intertwined bodies occupy the middle distance while open alley space in the foreground creates an uneasy observational gap for the eventual approach.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 도시 뒷골목의 한국인 남자 in the middle-left of the frame, midground; 도시 뒷골목의 한국인 여자 in the middle-right of the frame, midground; 전봇대 in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 전봇대 (Both figures are pressed against it) — The alley-facing side is visible behind the pair; used as Shared vertical anchor against which both figures are compressed; 도시 뒷골목 (Dimly visible at night); used as Provides the broad foreground approach corridor from the camera to the pair; 가로등 (On, casting faint light) — Its street-facing side is angled toward the pair; used as Established illumination above the figures without revealing their action clearly.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Faint streetlamp light keeps the night alley desaturated and underexposed, withholding the nature of the embrace.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 도시 뒷골목의 한국인 남자 (포식 본능 활성화 상태) (한국인 남성, 성인의 얼굴, 평범한 얼굴선, 짧은 검은 머리); 도시 뒷골목의 한국인 여자 (한국인 여성, 성인의 얼굴, 평범한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S2sh1__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S2sh1.png"
    },
    {
     "label": "CHARACTER REFERENCE — 도시 뒷골목의 한국인 남자 (포식 본능 활성화 상태): the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:915386>"
    },
    {
     "label": "CHARACTER REFERENCE — 도시 뒷골목의 한국인 여자: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:722947>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_도시_뒷골목_0238ec.png"
    },
    {
     "label": "CHARACTER REFERENCE — 도시 뒷골목의 한국인 남자 (포식 본능 활성화 상태): the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:915386>"
    },
    {
     "label": "CHARACTER REFERENCE — 도시 뒷골목의 한국인 여자: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:722947>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시문의 의도대로 두 인물이 껴안은 채 전봇대에 밀착한 모습을 잘 구현했으며 배경과 의상도 정확하게 일치합니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "캐릭터와 배경은 일치하나, 서로 떨어져 서 있어 '얽힌 몸(intertwined bodies)'과 포옹하는 행동 지시를 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시문의 의도대로 두 인물이 껴안은 채 전봇대에 밀착한 모습을 잘 구현했으며 배경과 의상도 정확하게 일치합니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "캐릭터와 배경은 일치하나, 서로 떨어져 서 있어 '얽힌 몸(intertwined bodies)'과 포옹하는 행동 지시를 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "두 인물이 전봇대에 함께 밀착하여 얽혀 있는 모습을 와이드 샷 프레임으로 정확하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "두 인물이 얽혀 있지 않고 전봇대를 사이에 두고 분리되어 있어 '몸을 밀착한 채'라는 핵심 연출 지시를 어겼습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "두 인물이 전봇대에 함께 밀착하여 얽혀 있는 모습을 와이드 샷 프레임으로 정확하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "두 인물이 얽혀 있지 않고 전봇대를 사이에 두고 분리되어 있어 '몸을 밀착한 채'라는 핵심 연출 지시를 어겼습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 8,
     "B": 14
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 8,
   "B": 14
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "지시문의 의도대로 두 인물이 껴안은 채 전봇대에 밀착한 모습을 잘 구현했으며 배경과 의상도 정확하게 일치합니다."
   },
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "캐릭터와 배경은 일치하나, 서로 떨어져 서 있어 '얽힌 몸(intertwined bodies)'과 포옹하는 행동 지시를 위반했습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_도시_뒷골목_0238ec.png"
   },
   {
    "label": "CHARACTER REFERENCE — 도시 뒷골목의 한국인 남자 (포식 본능 활성화 상태): the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:915386>"
   },
   {
    "label": "CHARACTER REFERENCE — 도시 뒷골목의 한국인 여자: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:722947>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "피사체와 전봇대가 모두 화면 우측에 배치되어, 위치 참조 사진의 구도를 피하고 지정된 레이아웃을 따르라는 지시를 위반했습니다.",
     "fix_en": "Center the composition to place the utility pole in the upper-center background, moving the man to the middle-left midground and the woman to the middle-right midground."
    },
    {
     "issue_ko": "여자만 전봇대의 측면에 기대어 있으며, 두 사람이 골목을 향하는 전봇대 앞면에 함께 밀착되어 있다는 지시와 다릅니다.",
     "fix_en": "Position both the man and the woman so they are jointly pressed against the front, alley-facing side of the utility pole."
    },
    {
     "issue_ko": "남자의 등(화면 좌측)을 붙잡고 있는 트렌치코트 소매의 손이 보이나, 여자의 위치와 자세상 불가능한 세 번째 팔에 해당합니다.",
     "fix_en": "Remove the anatomically incorrect hand and trench coat sleeve wrapping around the left side of the man's back."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Center the composition to place the utility pole in the upper-center background, moving the man to the middle-left midground and the woman to the middle-right midground.\n- Position both the man and the woman so they are jointly pressed against the front, alley-facing side of the utility pole.\n- Remove the anatomically incorrect hand and trench coat sleeve wrapping around the left side of the man's back.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 샷 크기와 프레이밍을 잘 준수하였으며, 레퍼런스의 의상과 배경을 충실히 반영하여 신체적 오류 없이 장면을 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "전봇대 우측에 누구의 것인지 알 수 없는 여분의 손이 생성되는 치명적인 신체 구조적 오류(하드 위반)가 발생했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 샷 크기와 프레이밍을 잘 준수하였으며, 레퍼런스의 의상과 배경을 충실히 반영하여 신체적 오류 없이 장면을 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "전봇대 우측에 누구의 것인지 알 수 없는 여분의 손이 생성되는 치명적인 신체 구조적 오류(하드 위반)가 발생했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 프레이밍과 배경 위치를 잘 따랐으며, 인물의 얼굴과 자세가 상대적으로 더 자연스럽게 구현되었습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "구도와 배경은 일치하나 여자의 얼굴과 목이 다소 왜곡되어 해부학적 자연스러움이 떨어집니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지정된 프레이밍과 배경 위치를 잘 따랐으며, 인물의 얼굴과 자세가 상대적으로 더 자연스럽게 구현되었습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "구도와 배경은 일치하나 여자의 얼굴과 목이 다소 왜곡되어 해부학적 자연스러움이 떨어집니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 13,
     "B": 11
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": false,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S2sh1__bgfirst_bg.png",
   "bg_asset_id": "ab96b13c-d203-43c0-9def-c06b4a627629",
   "bg_record_key": "S2sh1::bgfirst_bg",
   "chain_winner": false,
   "authority": "groupbg",
   "group_key": "도시 뒷골목",
   "groupbg_asset_id": "705dca19-894c-4b77-9bde-9cf9484ca2e6"
  },
  "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S2sh3::variants": {
  "author_fp": "992b8285b0c1ddfc",
  "author": {
   "variants": [
    {
     "approach_ko": "압축된 정측면 구도와 절제된 측광으로 물린 지점의 정지된 긴장감을 냉정하게 강조한다.",
     "prompt_en": "Use a compressed, nearly orthographic lens feel from the locked clean side axis, holding both profile fragments in a severe, flattened arrangement around the bite. Keep the teeth and neck contact critically sharp while both faces and the narrow utility-pole strip recede quickly into softness. Let the faint streetlamp skim laterally across skin, hair, and fabric with restrained shadow detail, leaving the red eye as the sole concentrated color accent."
    },
    {
     "approach_ko": "근접 초점의 강한 원근감과 얕은 초점층으로 피부와 치아의 물리적 접촉을 본능적이고 촉각적으로 포착한다.",
     "prompt_en": "Use an intimate close-focus lens feel while preserving the exact upper-body-height side axis and prescribed left-right placement, allowing subtle perspective depth between the two profile fragments and the point of contact. Concentrate the shallow focal plane on the embedded teeth and compressed neck surface, with facial contours falling gently out of focus toward the frame edges and the utility pole surviving only as a narrow soft brace behind her. Shape the subdued streetlamp illumination as a faint grazing highlight over the contact textures, surrounded by dense natural alley shadow and anchored by the man's deep-red eye."
    }
   ]
  },
  "reused": true
 },
 "S2sh3": {
  "input_fingerprint": "69a2a9b6c43577d7",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 한국인 남자의 이빨이 한국인 여자의 목덜미에 깊게 박힌 채 멈춘 측면.\n\nLOCATION (lock): The shadowed space against a utility pole in a dim urban back alley. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At upper-body height, the tracking move reaches a clean side axis and tightens until 도시 뒷골목의 한국인 남자 (포식 본능 활성화 상태)'s teeth, red eye, and 도시 뒷골목의 한국인 여자's neck occupy most of the close frame without losing fragments of both profiles. Shallow focus isolates the bite while a narrow strip of the utility pole remains behind the woman as a scale and support reference.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 도시 뒷골목의 한국인 남자 (포식 본능 활성화 상태) in the middle-left of the frame, foreground; 도시 뒷골목의 한국인 여자 in the middle-right of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 전봇대 (Supporting the woman's weight) — Only its side adjacent to the woman's shoulder is visible; used as A narrow background strip braces the woman's neck and maintains continuity with the wider alley view; 도시 뒷골목 (Dark at night); used as Falls beyond the shallow focus as the camera reaches the side profile.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The faint streetlamp illumination remains subdued, with the man's explicitly red eye forming the only concentrated deep-red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the dim streetlamp lighting, narrow urban alley surfaces, and the utility pole immediately behind the pair. EXCLUDE the earlier upright embrace and replace it with the side view of the man biting the woman's neck.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 도시 뒷골목의 한국인 남자 (포식 본능 활성화 상태) (한국인 남성, 성인의 얼굴, 평범한 얼굴선, 짧은 검은 머리); 도시 뒷골목의 한국인 여자 (한국인 여성, 성인의 얼굴, 평범한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 한국인 남자의 이빨이 한국인 여자의 목덜미에 깊게 박힌 채 멈춘 측면.\n\nLOCATION (lock): The shadowed space against a utility pole in a dim urban back alley. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At upper-body height, the tracking move reaches a clean side axis and tightens until 도시 뒷골목의 한국인 남자 (포식 본능 활성화 상태)'s teeth, red eye, and 도시 뒷골목의 한국인 여자's neck occupy most of the close frame without losing fragments of both profiles. Shallow focus isolates the bite while a narrow strip of the utility pole remains behind the woman as a scale and support reference.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 도시 뒷골목의 한국인 남자 (포식 본능 활성화 상태) in the middle-left of the frame, foreground; 도시 뒷골목의 한국인 여자 in the middle-right of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 전봇대 (Supporting the woman's weight) — Only its side adjacent to the woman's shoulder is visible; used as A narrow background strip braces the woman's neck and maintains continuity with the wider alley view; 도시 뒷골목 (Dark at night); used as Falls beyond the shallow focus as the camera reaches the side profile.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The faint streetlamp illumination remains subdued, with the man's explicitly red eye forming the only concentrated deep-red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the dim streetlamp lighting, narrow urban alley surfaces, and the utility pole immediately behind the pair. EXCLUDE the earlier upright embrace and replace it with the side view of the man biting the woman's neck.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 도시 뒷골목의 한국인 남자 (포식 본능 활성화 상태) (한국인 남성, 성인의 얼굴, 평범한 얼굴선, 짧은 검은 머리); 도시 뒷골목의 한국인 여자 (한국인 여성, 성인의 얼굴, 평범한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a compressed, nearly orthographic lens feel from the locked clean side axis, holding both profile fragments in a severe, flattened arrangement around the bite. Keep the teeth and neck contact critically sharp while both faces and the narrow utility-pole strip recede quickly into softness. Let the faint streetlamp skim laterally across skin, hair, and fabric with restrained shadow detail, leaving the red eye as the sole concentrated color accent.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 한국인 남자의 이빨이 한국인 여자의 목덜미에 깊게 박힌 채 멈춘 측면.\n\nLOCATION (lock): The shadowed space against a utility pole in a dim urban back alley. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At upper-body height, the tracking move reaches a clean side axis and tightens until 도시 뒷골목의 한국인 남자 (포식 본능 활성화 상태)'s teeth, red eye, and 도시 뒷골목의 한국인 여자's neck occupy most of the close frame without losing fragments of both profiles. Shallow focus isolates the bite while a narrow strip of the utility pole remains behind the woman as a scale and support reference.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 도시 뒷골목의 한국인 남자 (포식 본능 활성화 상태) in the middle-left of the frame, foreground; 도시 뒷골목의 한국인 여자 in the middle-right of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 전봇대 (Supporting the woman's weight) — Only its side adjacent to the woman's shoulder is visible; used as A narrow background strip braces the woman's neck and maintains continuity with the wider alley view; 도시 뒷골목 (Dark at night); used as Falls beyond the shallow focus as the camera reaches the side profile.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The faint streetlamp illumination remains subdued, with the man's explicitly red eye forming the only concentrated deep-red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the dim streetlamp lighting, narrow urban alley surfaces, and the utility pole immediately behind the pair. EXCLUDE the earlier upright embrace and replace it with the side view of the man biting the woman's neck.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 도시 뒷골목의 한국인 남자 (포식 본능 활성화 상태) (한국인 남성, 성인의 얼굴, 평범한 얼굴선, 짧은 검은 머리); 도시 뒷골목의 한국인 여자 (한국인 여성, 성인의 얼굴, 평범한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an intimate close-focus lens feel while preserving the exact upper-body-height side axis and prescribed left-right placement, allowing subtle perspective depth between the two profile fragments and the point of contact. Concentrate the shallow focal plane on the embedded teeth and compressed neck surface, with facial contours falling gently out of focus toward the frame edges and the utility pole surviving only as a narrow soft brace behind her. Shape the subdued streetlamp illumination as a faint grazing highlight over the contact textures, surrounded by dense natural alley shadow and anchored by the man's deep-red eye.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "B": 7,
   "A": 3
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "지시된 클로즈업 샷 크기를 잘 따랐으며, 남자가 여자의 목을 무는 핵심 행동과 붉은 눈의 디테일을 충실히 구현했습니다."
   },
   {
    "label": "A",
    "score": 3,
    "verdict_ko": "캐릭터와 배경의 일치도는 높으나, 남자가 여자의 목덜미를 무는 가장 중요한 핵심 행동을 누락하고 단순히 마주보고 있습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S2sh1_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 도시 뒷골목의 한국인 남자 (포식 본능 활성화 상태): the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:915386>"
   },
   {
    "label": "CHARACTER REFERENCE — 도시 뒷골목의 한국인 여자: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:1138559>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "남자가 여자의 목덜미가 아닌 턱과 입 주변을 물고 있습니다.",
     "fix_en": "Lower the man's head so his mouth and teeth are embedded deeply into the side of the woman's neck."
    }
   ]
  },
  "ref_mode": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S2sh1"
  }
 },
 "groupbg::감식 현장": {
  "input_fingerprint": "dac9a4dd81ac1871",
  "meta": {
   "model": "gpt-image-2",
   "size": "1536x864",
   "pack": "9.202607211540",
   "contract": "bgfirst_full_v1",
   "group_sig": {
    "key": "감식 현장",
    "tags": [
     "S3sh3"
    ]
   }
  },
  "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\n\nTHE LOCATION — South Korea, 2026; all people are Korean unless stated: The police-controlled center of an outdoor crime scene, within the cordoned investigative area.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.",
  "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\n\nTHE LOCATION — South Korea, 2026; all people are Korean unless stated: The police-controlled center of an outdoor crime scene, within the cordoned investigative area.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_감식_현장_96a146.png",
  "asset_id": "213ce5b9-c1bc-4179-89e1-ef1fe0ea12a9",
  "input_asset_ids": [
   "d9e9e79a-454d-40ca-996c-8a4c67276308"
  ],
  "origin_tag": "S3sh3",
  "place_text": "The police-controlled center of an outdoor crime scene, within the cordoned investigative area."
 },
 "S3sh3::bgfirst_bg": {
  "input_fingerprint": "0ca9986d66dbd6a7",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 경찰 무리 중앙에서 팔짱을 낀 채 서서 현장을 주시하는 혜수.\n\nLOCATION (lock): The police-controlled center of an outdoor crime scene, within the cordoned investigative area.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Presented as local-news footage, an eye-level dolly-in settles at medium distance on 혜수 at the center of the detective group, her crossed arms and fixed attention organizing the frame. The surrounding officers remain at the lateral edges and slightly behind her, creating a restrained command portrait rather than a frontal address to the viewer.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 혜수 in the middle-center of the frame, midground; 형사 무리 in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 지역 뉴스 화면 (Showing the field investigation) — Its visible front face carries the live crime-scene image centered on 혜수 and the detectives; used as Provides the mediated visual field containing the crime-scene coverage; 폴리스 라인 (Set around the scene) — Its camera-facing side runs laterally behind the detective group as a boundary marker; used as Creates a visible boundary deeper in the crime scene; 현장 감식 경찰 (Examining the scene and recovering bodies); used as Supplies peripheral investigative activity while the move isolates 혜수.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight appropriate to the exterior location is rendered through a subdued, desaturated broadcast image with firm investigative contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 경찰 무리 중앙에서 팔짱을 낀 채 서서 현장을 주시하는 혜수.\n\nLOCATION (lock): The police-controlled center of an outdoor crime scene, within the cordoned investigative area.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Presented as local-news footage, an eye-level dolly-in settles at medium distance on 혜수 at the center of the detective group, her crossed arms and fixed attention organizing the frame. The surrounding officers remain at the lateral edges and slightly behind her, creating a restrained command portrait rather than a frontal address to the viewer.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 혜수 in the middle-center of the frame, midground; 형사 무리 in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 지역 뉴스 화면 (Showing the field investigation) — Its visible front face carries the live crime-scene image centered on 혜수 and the detectives; used as Provides the mediated visual field containing the crime-scene coverage; 폴리스 라인 (Set around the scene) — Its camera-facing side runs laterally behind the detective group as a boundary marker; used as Creates a visible boundary deeper in the crime scene; 현장 감식 경찰 (Examining the scene and recovering bodies); used as Supplies peripheral investigative activity while the move isolates 혜수.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight appropriate to the exterior location is rendered through a subdued, desaturated broadcast image with firm investigative contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S3sh3__bgfirst_bg.png",
  "asset_id": "9dbab1b6-3549-4454-8a0e-d16c14d5a4c2",
  "input_asset_ids": [
   "d9e9e79a-454d-40ca-996c-8a4c67276308",
   "213ce5b9-c1bc-4179-89e1-ef1fe0ea12a9"
  ]
 },
 "S3sh3": {
  "input_fingerprint": "177abe1556d8e1be",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 경찰 무리 중앙에서 팔짱을 낀 채 서서 현장을 주시하는 혜수.\n\nLOCATION (lock): The police-controlled center of an outdoor crime scene, within the cordoned investigative area. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Presented as local-news footage, an eye-level dolly-in settles at medium distance on 혜수 at the center of the detective group, her crossed arms and fixed attention organizing the frame. The surrounding officers remain at the lateral edges and slightly behind her, creating a restrained command portrait rather than a frontal address to the viewer.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 혜수 in the middle-center of the frame, midground; 형사 무리 in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 지역 뉴스 화면 (Showing the field investigation) — Its visible front face carries the live crime-scene image centered on 혜수 and the detectives; used as Provides the mediated visual field containing the crime-scene coverage; 폴리스 라인 (Set around the scene) — Its camera-facing side runs laterally behind the detective group as a boundary marker; used as Creates a visible boundary deeper in the crime scene; 현장 감식 경찰 (Examining the scene and recovering bodies); used as Supplies peripheral investigative activity while the move isolates 혜수.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight appropriate to the exterior location is rendered through a subdued, desaturated broadcast image with firm investigative contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 경찰 무리 중앙에서 팔짱을 낀 채 서서 현장을 주시하는 혜수.\n\nLOCATION (lock): The police-controlled center of an outdoor crime scene, within the cordoned investigative area. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Presented as local-news footage, an eye-level dolly-in settles at medium distance on 혜수 at the center of the detective group, her crossed arms and fixed attention organizing the frame. The surrounding officers remain at the lateral edges and slightly behind her, creating a restrained command portrait rather than a frontal address to the viewer.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 혜수 in the middle-center of the frame, midground; 형사 무리 in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 지역 뉴스 화면 (Showing the field investigation) — Its visible front face carries the live crime-scene image centered on 혜수 and the detectives; used as Provides the mediated visual field containing the crime-scene coverage; 폴리스 라인 (Set around the scene) — Its camera-facing side runs laterally behind the detective group as a boundary marker; used as Creates a visible boundary deeper in the crime scene; 현장 감식 경찰 (Examining the scene and recovering bodies); used as Supplies peripheral investigative activity while the move isolates 혜수.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight appropriate to the exterior location is rendered through a subdued, desaturated broadcast image with firm investigative contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 경찰 무리 중앙에서 팔짱을 낀 채 서서 현장을 주시하는 혜수.\n\nLOCATION (lock): The police-controlled center of an outdoor crime scene, within the cordoned investigative area. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Presented as local-news footage, an eye-level dolly-in settles at medium distance on 혜수 at the center of the detective group, her crossed arms and fixed attention organizing the frame. The surrounding officers remain at the lateral edges and slightly behind her, creating a restrained command portrait rather than a frontal address to the viewer.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 혜수 in the middle-center of the frame, midground; 형사 무리 in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 지역 뉴스 화면 (Showing the field investigation) — Its visible front face carries the live crime-scene image centered on 혜수 and the detectives; used as Provides the mediated visual field containing the crime-scene coverage; 폴리스 라인 (Set around the scene) — Its camera-facing side runs laterally behind the detective group as a boundary marker; used as Creates a visible boundary deeper in the crime scene; 현장 감식 경찰 (Examining the scene and recovering bodies); used as Supplies peripheral investigative activity while the move isolates 혜수.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight appropriate to the exterior location is rendered through a subdued, desaturated broadcast image with firm investigative contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S3sh3__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S3sh3.png"
    },
    {
     "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:796814>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_감식_현장_96a146.png"
    },
    {
     "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:796814>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 프레이밍(중앙-우측 배치)과 정면을 피한 시선 처리를 잘 구현했으나, 좌측 TV 화면이 중복 생성된 흠이 있습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "프레이밍 지시를 어긴 전신 샷이며, 피해야 할 정면 응시를 연출하고 좌측에 프롬프트에 없는 인물들을 임의로 추가하여 감점되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 프레이밍(중앙-우측 배치)과 정면을 피한 시선 처리를 잘 구현했으나, 좌측 TV 화면이 중복 생성된 흠이 있습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "프레이밍 지시를 어긴 전신 샷이며, 피해야 할 정면 응시를 연출하고 좌측에 프롬프트에 없는 인물들을 임의로 추가하여 감점되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 샷 크기(미디엄 샷)와 인물 레이아웃(혜수 중앙, 형사 무리 우측 배경)을 가장 정확하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "지시된 미디엄 샷 대신 전신이 노출되었으며, 형사 무리의 우측 배치 조건을 어기고 지문에 없는 인물(기자)을 임의로 추가했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 샷 크기(미디엄 샷)와 인물 레이아웃(혜수 중앙, 형사 무리 우측 배경)을 가장 정확하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "지시된 미디엄 샷 대신 전신이 노출되었으며, 형사 무리의 우측 배치 조건을 어기고 지문에 없는 인물(기자)을 임의로 추가했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 14,
     "B": 6
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 14,
   "B": 6
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "지시된 프레이밍(중앙-우측 배치)과 정면을 피한 시선 처리를 잘 구현했으나, 좌측 TV 화면이 중복 생성된 흠이 있습니다."
   },
   {
    "label": "B",
    "score": 3,
    "verdict_ko": "프레이밍 지시를 어긴 전신 샷이며, 피해야 할 정면 응시를 연출하고 좌측에 프롬프트에 없는 인물들을 임의로 추가하여 감점되었습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_감식_현장_96a146.png"
   },
   {
    "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:796814>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "배경 원본의 스탠드에는 TV 모니터가 하나만 있으나, 생성된 이미지에서는 그 위에 또 다른 모니터가 추가되어 고정된 배경 요소가 변형되었습니다.",
     "fix_en": "Remove the top TV monitor so only the single original monitor on the stand remains."
    },
    {
     "issue_ko": "위쪽 TV 화면 내에 프롬프트에서 명시적으로 금지한 텍스트(뉴스 자막 형태)가 포함되어 있습니다.",
     "fix_en": "Remove all text and graphical captions from the TV screen."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the top TV monitor so only the single original monitor on the stand remains.\n- Remove all text and graphical captions from the TV screen.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 구도와 인물 특징을 잘 재현했으며, 텍스트 금지 지시를 준수하여 깔끔한 화면을 완성했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "전반적인 배치는 좋으나, 절대 금지된 텍스트(뉴스 자막)를 모니터 화면에 임의로 생성하여 감점되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 구도와 인물 특징을 잘 재현했으며, 텍스트 금지 지시를 준수하여 깔끔한 화면을 완성했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "전반적인 배치는 좋으나, 절대 금지된 텍스트(뉴스 자막)를 모니터 화면에 임의로 생성하여 감점되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지시된 레이아웃과 구도를 훌륭하게 따랐으며, 금지된 텍스트나 마커 유출 없이 배경 요소를 자연스럽게 배치했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "좌측 뉴스 모니터 화면에 지시되지 않은 문자가 생성되는 치명적인 오류(Hard Violation)가 발생했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지시된 레이아웃과 구도를 훌륭하게 따랐으며, 금지된 텍스트나 마커 유출 없이 배경 요소를 자연스럽게 배치했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "좌측 뉴스 모니터 화면에 지시되지 않은 문자가 생성되는 치명적인 오류(Hard Violation)가 발생했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 9,
     "B": 15
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S3sh3__bgfirst_bg.png",
   "bg_asset_id": "9dbab1b6-3549-4454-8a0e-d16c14d5a4c2",
   "bg_record_key": "S3sh3::bgfirst_bg",
   "chain_winner": true,
   "authority": "groupbg",
   "group_key": "감식 현장",
   "groupbg_asset_id": "213ce5b9-c1bc-4179-89e1-ef1fe0ea12a9"
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S4sh1::variants": {
  "author_fp": "528d5a4c4c763bce",
  "author": {
   "variants": [
    {
     "approach_ko": "수직선과 옥상 평면을 또렷하게 정돈한 깊은 초점의 정적인 건축 관찰숏.",
     "prompt_en": "Use a restrained, near-orthographic lens feel with clean verticals and minimal perspective distortion, turning the rooftop planes into a calm, orderly geometry. Hold deep focus from the rooftop perimeter through the centered structure and into the distant outskirts, with crisp but natural material detail. Let the existing daylight fall evenly and softly, keeping the desaturated palette serene while subtle wear and empty space create unease."
    },
    {
     "approach_ko": "망원 압축과 완만한 주변부 초점 저하로 고립된 옥탑방에 시선을 모으는 관조적 전경.",
     "prompt_en": "Photograph the overview with gentle telephoto compression, visually drawing the surrounding rooftop boundary and distant urban layers closer around the centered rooftop room without altering their true scale. Keep the structure fully legible and sharply resolved while allowing a very gradual softness toward the far edges and deepest background. Preserve the natural daylight direction and muted color, using compressed spatial layers and stillness to make the peaceful isolation feel quietly tense."
    }
   ]
  },
  "reused": false
 },
 "S4sh1": {
  "input_fingerprint": "79dfc0213dc5aa64",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 오래된 빌라 옥상 위 평화로운 옥탑방 건물을 위에서 내려다본 외부 전경.\n\nLOCATION (lock): The open rooftop of an old multi-unit residential building, viewed from above with its rooftop room in full view. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From above the old villa, a downward-looking dolly-in settles into a centered wide overview of the rooftop room, retaining enough rooftop perimeter to establish its isolation and scale. The high angle remains calm and geometrically stable, with the rooftop room held in the middle rather than pushed into dramatic perspective.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 옥탑방 건물 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 옥탑방 건물 (Still in daylight) — The camera sees its upper face and exterior sides from above; used as Primary architectural focus centered within the rooftop overview; 오래된 빌라 옥상 (Supporting the rooftop room) — Its upper plane faces the downward-looking camera; used as Provides the surrounding boundary and scale reference for the rooftop room; 인천 변두리 (Seen in daylight); used as Remains visible beyond the villa to establish the stated outskirts location.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural daylight appropriate to the exterior is restrained and desaturated, preserving the peaceful overview while retaining quiet unease.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 오래된 빌라 옥상 위 평화로운 옥탑방 건물을 위에서 내려다본 외부 전경.\n\nLOCATION (lock): The open rooftop of an old multi-unit residential building, viewed from above with its rooftop room in full view. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From above the old villa, a downward-looking dolly-in settles into a centered wide overview of the rooftop room, retaining enough rooftop perimeter to establish its isolation and scale. The high angle remains calm and geometrically stable, with the rooftop room held in the middle rather than pushed into dramatic perspective.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 옥탑방 건물 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 옥탑방 건물 (Still in daylight) — The camera sees its upper face and exterior sides from above; used as Primary architectural focus centered within the rooftop overview; 오래된 빌라 옥상 (Supporting the rooftop room) — Its upper plane faces the downward-looking camera; used as Provides the surrounding boundary and scale reference for the rooftop room; 인천 변두리 (Seen in daylight); used as Remains visible beyond the villa to establish the stated outskirts location.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural daylight appropriate to the exterior is restrained and desaturated, preserving the peaceful overview while retaining quiet unease.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained, near-orthographic lens feel with clean verticals and minimal perspective distortion, turning the rooftop planes into a calm, orderly geometry. Hold deep focus from the rooftop perimeter through the centered structure and into the distant outskirts, with crisp but natural material detail. Let the existing daylight fall evenly and softly, keeping the desaturated palette serene while subtle wear and empty space create unease.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 오래된 빌라 옥상 위 평화로운 옥탑방 건물을 위에서 내려다본 외부 전경.\n\nLOCATION (lock): The open rooftop of an old multi-unit residential building, viewed from above with its rooftop room in full view. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From above the old villa, a downward-looking dolly-in settles into a centered wide overview of the rooftop room, retaining enough rooftop perimeter to establish its isolation and scale. The high angle remains calm and geometrically stable, with the rooftop room held in the middle rather than pushed into dramatic perspective.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 옥탑방 건물 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 옥탑방 건물 (Still in daylight) — The camera sees its upper face and exterior sides from above; used as Primary architectural focus centered within the rooftop overview; 오래된 빌라 옥상 (Supporting the rooftop room) — Its upper plane faces the downward-looking camera; used as Provides the surrounding boundary and scale reference for the rooftop room; 인천 변두리 (Seen in daylight); used as Remains visible beyond the villa to establish the stated outskirts location.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural daylight appropriate to the exterior is restrained and desaturated, preserving the peaceful overview while retaining quiet unease.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the overview with gentle telephoto compression, visually drawing the surrounding rooftop boundary and distant urban layers closer around the centered rooftop room without altering their true scale. Keep the structure fully legible and sharply resolved while allowing a very gradual softness toward the far edges and deepest background. Preserve the natural daylight direction and muted color, using compressed spatial layers and stillness to make the peaceful isolation feel quietly tense.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 7,
   "B": 4
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "지정된 앵글(정중앙 배치, 기하학적으로 안정적인 구도)을 성공적으로 구현하여 프레이밍 지침을 가장 잘 따랐습니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "참조 이미지의 카메라 구도와 과장된 투시를 그대로 복사하여 안정적인 정중앙 배치라는 프레이밍 지침을 위반했습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L04B02.png"
   },
   {
    "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/background_chain/seed_bg_rooftop_villa_sel.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "옥탑방 전면 벽의 창문이 정중앙에 위치해 있습니다 (레퍼런스는 왼쪽으로 치우쳐 있음).",
     "fix_en": "Move the window on the front wall of the rooftop room to the left side."
    },
    {
     "issue_ko": "옥탑방 우측 벽면에 레퍼런스에 있는 초록색 출입문이 누락되어 빈 벽으로 보입니다.",
     "fix_en": "Add the green door to the visible right-side wall of the rooftop room."
    },
    {
     "issue_ko": "화면 하단 옥상 난간 바로 아래에 레퍼런스의 구조와 다르게 창문이 달린 층이 추가되었습니다.",
     "fix_en": "Remove the window and wall section directly below the foreground rooftop parapet, replacing it with the exterior drop-off seen in the reference."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Move the window on the front wall of the rooftop room to the left side.\n- Add the green door to the visible right-side wall of the rooftop room.\n- Remove the window and wall section directly below the foreground rooftop parapet, replacing it with the exterior drop-off seen in the reference.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "구조물 외형(STRUCTURE LOOK) 레퍼런스에 따라 옥탑방 전면에 초록색 문과 창문을 정확히 배치하여 지침을 충실히 이행했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "구조물 외형 레퍼런스의 우선순위를 무시하고 건물 전면에 있어야 할 초록색 문을 누락했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "구조물 외형(STRUCTURE LOOK) 레퍼런스에 따라 옥탑방 전면에 초록색 문과 창문을 정확히 배치하여 지침을 충실히 이행했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "구조물 외형 레퍼런스의 우선순위를 무시하고 건물 전면에 있어야 할 초록색 문을 누락했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "참조 이미지의 구조물 정면(창문, 계량기)을 정확하게 반영했으며, 지시된 하이앵글 와이드 샷의 구도를 잘 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "구조물의 우측면에 있어야 할 초록색 문을 정면에 잘못 배치하여 공간 구조 및 형태적 일관성을 심각하게 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "참조 이미지의 구조물 정면(창문, 계량기)을 정확하게 반영했으며, 지시된 하이앵글 와이드 샷의 구도를 잘 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "구조물의 우측면에 있어야 할 초록색 문을 정면에 잘못 배치하여 공간 구조 및 형태적 일관성을 심각하게 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 11,
     "B": 11
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": false,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "플레이트+seed만 (배경 전용)",
  "share_plan": {
   "ref_plan": "background"
  },
  "lane_policy": "ab_select_bypass:bg_only",
  "plate_select": {
   "candidates": {
    "A": "L04B01",
    "B": "L04B02",
    "C": "L04B03",
    "D": "L04B04"
   },
   "assigned": "L04B02",
   "choice": "Candidate B",
   "confident": true,
   "reason_ko": "\"위에서 내려다본 외부 전경\"이라는 설명에 부합하는 부감(하이 앵글) 구도이며, 특별한 시간대나 조명 변화에 대한 언급이 없으므로 현재 할당된 이미지를 유지합니다.",
   "kept": "L04B02"
  }
 },
 "S5sh2::bgfirst_bg": {
  "input_fingerprint": "cd7cfae81acc1f41",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 앞치마를 입고 거실에 서서 어두운 표정으로 TV 화면을 응시하는 강민숙(민숙).\n\nLOCATION (lock): The compact living room of a rooftop apartment, with daylight entering through a small window and a television in view.\n\nTIME OF DAY (lock): morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height on 강민숙's three-quarter front, the camera dollies inward to a medium frame that holds her apron, tense upper body, and dark expression while her TV sightline crosses toward one side of the image. The TV remains visible beyond that sightline with its news-facing screen readable but smaller than 강민숙, keeping her reaction as the primary focus.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 강민숙 in the middle-center of the frame, midground, looks toward TV news screen; TV news screen in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: TV (On) — The front screen faces partly toward camera and displays the ongoing news report; used as Receives 강민숙's fixed sightline and supplies the in-scene news content; 거실의 작은 창 (Sunlight entering) — Sunlight enters toward the living room; the exterior beyond is not resolved; used as Appears along one side of the living area and supplies the stated narrow daylight; 싱크대 (Water running from the faucet) — Its working side faces into the room; used as Adds grounded domestic activity behind the reaction shot; 냄비 (Water boiling with steam rising) — Its upper opening is partially visible from the living-area angle; used as Provides a secondary background cue from the cooking area.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Narrow sunlight from the small living-room window meets the TV image in a subdued, desaturated interior with tense facial contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 앞치마를 입고 거실에 서서 어두운 표정으로 TV 화면을 응시하는 강민숙(민숙).\n\nLOCATION (lock): The compact living room of a rooftop apartment, with daylight entering through a small window and a television in view.\n\nTIME OF DAY (lock): morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height on 강민숙's three-quarter front, the camera dollies inward to a medium frame that holds her apron, tense upper body, and dark expression while her TV sightline crosses toward one side of the image. The TV remains visible beyond that sightline with its news-facing screen readable but smaller than 강민숙, keeping her reaction as the primary focus.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 강민숙 in the middle-center of the frame, midground, looks toward TV news screen; TV news screen in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: TV (On) — The front screen faces partly toward camera and displays the ongoing news report; used as Receives 강민숙's fixed sightline and supplies the in-scene news content; 거실의 작은 창 (Sunlight entering) — Sunlight enters toward the living room; the exterior beyond is not resolved; used as Appears along one side of the living area and supplies the stated narrow daylight; 싱크대 (Water running from the faucet) — Its working side faces into the room; used as Adds grounded domestic activity behind the reaction shot; 냄비 (Water boiling with steam rising) — Its upper opening is partially visible from the living-area angle; used as Provides a secondary background cue from the cooking area.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Narrow sunlight from the small living-room window meets the TV image in a subdued, desaturated interior with tense facial contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S5sh2__bgfirst_bg.png",
  "asset_id": "7d269369-386e-46f3-90dd-ea98dcc35ebd",
  "input_asset_ids": [
   "b39fa88a-1f77-46c7-96f4-5488f924644d",
   "53db5d48-d1cb-49c1-b19f-5d8945bf5415"
  ]
 },
 "S5sh2": {
  "input_fingerprint": "ba3b3985185d74ee",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning.\n\nSHOT TEXT (authoritative, Korean): 앞치마를 입고 거실에 서서 어두운 표정으로 TV 화면을 응시하는 강민숙(민숙).\n\nLOCATION (lock): The compact living room of a rooftop apartment, with daylight entering through a small window and a television in view. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height on 강민숙's three-quarter front, the camera dollies inward to a medium frame that holds her apron, tense upper body, and dark expression while her TV sightline crosses toward one side of the image. The TV remains visible beyond that sightline with its news-facing screen readable but smaller than 강민숙, keeping her reaction as the primary focus.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 강민숙 in the middle-center of the frame, midground, looks toward TV news screen; TV news screen in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: TV (On) — The front screen faces partly toward camera and displays the ongoing news report; used as Receives 강민숙's fixed sightline and supplies the in-scene news content; 거실의 작은 창 (Sunlight entering) — Sunlight enters toward the living room; the exterior beyond is not resolved; used as Appears along one side of the living area and supplies the stated narrow daylight; 싱크대 (Water running from the faucet) — Its working side faces into the room; used as Adds grounded domestic activity behind the reaction shot; 냄비 (Water boiling with steam rising) — Its upper opening is partially visible from the living-area angle; used as Provides a secondary background cue from the cooking area.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Narrow sunlight from the small living-room window meets the TV image in a subdued, desaturated interior with tense facial contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk continues to wear her apron, and the small circular mark remains on her wrist as it has since Suryoung's childhood.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 여성, 40대 초반의 중년 얼굴, 부드러운 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning.\n\nSHOT TEXT (authoritative, Korean): 앞치마를 입고 거실에 서서 어두운 표정으로 TV 화면을 응시하는 강민숙(민숙).\n\nLOCATION (lock): The compact living room of a rooftop apartment, with daylight entering through a small window and a television in view. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height on 강민숙's three-quarter front, the camera dollies inward to a medium frame that holds her apron, tense upper body, and dark expression while her TV sightline crosses toward one side of the image. The TV remains visible beyond that sightline with its news-facing screen readable but smaller than 강민숙, keeping her reaction as the primary focus.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 강민숙 in the middle-center of the frame, midground, looks toward TV news screen; TV news screen in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: TV (On) — The front screen faces partly toward camera and displays the ongoing news report; used as Receives 강민숙's fixed sightline and supplies the in-scene news content; 거실의 작은 창 (Sunlight entering) — Sunlight enters toward the living room; the exterior beyond is not resolved; used as Appears along one side of the living area and supplies the stated narrow daylight; 싱크대 (Water running from the faucet) — Its working side faces into the room; used as Adds grounded domestic activity behind the reaction shot; 냄비 (Water boiling with steam rising) — Its upper opening is partially visible from the living-area angle; used as Provides a secondary background cue from the cooking area.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Narrow sunlight from the small living-room window meets the TV image in a subdued, desaturated interior with tense facial contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk continues to wear her apron, and the small circular mark remains on her wrist as it has since Suryoung's childhood.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 여성, 40대 초반의 중년 얼굴, 부드러운 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning.\n\nSHOT TEXT (authoritative, Korean): 앞치마를 입고 거실에 서서 어두운 표정으로 TV 화면을 응시하는 강민숙(민숙).\n\nLOCATION (lock): The compact living room of a rooftop apartment, with daylight entering through a small window and a television in view. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height on 강민숙's three-quarter front, the camera dollies inward to a medium frame that holds her apron, tense upper body, and dark expression while her TV sightline crosses toward one side of the image. The TV remains visible beyond that sightline with its news-facing screen readable but smaller than 강민숙, keeping her reaction as the primary focus.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 강민숙 in the middle-center of the frame, midground, looks toward TV news screen; TV news screen in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: TV (On) — The front screen faces partly toward camera and displays the ongoing news report; used as Receives 강민숙's fixed sightline and supplies the in-scene news content; 거실의 작은 창 (Sunlight entering) — Sunlight enters toward the living room; the exterior beyond is not resolved; used as Appears along one side of the living area and supplies the stated narrow daylight; 싱크대 (Water running from the faucet) — Its working side faces into the room; used as Adds grounded domestic activity behind the reaction shot; 냄비 (Water boiling with steam rising) — Its upper opening is partially visible from the living-area angle; used as Provides a secondary background cue from the cooking area.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Narrow sunlight from the small living-room window meets the TV image in a subdued, desaturated interior with tense facial contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk continues to wear her apron, and the small circular mark remains on her wrist as it has since Suryoung's childhood.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 여성, 40대 초반의 중년 얼굴, 부드러운 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S5sh2__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S5sh2.png"
    },
    {
     "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:967745>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L05B02.png"
    },
    {
     "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:967745>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지시된 미디엄 샷 프레이밍을 정확히 따랐으며, TV를 배경(우측 중앙)에 배치하라는 레이아웃 지침을 충실히 구현했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "프레임이 너무 넓게 잡혀 미디엄 샷 지시를 어겼으며, TV가 배경이 아닌 전경에 배치되어 프레임 레이아웃 규정을 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지시된 미디엄 샷 프레이밍을 정확히 따랐으며, TV를 배경(우측 중앙)에 배치하라는 레이아웃 지침을 충실히 구현했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "프레임이 너무 넓게 잡혀 미디엄 샷 지시를 어겼으며, TV가 배경이 아닌 전경에 배치되어 프레임 레이아웃 규정을 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "지정된 미디엄 샷 프레이밍을 정확히 구현했으며, TV를 배경에 적절히 배치하고 수돗물, 끓는 냄비, 손목의 표식 등 모든 디테일을 훌륭하게 살렸습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "프레이밍이 지정된 미디엄 샷보다 너무 넓게 잡혔고, 배경에 있어야 할 TV가 전경에 잘못 배치되어 공간적 왜곡이 발생했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지정된 미디엄 샷 프레이밍을 정확히 구현했으며, TV를 배경에 적절히 배치하고 수돗물, 끓는 냄비, 손목의 표식 등 모든 디테일을 훌륭하게 살렸습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "프레이밍이 지정된 미디엄 샷보다 너무 넓게 잡혔고, 배경에 있어야 할 TV가 전경에 잘못 배치되어 공간적 왜곡이 발생했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 10,
     "B": 17
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 10,
   "B": 17
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 8,
    "verdict_ko": "지시된 미디엄 샷 프레이밍을 정확히 따랐으며, TV를 배경(우측 중앙)에 배치하라는 레이아웃 지침을 충실히 구현했습니다."
   },
   {
    "label": "A",
    "score": 5,
    "verdict_ko": "프레임이 너무 넓게 잡혀 미디엄 샷 지시를 어겼으며, TV가 배경이 아닌 전경에 배치되어 프레임 레이아웃 규정을 위반했습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L05B02.png"
   },
   {
    "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:967745>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "모든 텍스트 생성을 금지한 지시문과 달리 TV 화면에 뉴스 자막 형태의 가짜 글자들이 나타나 있습니다.",
     "fix_en": "Remove all text, banners, and captions from the TV screen."
    },
    {
     "issue_ko": "로케이션 레퍼런스 사진의 벽면 부착형 수도꼭지가 싱크대 데크 부착형으로 다르게 렌더링되어 고정된 공간 특징을 위반했습니다.",
     "fix_en": "Change the kitchen faucet to a wall-mounted fixture matching the exact plumbing layout of the location reference."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all text, banners, and captions from the TV screen.\n- Change the kitchen faucet to a wall-mounted fixture matching the exact plumbing layout of the location reference.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지시된 구도, 인물, 배경 요소를 모두 충실히 구현했으며, TV 뉴스 화면이 부자연스러운 텍스트 없이 자연스럽게 연출되었습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요구된 모든 세부 사항을 잘 반영했으나, TV 화면에 생성된 부자연스러운 AI 텍스트가 시각적 완성도를 다소 떨어뜨립니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지시된 구도, 인물, 배경 요소를 모두 충실히 구현했으며, TV 뉴스 화면이 부자연스러운 텍스트 없이 자연스럽게 연출되었습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요구된 모든 세부 사항을 잘 반영했으나, TV 화면에 생성된 부자연스러운 AI 텍스트가 시각적 완성도를 다소 떨어뜨립니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지시된 구도, 인물의 감정선, 싱크대와 냄비 등 배경 요소의 배치가 완벽하며, TV 뉴스의 화면 디테일이 더 사실적입니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "전반적인 프레이밍과 요소 배치는 우수하나, TV 뉴스 화면에 텍스트가 없어 지시된 디테일 측면에서 약간 아쉽습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지시된 구도, 인물의 감정선, 싱크대와 냄비 등 배경 요소의 배치가 완벽하며, TV 뉴스의 화면 디테일이 더 사실적입니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "전반적인 프레이밍과 요소 배치는 우수하나, TV 뉴스 화면에 텍스트가 없어 지시된 디테일 측면에서 약간 아쉽습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 15,
     "B": 15
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": false,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S5sh2__bgfirst_bg.png",
   "bg_asset_id": "7d269369-386e-46f3-90dd-ea98dcc35ebd",
   "bg_record_key": "S5sh2::bgfirst_bg",
   "chain_winner": false,
   "authority": "plate"
  },
  "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L05B01",
    "B": "L05B02",
    "C": "L05B03",
    "D": "L05B04",
    "E": "L05B05",
    "F": "L05B06",
    "G": "L05B07"
   },
   "assigned": "L05B02",
   "choice": "Candidate B",
   "confident": true,
   "reason_ko": "\"거실에 서서... TV 화면을 응시하는\"이라는 지문에 부합하게 거실 공간과 TV가 보이는 배경입니다.",
   "kept": "L05B02"
  }
 },
 "S5sh8::variants": {
  "author_fp": "2722b6daca50f295",
  "author": {
   "variants": [
    {
     "approach_ko": "얕은 심도로 수리영의 코와 미소가 민숙의 어깨에 닿는 미세한 접촉을 강조한 압축된 친밀한 측면 클로즈업.",
     "prompt_en": "Use a gently compressed, shallow-focus profile close-up, holding the point where Suriyoung’s nose meets Minsuk’s shoulder and the fine movement of her smile in precise focus. Let Minsuk’s adjacent profile remain softly legible while the entrance dissolves into subdued residential texture; soft window daylight grazes their faces laterally, with restrained warmth in skin and quiet, desaturated shadows."
    },
    {
     "approach_ko": "두 사람의 옆얼굴과 현관의 공간선을 함께 읽히게 하는 자연스러운 원근과 깊은 초점의 긴장감 있는 클로즈업.",
     "prompt_en": "Photograph the close profile with natural perspective and moderately deep focus, keeping both women’s heads and shoulders clearly articulated while the entrance geometry remains recognizable as a soft second plane. Use the doorway lines and the dimmer interior around them to contain the central pairing, allowing the small window’s diffuse morning light to separate their profiles without romantic glow; preserve tactile wear, muted color, and a quiet undertone of tension."
    }
   ]
  },
  "reused": false
 },
 "S5sh8": {
  "input_fingerprint": "ac2cd280a736866e",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning.\n\nSHOT TEXT (authoritative, Korean): 수리영이 서 있는 강민숙(민숙)의 어깨 쪽에 코를 댄 채 미소 짓는 측면.\n\nLOCATION (lock): The interior entryway of the rooftop apartment, where shoes are put on beside the front door. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At shoulder height near the entrance, the side-tracking camera settles into a clean close profile as 수리영 brings her nose to 강민숙's shoulder and smiles. Their heads and shoulders share the central frame, with 수리영 slightly nearer the lens and the entrance area left softly present behind them before the camera moves down toward the wrist.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 현관 공간 (Visible behind the pair); used as Soft background context locating the affectionate exchange near departure; 거실의 작은 창 (Sunlight entering) — Its opening faces into the living area; no exterior detail is resolved; used as Provides limited interior daylight without displacing the two profiles.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The small window's daylight is kept soft and desaturated, allowing restrained warmth in the intimate profiles without losing the underlying tension.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the compact rooftop-home interior, worn residential materials, daylight filtering through the small window, and the woman's apron. EXCLUDE the television-watching action and kitchen activity; frame the daughter leaning toward the woman's shoulder.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk still bears the small circular mark on her wrist.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리); 강민숙 (한국인 여성, 40대 초반의 중년 얼굴, 부드러운 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning.\n\nSHOT TEXT (authoritative, Korean): 수리영이 서 있는 강민숙(민숙)의 어깨 쪽에 코를 댄 채 미소 짓는 측면.\n\nLOCATION (lock): The interior entryway of the rooftop apartment, where shoes are put on beside the front door. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At shoulder height near the entrance, the side-tracking camera settles into a clean close profile as 수리영 brings her nose to 강민숙's shoulder and smiles. Their heads and shoulders share the central frame, with 수리영 slightly nearer the lens and the entrance area left softly present behind them before the camera moves down toward the wrist.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 현관 공간 (Visible behind the pair); used as Soft background context locating the affectionate exchange near departure; 거실의 작은 창 (Sunlight entering) — Its opening faces into the living area; no exterior detail is resolved; used as Provides limited interior daylight without displacing the two profiles.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The small window's daylight is kept soft and desaturated, allowing restrained warmth in the intimate profiles without losing the underlying tension.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the compact rooftop-home interior, worn residential materials, daylight filtering through the small window, and the woman's apron. EXCLUDE the television-watching action and kitchen activity; frame the daughter leaning toward the woman's shoulder.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk still bears the small circular mark on her wrist.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리); 강민숙 (한국인 여성, 40대 초반의 중년 얼굴, 부드러운 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a gently compressed, shallow-focus profile close-up, holding the point where Suriyoung’s nose meets Minsuk’s shoulder and the fine movement of her smile in precise focus. Let Minsuk’s adjacent profile remain softly legible while the entrance dissolves into subdued residential texture; soft window daylight grazes their faces laterally, with restrained warmth in skin and quiet, desaturated shadows.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning.\n\nSHOT TEXT (authoritative, Korean): 수리영이 서 있는 강민숙(민숙)의 어깨 쪽에 코를 댄 채 미소 짓는 측면.\n\nLOCATION (lock): The interior entryway of the rooftop apartment, where shoes are put on beside the front door. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At shoulder height near the entrance, the side-tracking camera settles into a clean close profile as 수리영 brings her nose to 강민숙's shoulder and smiles. Their heads and shoulders share the central frame, with 수리영 slightly nearer the lens and the entrance area left softly present behind them before the camera moves down toward the wrist.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 현관 공간 (Visible behind the pair); used as Soft background context locating the affectionate exchange near departure; 거실의 작은 창 (Sunlight entering) — Its opening faces into the living area; no exterior detail is resolved; used as Provides limited interior daylight without displacing the two profiles.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The small window's daylight is kept soft and desaturated, allowing restrained warmth in the intimate profiles without losing the underlying tension.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the compact rooftop-home interior, worn residential materials, daylight filtering through the small window, and the woman's apron. EXCLUDE the television-watching action and kitchen activity; frame the daughter leaning toward the woman's shoulder.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk still bears the small circular mark on her wrist.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리); 강민숙 (한국인 여성, 40대 초반의 중년 얼굴, 부드러운 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the close profile with natural perspective and moderately deep focus, keeping both women’s heads and shoulders clearly articulated while the entrance geometry remains recognizable as a soft second plane. Use the doorway lines and the dimmer interior around them to contain the central pairing, allowing the small window’s diffuse morning light to separate their profiles without romantic glow; preserve tactile wear, muted color, and a quiet undertone of tension.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "B": 6,
   "A": 4
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 6,
    "verdict_ko": "수리영의 의상 오류가 있으나, 지시된 헤드 앤 숄더 클로즈업 프레이밍과 카메라와의 거리 렌즈 배치를 정확히 구현하여 규정상 우위를 점합니다."
   },
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "의상과 소품은 정확하지만, 클로즈업 지시를 어기고 피사체보다 공간이 더 많이 보이는 미디엄 샷으로 렌더링되어 프레이밍 우선순위에서 감점되었습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S5sh2_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   },
   {
    "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:967745>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "이전 컷에서 자연스럽게 풀려 있던 민숙의 머리가 머리끈으로 묶여 있습니다.",
     "fix_en": "Remove the black hair tie and render Minsuk's hair hanging completely loose to match the previous shot."
    },
    {
     "issue_ko": "수리영이 렌즈에 더 가깝게 위치해야 한다는 프레이밍 지시와 달리 민숙의 어깨 뒤편에 가려져 있습니다.",
     "fix_en": "Position Suriyoung so she overlaps the camera-facing side of Minsuk's shoulder, making her nearer to the lens."
    },
    {
     "issue_ko": "이전 컷의 공간 구조에는 없는 현관문 형태의 문이 인물들 뒤(화면 왼쪽)에 추가로 생성되어 있습니다.",
     "fix_en": "Remove the open door and its hardware on the left side of the frame, showing the entryway space opening directly into the living area."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the black hair tie and render Minsuk's hair hanging completely loose to match the previous shot.\n- Position Suriyoung so she overlaps the camera-facing side of Minsuk's shoulder, making her nearer to the lens.\n- Remove the open door and its hardware on the left side of the frame, showing the entryway space opening directly into the living area.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "측면 구도, 어깨에 코를 대고 미소 짓는 동작, 손목의 표식 등 모든 조건을 완벽히 충족함."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "수리영이 카메라 정면을 응시하여 명시된 '측면' 구도 지시를 위반함."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "측면 구도, 어깨에 코를 대고 미소 짓는 동작, 손목의 표식 등 모든 조건을 완벽히 충족함."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "수리영이 카메라 정면을 응시하여 명시된 '측면' 구도 지시를 위반함."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "수리영이 카메라를 응시하고 있어 '어깨에 코를 댄 측면'이라는 핵심 행동 및 구도 지시를 위반했습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "두 인물이 완벽한 측면 구도를 유지하며, 수리영이 강민숙의 어깨에 코를 대고 미소 짓는 모습을 프롬프트대로 정확하게 구현했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "수리영이 카메라를 응시하고 있어 '어깨에 코를 댄 측면'이라는 핵심 행동 및 구도 지시를 위반했습니다."
     },
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "두 인물이 완벽한 측면 구도를 유지하며, 수리영이 강민숙의 어깨에 코를 대고 미소 짓는 모습을 프롬프트대로 정확하게 구현했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 15,
     "B": 8
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S5sh2"
  }
 },
 "S5sh9::variants": {
  "author_fp": "f51935fa631b98d9",
  "author": {
   "variants": [
    {
     "approach_ko": "손목 표면에 거의 정면으로 밀착해 표식의 형태를 차분하고 명료하게 읽히게 하는 관찰적 인서트.",
     "prompt_en": "Photograph the wrist from a nearly square-on, level viewpoint, holding the small circular mark in crisp focus at middle-center while the adjacent clothing edge stays softly legible. Use shallow depth of field so the entrance surroundings dissolve into subdued, natural blur, with broad soft daylight revealing skin texture and the mark’s adherence to the wrist’s curvature."
    },
    {
     "approach_ko": "손목을 비스듬히 스치는 시선과 얕은 사광으로 피부의 곡면과 표식의 물성을 강조하는 긴장된 인서트.",
     "prompt_en": "Take a slightly oblique, wrist-height view along the arm, letting the wrist’s curved plane lead directly to the centered circular mark while the clothing edge forms a restrained peripheral layer. Favor gentle grazing daylight and a modestly deeper focus across the wrist, emphasizing pores, fine tonal variation, and the physical integration of the mark as the entrance recedes into dim, desaturated softness."
    }
   ]
  },
  "reused": false
 },
 "S5sh9": {
  "input_fingerprint": "b7c4ac7caaa5a547",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning.\n\nSHOT TEXT (authoritative, Korean): 강민숙(민숙)의 손목에 있는 또렷한 작은 원형 표식 클로즈업.\n\nLOCATION (lock): The rooftop apartment’s interior entry area beside the front door, lit by the home’s daytime ambient light. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From 수리영's side of the interaction, the camera finishes its downward move at wrist height and tightens into an insert of 강민숙's small circular mark. The wrist and adjacent edge of clothing remain naturally scaled within the frame, while the entrance surroundings fall away without turning the mark into an oversized object.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: 강민숙의 손목 in the middle-center of the frame, foreground.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Soft interior daylight renders the circular mark with restrained contrast and otherwise desaturated color.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the same interior light, the woman's clothing, and her arm position from the reference. EXCLUDE the daughter's face and the broader room composition; isolate the small circular mark on the wrist.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The same small, unusually distinct circular mark remains on Minsuk's wrist.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 여성, 40대 초반의 중년 얼굴, 부드러운 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning.\n\nSHOT TEXT (authoritative, Korean): 강민숙(민숙)의 손목에 있는 또렷한 작은 원형 표식 클로즈업.\n\nLOCATION (lock): The rooftop apartment’s interior entry area beside the front door, lit by the home’s daytime ambient light. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From 수리영's side of the interaction, the camera finishes its downward move at wrist height and tightens into an insert of 강민숙's small circular mark. The wrist and adjacent edge of clothing remain naturally scaled within the frame, while the entrance surroundings fall away without turning the mark into an oversized object.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: 강민숙의 손목 in the middle-center of the frame, foreground.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Soft interior daylight renders the circular mark with restrained contrast and otherwise desaturated color.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the same interior light, the woman's clothing, and her arm position from the reference. EXCLUDE the daughter's face and the broader room composition; isolate the small circular mark on the wrist.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The same small, unusually distinct circular mark remains on Minsuk's wrist.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 여성, 40대 초반의 중년 얼굴, 부드러운 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the wrist from a nearly square-on, level viewpoint, holding the small circular mark in crisp focus at middle-center while the adjacent clothing edge stays softly legible. Use shallow depth of field so the entrance surroundings dissolve into subdued, natural blur, with broad soft daylight revealing skin texture and the mark’s adherence to the wrist’s curvature.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning.\n\nSHOT TEXT (authoritative, Korean): 강민숙(민숙)의 손목에 있는 또렷한 작은 원형 표식 클로즈업.\n\nLOCATION (lock): The rooftop apartment’s interior entry area beside the front door, lit by the home’s daytime ambient light. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From 수리영's side of the interaction, the camera finishes its downward move at wrist height and tightens into an insert of 강민숙's small circular mark. The wrist and adjacent edge of clothing remain naturally scaled within the frame, while the entrance surroundings fall away without turning the mark into an oversized object.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: 강민숙의 손목 in the middle-center of the frame, foreground.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Soft interior daylight renders the circular mark with restrained contrast and otherwise desaturated color.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the same interior light, the woman's clothing, and her arm position from the reference. EXCLUDE the daughter's face and the broader room composition; isolate the small circular mark on the wrist.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The same small, unusually distinct circular mark remains on Minsuk's wrist.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 여성, 40대 초반의 중년 얼굴, 부드러운 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nTake a slightly oblique, wrist-height view along the arm, letting the wrist’s curved plane lead directly to the centered circular mark while the clothing edge forms a restrained peripheral layer. Favor gentle grazing daylight and a modestly deeper focus across the wrist, emphasizing pores, fine tonal variation, and the physical integration of the mark as the entrance recedes into dim, desaturated softness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "B": 7,
   "A": 3
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "지시된 클로즈업 프레이밍을 잘 구현했으며 이전 샷의 팔 자세와 연속성을 정확히 유지했습니다."
   },
   {
    "label": "A",
    "score": 3,
    "verdict_ko": "팔을 위로 뻗은 자세로 변경하여 이전 샷의 팔 위치를 유지하라는 지시를 위반했습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S5sh8_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:967745>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "레퍼런스 이미지에서는 손이 문 표면을 짚고 있으나, 생성된 이미지에서는 문 가장자리를 움켜쥐고 있어 지시된 팔의 자세(arm position)와 다르며, 안쪽 팔목이 보이는 상태에서 손가락이 반대로 감싸고 있어 해부학적으로 어긋납니다.",
     "fix_en": "Change the hand to press against the flat surface of the door as in the reference, correcting the anatomical orientation of the wrist and fingers."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the hand to press against the flat surface of the door as in the reference, correcting the anatomical orientation of the wrist and fingers.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 의상과 손목의 작은 원형 표식을 중앙에 정확히 배치했으며, 손의 해부학적 구조가 자연스럽게 구현되었습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "손가락 관절과 개수가 해부학적으로 불가능한 형태(여분의 손가락)를 띠어 심각한 오류가 있습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 의상과 손목의 작은 원형 표식을 중앙에 정확히 배치했으며, 손의 해부학적 구조가 자연스럽게 구현되었습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "손가락 관절과 개수가 해부학적으로 불가능한 형태(여분의 손가락)를 띠어 심각한 오류가 있습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 인서트 클로즈업 구도와 원형 표식을 충실히 구현했으며, B에 비해 손의 해부학적 구조가 훨씬 자연스럽습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "요구된 샷 크기와 피사체는 반영했으나, 손가락 개수가 많고 관절 형태가 왜곡되는 심각한 해부학적 오류가 발생했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 인서트 클로즈업 구도와 원형 표식을 충실히 구현했으며, B에 비해 손의 해부학적 구조가 훨씬 자연스럽습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "요구된 샷 크기와 피사체는 반영했으나, 손가락 개수가 많고 관절 형태가 왜곡되는 심각한 해부학적 오류가 발생했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 7,
     "B": 14
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S5sh8"
  }
 },
 "groupbg::해안 도로": {
  "input_fingerprint": "1ce9977c83f45d5f",
  "meta": {
   "model": "gpt-image-2",
   "size": "1536x864",
   "pack": "9.202607211540",
   "contract": "bgfirst_full_v1",
   "group_sig": {
    "key": "해안 도로",
    "tags": [
     "S24sh2",
     "S6sh1",
     "S9sh1"
    ]
   }
  },
  "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\n\nTHE LOCATION — South Korea, 2026; all people are Korean unless stated: An open coastal roadway beside the sea in daylight.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.",
  "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\n\nTHE LOCATION — South Korea, 2026; all people are Korean unless stated: An open coastal roadway beside the sea in daylight.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_해안_도로_8394b3.png",
  "asset_id": "519694bd-31b4-4c58-9c2d-45ff2fb4767c",
  "input_asset_ids": [
   "49fc56b9-157e-4e4a-acec-45068c3cc30e"
  ],
  "origin_tag": "S6sh1",
  "place_text": "An open coastal roadway beside the sea in daylight."
 },
 "S6sh1::bgfirst_bg": {
  "input_fingerprint": "724515f0075eb2ef",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 이어폰을 낀 채 밝은 표정으로 해안가 도로 위 자전거 안장에 앉아 한쪽 무릎을 굽혀 페달을 밟고 있는 중간 동작, 옷자락이 뒤로 흩날리는 수리영.\n\nLOCATION (lock): An open coastal roadway beside the sea in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 수리영's torso height, the camera tracks beside and slightly ahead of the bicycle at medium distance, looking back on her three-quarter profile as she pedals along the coastal road. Her bright face occupies the upper middle of the frame while the flexed knee and pedal cycle remain visible below, and her clothing trails toward the rear edge to reinforce forward travel.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward road ahead at frame right; 해안가 도로 진행 방향 in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 해안가 도로 (수리영 is riding along it); used as Defines the lateral direction of the parallel tracking move and remains a scale reference beneath the bicycle; 바다 (Visible beside the road); used as Provides the coastal context beyond the moving rider; 자전거 (Moving forward) — Its near side faces the parallel-tracking camera; used as Keeps the seated posture and active pedal motion legible within the lower frame.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural daylight appropriate to the coastal road is kept restrained and desaturated while preserving 수리영's bright, energetic expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 이어폰을 낀 채 밝은 표정으로 해안가 도로 위 자전거 안장에 앉아 한쪽 무릎을 굽혀 페달을 밟고 있는 중간 동작, 옷자락이 뒤로 흩날리는 수리영.\n\nLOCATION (lock): An open coastal roadway beside the sea in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 수리영's torso height, the camera tracks beside and slightly ahead of the bicycle at medium distance, looking back on her three-quarter profile as she pedals along the coastal road. Her bright face occupies the upper middle of the frame while the flexed knee and pedal cycle remain visible below, and her clothing trails toward the rear edge to reinforce forward travel.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward road ahead at frame right; 해안가 도로 진행 방향 in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 해안가 도로 (수리영 is riding along it); used as Defines the lateral direction of the parallel tracking move and remains a scale reference beneath the bicycle; 바다 (Visible beside the road); used as Provides the coastal context beyond the moving rider; 자전거 (Moving forward) — Its near side faces the parallel-tracking camera; used as Keeps the seated posture and active pedal motion legible within the lower frame.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural daylight appropriate to the coastal road is kept restrained and desaturated while preserving 수리영's bright, energetic expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S6sh1__bgfirst_bg.png",
  "asset_id": "8e08cb34-e8ec-429f-af33-4ae4dcac5c09",
  "input_asset_ids": [
   "49fc56b9-157e-4e4a-acec-45068c3cc30e",
   "519694bd-31b4-4c58-9c2d-45ff2fb4767c"
  ]
 },
 "S6sh1": {
  "input_fingerprint": "7bbdd772ad42242b",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 이어폰을 낀 채 밝은 표정으로 해안가 도로 위 자전거 안장에 앉아 한쪽 무릎을 굽혀 페달을 밟고 있는 중간 동작, 옷자락이 뒤로 흩날리는 수리영.\n\nLOCATION (lock): An open coastal roadway beside the sea in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 수리영's torso height, the camera tracks beside and slightly ahead of the bicycle at medium distance, looking back on her three-quarter profile as she pedals along the coastal road. Her bright face occupies the upper middle of the frame while the flexed knee and pedal cycle remain visible below, and her clothing trails toward the rear edge to reinforce forward travel.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward road ahead at frame right; 해안가 도로 진행 방향 in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 해안가 도로 (수리영 is riding along it); used as Defines the lateral direction of the parallel tracking move and remains a scale reference beneath the bicycle; 바다 (Visible beside the road); used as Provides the coastal context beyond the moving rider; 자전거 (Moving forward) — Its near side faces the parallel-tracking camera; used as Keeps the seated posture and active pedal motion legible within the lower frame.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural daylight appropriate to the coastal road is kept restrained and desaturated while preserving 수리영's bright, energetic expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 이어폰을 낀 채 밝은 표정으로 해안가 도로 위 자전거 안장에 앉아 한쪽 무릎을 굽혀 페달을 밟고 있는 중간 동작, 옷자락이 뒤로 흩날리는 수리영.\n\nLOCATION (lock): An open coastal roadway beside the sea in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 수리영's torso height, the camera tracks beside and slightly ahead of the bicycle at medium distance, looking back on her three-quarter profile as she pedals along the coastal road. Her bright face occupies the upper middle of the frame while the flexed knee and pedal cycle remain visible below, and her clothing trails toward the rear edge to reinforce forward travel.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward road ahead at frame right; 해안가 도로 진행 방향 in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 해안가 도로 (수리영 is riding along it); used as Defines the lateral direction of the parallel tracking move and remains a scale reference beneath the bicycle; 바다 (Visible beside the road); used as Provides the coastal context beyond the moving rider; 자전거 (Moving forward) — Its near side faces the parallel-tracking camera; used as Keeps the seated posture and active pedal motion legible within the lower frame.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural daylight appropriate to the coastal road is kept restrained and desaturated while preserving 수리영's bright, energetic expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 이어폰을 낀 채 밝은 표정으로 해안가 도로 위 자전거 안장에 앉아 한쪽 무릎을 굽혀 페달을 밟고 있는 중간 동작, 옷자락이 뒤로 흩날리는 수리영.\n\nLOCATION (lock): An open coastal roadway beside the sea in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 수리영's torso height, the camera tracks beside and slightly ahead of the bicycle at medium distance, looking back on her three-quarter profile as she pedals along the coastal road. Her bright face occupies the upper middle of the frame while the flexed knee and pedal cycle remain visible below, and her clothing trails toward the rear edge to reinforce forward travel.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward road ahead at frame right; 해안가 도로 진행 방향 in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 해안가 도로 (수리영 is riding along it); used as Defines the lateral direction of the parallel tracking move and remains a scale reference beneath the bicycle; 바다 (Visible beside the road); used as Provides the coastal context beyond the moving rider; 자전거 (Moving forward) — Its near side faces the parallel-tracking camera; used as Keeps the seated posture and active pedal motion legible within the lower frame.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural daylight appropriate to the coastal road is kept restrained and desaturated while preserving 수리영's bright, energetic expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S6sh1__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S6sh1.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_해안_도로_8394b3.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "피사체가 프레임 우측을 향해 이동하는 측면 구도와 해안도로의 배치를 지시문대로 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "자전거가 프레임 우측이 아닌 정면을 향해 주행하고 있어 요구된 프레임 레이아웃과 카메라 앵글을 어겼습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "피사체가 프레임 우측을 향해 이동하는 측면 구도와 해안도로의 배치를 지시문대로 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "자전거가 프레임 우측이 아닌 정면을 향해 주행하고 있어 요구된 프레임 레이아웃과 카메라 앵글을 어겼습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "카메라가 자전거 측면 및 약간 앞쪽에서 3/4 측면 얼굴을 포착하며 화면 우측으로 진행하는 명시된 앵글과 구도를 정확히 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "정면 앵글로 촬영되어 '자전거의 측면이 카메라를 향해야 한다'는 지시와 '화면 우측 진행 구도'를 따르지 않았습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "카메라가 자전거 측면 및 약간 앞쪽에서 3/4 측면 얼굴을 포착하며 화면 우측으로 진행하는 명시된 앵글과 구도를 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "정면 앵글로 촬영되어 '자전거의 측면이 카메라를 향해야 한다'는 지시와 '화면 우측 진행 구도'를 따르지 않았습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 14,
     "B": 9
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 14,
   "B": 9
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "피사체가 프레임 우측을 향해 이동하는 측면 구도와 해안도로의 배치를 지시문대로 정확히 구현했습니다."
   },
   {
    "label": "B",
    "score": 5,
    "verdict_ko": "자전거가 프레임 우측이 아닌 정면을 향해 주행하고 있어 요구된 프레임 레이아웃과 카메라 앵글을 어겼습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_해안_도로_8394b3.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "자전거 바구니 앞면에 금지된 텍스트 형태의 자국이 있습니다.",
     "fix_en": "Remove the text-like marks from the front of the bicycle basket, leaving only the plain wire mesh."
    },
    {
     "issue_ko": "페달을 밟고 있어야 한다는 지시와 달리, 굽힌 오른쪽 발 아래에 페달이 없어 허공에 떠 있습니다.",
     "fix_en": "Place a bicycle pedal directly under her right foot to show her actively pressing down on it."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the text-like marks from the front of the bicycle basket, leaving only the plain wire mesh.\n- Place a bicycle pedal directly under her right foot to show her actively pressing down on it.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "굽힌 무릎과 페달을 밟는 발이 프레임 안에 명확히 나타나며, 모든 프롬프트와 레퍼런스 조건을 충실히 반영했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "자전거 바구니에 불필요한 라벨이 생성되었고, 페달을 밟는 발이 프레임에서 잘려 지시된 디테일을 완벽히 보여주지 못합니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "굽힌 무릎과 페달을 밟는 발이 프레임 안에 명확히 나타나며, 모든 프롬프트와 레퍼런스 조건을 충실히 반영했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "자전거 바구니에 불필요한 라벨이 생성되었고, 페달을 밟는 발이 프레임에서 잘려 지시된 디테일을 완벽히 보여주지 못합니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 배경과 스케치의 인물 포즈를 정확히 반영하였으며, 불필요한 시각적 오류나 텍스트 유출 없이 지침을 충실히 이행했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "자전거 바구니 측면에 불필요한 문양(텍스트/그래픽 오버레이)이 유출되어 하드 위반에 해당합니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 배경과 스케치의 인물 포즈를 정확히 반영하였으며, 불필요한 시각적 오류나 텍스트 유출 없이 지침을 충실히 이행했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "자전거 바구니 측면에 불필요한 문양(텍스트/그래픽 오버레이)이 유출되어 하드 위반에 해당합니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 9,
     "B": 14
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S6sh1__bgfirst_bg.png",
   "bg_asset_id": "8e08cb34-e8ec-429f-af33-4ae4dcac5c09",
   "bg_record_key": "S6sh1::bgfirst_bg",
   "chain_winner": true,
   "authority": "groupbg",
   "group_key": "해안 도로",
   "groupbg_asset_id": "519694bd-31b4-4c58-9c2d-45ff2fb4767c"
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "groupbg::보호원 철창 구역": {
  "input_fingerprint": "f53ebbed47627017",
  "meta": {
   "model": "gpt-image-2",
   "size": "1536x864",
   "pack": "9.202607211540",
   "contract": "bgfirst_full_v1",
   "group_sig": {
    "key": "보호원 철창 구역",
    "tags": [
     "S7sh2",
     "S7sh5"
    ]
   }
  },
  "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\n\nTHE LOCATION — South Korea, 2026; all people are Korean unless stated: An indoor kennel area of a stray-dog shelter, directly in front of the Rottweiler’s enclosure.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.",
  "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\n\nTHE LOCATION — South Korea, 2026; all people are Korean unless stated: An indoor kennel area of a stray-dog shelter, directly in front of the Rottweiler’s enclosure.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_보호원_철창_구역_2d438c.png",
  "asset_id": "cbf215e1-b6e2-464c-8366-3e0dfffbd74c",
  "input_asset_ids": [
   "dac13ac6-6bfd-4601-89bf-88a1c88718e1"
  ],
  "origin_tag": "S7sh2",
  "place_text": "An indoor kennel area of a stray-dog shelter, directly in front of the Rottweiler’s enclosure."
 },
 "S7sh2::bgfirst_bg": {
  "input_fingerprint": "aa78c4be8249a8fc",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 미간에 힘을 주고 로트와일러의 눈을 매섭게 응시하는 수리영의 정면.\n\nLOCATION (lock): An indoor kennel area of a stray-dog shelter, directly in front of the Rottweiler’s enclosure.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 수리영’s eye level, the camera completes the prescribed push-in and holds a tight frontal close-up, with her face occupying most of the center frame and the 로트와일러 remaining off screen on the lens axis. Her tightened brow and unwavering eyes carry the frame, while shallow focus suppresses the shelter context without changing the established eyeline.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural ambient light appropriate to the daytime setting is kept subdued and desaturated, with firm contrast around her brow and eyes.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 미간에 힘을 주고 로트와일러의 눈을 매섭게 응시하는 수리영의 정면.\n\nLOCATION (lock): An indoor kennel area of a stray-dog shelter, directly in front of the Rottweiler’s enclosure.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 수리영’s eye level, the camera completes the prescribed push-in and holds a tight frontal close-up, with her face occupying most of the center frame and the 로트와일러 remaining off screen on the lens axis. Her tightened brow and unwavering eyes carry the frame, while shallow focus suppresses the shelter context without changing the established eyeline.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural ambient light appropriate to the daytime setting is kept subdued and desaturated, with firm contrast around her brow and eyes.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S7sh2__bgfirst_bg.png",
  "asset_id": "ab344427-3bba-4a25-a6d9-1de6da2fb4ec",
  "input_asset_ids": [
   "dac13ac6-6bfd-4601-89bf-88a1c88718e1",
   "cbf215e1-b6e2-464c-8366-3e0dfffbd74c"
  ]
 },
 "S7sh2": {
  "input_fingerprint": "75c7d96a60c50600",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 미간에 힘을 주고 로트와일러의 눈을 매섭게 응시하는 수리영의 정면.\n\nLOCATION (lock): An indoor kennel area of a stray-dog shelter, directly in front of the Rottweiler’s enclosure. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 수리영’s eye level, the camera completes the prescribed push-in and holds a tight frontal close-up, with her face occupying most of the center frame and the 로트와일러 remaining off screen on the lens axis. Her tightened brow and unwavering eyes carry the frame, while shallow focus suppresses the shelter context without changing the established eyeline.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural ambient light appropriate to the daytime setting is kept subdued and desaturated, with firm contrast around her brow and eyes.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 미간에 힘을 주고 로트와일러의 눈을 매섭게 응시하는 수리영의 정면.\n\nLOCATION (lock): An indoor kennel area of a stray-dog shelter, directly in front of the Rottweiler’s enclosure. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 수리영’s eye level, the camera completes the prescribed push-in and holds a tight frontal close-up, with her face occupying most of the center frame and the 로트와일러 remaining off screen on the lens axis. Her tightened brow and unwavering eyes carry the frame, while shallow focus suppresses the shelter context without changing the established eyeline.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural ambient light appropriate to the daytime setting is kept subdued and desaturated, with firm contrast around her brow and eyes.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 미간에 힘을 주고 로트와일러의 눈을 매섭게 응시하는 수리영의 정면.\n\nLOCATION (lock): An indoor kennel area of a stray-dog shelter, directly in front of the Rottweiler’s enclosure. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 수리영’s eye level, the camera completes the prescribed push-in and holds a tight frontal close-up, with her face occupying most of the center frame and the 로트와일러 remaining off screen on the lens axis. Her tightened brow and unwavering eyes carry the frame, while shallow focus suppresses the shelter context without changing the established eyeline.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural ambient light appropriate to the daytime setting is kept subdued and desaturated, with firm contrast around her brow and eyes.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S7sh2__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S7sh2.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_보호원_철창_구역_2d438c.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지시된 프레이밍과 렌즈 축(정면)을 향한 정확한 시선을 보여주며, 배경과 표정 묘사가 훌륭함."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "표정과 프레이밍은 좋으나, 시선이 렌즈 축(정면)을 벗어나 측면을 향하고 있어 프롬프트 지시와 어긋남."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지시된 프레이밍과 렌즈 축(정면)을 향한 정확한 시선을 보여주며, 배경과 표정 묘사가 훌륭함."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "표정과 프레이밍은 좋으나, 시선이 렌즈 축(정면)을 벗어나 측면을 향하고 있어 프롬프트 지시와 어긋남."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "렌즈 축을 정면으로 응시하는 강렬한 시선과 표정은 지시를 잘 따랐으나, 얕은 심도 요청과 달리 배경이 너무 선명하게 묘사되었습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "요청된 얕은 심도와 타이트한 프레이밍은 훌륭히 구현했으나, 카메라 렌즈 축을 향해야 할 시선이 측면으로 빗나가 핵심 연출을 놓쳤습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "렌즈 축을 정면으로 응시하는 강렬한 시선과 표정은 지시를 잘 따랐으나, 얕은 심도 요청과 달리 배경이 너무 선명하게 묘사되었습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "요청된 얕은 심도와 타이트한 프레이밍은 훌륭히 구현했으나, 카메라 렌즈 축을 향해야 할 시선이 측면으로 빗나가 핵심 연출을 놓쳤습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 15,
     "B": 11
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 15,
   "B": 11
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 8,
    "verdict_ko": "지시된 프레이밍과 렌즈 축(정면)을 향한 정확한 시선을 보여주며, 배경과 표정 묘사가 훌륭함."
   },
   {
    "label": "B",
    "score": 6,
    "verdict_ko": "표정과 프레이밍은 좋으나, 시선이 렌즈 축(정면)을 벗어나 측면을 향하고 있어 프롬프트 지시와 어긋남."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_보호원_철창_구역_2d438c.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "레퍼런스 이미지에서 인물이 착용하고 있는 베이지색 재킷이 누락되었고 흰색 티셔츠만 보입니다.",
     "fix_en": "Add the beige jacket from the character reference over the character's white t-shirt."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add the beige jacket from the character reference over the character's white t-shirt.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 클로즈업 프레이밍과 스케치의 인물 크기를 정확히 준수하여 얼굴에 집중도를 높였습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "지정된 클로즈업 샷보다 프레임을 넓혀 의상을 더 많이 노출하는 구도적 오류가 있습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 클로즈업 프레이밍과 스케치의 인물 크기를 정확히 준수하여 얼굴에 집중도를 높였습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "지정된 클로즈업 샷보다 프레임을 넓혀 의상을 더 많이 노출하는 구도적 오류가 있습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지정된 타이트 클로즈업 앵글과 스케치의 인물 크기를 완벽하게 구현했으며, 미간에 힘을 준 표정이 잘 표현되었습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "표정과 배경은 좋으나, 프레이밍이 지시사항 및 스케치보다 넓게 잡혀 흉부까지 노출된 점이 아쉽습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 타이트 클로즈업 앵글과 스케치의 인물 크기를 완벽하게 구현했으며, 미간에 힘을 준 표정이 잘 표현되었습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "표정과 배경은 좋으나, 프레이밍이 지시사항 및 스케치보다 넓게 잡혀 흉부까지 노출된 점이 아쉽습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 15,
     "B": 10
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S7sh2__bgfirst_bg.png",
   "bg_asset_id": "ab344427-3bba-4a25-a6d9-1de6da2fb4ec",
   "bg_record_key": "S7sh2::bgfirst_bg",
   "chain_winner": true,
   "authority": "groupbg",
   "group_key": "보호원 철창 구역",
   "groupbg_asset_id": "cbf215e1-b6e2-464c-8366-3e0dfffbd74c"
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S7sh5::variants": {
  "author_fp": "caa420a62d93dd9c",
  "author": {
   "variants": [
    {
     "approach_ko": "정면에 가까운 깊은 초점과 철창 격자의 반복으로 수리영의 침묵을 냉정하게 가두는 접근.",
     "prompt_en": "Use a restrained, near-frontal lens perspective with moderately deep focus, keeping Suri-young’s face, phone hand, and the repeated cage grid clearly legible in layered depth. Let the rigid verticals and horizontals surround her slightly off-center figure, while soft daytime ambient light falls evenly and without glamour, emphasizing her pressed lips, fixed distant gaze, and held stillness."
    },
    {
     "approach_ko": "얕은 초점과 압축된 시선으로 입술과 전화기 주변의 미세한 긴장을 고립시키는 접근.",
     "prompt_en": "Use a gently compressed lens feel and shallow focus to isolate Suri-young against the softened cage frontage, holding crisp attention on the phone beside her face, her tightly pressed lips, and the minute tension in her expression. Keep the medium frame quiet and spacious around her gaze, with subdued side-biased ambient daylight shaping her face softly as the confined dogs and repeating bars recede into muted depth."
    }
   ]
  },
  "reused": false
 },
 "S7sh5": {
  "input_fingerprint": "fa997dcedef7006f",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 철창 앞에서 스마트폰을 귀에 댄 채 입술을 꾹 다문 수리영.\n\nLOCATION (lock): The indoor kennel aisle of a stray-dog shelter, lined with enclosures holding dogs. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height beside the cage line, hold a static medium frame with 수리영 slightly right of center and the cage front directly behind her. Her phone hand rises along the side of her face while the composition reserves enough space around her pressed lips and distant gaze to make the unanswered silence palpable.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 철창과 개장 줄 (Dogs remain confined behind the cage fronts) — The cage fronts face the camera behind 수리영, with confined dogs beyond them; used as The repeated cage frontage forms a rigid background grid behind 수리영; 스마트폰 (Held to her ear during an unanswered call) — The screen face is turned toward 수리영 and largely away from the camera, so no screen content is legible; used as Held beside 수리영’s face as the narrative focal prop.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural ambient light appropriate to the daytime shelter is restrained and desaturated, with subdued luminance around her face.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 철창 앞에서 스마트폰을 귀에 댄 채 입술을 꾹 다문 수리영.\n\nLOCATION (lock): The indoor kennel aisle of a stray-dog shelter, lined with enclosures holding dogs. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height beside the cage line, hold a static medium frame with 수리영 slightly right of center and the cage front directly behind her. Her phone hand rises along the side of her face while the composition reserves enough space around her pressed lips and distant gaze to make the unanswered silence palpable.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 철창과 개장 줄 (Dogs remain confined behind the cage fronts) — The cage fronts face the camera behind 수리영, with confined dogs beyond them; used as The repeated cage frontage forms a rigid background grid behind 수리영; 스마트폰 (Held to her ear during an unanswered call) — The screen face is turned toward 수리영 and largely away from the camera, so no screen content is legible; used as Held beside 수리영’s face as the narrative focal prop.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural ambient light appropriate to the daytime shelter is restrained and desaturated, with subdued luminance around her face.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained, near-frontal lens perspective with moderately deep focus, keeping Suri-young’s face, phone hand, and the repeated cage grid clearly legible in layered depth. Let the rigid verticals and horizontals surround her slightly off-center figure, while soft daytime ambient light falls evenly and without glamour, emphasizing her pressed lips, fixed distant gaze, and held stillness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 철창 앞에서 스마트폰을 귀에 댄 채 입술을 꾹 다문 수리영.\n\nLOCATION (lock): The indoor kennel aisle of a stray-dog shelter, lined with enclosures holding dogs. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height beside the cage line, hold a static medium frame with 수리영 slightly right of center and the cage front directly behind her. Her phone hand rises along the side of her face while the composition reserves enough space around her pressed lips and distant gaze to make the unanswered silence palpable.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 철창과 개장 줄 (Dogs remain confined behind the cage fronts) — The cage fronts face the camera behind 수리영, with confined dogs beyond them; used as The repeated cage frontage forms a rigid background grid behind 수리영; 스마트폰 (Held to her ear during an unanswered call) — The screen face is turned toward 수리영 and largely away from the camera, so no screen content is legible; used as Held beside 수리영’s face as the narrative focal prop.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural ambient light appropriate to the daytime shelter is restrained and desaturated, with subdued luminance around her face.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a gently compressed lens feel and shallow focus to isolate Suri-young against the softened cage frontage, holding crisp attention on the phone beside her face, her tightly pressed lips, and the minute tension in her expression. Keep the medium frame quiet and spacious around her gaze, with subdued side-biased ambient daylight shaping her face softly as the confined dogs and repeating bars recede into muted depth.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 9,
   "B": 5
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 9,
    "verdict_ko": "지정된 미디엄 샷 앵글, 우측으로 치우친 인물 배치, 입술을 다문 표정을 정확히 구현했으며 인체 비례와 손의 형태도 자연스럽습니다."
   },
   {
    "label": "B",
    "score": 5,
    "verdict_ko": "지시된 구도와 인물의 위치는 잘 따랐으나, 스마트폰을 쥔 오른손에 손가락이 기형적으로 묘사되는 치명적인 해부학적 오류가 있습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S7sh2_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "이전 숏 스틸컷에서 인물은 흰색 티셔츠만 입고 있는 것으로 확정되었으나, 현재 이미지에서는 베이지색 재킷을 입고 있어 의상 연속성에 어긋납니다.",
     "fix_en": "Remove the beige jacket, leaving the character wearing only the plain white t-shirt to match the previous shot still."
    },
    {
     "issue_ko": "스마트폰의 후면 카메라 모듈이 손 아래쪽을 향하고 있어 전화기가 위아래로 뒤집힌 상태입니다.",
     "fix_en": "Rotate the smartphone 180 degrees so the rear camera module is at the top of the device."
    },
    {
     "issue_ko": "지시문은 '입술을 꾹 다문' 표정을 요구했으나, 인물의 입술이 이완되어 살짝 벌어져 있습니다.",
     "fix_en": "Modify the character's mouth so her lips are pressed tightly together."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the beige jacket, leaving the character wearing only the plain white t-shirt to match the previous shot still.\n- Rotate the smartphone 180 degrees so the rear camera module is at the top of the device.\n- Modify the character's mouth so her lips are pressed tightly together.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "이전 컷(Reference 1)에서 설정된 의상(흰색 티셔츠)을 정확히 유지했으며, 프레이밍, 배경, 인물의 행동과 표정 모두 프롬프트의 요구사항을 훌륭하게 충족합니다."
     },
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "프레이밍과 배경 및 연기는 프롬프트와 잘 일치하지만, 이전 컷에서 벗고 있던 재킷을 입고 있어 의상 연속성에서 감점되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "이전 컷(Reference 1)에서 설정된 의상(흰색 티셔츠)을 정확히 유지했으며, 프레이밍, 배경, 인물의 행동과 표정 모두 프롬프트의 요구사항을 훌륭하게 충족합니다."
     },
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "프레이밍과 배경 및 연기는 프롬프트와 잘 일치하지만, 이전 컷에서 벗고 있던 재킷을 입고 있어 의상 연속성에서 감점되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "이전 샷의 흰색 티셔츠 의상 연속성을 완벽하게 유지하며, 명시된 구도와 동작 지침을 정확하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "구도와 동작은 훌륭하지만, 이전 샷 캡처에 등장하지 않았던 재킷을 추가하여 의상 연속성 지침을 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "이전 샷의 흰색 티셔츠 의상 연속성을 완벽하게 유지하며, 명시된 구도와 동작 지침을 정확하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "구도와 동작은 훌륭하지만, 이전 샷 캡처에 등장하지 않았던 재킷을 추가하여 의상 연속성 지침을 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 16,
     "B": 20
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S7sh2"
  }
 },
 "S8sh1::bgfirst_bg": {
  "input_fingerprint": "040cc605f31fc6f5",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 화려한 네온사인과 포장마차가 있는 거리에서 앞으로 나아가는 자전거를 탄 채 바람에 머리카락이 뒤로 흩날리며 손잡이를 단단히 쥔 수리영의 상체.\n\nLOCATION (lock): A busy harbor-side commercial street at dusk, running between neon signs and street-food stalls.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height ahead of 수리영, track backward at her speed and tighten to her upper body while retaining the handlebar and a narrow band of the moving street below. She stays near center in a slight three-quarter approach, hair streaming behind her as passing neon signs and street stalls slide laterally through the background.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward forward street route.\n- KEY BACKGROUND ELEMENTS: 네온사인 (Illuminated) — Their sign faces turn toward the street and pass behind 수리영 with illuminated commercial graphics visible; used as Passing illuminated graphics establish the commercial street without overtaking the rider; 길거리 포장마차 (Present along the busy street) — Their street-facing and side elevations pass behind the bicycle as the camera retreats; used as Create intermittent layers of street activity in the rear and side background; 자전거 (In motion) — The handlebar front is angled toward the camera while the bicycle travels along the camera’s retreating axis; used as Provides moving scale beneath 수리영’s upper body without occupying more than a lower portion of frame.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk ambient light mixes with the explicitly present neon signage in a restrained, desaturated palette with intermittent warm accents.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 화려한 네온사인과 포장마차가 있는 거리에서 앞으로 나아가는 자전거를 탄 채 바람에 머리카락이 뒤로 흩날리며 손잡이를 단단히 쥔 수리영의 상체.\n\nLOCATION (lock): A busy harbor-side commercial street at dusk, running between neon signs and street-food stalls.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height ahead of 수리영, track backward at her speed and tighten to her upper body while retaining the handlebar and a narrow band of the moving street below. She stays near center in a slight three-quarter approach, hair streaming behind her as passing neon signs and street stalls slide laterally through the background.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward forward street route.\n- KEY BACKGROUND ELEMENTS: 네온사인 (Illuminated) — Their sign faces turn toward the street and pass behind 수리영 with illuminated commercial graphics visible; used as Passing illuminated graphics establish the commercial street without overtaking the rider; 길거리 포장마차 (Present along the busy street) — Their street-facing and side elevations pass behind the bicycle as the camera retreats; used as Create intermittent layers of street activity in the rear and side background; 자전거 (In motion) — The handlebar front is angled toward the camera while the bicycle travels along the camera’s retreating axis; used as Provides moving scale beneath 수리영’s upper body without occupying more than a lower portion of frame.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk ambient light mixes with the explicitly present neon signage in a restrained, desaturated palette with intermittent warm accents.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S8sh1__bgfirst_bg.png",
  "asset_id": "71098bd3-d9dd-4dc0-92fe-cc4b551cdce1",
  "input_asset_ids": [
   "c5b9a41e-951f-4c05-a0d7-7bd4a8ae99a8",
   "5bc33568-aeb5-434c-a463-d1aadbe87922"
  ]
 },
 "S8sh1": {
  "input_fingerprint": "6e88b29f7dbb7cc8",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 화려한 네온사인과 포장마차가 있는 거리에서 앞으로 나아가는 자전거를 탄 채 바람에 머리카락이 뒤로 흩날리며 손잡이를 단단히 쥔 수리영의 상체.\n\nLOCATION (lock): A busy harbor-side commercial street at dusk, running between neon signs and street-food stalls. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height ahead of 수리영, track backward at her speed and tighten to her upper body while retaining the handlebar and a narrow band of the moving street below. She stays near center in a slight three-quarter approach, hair streaming behind her as passing neon signs and street stalls slide laterally through the background.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward forward street route.\n- KEY BACKGROUND ELEMENTS: 네온사인 (Illuminated) — Their sign faces turn toward the street and pass behind 수리영 with illuminated commercial graphics visible; used as Passing illuminated graphics establish the commercial street without overtaking the rider; 길거리 포장마차 (Present along the busy street) — Their street-facing and side elevations pass behind the bicycle as the camera retreats; used as Create intermittent layers of street activity in the rear and side background; 자전거 (In motion) — The handlebar front is angled toward the camera while the bicycle travels along the camera’s retreating axis; used as Provides moving scale beneath 수리영’s upper body without occupying more than a lower portion of frame.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk ambient light mixes with the explicitly present neon signage in a restrained, desaturated palette with intermittent warm accents.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 화려한 네온사인과 포장마차가 있는 거리에서 앞으로 나아가는 자전거를 탄 채 바람에 머리카락이 뒤로 흩날리며 손잡이를 단단히 쥔 수리영의 상체.\n\nLOCATION (lock): A busy harbor-side commercial street at dusk, running between neon signs and street-food stalls. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height ahead of 수리영, track backward at her speed and tighten to her upper body while retaining the handlebar and a narrow band of the moving street below. She stays near center in a slight three-quarter approach, hair streaming behind her as passing neon signs and street stalls slide laterally through the background.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward forward street route.\n- KEY BACKGROUND ELEMENTS: 네온사인 (Illuminated) — Their sign faces turn toward the street and pass behind 수리영 with illuminated commercial graphics visible; used as Passing illuminated graphics establish the commercial street without overtaking the rider; 길거리 포장마차 (Present along the busy street) — Their street-facing and side elevations pass behind the bicycle as the camera retreats; used as Create intermittent layers of street activity in the rear and side background; 자전거 (In motion) — The handlebar front is angled toward the camera while the bicycle travels along the camera’s retreating axis; used as Provides moving scale beneath 수리영’s upper body without occupying more than a lower portion of frame.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk ambient light mixes with the explicitly present neon signage in a restrained, desaturated palette with intermittent warm accents.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 화려한 네온사인과 포장마차가 있는 거리에서 앞으로 나아가는 자전거를 탄 채 바람에 머리카락이 뒤로 흩날리며 손잡이를 단단히 쥔 수리영의 상체.\n\nLOCATION (lock): A busy harbor-side commercial street at dusk, running between neon signs and street-food stalls. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height ahead of 수리영, track backward at her speed and tighten to her upper body while retaining the handlebar and a narrow band of the moving street below. She stays near center in a slight three-quarter approach, hair streaming behind her as passing neon signs and street stalls slide laterally through the background.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward forward street route.\n- KEY BACKGROUND ELEMENTS: 네온사인 (Illuminated) — Their sign faces turn toward the street and pass behind 수리영 with illuminated commercial graphics visible; used as Passing illuminated graphics establish the commercial street without overtaking the rider; 길거리 포장마차 (Present along the busy street) — Their street-facing and side elevations pass behind the bicycle as the camera retreats; used as Create intermittent layers of street activity in the rear and side background; 자전거 (In motion) — The handlebar front is angled toward the camera while the bicycle travels along the camera’s retreating axis; used as Provides moving scale beneath 수리영’s upper body without occupying more than a lower portion of frame.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk ambient light mixes with the explicitly present neon signage in a restrained, desaturated palette with intermittent warm accents.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S8sh1__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S8sh1.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L09B01.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "지시문에 명시되지 않은 배경 인물들을 다수 임의로 추가하여 심각한 규칙 위반(발명된 인물)이 발생했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "불필요한 인물을 추가하지 않고 지정된 구도와 장소를 잘 구현했으나, 캐릭터 설정인 '묶은 머리' 대신 푼 머리로 묘사된 점이 감점 요인입니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "지시문에 명시되지 않은 배경 인물들을 다수 임의로 추가하여 심각한 규칙 위반(발명된 인물)이 발생했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "불필요한 인물을 추가하지 않고 지정된 구도와 장소를 잘 구현했으나, 캐릭터 설정인 '묶은 머리' 대신 푼 머리로 묘사된 점이 감점 요인입니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 프롬프트에 따라 주인공 외의 다른 인물을 추가하지 않았고 카메라 구도와 배경의 분위기를 훌륭하게 구현했으나, 기준 이미지의 묶은 머리 대신 푼 머리가 적용된 점은 아쉽습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "주인공의 머리 스타일은 기준과 더 일치하지만, 프롬프트에 명시되지 않은 여러 행인들을 배경에 임의로 추가하여 엄격한 금지 사항(발명된 인물)을 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 프롬프트에 따라 주인공 외의 다른 인물을 추가하지 않았고 카메라 구도와 배경의 분위기를 훌륭하게 구현했으나, 기준 이미지의 묶은 머리 대신 푼 머리가 적용된 점은 아쉽습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "주인공의 머리 스타일은 기준과 더 일치하지만, 프롬프트에 명시되지 않은 여러 행인들을 배경에 임의로 추가하여 엄격한 금지 사항(발명된 인물)을 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 6,
     "B": 13
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 6,
   "B": 13
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 3,
    "verdict_ko": "지시문에 명시되지 않은 배경 인물들을 다수 임의로 추가하여 심각한 규칙 위반(발명된 인물)이 발생했습니다."
   },
   {
    "label": "B",
    "score": 6,
    "verdict_ko": "불필요한 인물을 추가하지 않고 지정된 구도와 장소를 잘 구현했으나, 캐릭터 설정인 '묶은 머리' 대신 푼 머리로 묘사된 점이 감점 요인입니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L09B01.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "지시된 '약간의 3/4 측면 구도(slight three-quarter approach)'와 달리 인물과 자전거가 완벽한 정면을 향하고 있습니다.",
     "fix_en": "Angle Suri-young and the bicycle into a slight three-quarter approach instead of a straight frontal view."
    },
    {
     "issue_ko": "인물 설정에 명시된 '묶은 검은 머리'와 달리 머리카락이 길게 풀려 있습니다.",
     "fix_en": "Change the character's hair to be tied back, matching her reference profile."
    },
    {
     "issue_ko": "자전거 브레이크 레버가 앞이 아닌 핸들바 중앙(안쪽)을 향해 잘못 연결되어 있습니다.",
     "fix_en": "Orient the bicycle brake levers properly so they face forward and parallel to the grips."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Angle Suri-young and the bicycle into a slight three-quarter approach instead of a straight frontal view.\n- Change the character's hair to be tied back, matching her reference profile.\n- Orient the bicycle brake levers properly so they face forward and parallel to the grips.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 약간 측면(3/4) 카메라 구도와 캐릭터의 묶은 머리 설정을 충실히 구현하여 프롬프트 부합도가 높습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "지정된 측면 접근 카메라 앵글 대신 완전한 정면 구도를 취했으며, 캐릭터의 기본 설정인 묶은 머리가 풀려 있어 감점되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 약간 측면(3/4) 카메라 구도와 캐릭터의 묶은 머리 설정을 충실히 구현하여 프롬프트 부합도가 높습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "지정된 측면 접근 카메라 앵글 대신 완전한 정면 구도를 취했으며, 캐릭터의 기본 설정인 묶은 머리가 풀려 있어 감점되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 4/3 측면 구도를 잘 살렸으며 캐릭터의 묶은 머리 설정을 유지했으나, 핸들을 쥔 손의 위치가 다소 어색합니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "정면 구도로 촬영되어 카메라 앵글 지침을 어겼으며, 캐릭터의 묶은 머리 설정이 무시되고 핸들의 형태와 손 위치가 부자연스럽습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 4/3 측면 구도를 잘 살렸으며 캐릭터의 묶은 머리 설정을 유지했으나, 핸들을 쥔 손의 위치가 다소 어색합니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "정면 구도로 촬영되어 카메라 앵글 지침을 어겼으며, 캐릭터의 묶은 머리 설정이 무시되고 핸들의 형태와 손 위치가 부자연스럽습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 9,
     "B": 14
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S8sh1__bgfirst_bg.png",
   "bg_asset_id": "71098bd3-d9dd-4dc0-92fe-cc4b551cdce1",
   "bg_record_key": "S8sh1::bgfirst_bg",
   "chain_winner": false,
   "authority": "plate"
  },
  "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L09B01",
    "B": "L09B02",
    "C": "L09B03"
   },
   "assigned": "L09B01",
   "choice": "Candidate A",
   "confident": true,
   "reason_ko": "지문에서 '포장마차가 있는 거리'를 명시하고 있으므로, 길거리에 천막과 노점들이 늘어서 있는 A가 가장 적합한 공간입니다.",
   "kept": "L09B01"
  }
 },
 "S8sh2::bgfirst_bg": {
  "input_fingerprint": "cbf5208d0852a013",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 대형 마트 내부 진열대 사이에서 한쪽으로 고개를 튼 채 멈춰 서서 시선을 고정하고 있는 수리영.\n\nLOCATION (lock): The sales floor inside a large supermarket, in an aisle between merchandise displays.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level across the aisle intersection, finish the pan in a wide, stable composition as 수리영 halts near center-left between the shelving lines. Her body remains caught in the forward search while only her head turns toward screen right, allowing the aisle geometry to carry her eyeline toward the unseen employees.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, looks toward employees off frame right.\n- KEY BACKGROUND ELEMENTS: 마트 진열대 (Arranged as retail aisles) — The shelving sides recede away from the aisle intersection, leaving a clear cross-aisle sightline toward screen right; used as Parallel shelving lines create converging search corridors on both sides of 수리영; 진열대 사이 교차 통로 (Clear enough for 수리영 to stop and look across it); used as Provides the open lateral path where the pan stops and her gaze exits frame.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient interior light appropriate to the mart is treated with subdued luminance and a restrained, desaturated color response.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 대형 마트 내부 진열대 사이에서 한쪽으로 고개를 튼 채 멈춰 서서 시선을 고정하고 있는 수리영.\n\nLOCATION (lock): The sales floor inside a large supermarket, in an aisle between merchandise displays.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level across the aisle intersection, finish the pan in a wide, stable composition as 수리영 halts near center-left between the shelving lines. Her body remains caught in the forward search while only her head turns toward screen right, allowing the aisle geometry to carry her eyeline toward the unseen employees.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, looks toward employees off frame right.\n- KEY BACKGROUND ELEMENTS: 마트 진열대 (Arranged as retail aisles) — The shelving sides recede away from the aisle intersection, leaving a clear cross-aisle sightline toward screen right; used as Parallel shelving lines create converging search corridors on both sides of 수리영; 진열대 사이 교차 통로 (Clear enough for 수리영 to stop and look across it); used as Provides the open lateral path where the pan stops and her gaze exits frame.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient interior light appropriate to the mart is treated with subdued luminance and a restrained, desaturated color response.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S8sh2__bgfirst_bg.png",
  "asset_id": "fb6d8224-d644-4bd5-adf9-7085fe6b7b86",
  "input_asset_ids": [
   "6d93d082-99ce-4190-8b65-595324fa347e",
   "c3dd43f9-eb19-4c44-8473-764af8bbf211"
  ]
 },
 "S8sh2": {
  "input_fingerprint": "a6203a7926e74a6a",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 대형 마트 내부 진열대 사이에서 한쪽으로 고개를 튼 채 멈춰 서서 시선을 고정하고 있는 수리영.\n\nLOCATION (lock): The sales floor inside a large supermarket, in an aisle between merchandise displays. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level across the aisle intersection, finish the pan in a wide, stable composition as 수리영 halts near center-left between the shelving lines. Her body remains caught in the forward search while only her head turns toward screen right, allowing the aisle geometry to carry her eyeline toward the unseen employees.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, looks toward employees off frame right.\n- KEY BACKGROUND ELEMENTS: 마트 진열대 (Arranged as retail aisles) — The shelving sides recede away from the aisle intersection, leaving a clear cross-aisle sightline toward screen right; used as Parallel shelving lines create converging search corridors on both sides of 수리영; 진열대 사이 교차 통로 (Clear enough for 수리영 to stop and look across it); used as Provides the open lateral path where the pan stops and her gaze exits frame.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient interior light appropriate to the mart is treated with subdued luminance and a restrained, desaturated color response.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 대형 마트 내부 진열대 사이에서 한쪽으로 고개를 튼 채 멈춰 서서 시선을 고정하고 있는 수리영.\n\nLOCATION (lock): The sales floor inside a large supermarket, in an aisle between merchandise displays. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level across the aisle intersection, finish the pan in a wide, stable composition as 수리영 halts near center-left between the shelving lines. Her body remains caught in the forward search while only her head turns toward screen right, allowing the aisle geometry to carry her eyeline toward the unseen employees.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, looks toward employees off frame right.\n- KEY BACKGROUND ELEMENTS: 마트 진열대 (Arranged as retail aisles) — The shelving sides recede away from the aisle intersection, leaving a clear cross-aisle sightline toward screen right; used as Parallel shelving lines create converging search corridors on both sides of 수리영; 진열대 사이 교차 통로 (Clear enough for 수리영 to stop and look across it); used as Provides the open lateral path where the pan stops and her gaze exits frame.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient interior light appropriate to the mart is treated with subdued luminance and a restrained, desaturated color response.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 대형 마트 내부 진열대 사이에서 한쪽으로 고개를 튼 채 멈춰 서서 시선을 고정하고 있는 수리영.\n\nLOCATION (lock): The sales floor inside a large supermarket, in an aisle between merchandise displays. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level across the aisle intersection, finish the pan in a wide, stable composition as 수리영 halts near center-left between the shelving lines. Her body remains caught in the forward search while only her head turns toward screen right, allowing the aisle geometry to carry her eyeline toward the unseen employees.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, looks toward employees off frame right.\n- KEY BACKGROUND ELEMENTS: 마트 진열대 (Arranged as retail aisles) — The shelving sides recede away from the aisle intersection, leaving a clear cross-aisle sightline toward screen right; used as Parallel shelving lines create converging search corridors on both sides of 수리영; 진열대 사이 교차 통로 (Clear enough for 수리영 to stop and look across it); used as Provides the open lateral path where the pan stops and her gaze exits frame.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient interior light appropriate to the mart is treated with subdued luminance and a restrained, desaturated color response.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S8sh2__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S8sh2.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L09B02.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "피사체가 진열대 사이에 위치하도록 카메라 구도를 새롭게 설정하였으며, 몸은 앞을 향한 채 고개만 오른쪽으로 돌린 포즈와 샷 크기를 충실히 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "레퍼런스 이미지의 구도를 그대로 복사하지 말라는 지시를 어겼으며, 피사체를 진열대 사이가 아닌 앞쪽 공터에 배치하여 프롬프트의 공간 연출을 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "피사체가 진열대 사이에 위치하도록 카메라 구도를 새롭게 설정하였으며, 몸은 앞을 향한 채 고개만 오른쪽으로 돌린 포즈와 샷 크기를 충실히 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "레퍼런스 이미지의 구도를 그대로 복사하지 말라는 지시를 어겼으며, 피사체를 진열대 사이가 아닌 앞쪽 공터에 배치하여 프롬프트의 공간 연출을 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "요구된 대로 진열대 사이의 통로 구도를 새롭게 연출하여 인물을 배치했으며, 양옆으로 진열대가 뻗어 있는 공간감을 잘 살렸습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "장소 레퍼런스 이미지의 카메라 구도와 프레이밍을 절대 복사하지 말라는 지침(Never copy its camera framing)을 명백히 위반하였습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "요구된 대로 진열대 사이의 통로 구도를 새롭게 연출하여 인물을 배치했으며, 양옆으로 진열대가 뻗어 있는 공간감을 잘 살렸습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "장소 레퍼런스 이미지의 카메라 구도와 프레이밍을 절대 복사하지 말라는 지침(Never copy its camera framing)을 명백히 위반하였습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 8,
     "B": 16
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 8,
   "B": 16
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 8,
    "verdict_ko": "피사체가 진열대 사이에 위치하도록 카메라 구도를 새롭게 설정하였으며, 몸은 앞을 향한 채 고개만 오른쪽으로 돌린 포즈와 샷 크기를 충실히 구현했습니다."
   },
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "레퍼런스 이미지의 구도를 그대로 복사하지 말라는 지시를 어겼으며, 피사체를 진열대 사이가 아닌 앞쪽 공터에 배치하여 프롬프트의 공간 연출을 위반했습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L09B02.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "수리영이 화면의 중앙-좌측이 아닌 정중앙에 위치해 있습니다.",
     "fix_en": "Move 수리영 to the middle-left of the frame."
    },
    {
     "issue_ko": "수리영이 멈춰 서 있지 않고 한쪽 다리를 앞으로 내밀며 걷고 있습니다.",
     "fix_en": "Change her pose so she is standing completely still with both feet firmly planted."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Move 수리영 to the middle-left of the frame.\n- Change her pose so she is standing completely still with both feet firmly planted.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 배경과 인물 외형을 잘 유지하면서, 프롬프트의 핵심인 '몸은 앞을 향한 채 고개를 화면 오른쪽으로 돌린' 포즈를 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "배경과 인물 디자인은 훌륭하게 일치하나, 고개를 옆으로 돌려 시선을 고정하라는 결정적인 행동 지시를 무시하고 카메라 정면을 응시하고 있습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 배경과 인물 외형을 잘 유지하면서, 프롬프트의 핵심인 '몸은 앞을 향한 채 고개를 화면 오른쪽으로 돌린' 포즈를 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "배경과 인물 디자인은 훌륭하게 일치하나, 고개를 옆으로 돌려 시선을 고정하라는 결정적인 행동 지시를 무시하고 카메라 정면을 응시하고 있습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "피사체가 화면 정중앙에 서서 정면을 응시하고 있어, 화면 좌측 중앙 배치 및 오른쪽으로 고개를 돌리는 핵심 연출 지시를 이행하지 못했습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "피사체가 화면 좌측 중앙 부근에 위치한 상태에서 몸은 앞을 향하되 고개만 오른쪽으로 돌려 시선을 던지는 지시를 매우 충실하게 구현했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "피사체가 화면 정중앙에 서서 정면을 응시하고 있어, 화면 좌측 중앙 배치 및 오른쪽으로 고개를 돌리는 핵심 연출 지시를 이행하지 못했습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "피사체가 화면 좌측 중앙 부근에 위치한 상태에서 몸은 앞을 향하되 고개만 오른쪽으로 돌려 시선을 던지는 지시를 매우 충실하게 구현했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 15,
     "B": 9
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S8sh2__bgfirst_bg.png",
   "bg_asset_id": "fb6d8224-d644-4bd5-adf9-7085fe6b7b86",
   "bg_record_key": "S8sh2::bgfirst_bg",
   "chain_winner": false,
   "authority": "plate"
  },
  "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L09B01",
    "B": "L09B02",
    "C": "L09B03"
   },
   "assigned": "L09B02",
   "choice": "Candidate B",
   "confident": true,
   "reason_ko": "\"대형 마트 내부 진열대 사이에서\"라는 지문에 부합하는 마트 내부 공간입니다.",
   "kept": "L09B02"
  }
 },
 "S8sh6::bgfirst_bg": {
  "input_fingerprint": "c7d7198289dd5132",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 마트 밖 입구에 서서 한 손에 쥔 스마트폰 화면을 내려다보는 수리영.\n\nLOCATION (lock): The exterior entrance of a large supermarket, immediately outside the exit doors at dusk.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Outside the entrance at chest height, hold a static medium frame lowered just enough to include 수리영’s hand and smartphone in the lower center while her face remains in the upper half. She leans slightly over the screen, isolated against the entrance as her downward eyeline closes the scene’s search movement into stillness.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 마트 입구 (In use as she exits) — The exterior-facing side of the entrance sits behind her, aligned with her exit path; used as Marks the threshold immediately behind 수리영 after she leaves the mart; 스마트폰 (Held in one hand during the unanswered call) — The screen face tilts upward toward 수리영 and remains only obliquely visible to the camera; used as Held in the lower center as the practical focus of the unanswered contact attempt.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk ambient light is kept subdued and desaturated, preserving restrained contrast between 수리영 and the mart entrance.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 마트 밖 입구에 서서 한 손에 쥔 스마트폰 화면을 내려다보는 수리영.\n\nLOCATION (lock): The exterior entrance of a large supermarket, immediately outside the exit doors at dusk.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Outside the entrance at chest height, hold a static medium frame lowered just enough to include 수리영’s hand and smartphone in the lower center while her face remains in the upper half. She leans slightly over the screen, isolated against the entrance as her downward eyeline closes the scene’s search movement into stillness.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 마트 입구 (In use as she exits) — The exterior-facing side of the entrance sits behind her, aligned with her exit path; used as Marks the threshold immediately behind 수리영 after she leaves the mart; 스마트폰 (Held in one hand during the unanswered call) — The screen face tilts upward toward 수리영 and remains only obliquely visible to the camera; used as Held in the lower center as the practical focus of the unanswered contact attempt.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk ambient light is kept subdued and desaturated, preserving restrained contrast between 수리영 and the mart entrance.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S8sh6__bgfirst_bg.png",
  "asset_id": "578c527d-8003-4748-a142-26c0fad75499",
  "input_asset_ids": [
   "5ec6c53d-c023-4985-95fd-1ee43d262b08",
   "0a787327-477f-4354-bedd-131c7a477983"
  ]
 },
 "S8sh6": {
  "input_fingerprint": "ec24d451001afe43",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 마트 밖 입구에 서서 한 손에 쥔 스마트폰 화면을 내려다보는 수리영.\n\nLOCATION (lock): The exterior entrance of a large supermarket, immediately outside the exit doors at dusk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Outside the entrance at chest height, hold a static medium frame lowered just enough to include 수리영’s hand and smartphone in the lower center while her face remains in the upper half. She leans slightly over the screen, isolated against the entrance as her downward eyeline closes the scene’s search movement into stillness.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 마트 입구 (In use as she exits) — The exterior-facing side of the entrance sits behind her, aligned with her exit path; used as Marks the threshold immediately behind 수리영 after she leaves the mart; 스마트폰 (Held in one hand during the unanswered call) — The screen face tilts upward toward 수리영 and remains only obliquely visible to the camera; used as Held in the lower center as the practical focus of the unanswered contact attempt.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk ambient light is kept subdued and desaturated, preserving restrained contrast between 수리영 and the mart entrance.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 마트 밖 입구에 서서 한 손에 쥔 스마트폰 화면을 내려다보는 수리영.\n\nLOCATION (lock): The exterior entrance of a large supermarket, immediately outside the exit doors at dusk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Outside the entrance at chest height, hold a static medium frame lowered just enough to include 수리영’s hand and smartphone in the lower center while her face remains in the upper half. She leans slightly over the screen, isolated against the entrance as her downward eyeline closes the scene’s search movement into stillness.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 마트 입구 (In use as she exits) — The exterior-facing side of the entrance sits behind her, aligned with her exit path; used as Marks the threshold immediately behind 수리영 after she leaves the mart; 스마트폰 (Held in one hand during the unanswered call) — The screen face tilts upward toward 수리영 and remains only obliquely visible to the camera; used as Held in the lower center as the practical focus of the unanswered contact attempt.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk ambient light is kept subdued and desaturated, preserving restrained contrast between 수리영 and the mart entrance.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 마트 밖 입구에 서서 한 손에 쥔 스마트폰 화면을 내려다보는 수리영.\n\nLOCATION (lock): The exterior entrance of a large supermarket, immediately outside the exit doors at dusk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Outside the entrance at chest height, hold a static medium frame lowered just enough to include 수리영’s hand and smartphone in the lower center while her face remains in the upper half. She leans slightly over the screen, isolated against the entrance as her downward eyeline closes the scene’s search movement into stillness.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 마트 입구 (In use as she exits) — The exterior-facing side of the entrance sits behind her, aligned with her exit path; used as Marks the threshold immediately behind 수리영 after she leaves the mart; 스마트폰 (Held in one hand during the unanswered call) — The screen face tilts upward toward 수리영 and remains only obliquely visible to the camera; used as Held in the lower center as the practical focus of the unanswered contact attempt.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk ambient light is kept subdued and desaturated, preserving restrained contrast between 수리영 and the mart entrance.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S8sh6__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S8sh6.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    },
    {
     "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.",
     "path": "<bytes:802115>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L09B03.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    },
    {
     "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.",
     "path": "<bytes:802115>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요청된 미디엄 샷 구도와 로케이션의 건축적 특징(마트 간판, 코너 구조 등)을 매우 정확하게 재현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "인물과 포즈는 적절하나, 레퍼런스의 핵심인 마트 간판과 코너 건축 구조가 완전히 왜곡되어 로케이션 일치도가 낮습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요청된 미디엄 샷 구도와 로케이션의 건축적 특징(마트 간판, 코너 구조 등)을 매우 정확하게 재현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "인물과 포즈는 적절하나, 레퍼런스의 핵심인 마트 간판과 코너 건축 구조가 완전히 왜곡되어 로케이션 일치도가 낮습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 미디엄 샷 구도를 잘 따랐으며, 마트 출입문 바로 앞이라는 지정된 위치에 인물을 올바르게 배치했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "지정된 출입문 바로 앞이 아닌 길가로 인물 위치가 벗어났으며, 배경이 지시보다 넓게 잡혔습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 미디엄 샷 구도를 잘 따랐으며, 마트 출입문 바로 앞이라는 지정된 위치에 인물을 올바르게 배치했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "지정된 출입문 바로 앞이 아닌 길가로 인물 위치가 벗어났으며, 배경이 지시보다 넓게 잡혔습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 12,
     "B": 11
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": false,
    "policy": 1
   }
  },
  "totals": {
   "A": 12,
   "B": 11
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "요청된 미디엄 샷 구도와 로케이션의 건축적 특징(마트 간판, 코너 구조 등)을 매우 정확하게 재현했습니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "인물과 포즈는 적절하나, 레퍼런스의 핵심인 마트 간판과 코너 건축 구조가 완전히 왜곡되어 로케이션 일치도가 낮습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L09B03.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   },
   {
    "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.",
    "path": "<bytes:802115>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "스마트폰의 화면이 카메라를 향해 비스듬히 보여야 하지만, 기기의 뒷면(카메라 렌즈)이 보입니다.",
     "fix_en": "Rotate the smartphone so its screen side is obliquely visible to the camera instead of showing the back."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Rotate the smartphone so its screen side is obliquely visible to the camera instead of showing the back.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "배경과 인물의 앵글이 정확하게 일치하며, 스마트폰의 화면이 인물을 향하도록 올바르게 쥐고 있습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "스마트폰의 기능적 화면이 카메라를 향해 있어 인물이 기기의 뒷면을 내려다보는 논리적 오류가 발생했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "배경과 인물의 앵글이 정확하게 일치하며, 스마트폰의 화면이 인물을 향하도록 올바르게 쥐고 있습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "스마트폰의 기능적 화면이 카메라를 향해 있어 인물이 기기의 뒷면을 내려다보는 논리적 오류가 발생했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "배경, 인물, 구도가 스케치와 완벽히 일치하며, 사용자가 화면을 보는 설정에 맞게 스마트폰의 뒷면이 카메라를 향하도록 방향 규칙을 정확히 구현했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "전반적인 구도와 배경은 훌륭하나, 스마트폰 표면의 렌더링이 모호하게 뭉개졌고 기기의 앞뒤 방향이 불분명하게 표현되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "배경, 인물, 구도가 스케치와 완벽히 일치하며, 사용자가 화면을 보는 설정에 맞게 스마트폰의 뒷면이 카메라를 향하도록 방향 규칙을 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "전반적인 구도와 배경은 훌륭하나, 스마트폰 표면의 렌더링이 모호하게 뭉개졌고 기기의 앞뒤 방향이 불분명하게 표현되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 14,
     "B": 8
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S8sh6__bgfirst_bg.png",
   "bg_asset_id": "578c527d-8003-4748-a142-26c0fad75499",
   "bg_record_key": "S8sh6::bgfirst_bg",
   "chain_winner": true,
   "authority": "plate"
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L09B01",
    "B": "L09B02",
    "C": "L09B03"
   },
   "assigned": "L09B03",
   "choice": "Candidate C",
   "confident": true,
   "reason_ko": "\"마트 밖 입구에 서서\"라는 지문에 따라 마트 외부 출입구가 보이는 공간이어야 합니다.",
   "kept": "L09B03"
  }
 },
 "S9sh1::bgfirst_bg": {
  "input_fingerprint": "c0b6e47a211bc0df",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 어두운 밤, 해안도로에서 자전거 안장에 앉아 상체를 숙인 채 한쪽 무릎을 아래로 굽혀 페달을 밟고 있는 수리영의 측면, 머리카락이 진행 반대 방향으로 거세게 휘날리는 모습.\n\nLOCATION (lock): A dark coastal road beside the sea at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track parallel at torso height in a perpendicular profile, framing 수리영 from her streaming hair through the lowered torso, saddle, and downward-driving knee. She rides toward screen right near the middle of frame while the coastal road slips horizontally behind her, preserving the matched-speed side observation.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward screen-right road ahead.\n- KEY BACKGROUND ELEMENTS: 해안도로 (The route along which she rides home); used as Carries 수리영 laterally through the night composition and supplies environmental scale; 자전거 (Moving rapidly along the road) — Its full side plane faces the camera as it travels toward screen right; used as Keeps the saddle, frame, and pedaling mechanism legible beneath her lowered body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low ambient illumination appropriate to the dark night setting is rendered with subdued luminance and desaturated, tense contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 어두운 밤, 해안도로에서 자전거 안장에 앉아 상체를 숙인 채 한쪽 무릎을 아래로 굽혀 페달을 밟고 있는 수리영의 측면, 머리카락이 진행 반대 방향으로 거세게 휘날리는 모습.\n\nLOCATION (lock): A dark coastal road beside the sea at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track parallel at torso height in a perpendicular profile, framing 수리영 from her streaming hair through the lowered torso, saddle, and downward-driving knee. She rides toward screen right near the middle of frame while the coastal road slips horizontally behind her, preserving the matched-speed side observation.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward screen-right road ahead.\n- KEY BACKGROUND ELEMENTS: 해안도로 (The route along which she rides home); used as Carries 수리영 laterally through the night composition and supplies environmental scale; 자전거 (Moving rapidly along the road) — Its full side plane faces the camera as it travels toward screen right; used as Keeps the saddle, frame, and pedaling mechanism legible beneath her lowered body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low ambient illumination appropriate to the dark night setting is rendered with subdued luminance and desaturated, tense contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S9sh1__bgfirst_bg.png",
  "asset_id": "fee703a6-8db9-4474-b0c5-57dbd9b09b6b",
  "input_asset_ids": [
   "80e2678c-760b-411e-a002-7cd08781d89c",
   "519694bd-31b4-4c58-9c2d-45ff2fb4767c"
  ]
 },
 "S9sh1": {
  "input_fingerprint": "4bde30e0e8b2f94e",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤, 해안도로에서 자전거 안장에 앉아 상체를 숙인 채 한쪽 무릎을 아래로 굽혀 페달을 밟고 있는 수리영의 측면, 머리카락이 진행 반대 방향으로 거세게 휘날리는 모습.\n\nLOCATION (lock): A dark coastal road beside the sea at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track parallel at torso height in a perpendicular profile, framing 수리영 from her streaming hair through the lowered torso, saddle, and downward-driving knee. She rides toward screen right near the middle of frame while the coastal road slips horizontally behind her, preserving the matched-speed side observation.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward screen-right road ahead.\n- KEY BACKGROUND ELEMENTS: 해안도로 (The route along which she rides home); used as Carries 수리영 laterally through the night composition and supplies environmental scale; 자전거 (Moving rapidly along the road) — Its full side plane faces the camera as it travels toward screen right; used as Keeps the saddle, frame, and pedaling mechanism legible beneath her lowered body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low ambient illumination appropriate to the dark night setting is rendered with subdued luminance and desaturated, tense contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤, 해안도로에서 자전거 안장에 앉아 상체를 숙인 채 한쪽 무릎을 아래로 굽혀 페달을 밟고 있는 수리영의 측면, 머리카락이 진행 반대 방향으로 거세게 휘날리는 모습.\n\nLOCATION (lock): A dark coastal road beside the sea at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track parallel at torso height in a perpendicular profile, framing 수리영 from her streaming hair through the lowered torso, saddle, and downward-driving knee. She rides toward screen right near the middle of frame while the coastal road slips horizontally behind her, preserving the matched-speed side observation.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward screen-right road ahead.\n- KEY BACKGROUND ELEMENTS: 해안도로 (The route along which she rides home); used as Carries 수리영 laterally through the night composition and supplies environmental scale; 자전거 (Moving rapidly along the road) — Its full side plane faces the camera as it travels toward screen right; used as Keeps the saddle, frame, and pedaling mechanism legible beneath her lowered body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low ambient illumination appropriate to the dark night setting is rendered with subdued luminance and desaturated, tense contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤, 해안도로에서 자전거 안장에 앉아 상체를 숙인 채 한쪽 무릎을 아래로 굽혀 페달을 밟고 있는 수리영의 측면, 머리카락이 진행 반대 방향으로 거세게 휘날리는 모습.\n\nLOCATION (lock): A dark coastal road beside the sea at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track parallel at torso height in a perpendicular profile, framing 수리영 from her streaming hair through the lowered torso, saddle, and downward-driving knee. She rides toward screen right near the middle of frame while the coastal road slips horizontally behind her, preserving the matched-speed side observation.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward screen-right road ahead.\n- KEY BACKGROUND ELEMENTS: 해안도로 (The route along which she rides home); used as Carries 수리영 laterally through the night composition and supplies environmental scale; 자전거 (Moving rapidly along the road) — Its full side plane faces the camera as it travels toward screen right; used as Keeps the saddle, frame, and pedaling mechanism legible beneath her lowered body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low ambient illumination appropriate to the dark night setting is rendered with subdued luminance and desaturated, tense contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S9sh1__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S9sh1.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_해안_도로_8394b3.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지시된 카메라 측면 앵글, 야간 해안도로의 배경, 역동적인 자세를 훌륭하게 표현했으며 레퍼런스의 의상(청바지)과 묶은 머리 설정까지 정확하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "카메라 구도와 야간 배경, 인물의 동작은 지시사항에 부합하나, 레퍼런스의 지정 의상(청바지) 대신 반바지를 입히고 묶은 머리 설정을 누락하여 감점되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지시된 카메라 측면 앵글, 야간 해안도로의 배경, 역동적인 자세를 훌륭하게 표현했으며 레퍼런스의 의상(청바지)과 묶은 머리 설정까지 정확하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "카메라 구도와 야간 배경, 인물의 동작은 지시사항에 부합하나, 레퍼런스의 지정 의상(청바지) 대신 반바지를 입히고 묶은 머리 설정을 누락하여 감점되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 미디엄 샷 프레이밍을 정확히 따랐으며, 레퍼런스의 복장(청바지)과 배경(철책선)을 충실하게 재현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "프레임이 너무 넓어 풀 샷에 가깝고, 레퍼런스와 달리 반바지를 입고 있으며 배경의 난간 형태도 일치하지 않습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지정된 미디엄 샷 프레이밍을 정확히 따랐으며, 레퍼런스의 복장(청바지)과 배경(철책선)을 충실하게 재현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "프레임이 너무 넓어 풀 샷에 가깝고, 레퍼런스와 달리 반바지를 입고 있으며 배경의 난간 형태도 일치하지 않습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 10,
     "B": 17
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 10,
   "B": 17
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 9,
    "verdict_ko": "지시된 카메라 측면 앵글, 야간 해안도로의 배경, 역동적인 자세를 훌륭하게 표현했으며 레퍼런스의 의상(청바지)과 묶은 머리 설정까지 정확하게 구현했습니다."
   },
   {
    "label": "A",
    "score": 6,
    "verdict_ko": "카메라 구도와 야간 배경, 인물의 동작은 지시사항에 부합하나, 레퍼런스의 지정 의상(청바지) 대신 반바지를 입히고 묶은 머리 설정을 누락하여 감점되었습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_해안_도로_8394b3.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "요구된 샷 크기는 머리카락, 상체, 안장, 무릎까지만 포함하는 미디엄 샷이나, 생성된 이미지는 자전거 바퀴와 지면까지 모두 보이는 풀샷으로 렌더링되었습니다.",
     "fix_en": "Change the camera scale to a medium shot, tightly framing the character from her streaming hair down to the saddle and her knee, excluding the bicycle wheels and the ground below."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the camera scale to a medium shot, tightly framing the character from her streaming hair down to the saddle and her knee, excluding the bicycle wheels and the ground below.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "지시된 미디엄 샷 프레이밍을 완벽하게 수행하여, 자전거 전체를 보여주기 위해 프레임을 넓히지 않고 흩날리는 머리카락부터 아래로 향한 무릎까지의 요청된 피사체만을 정확히 포착했습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "프레임을 불필요하게 넓혀 자전거 전체와 지면까지 보여줌으로써, 아래로 향한 무릎까지만 프레이밍하라는 엄격한 미디엄 샷 크기 제약을 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "지시된 미디엄 샷 프레이밍을 완벽하게 수행하여, 자전거 전체를 보여주기 위해 프레임을 넓히지 않고 흩날리는 머리카락부터 아래로 향한 무릎까지의 요청된 피사체만을 정확히 포착했습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "프레임을 불필요하게 넓혀 자전거 전체와 지면까지 보여줌으로써, 아래로 향한 무릎까지만 프레이밍하라는 엄격한 미디엄 샷 크기 제약을 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "지정된 미디엄 샷 프레이밍을 정확히 준수하여 머리카락부터 안장, 굽힌 무릎까지만 화면에 담아냈으며 야간 해안도로의 분위기와 역동적인 자세를 훌륭하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 미디엄 샷 구도를 벗어나 프레임을 넓게 잡아 자전거 바퀴 전체와 발까지 모두 노출했으므로 프레이밍 규칙 위반입니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지정된 미디엄 샷 프레이밍을 정확히 준수하여 머리카락부터 안장, 굽힌 무릎까지만 화면에 담아냈으며 야간 해안도로의 분위기와 역동적인 자세를 훌륭하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요구된 미디엄 샷 구도를 벗어나 프레임을 넓게 잡아 자전거 바퀴 전체와 발까지 모두 노출했으므로 프레이밍 규칙 위반입니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 14,
     "B": 19
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S9sh1__bgfirst_bg.png",
   "bg_asset_id": "fee703a6-8db9-4474-b0c5-57dbd9b09b6b",
   "bg_record_key": "S9sh1::bgfirst_bg",
   "chain_winner": false,
   "authority": "groupbg",
   "group_key": "해안 도로",
   "groupbg_asset_id": "519694bd-31b4-4c58-9c2d-45ff2fb4767c"
  },
  "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S10sh4::bgfirst_bg": {
  "input_fingerprint": "31065e1c0834938f",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 남자의 손가락이 광고 전단지를 문틈 사이로 꾹 밀어 넣은 채 멈춘 클로즈업.\n\nLOCATION (lock): The rooftop landing directly outside the rooftop apartment’s metal entrance door.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At hand height beside the entrance, complete the dolly-in and freeze on the 옥탑방 침입 남자’s finger braced against the leaflet, with the door gap running vertically through the center third. The finger is the human-scale focal detail; the leaflet and door remain partial contextual elements rather than oversized objects, and the man’s face stays entirely outside frame.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: door gap in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 옥탑방 철문과 문틈 (Closed at the held instant before the lock releases) — The entrance-facing side is seen obliquely, with the closed edges forming a narrow vertical gap toward camera; used as Creates the narrow central channel into which the leaflet is pressed; 광고 전단지 (Being pressed firmly into the door gap) — Only a narrow portion of the printed face is visible while most of the leaflet is turned into the gap; used as Wedged into the door gap beneath the man’s pressing finger.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low ambient illumination appropriate to the night setting keeps the hand and door gap in subdued, tense contrast without specifying an additional source.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 남자의 손가락이 광고 전단지를 문틈 사이로 꾹 밀어 넣은 채 멈춘 클로즈업.\n\nLOCATION (lock): The rooftop landing directly outside the rooftop apartment’s metal entrance door.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At hand height beside the entrance, complete the dolly-in and freeze on the 옥탑방 침입 남자’s finger braced against the leaflet, with the door gap running vertically through the center third. The finger is the human-scale focal detail; the leaflet and door remain partial contextual elements rather than oversized objects, and the man’s face stays entirely outside frame.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: door gap in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 옥탑방 철문과 문틈 (Closed at the held instant before the lock releases) — The entrance-facing side is seen obliquely, with the closed edges forming a narrow vertical gap toward camera; used as Creates the narrow central channel into which the leaflet is pressed; 광고 전단지 (Being pressed firmly into the door gap) — Only a narrow portion of the printed face is visible while most of the leaflet is turned into the gap; used as Wedged into the door gap beneath the man’s pressing finger.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low ambient illumination appropriate to the night setting keeps the hand and door gap in subdued, tense contrast without specifying an additional source.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S10sh4__bgfirst_bg.png",
  "asset_id": "332366f8-dd7c-44cc-b9ee-62efb1d18c9a",
  "input_asset_ids": [
   "a74ed9f5-f800-42a9-bea6-78248f5ec293",
   "1f119228-59af-4b02-aedc-263b35c8eb50",
   "32be4774-9c13-4ca3-8108-3c20400f84a1"
  ]
 },
 "S10sh4": {
  "input_fingerprint": "4689a83d1aa79244",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 남자의 손가락이 광고 전단지를 문틈 사이로 꾹 밀어 넣은 채 멈춘 클로즈업.\n\nLOCATION (lock): The rooftop landing directly outside the rooftop apartment’s metal entrance door. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At hand height beside the entrance, complete the dolly-in and freeze on the 옥탑방 침입 남자’s finger braced against the leaflet, with the door gap running vertically through the center third. The finger is the human-scale focal detail; the leaflet and door remain partial contextual elements rather than oversized objects, and the man’s face stays entirely outside frame.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: door gap in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 옥탑방 철문과 문틈 (Closed at the held instant before the lock releases) — The entrance-facing side is seen obliquely, with the closed edges forming a narrow vertical gap toward camera; used as Creates the narrow central channel into which the leaflet is pressed; 광고 전단지 (Being pressed firmly into the door gap) — Only a narrow portion of the printed face is visible while most of the leaflet is turned into the gap; used as Wedged into the door gap beneath the man’s pressing finger.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low ambient illumination appropriate to the night setting keeps the hand and door gap in subdued, tense contrast without specifying an additional source.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 옥탑방 침입 남자 (한국인 남성, 성인형 머리 윤곽, 검은 머리, 가려진 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 남자의 손가락이 광고 전단지를 문틈 사이로 꾹 밀어 넣은 채 멈춘 클로즈업.\n\nLOCATION (lock): The rooftop landing directly outside the rooftop apartment’s metal entrance door. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At hand height beside the entrance, complete the dolly-in and freeze on the 옥탑방 침입 남자’s finger braced against the leaflet, with the door gap running vertically through the center third. The finger is the human-scale focal detail; the leaflet and door remain partial contextual elements rather than oversized objects, and the man’s face stays entirely outside frame.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: door gap in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 옥탑방 철문과 문틈 (Closed at the held instant before the lock releases) — The entrance-facing side is seen obliquely, with the closed edges forming a narrow vertical gap toward camera; used as Creates the narrow central channel into which the leaflet is pressed; 광고 전단지 (Being pressed firmly into the door gap) — Only a narrow portion of the printed face is visible while most of the leaflet is turned into the gap; used as Wedged into the door gap beneath the man’s pressing finger.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low ambient illumination appropriate to the night setting keeps the hand and door gap in subdued, tense contrast without specifying an additional source.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 옥탑방 침입 남자 (한국인 남성, 성인형 머리 윤곽, 검은 머리, 가려진 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 남자의 손가락이 광고 전단지를 문틈 사이로 꾹 밀어 넣은 채 멈춘 클로즈업.\n\nLOCATION (lock): The rooftop landing directly outside the rooftop apartment’s metal entrance door. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At hand height beside the entrance, complete the dolly-in and freeze on the 옥탑방 침입 남자’s finger braced against the leaflet, with the door gap running vertically through the center third. The finger is the human-scale focal detail; the leaflet and door remain partial contextual elements rather than oversized objects, and the man’s face stays entirely outside frame.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: door gap in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 옥탑방 철문과 문틈 (Closed at the held instant before the lock releases) — The entrance-facing side is seen obliquely, with the closed edges forming a narrow vertical gap toward camera; used as Creates the narrow central channel into which the leaflet is pressed; 광고 전단지 (Being pressed firmly into the door gap) — Only a narrow portion of the printed face is visible while most of the leaflet is turned into the gap; used as Wedged into the door gap beneath the man’s pressing finger.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low ambient illumination appropriate to the night setting keeps the hand and door gap in subdued, tense contrast without specifying an additional source.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 옥탑방 침입 남자 (한국인 남성, 성인형 머리 윤곽, 검은 머리, 가려진 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S10sh4__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S10sh4.png"
    },
    {
     "label": "CHARACTER REFERENCE — 옥탑방 침입 남자: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:868821>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L04B04.png"
    },
    {
     "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/background_chain/seed_bg_rooftop_villa_sel.png"
    },
    {
     "label": "CHARACTER REFERENCE — 옥탑방 침입 남자: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:868821>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "물리적 공간은 자연스럽게 구현되었으나, 프레임이 너무 넓고 전단지가 문틈에 끼워진 상태가 아닌 문 표면에 붙어 있어 연출 지시에서 감점되었습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "문틈과 프레임이 건물 벽에서 분리된 채 배경 위에 떠 있는 형태의 물리적으로 불가능한 공간 합성 오류(Hard Violation)로 탈락입니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "물리적 공간은 자연스럽게 구현되었으나, 프레임이 너무 넓고 전단지가 문틈에 끼워진 상태가 아닌 문 표면에 붙어 있어 연출 지시에서 감점되었습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "문틈과 프레임이 건물 벽에서 분리된 채 배경 위에 떠 있는 형태의 물리적으로 불가능한 공간 합성 오류(Hard Violation)로 탈락입니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "전단지를 문틈에 밀어 넣는 손가락의 핵심 동작과 중앙 세로 프레이밍을 지시대로 잘 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "문틈이 프레임 우측에 치우쳐 있으며, 전단지를 틈새로 밀어 넣는 대신 겉면에 대고 가리키는 동작에 그쳤습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "전단지를 문틈에 밀어 넣는 손가락의 핵심 동작과 중앙 세로 프레이밍을 지시대로 잘 구현했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "문틈이 프레임 우측에 치우쳐 있으며, 전단지를 틈새로 밀어 넣는 대신 겉면에 대고 가리키는 동작에 그쳤습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 8,
     "B": 10
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": false,
    "policy": 1
   }
  },
  "totals": {
   "A": 8,
   "B": 10
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 5,
    "verdict_ko": "물리적 공간은 자연스럽게 구현되었으나, 프레임이 너무 넓고 전단지가 문틈에 끼워진 상태가 아닌 문 표면에 붙어 있어 연출 지시에서 감점되었습니다."
   },
   {
    "label": "B",
    "score": 3,
    "verdict_ko": "문틈과 프레임이 건물 벽에서 분리된 채 배경 위에 떠 있는 형태의 물리적으로 불가능한 공간 합성 오류(Hard Violation)로 탈락입니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L04B04.png"
   },
   {
    "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/background_chain/seed_bg_rooftop_villa_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 옥탑방 침입 남자: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:868821>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "프롬프트의 금지 조건('no text')을 위반하여 전단지에 텍스트(한국어)가 명확하게 포함되어 있음.",
     "fix_en": "Remove all readable text from the advertisement leaflet, leaving only non-textual graphic elements or a blank surface."
    },
    {
     "issue_ko": "손등이 카메라를 향하고 있음에도 뻗은 검지손가락의 손톱이 측면을 향하고 있어 해부학적 구조가 어긋남.",
     "fix_en": "Reposition the fingernail on the extended index finger to face the camera, matching the anatomical orientation of the back of the hand."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all readable text from the advertisement leaflet, leaving only non-textual graphic elements or a blank surface.\n- Reposition the fingernail on the extended index finger to face the camera, matching the anatomical orientation of the back of the hand.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "텍스트 생성 오류 없이 지정된 카메라 구도와 손가락, 전단지의 질감을 사실적으로 잘 구현했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "구도는 좋으나, 전단지에 프롬프트 단어가 부자연스러운 디지털 그래픽처럼 생성되어 텍스트 및 재질감 제한을 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "텍스트 생성 오류 없이 지정된 카메라 구도와 손가락, 전단지의 질감을 사실적으로 잘 구현했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "구도는 좋으나, 전단지에 프롬프트 단어가 부자연스러운 디지털 그래픽처럼 생성되어 텍스트 및 재질감 제한을 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "완벽한 클로즈업 구도와 야간 조명을 보여주며, 문틈에 끼워진 전단지에 한글 텍스트가 있어 '광고 전단지'라는 지시사항을 매우 정확히 구현했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "클로즈업 구도와 배경 설정은 지시를 잘 따랐으나, 전단지에 인쇄된 내용이 없어 광고지로서의 디테일이 부족합니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "완벽한 클로즈업 구도와 야간 조명을 보여주며, 문틈에 끼워진 전단지에 한글 텍스트가 있어 '광고 전단지'라는 지시사항을 매우 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "클로즈업 구도와 배경 설정은 지시를 잘 따랐으나, 전단지에 인쇄된 내용이 없어 광고지로서의 디테일이 부족합니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 13,
     "B": 13
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": false,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S10sh4__bgfirst_bg.png",
   "bg_asset_id": "332366f8-dd7c-44cc-b9ee-62efb1d18c9a",
   "bg_record_key": "S10sh4::bgfirst_bg",
   "chain_winner": false,
   "authority": "plate",
   "seed_attached": true
  },
  "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "lane_policy": "ab_select_ready",
  "plate_select": {
   "candidates": {
    "A": "L04B01",
    "B": "L04B02",
    "C": "L04B03",
    "D": "L04B04"
   },
   "assigned": "L04B04",
   "choice": "Candidate D",
   "confident": true,
   "reason_ko": "전단지를 밀어 넣는 '문틈'을 촬영하기 위해 옥탑방 현관문에 가장 가까이 접근한 앵글입니다.",
   "kept": "L04B04"
  }
 },
 "S10sh6::variants": {
  "author_fp": "d7cbb8aaf267070b",
  "author": {
   "variants": [
    {
     "approach_ko": "창유리를 또렷한 관찰면으로 삼아 겹친 두 인물의 자세와 물리적 접촉을 냉정하게 포착하는 정면 구도.",
     "prompt_en": "Use a restrained, observational lens feel and a rigidly square-on composition, with the window glass itself in crisp focus. Let the overlapping reflections remain subdued but physically legible as two fully formed human figures, preserving the unreadable face and disturbing shoulder-level merger. Keep the pane edges and surrounding room understated, with soft low ambient light, desaturated color, controlled blacks, and no theatrical highlights."
    },
    {
     "approach_ko": "주변을 깊은 어둠에 가라앉히고 압축된 반사와 좁은 초점으로 공격 순간의 밀착감과 모호함을 강조하는 구도.",
     "prompt_en": "Favor a compressed, claustrophobic lens feel that makes the central reflected figures seem tightly locked together while retaining the prescribed medium scale and frontal eye-level viewpoint. Hold focus narrowly on the dim reflection plane, allowing the glass surface and room beyond it to recede into soft darkness without losing realistic material depth. Underexpose gently so the silhouettes emerge through delicate tonal separation rather than hard outlines, emphasizing the grounded ambiguity of the apparent attack."
    }
   ]
  },
  "reused": false
 },
 "S10sh6": {
  "input_fingerprint": "47df1570698a49ae",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 실내 창문에 비친 그림자, 남자가 강민숙(민숙)의 어깨를 입으로 덮친 형상으로 겹쳐진 실루엣.\n\nLOCATION (lock): Inside the rooftop apartment beside a window, where the attack is visible only as overlapping silhouettes on the glass. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From standing eye level several meters inside the room, face the window squarely and isolate the overlapping reflected silhouettes within the central portion of the frame. The 옥탑방 침입 남자’s head merges with 강민숙’s shoulder line as if attacking it, but his face remains unreadable and the reflection preserves grounded ambiguity.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 옥탑방 침입 남자 reflected silhouette in the middle-center of the frame, midground; 강민숙 reflected silhouette in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 실내 창문 (Reflecting the attack inside the room) — Its surface reflects the overlapping silhouettes of 옥탑방 침입 남자 and 강민숙, with his head covering her shoulder line; used as Serves as the sole perception plane on which the attack is revealed.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low night-appropriate ambient illumination renders the window reflection as subdued silhouettes with tense, desaturated contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The intruder has forced the rooftop-room door open and entered through it.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 실내 창문에 비친 그림자, 남자가 강민숙(민숙)의 어깨를 입으로 덮친 형상으로 겹쳐진 실루엣.\n\nLOCATION (lock): Inside the rooftop apartment beside a window, where the attack is visible only as overlapping silhouettes on the glass. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From standing eye level several meters inside the room, face the window squarely and isolate the overlapping reflected silhouettes within the central portion of the frame. The 옥탑방 침입 남자’s head merges with 강민숙’s shoulder line as if attacking it, but his face remains unreadable and the reflection preserves grounded ambiguity.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 옥탑방 침입 남자 reflected silhouette in the middle-center of the frame, midground; 강민숙 reflected silhouette in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 실내 창문 (Reflecting the attack inside the room) — Its surface reflects the overlapping silhouettes of 옥탑방 침입 남자 and 강민숙, with his head covering her shoulder line; used as Serves as the sole perception plane on which the attack is revealed.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low night-appropriate ambient illumination renders the window reflection as subdued silhouettes with tense, desaturated contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The intruder has forced the rooftop-room door open and entered through it.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained, observational lens feel and a rigidly square-on composition, with the window glass itself in crisp focus. Let the overlapping reflections remain subdued but physically legible as two fully formed human figures, preserving the unreadable face and disturbing shoulder-level merger. Keep the pane edges and surrounding room understated, with soft low ambient light, desaturated color, controlled blacks, and no theatrical highlights.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 실내 창문에 비친 그림자, 남자가 강민숙(민숙)의 어깨를 입으로 덮친 형상으로 겹쳐진 실루엣.\n\nLOCATION (lock): Inside the rooftop apartment beside a window, where the attack is visible only as overlapping silhouettes on the glass. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From standing eye level several meters inside the room, face the window squarely and isolate the overlapping reflected silhouettes within the central portion of the frame. The 옥탑방 침입 남자’s head merges with 강민숙’s shoulder line as if attacking it, but his face remains unreadable and the reflection preserves grounded ambiguity.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 옥탑방 침입 남자 reflected silhouette in the middle-center of the frame, midground; 강민숙 reflected silhouette in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 실내 창문 (Reflecting the attack inside the room) — Its surface reflects the overlapping silhouettes of 옥탑방 침입 남자 and 강민숙, with his head covering her shoulder line; used as Serves as the sole perception plane on which the attack is revealed.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low night-appropriate ambient illumination renders the window reflection as subdued silhouettes with tense, desaturated contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The intruder has forced the rooftop-room door open and entered through it.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFavor a compressed, claustrophobic lens feel that makes the central reflected figures seem tightly locked together while retaining the prescribed medium scale and frontal eye-level viewpoint. Hold focus narrowly on the dim reflection plane, allowing the glass surface and room beyond it to recede into soft darkness without losing realistic material depth. Underexpose gently so the silhouettes emerge through delicate tonal separation rather than hard outlines, emphasizing the grounded ambiguity of the apparent attack.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 8,
   "B": 5
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 8,
    "verdict_ko": "레퍼런스의 옥탑방 배경(빨래건조대, 실외기 등)을 정확하게 유지하면서 어깨를 덮치는 남자의 실루엣을 창문 반사로 훌륭하게 구현했습니다."
   },
   {
    "label": "B",
    "score": 5,
    "verdict_ko": "실루엣의 행동이 지시문만큼 명확하지 않으며, 창밖으로 보이는 레퍼런스의 배경 요소들이 심하게 왜곡되고 생략되었습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S10sh4_sel.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "남자의 실루엣이 여자의 어깨를 입으로 덮치는 형태가 아니라 여자의 얼굴 높이에 맞춰 마주보고 있습니다.",
     "fix_en": "Lower the man's head silhouette so his mouth covers the woman's shoulder line."
    },
    {
     "issue_ko": "화면 왼쪽 가장자리에 있는 종이 상자에 텍스트가 포함되어 있습니다.",
     "fix_en": "Remove all text from the cardboard box on the left edge."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Lower the man's head silhouette so his mouth covers the woman's shoulder line.\n- Remove all text from the cardboard box on the left edge.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "남자가 여자의 어깨를 입으로 덮치는 듯이 겹쳐진 실루엣의 형태와 프레이밍을 매우 정확하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "프레이밍과 배경은 잘 일치하지만, 남자의 실루엣이 어깨를 덮치기보다는 목을 향하고 있어 묘사가 조금 아쉽습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "남자가 여자의 어깨를 입으로 덮치는 듯이 겹쳐진 실루엣의 형태와 프레이밍을 매우 정확하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "프레이밍과 배경은 잘 일치하지만, 남자의 실루엣이 어깨를 덮치기보다는 목을 향하고 있어 묘사가 조금 아쉽습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "남자가 여자의 어깨 쪽으로 고개를 숙여 입으로 덮치는 듯한 실루엣의 동작을 프롬프트 지시에 맞게 가장 정확히 연출했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "창문에 비친 실루엣과 배경은 훌륭하나, 남자의 고개가 여자의 얼굴 쪽을 향하고 있어 '어깨를 입으로 덮친 형상'이라는 핵심 지시를 충족하지 못했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "남자가 여자의 어깨 쪽으로 고개를 숙여 입으로 덮치는 듯한 실루엣의 동작을 프롬프트 지시에 맞게 가장 정확히 연출했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "창문에 비친 실루엣과 배경은 훌륭하나, 남자의 고개가 여자의 얼굴 쪽을 향하고 있어 '어깨를 입으로 덮친 형상'이라는 핵심 지시를 충족하지 못했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 12,
     "B": 15
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev만 (배경 전용·공유 계획)",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S10sh4"
  },
  "lane_policy": "ab_select_bypass:bg_only:share_plan_prev_bgonly",
  "plate_select": {
   "candidates": {
    "A": "L04B01",
    "B": "L04B02",
    "C": "L04B03",
    "D": "L04B04"
   },
   "assigned": "L04B04",
   "choice": "Candidate A",
   "confident": true,
   "reason_ko": "원문의 '실내 창문에 비친 그림자'와 겹쳐진 실루엣을 표현하려면 외부에서 바라본 불 켜진 창문이 필요하며, Candidate A가 이에 정확히 부합합니다.",
   "kept": "L04B04"
  }
 },
 "S11sh2::bgfirst_bg": {
  "input_fingerprint": "0c4e5efad9f1e39b",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 계단을 급히 오르던 중 한 발을 윗계단에 디딘 역동적인 자세로, 코를 찡그린 채 굳은 표정을 한 수리영.\n\nLOCATION (lock): The exposed exterior stairway connecting the residential building to its rooftop apartment.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height ahead of 수리영, retreat upward along the staircase and hold a slightly high-angle medium frame as she plants one foot on the higher step. Her torso pitches into the climb in the lower-middle frame, but her nose and expression tighten at the sudden smell, interrupting the upward momentum before the rooftop pivot.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-center of the frame, midground, moves toward upper staircase and rooftop; upper staircase in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 옥탑방 연결 계단 (Being climbed rapidly) — The steps rise away from 수리영 toward the camera and continue into the upper center; used as Forms the steep movement axis from the lower frame toward the rooftop; 빌라 마당 (The area where she has just left the bicycle); used as Remains visible below the stairs as depth behind 수리영’s ascent.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low ambient illumination appropriate to the night exterior is restrained and desaturated, with tense contrast held on her tightened expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 계단을 급히 오르던 중 한 발을 윗계단에 디딘 역동적인 자세로, 코를 찡그린 채 굳은 표정을 한 수리영.\n\nLOCATION (lock): The exposed exterior stairway connecting the residential building to its rooftop apartment.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height ahead of 수리영, retreat upward along the staircase and hold a slightly high-angle medium frame as she plants one foot on the higher step. Her torso pitches into the climb in the lower-middle frame, but her nose and expression tighten at the sudden smell, interrupting the upward momentum before the rooftop pivot.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-center of the frame, midground, moves toward upper staircase and rooftop; upper staircase in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 옥탑방 연결 계단 (Being climbed rapidly) — The steps rise away from 수리영 toward the camera and continue into the upper center; used as Forms the steep movement axis from the lower frame toward the rooftop; 빌라 마당 (The area where she has just left the bicycle); used as Remains visible below the stairs as depth behind 수리영’s ascent.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low ambient illumination appropriate to the night exterior is restrained and desaturated, with tense contrast held on her tightened expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S11sh2__bgfirst_bg.png",
  "asset_id": "e09b76bb-1761-4475-b405-acba181feb74",
  "input_asset_ids": [
   "b3709c4a-8e65-4fa9-814f-8186bcc31c04",
   "7335ed1f-8f40-4a2d-a242-1aabb1047049",
   "32be4774-9c13-4ca3-8108-3c20400f84a1"
  ]
 },
 "S11sh2": {
  "input_fingerprint": "b397cdbb9442d79d",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 계단을 급히 오르던 중 한 발을 윗계단에 디딘 역동적인 자세로, 코를 찡그린 채 굳은 표정을 한 수리영.\n\nLOCATION (lock): The exposed exterior stairway connecting the residential building to its rooftop apartment. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height ahead of 수리영, retreat upward along the staircase and hold a slightly high-angle medium frame as she plants one foot on the higher step. Her torso pitches into the climb in the lower-middle frame, but her nose and expression tighten at the sudden smell, interrupting the upward momentum before the rooftop pivot.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-center of the frame, midground, moves toward upper staircase and rooftop; upper staircase in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 옥탑방 연결 계단 (Being climbed rapidly) — The steps rise away from 수리영 toward the camera and continue into the upper center; used as Forms the steep movement axis from the lower frame toward the rooftop; 빌라 마당 (The area where she has just left the bicycle); used as Remains visible below the stairs as depth behind 수리영’s ascent.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low ambient illumination appropriate to the night exterior is restrained and desaturated, with tense contrast held on her tightened expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room's iron entrance door remains half open from the intrusion.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 계단을 급히 오르던 중 한 발을 윗계단에 디딘 역동적인 자세로, 코를 찡그린 채 굳은 표정을 한 수리영.\n\nLOCATION (lock): The exposed exterior stairway connecting the residential building to its rooftop apartment. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height ahead of 수리영, retreat upward along the staircase and hold a slightly high-angle medium frame as she plants one foot on the higher step. Her torso pitches into the climb in the lower-middle frame, but her nose and expression tighten at the sudden smell, interrupting the upward momentum before the rooftop pivot.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-center of the frame, midground, moves toward upper staircase and rooftop; upper staircase in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 옥탑방 연결 계단 (Being climbed rapidly) — The steps rise away from 수리영 toward the camera and continue into the upper center; used as Forms the steep movement axis from the lower frame toward the rooftop; 빌라 마당 (The area where she has just left the bicycle); used as Remains visible below the stairs as depth behind 수리영’s ascent.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low ambient illumination appropriate to the night exterior is restrained and desaturated, with tense contrast held on her tightened expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room's iron entrance door remains half open from the intrusion.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 계단을 급히 오르던 중 한 발을 윗계단에 디딘 역동적인 자세로, 코를 찡그린 채 굳은 표정을 한 수리영.\n\nLOCATION (lock): The exposed exterior stairway connecting the residential building to its rooftop apartment. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height ahead of 수리영, retreat upward along the staircase and hold a slightly high-angle medium frame as she plants one foot on the higher step. Her torso pitches into the climb in the lower-middle frame, but her nose and expression tighten at the sudden smell, interrupting the upward momentum before the rooftop pivot.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-center of the frame, midground, moves toward upper staircase and rooftop; upper staircase in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 옥탑방 연결 계단 (Being climbed rapidly) — The steps rise away from 수리영 toward the camera and continue into the upper center; used as Forms the steep movement axis from the lower frame toward the rooftop; 빌라 마당 (The area where she has just left the bicycle); used as Remains visible below the stairs as depth behind 수리영’s ascent.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low ambient illumination appropriate to the night exterior is restrained and desaturated, with tense contrast held on her tightened expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room's iron entrance door remains half open from the intrusion.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S11sh2__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S11sh2.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L13B01.png"
    },
    {
     "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/background_chain/seed_bg_rooftop_villa_sel.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 찡그린 표정과 역동적인 자세를 잘 표현했으며, 난간을 잡은 모습이 더 사실적이고 미디엄 샷 프레이밍에 상대적으로 더 가깝습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "표정과 의상, 배경 요소는 잘 반영되었으나, 프레이밍이 미디엄 샷보다 넓고 인물과 계단의 투시가 다소 어색합니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 찡그린 표정과 역동적인 자세를 잘 표현했으며, 난간을 잡은 모습이 더 사실적이고 미디엄 샷 프레이밍에 상대적으로 더 가깝습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "표정과 의상, 배경 요소는 잘 반영되었으나, 프레이밍이 미디엄 샷보다 넓고 인물과 계단의 투시가 다소 어색합니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "요구된 미디엄 샷 구도 내에서 코를 찡그린 표정과 역동적인 자세를 정확히 구현했으며, 해부학적 왜곡 없이 자연스럽습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "지시된 표정과 계단을 오르는 상황은 잘 표현되었으나, 화면 왼쪽 인물의 손이 벽과 융합된 것처럼 묘사되어 해부학적 오류가 있습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "요구된 미디엄 샷 구도 내에서 코를 찡그린 표정과 역동적인 자세를 정확히 구현했으며, 해부학적 왜곡 없이 자연스럽습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "지시된 표정과 계단을 오르는 상황은 잘 표현되었으나, 화면 왼쪽 인물의 손이 벽과 융합된 것처럼 묘사되어 해부학적 오류가 있습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 12,
     "B": 15
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 12,
   "B": 15
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "지시된 찡그린 표정과 역동적인 자세를 잘 표현했으며, 난간을 잡은 모습이 더 사실적이고 미디엄 샷 프레이밍에 상대적으로 더 가깝습니다."
   },
   {
    "label": "A",
    "score": 6,
    "verdict_ko": "표정과 의상, 배경 요소는 잘 반영되었으나, 프레이밍이 미디엄 샷보다 넓고 인물과 계단의 투시가 다소 어색합니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L13B01.png"
   },
   {
    "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/background_chain/seed_bg_rooftop_villa_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "카메라가 피사체 아래에서 옥상을 올려다보는 로우 앵글로 배치되어, 인물이 계단을 내려오는 방향이 되었고 배경에 마당 대신 옥탑방이 나타납니다.",
     "fix_en": "Position the camera above the character on the stairs, looking down at a slightly high angle so she is climbing up toward the lens, with the lower stairs and villa yard visible in the background."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Position the camera above the character on the stairs, looking down at a slightly high angle so she is climbing up toward the lens, with the lower stairs and villa yard visible in the background.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "코를 찡그린 굳은 표정과 계단을 딛는 역동적인 자세 등 프롬프트의 세부 지시를 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "코를 찡그리는 구체적인 표정 지시가 누락되었으며, 계단을 오르는 자세가 상대적으로 덜 역동적입니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "코를 찡그린 굳은 표정과 계단을 딛는 역동적인 자세 등 프롬프트의 세부 지시를 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "코를 찡그리는 구체적인 표정 지시가 누락되었으며, 계단을 오르는 자세가 상대적으로 덜 역동적입니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "제시된 '코를 찡그린 채 굳은 표정'을 정확하게 묘사했으며, 배경 및 캐릭터 디테일도 충실히 구현했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "캐릭터와 배경의 기본 설정은 맞췄으나 핵심적인 표정 연기(코를 찡그림)가 누락되어 밋밋한 표정입니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "제시된 '코를 찡그린 채 굳은 표정'을 정확하게 묘사했으며, 배경 및 캐릭터 디테일도 충실히 구현했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "캐릭터와 배경의 기본 설정은 맞췄으나 핵심적인 표정 연기(코를 찡그림)가 누락되어 밋밋한 표정입니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 14,
     "B": 9
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S11sh2__bgfirst_bg.png",
   "bg_asset_id": "e09b76bb-1761-4475-b405-acba181feb74",
   "bg_record_key": "S11sh2::bgfirst_bg",
   "chain_winner": false,
   "authority": "plate",
   "seed_attached": true
  },
  "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "lane_policy": "ab_select_ready"
 },
 "S12sh6::bgfirst_bg": {
  "input_fingerprint": "5acdfc2a1f297637",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 살점이 뜯긴 어깨와 쇄골에 피를 흘리며 고개를 푹 숙인 채 앉아 있는 강민숙(민숙)의 끔찍한 형상.\n\nLOCATION (lock): A bedroom inside the rooftop apartment, on the floor behind the window curtain.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From shoulder height beside 수리영’s former approach axis, complete the dolly-in to a close view of 강민숙’s bowed head, torn shoulder, and exposed collarbone while keeping enough of her seated upper body to preserve the reveal. She occupies the center of frame alone, with the pulled curtain at one edge and the blood-marked floor only partially visible below.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 강민숙 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 커튼 (Pulled aside); used as Forms a narrow edge frame that marks where the body was concealed; 바닥의 혈흔과 피 묻은 발자국 (Blood is pooled on the floor, with footprints leading to the curtain); used as Appears only along the lower edge as evidence leading to the revealed body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued interior illumination and a desaturated palette hold the dead figure in tense contrast, with the explicitly bloody wounds providing the restrained deep-red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 살점이 뜯긴 어깨와 쇄골에 피를 흘리며 고개를 푹 숙인 채 앉아 있는 강민숙(민숙)의 끔찍한 형상.\n\nLOCATION (lock): A bedroom inside the rooftop apartment, on the floor behind the window curtain.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From shoulder height beside 수리영’s former approach axis, complete the dolly-in to a close view of 강민숙’s bowed head, torn shoulder, and exposed collarbone while keeping enough of her seated upper body to preserve the reveal. She occupies the center of frame alone, with the pulled curtain at one edge and the blood-marked floor only partially visible below.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 강민숙 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 커튼 (Pulled aside); used as Forms a narrow edge frame that marks where the body was concealed; 바닥의 혈흔과 피 묻은 발자국 (Blood is pooled on the floor, with footprints leading to the curtain); used as Appears only along the lower edge as evidence leading to the revealed body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued interior illumination and a desaturated palette hold the dead figure in tense contrast, with the explicitly bloody wounds providing the restrained deep-red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S12sh6__bgfirst_bg.png",
  "asset_id": "2cddadf3-fe69-4cbe-acde-525894eeaa31",
  "input_asset_ids": [
   "fc167c2a-5532-47bd-b62a-c23658817ce8",
   "162654ce-53bb-4786-8f45-c051f5458517"
  ]
 },
 "S12sh6": {
  "input_fingerprint": "36965648a8e46880",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 살점이 뜯긴 어깨와 쇄골에 피를 흘리며 고개를 푹 숙인 채 앉아 있는 강민숙(민숙)의 끔찍한 형상.\n\nLOCATION (lock): A bedroom inside the rooftop apartment, on the floor behind the window curtain. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From shoulder height beside 수리영’s former approach axis, complete the dolly-in to a close view of 강민숙’s bowed head, torn shoulder, and exposed collarbone while keeping enough of her seated upper body to preserve the reveal. She occupies the center of frame alone, with the pulled curtain at one edge and the blood-marked floor only partially visible below.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 강민숙 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 커튼 (Pulled aside); used as Forms a narrow edge frame that marks where the body was concealed; 바닥의 혈흔과 피 묻은 발자국 (Blood is pooled on the floor, with footprints leading to the curtain); used as Appears only along the lower edge as evidence leading to the revealed body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued interior illumination and a desaturated palette hold the dead figure in tense contrast, with the explicitly bloody wounds providing the restrained deep-red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She remains seated behind the curtain with her torso slumped forward and her head hanging fully down. One hand is closed around a crumpled photograph, while her bitten shoulder and collarbone are exposed, torn, and bleeding.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk remains dead, seated behind the curtain with her torso slumped forward and her head hanging fully down. One hand is closed around a crumpled photograph, while her bitten shoulder and collarbone are exposed, torn, and bleeding.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 여성, 40대 초반의 중년 얼굴, 부드러운 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 살점이 뜯긴 어깨와 쇄골에 피를 흘리며 고개를 푹 숙인 채 앉아 있는 강민숙(민숙)의 끔찍한 형상.\n\nLOCATION (lock): A bedroom inside the rooftop apartment, on the floor behind the window curtain. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From shoulder height beside 수리영’s former approach axis, complete the dolly-in to a close view of 강민숙’s bowed head, torn shoulder, and exposed collarbone while keeping enough of her seated upper body to preserve the reveal. She occupies the center of frame alone, with the pulled curtain at one edge and the blood-marked floor only partially visible below.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 강민숙 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 커튼 (Pulled aside); used as Forms a narrow edge frame that marks where the body was concealed; 바닥의 혈흔과 피 묻은 발자국 (Blood is pooled on the floor, with footprints leading to the curtain); used as Appears only along the lower edge as evidence leading to the revealed body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued interior illumination and a desaturated palette hold the dead figure in tense contrast, with the explicitly bloody wounds providing the restrained deep-red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She remains seated behind the curtain with her torso slumped forward and her head hanging fully down. One hand is closed around a crumpled photograph, while her bitten shoulder and collarbone are exposed, torn, and bleeding.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk remains dead, seated behind the curtain with her torso slumped forward and her head hanging fully down. One hand is closed around a crumpled photograph, while her bitten shoulder and collarbone are exposed, torn, and bleeding.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 여성, 40대 초반의 중년 얼굴, 부드러운 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 살점이 뜯긴 어깨와 쇄골에 피를 흘리며 고개를 푹 숙인 채 앉아 있는 강민숙(민숙)의 끔찍한 형상.\n\nLOCATION (lock): A bedroom inside the rooftop apartment, on the floor behind the window curtain. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From shoulder height beside 수리영’s former approach axis, complete the dolly-in to a close view of 강민숙’s bowed head, torn shoulder, and exposed collarbone while keeping enough of her seated upper body to preserve the reveal. She occupies the center of frame alone, with the pulled curtain at one edge and the blood-marked floor only partially visible below.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 강민숙 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 커튼 (Pulled aside); used as Forms a narrow edge frame that marks where the body was concealed; 바닥의 혈흔과 피 묻은 발자국 (Blood is pooled on the floor, with footprints leading to the curtain); used as Appears only along the lower edge as evidence leading to the revealed body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued interior illumination and a desaturated palette hold the dead figure in tense contrast, with the explicitly bloody wounds providing the restrained deep-red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She remains seated behind the curtain with her torso slumped forward and her head hanging fully down. One hand is closed around a crumpled photograph, while her bitten shoulder and collarbone are exposed, torn, and bleeding.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk remains dead, seated behind the curtain with her torso slumped forward and her head hanging fully down. One hand is closed around a crumpled photograph, while her bitten shoulder and collarbone are exposed, torn, and bleeding.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 여성, 40대 초반의 중년 얼굴, 부드러운 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S12sh6__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S12sh6.png"
    },
    {
     "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:1297056>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L05B07.png"
    },
    {
     "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:1297056>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "지정된 클로즈업 샷보다 프레이밍이 너무 넓고 손이 들려 있어 중력 규칙을 위반했으나, 찢어진 어깨와 출혈 등 핵심 묘사는 잘 반영되었습니다."
     },
     {
      "label": "B",
      "score": 2,
      "verdict_ko": "자세는 중력 규칙을 따르지만 숏 텍스트의 핵심인 어깨의 상처와 유혈 묘사가 완전히 누락되었으며 프레이밍도 지시와 다릅니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "지정된 클로즈업 샷보다 프레이밍이 너무 넓고 손이 들려 있어 중력 규칙을 위반했으나, 찢어진 어깨와 출혈 등 핵심 묘사는 잘 반영되었습니다."
     },
     {
      "label": "B",
      "score": 2,
      "verdict_ko": "자세는 중력 규칙을 따르지만 숏 텍스트의 핵심인 어깨의 상처와 유혈 묘사가 완전히 누락되었으며 프레이밍도 지시와 다릅니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "지시된 클로즈업 샷보다 프레임이 넓게 잡혔으나, 찢긴 어깨, 쇄골의 혈흔, 구겨진 사진을 쥔 채 고개를 숙인 핵심 묘사를 충실히 구현했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "프레이밍이 지시된 것보다 넓게 설정되었으며, 가장 중요한 찢긴 어깨와 쇄골의 상처 및 혈흔 묘사가 완전히 누락되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "지시된 클로즈업 샷보다 프레임이 넓게 잡혔으나, 찢긴 어깨, 쇄골의 혈흔, 구겨진 사진을 쥔 채 고개를 숙인 핵심 묘사를 충실히 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "프레이밍이 지시된 것보다 넓게 설정되었으며, 가장 중요한 찢긴 어깨와 쇄골의 상처 및 혈흔 묘사가 완전히 누락되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 10,
     "B": 5
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 10,
   "B": 5
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "지정된 클로즈업 샷보다 프레이밍이 너무 넓고 손이 들려 있어 중력 규칙을 위반했으나, 찢어진 어깨와 출혈 등 핵심 묘사는 잘 반영되었습니다."
   },
   {
    "label": "B",
    "score": 2,
    "verdict_ko": "자세는 중력 규칙을 따르지만 숏 텍스트의 핵심인 어깨의 상처와 유혈 묘사가 완전히 누락되었으며 프레이밍도 지시와 다릅니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L05B07.png"
   },
   {
    "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:1297056>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "죽은 인물의 손이 허공에 들려 사진을 쥐고 있습니다. 무의식 상태의 중력 규칙에 따라 손은 허공에 떠 있지 않고 다리 위에 완전히 힘없이 늘어져 있어야 합니다.",
     "fix_en": "Drop the hand holding the crumpled photograph so it rests completely limp and fully supported on her lap, obeying gravity."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Drop the hand holding the crumpled photograph so it rests completely limp and fully supported on her lap, obeying gravity.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "중력 지침에 따라 구겨진 사진을 쥔 손이 다리 위에 완전히 지지되어 있어 물리적 사실성을 완벽히 충족합니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "시체 상태임에도 불구하고 사진을 쥔 손이 허공에 들려 있어 중력과 근육 이완에 대한 지침을 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "중력 지침에 따라 구겨진 사진을 쥔 손이 다리 위에 완전히 지지되어 있어 물리적 사실성을 완벽히 충족합니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "시체 상태임에도 불구하고 사진을 쥔 손이 허공에 들려 있어 중력과 근육 이완에 대한 지침을 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "구겨진 사진을 꽉 쥔 손의 형태가 지시문과 스케치에 정확히 부합하며, 전반적인 분위기와 구도도 훌륭하게 구현되었습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "지시문(움켜쥔 손) 및 스케치와 달리 손을 펴고 있어 감각적 디테일에서 아쉬움이 있으며 손가락 묘사도 다소 부자연스럽습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "구겨진 사진을 꽉 쥔 손의 형태가 지시문과 스케치에 정확히 부합하며, 전반적인 분위기와 구도도 훌륭하게 구현되었습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "지시문(움켜쥔 손) 및 스케치와 달리 손을 펴고 있어 감각적 디테일에서 아쉬움이 있으며 손가락 묘사도 다소 부자연스럽습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 13,
     "B": 15
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": false,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S12sh6__bgfirst_bg.png",
   "bg_asset_id": "2cddadf3-fe69-4cbe-acde-525894eeaa31",
   "bg_record_key": "S12sh6::bgfirst_bg",
   "chain_winner": true,
   "authority": "plate"
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L05B01",
    "B": "L05B02",
    "C": "L05B03",
    "D": "L05B04",
    "E": "L05B05",
    "F": "L05B06",
    "G": "L05B07"
   },
   "assigned": "L05B06",
   "choice": "Candidate F",
   "confident": false,
   "reason_ko": "지문에서 인물이 특정 방이나 가구(침대, 커튼 등)와 함께 묘사되지 않고 단순히 앉아 있는 모습만 설명되어 있어, 서브공간을 명확히 특정할 근거가 부족하므로 기존 할당을 유지함.",
   "kept": "L05B06"
  }
 },
 "S12sh11::variants": {
  "author_fp": "74b88d9a6c38886f",
  "author": {
   "variants": [
    {
     "approach_ko": "손목 표면을 정면에 가깝게 포착해 문신과 피부 질감을 냉정하고 법의학적으로 강조한다.",
     "prompt_en": "Photograph the wrist with a restrained, near-orthographic macro feel, making the circular tattoo read with stark forensic clarity at the center. Gentle raking illumination reveals pores, fine skin texture, and the red pigment conforming naturally to the wrist’s curvature, while the narrow photograph fragment and bed area dissolve into soft, unobtrusive blur."
    },
    {
     "approach_ko": "얕은 초점과 급격한 명암 저하로 문신만 어둠 속에서 압박감 있게 떠오르게 한다.",
     "prompt_en": "Use an intimate, compressed close-focus treatment with an extremely narrow plane of focus locked on the darkest red portion of the circular tattoo. Let the wrist edges fall quickly out of focus and into subdued shadow, creating a claustrophobic sense of discovery; keep the photograph fragment barely legible as a crumpled physical edge at the bottom, with everything beyond it reduced to indistinct depth."
    }
   ]
  },
  "reused": false
 },
 "S12sh11": {
  "input_fingerprint": "05c9d54995d2d0f0",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 강민숙(민숙)의 창백한 손목 위, 붉은색으로 짙게 물든 원형 문신 클로즈업.\n\nLOCATION (lock): The bloodstained bedroom inside the rooftop apartment, beside the body behind the curtain. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Complete the stage-prescribed dolly-in at hand height, settling into an extreme close view centered on the red circular tattoo across 강민숙's pale wrist; most of her hand is cropped away, while only a narrow fragment of the crumpled photograph remains at the lower edge. The shallow focus isolates the tattoo and leaves the adjacent bed area indistinct, preparing the upward tilt; compared with the preceding beat, only camera distance tightens.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 구겨진 사진 (구겨져 있음) — Only a narrow portion of the image-bearing face is visible at the lower frame edge, with its subject cropped beyond recognition; used as A small edge fragment provides context for what had been held without competing with the tattoo.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued, tense illumination holds the wrist in restrained tones while the deep red tattoo supplies the only forceful color contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the dim nighttime room lighting, the dead woman's seated position, bloodied condition, and unchanged clothing from the reference. EXCLUDE the wider view of her torn shoulder and the surrounding room; frame only the pale wrist and the dark red circular mark.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She remains seated behind the curtain with her torso slumped forward and her head hanging fully down. One hand is closed around a crumpled photograph, while her bitten shoulder and collarbone are exposed, torn, and bleeding.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk's corpse remains seated and wounded amid the blood, while the circular mark on her wrist has turned red; one hand is closed around a crumpled photograph.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 여성, 40대 초반의 중년 얼굴, 부드러운 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 강민숙(민숙)의 창백한 손목 위, 붉은색으로 짙게 물든 원형 문신 클로즈업.\n\nLOCATION (lock): The bloodstained bedroom inside the rooftop apartment, beside the body behind the curtain. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Complete the stage-prescribed dolly-in at hand height, settling into an extreme close view centered on the red circular tattoo across 강민숙's pale wrist; most of her hand is cropped away, while only a narrow fragment of the crumpled photograph remains at the lower edge. The shallow focus isolates the tattoo and leaves the adjacent bed area indistinct, preparing the upward tilt; compared with the preceding beat, only camera distance tightens.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 구겨진 사진 (구겨져 있음) — Only a narrow portion of the image-bearing face is visible at the lower frame edge, with its subject cropped beyond recognition; used as A small edge fragment provides context for what had been held without competing with the tattoo.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued, tense illumination holds the wrist in restrained tones while the deep red tattoo supplies the only forceful color contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the dim nighttime room lighting, the dead woman's seated position, bloodied condition, and unchanged clothing from the reference. EXCLUDE the wider view of her torn shoulder and the surrounding room; frame only the pale wrist and the dark red circular mark.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She remains seated behind the curtain with her torso slumped forward and her head hanging fully down. One hand is closed around a crumpled photograph, while her bitten shoulder and collarbone are exposed, torn, and bleeding.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk's corpse remains seated and wounded amid the blood, while the circular mark on her wrist has turned red; one hand is closed around a crumpled photograph.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 여성, 40대 초반의 중년 얼굴, 부드러운 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the wrist with a restrained, near-orthographic macro feel, making the circular tattoo read with stark forensic clarity at the center. Gentle raking illumination reveals pores, fine skin texture, and the red pigment conforming naturally to the wrist’s curvature, while the narrow photograph fragment and bed area dissolve into soft, unobtrusive blur.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 강민숙(민숙)의 창백한 손목 위, 붉은색으로 짙게 물든 원형 문신 클로즈업.\n\nLOCATION (lock): The bloodstained bedroom inside the rooftop apartment, beside the body behind the curtain. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Complete the stage-prescribed dolly-in at hand height, settling into an extreme close view centered on the red circular tattoo across 강민숙's pale wrist; most of her hand is cropped away, while only a narrow fragment of the crumpled photograph remains at the lower edge. The shallow focus isolates the tattoo and leaves the adjacent bed area indistinct, preparing the upward tilt; compared with the preceding beat, only camera distance tightens.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 구겨진 사진 (구겨져 있음) — Only a narrow portion of the image-bearing face is visible at the lower frame edge, with its subject cropped beyond recognition; used as A small edge fragment provides context for what had been held without competing with the tattoo.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued, tense illumination holds the wrist in restrained tones while the deep red tattoo supplies the only forceful color contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the dim nighttime room lighting, the dead woman's seated position, bloodied condition, and unchanged clothing from the reference. EXCLUDE the wider view of her torn shoulder and the surrounding room; frame only the pale wrist and the dark red circular mark.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She remains seated behind the curtain with her torso slumped forward and her head hanging fully down. One hand is closed around a crumpled photograph, while her bitten shoulder and collarbone are exposed, torn, and bleeding.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk's corpse remains seated and wounded amid the blood, while the circular mark on her wrist has turned red; one hand is closed around a crumpled photograph.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 여성, 40대 초반의 중년 얼굴, 부드러운 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an intimate, compressed close-focus treatment with an extremely narrow plane of focus locked on the darkest red portion of the circular tattoo. Let the wrist edges fall quickly out of focus and into subdued shadow, creating a claustrophobic sense of discovery; keep the photograph fragment barely legible as a crumpled physical edge at the bottom, with everything beyond it reduced to indistinct depth.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "B": 7,
   "A": 3
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "손목의 원형 문신에 집중된 클로즈업 구도를 잘 따랐으며, 팔이 중력에 의해 자연스럽게 놓여 있는 시체의 자세 지침을 충실히 반영했습니다."
   },
   {
    "label": "A",
    "score": 3,
    "verdict_ko": "팔이 중력을 거슬러 공중에 수직으로 들려 있어 '근육의 힘이 들어가지 않은 자세' 지침을 위반했으며, 사진 파편이 팔 위에 평면적으로 붙어 있어 어색합니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S12sh6_sel.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "손이 텅 빈 채로 펼쳐져 있으며, 지시된 대로 구겨진 사진을 쥐고 있지 않습니다(사진 조각이 손과 완전히 분리되어 하단에 위치함).",
     "fix_en": "Redraw the hand extending toward the bottom edge so the fingers are closed firmly around the crumpled photograph fragment."
    },
    {
     "issue_ko": "죽은 사람의 팔과 손이 허공에 뻣뻣하게 들려 있어, 중력에 의해 지지대에 완전히 늘어져 있어야 한다는 규칙을 위반했습니다.",
     "fix_en": "Render the arm and hand resting heavily and flatly against a dark, fabric-covered surface directly beneath them."
    },
    {
     "issue_ko": "손목의 붉은 원형 문신이 피부에 스며든 잉크가 아니라 위로 두껍게 튀어나온 딱지나 스티커처럼 묘사되었습니다.",
     "fix_en": "Flatten the red circle's texture so it appears as smooth ink embedded within the skin rather than a raised physical patch."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Redraw the hand extending toward the bottom edge so the fingers are closed firmly around the crumpled photograph fragment.\n- Render the arm and hand resting heavily and flatly against a dark, fabric-covered surface directly beneath them.\n- Flatten the red circle's texture so it appears as smooth ink embedded within the skin rather than a raised physical patch.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "손의 대부분이 잘리고 사진의 일부만 하단 가장자리에 남아야 한다는 프레이밍 지시를 정확히 따랐습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "구겨진 사진을 쥔 또 다른 손이 프레임에 온전히 등장하여 '사진의 좁은 조각만 보여야 한다'는 지시를 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "손의 대부분이 잘리고 사진의 일부만 하단 가장자리에 남아야 한다는 프레이밍 지시를 정확히 따랐습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "구겨진 사진을 쥔 또 다른 손이 프레임에 온전히 등장하여 '사진의 좁은 조각만 보여야 한다'는 지시를 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "요청된 프레이밍(손은 대부분 잘리고 사진의 좁은 파편만 하단에 보임)을 무시하고 주먹 쥔 손 전체를 화면에 노출하여 지시를 위반했습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "타투를 중앙에 배치하면서 손을 크롭하고 구겨진 사진의 가장자리만 하단에 남기는 익스트림 클로즈업 프레이밍을 정확히 구현했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "요청된 프레이밍(손은 대부분 잘리고 사진의 좁은 파편만 하단에 보임)을 무시하고 주먹 쥔 손 전체를 화면에 노출하여 지시를 위반했습니다."
     },
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "타투를 중앙에 배치하면서 손을 크롭하고 구겨진 사진의 가장자리만 하단에 남기는 익스트림 클로즈업 프레이밍을 정확히 구현했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 16,
     "B": 6
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S12sh6"
  },
  "locked_char_refs_excluded": [
   "강민숙(C02)"
  ]
 },
 "S13sh2::bgfirst_bg": {
  "input_fingerprint": "08470036b9e95b6b",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 번쩍이는 경찰차 경광등 불빛 아래, 차 문을 열고 내린 채 굳은 표정으로 서 있는 혜수.\n\nLOCATION (lock): The outdoor arrival area at the base of the old residential building, beside a police car flashing at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the prescribed chest-height three-quarter position beside the vehicle, hold a static medium frame as 혜수 locks her weight beside the newly opened door, occupying the middle-right while the vehicle body and door define the left side. Keep her gaze projected beyond the frame toward 수리영, with the intermittent emergency-light pulses registering across her rigid expression.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 경찰차 (문이 열려 있음) — The vehicle's side and the open door face the camera at a three-quarter angle; used as Its open door creates a hard lateral frame beside 혜수 and establishes her point of arrival; 경찰차 경광등 (번쩍이고 있음) — The camera-facing portions of the light assembly remain visible above the vehicle body; used as The active lights intermittently reshape the contrast across 혜수 and the vehicle.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Flashing police lights create intermittent hard contrast over otherwise subdued nighttime exposure.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 번쩍이는 경찰차 경광등 불빛 아래, 차 문을 열고 내린 채 굳은 표정으로 서 있는 혜수.\n\nLOCATION (lock): The outdoor arrival area at the base of the old residential building, beside a police car flashing at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the prescribed chest-height three-quarter position beside the vehicle, hold a static medium frame as 혜수 locks her weight beside the newly opened door, occupying the middle-right while the vehicle body and door define the left side. Keep her gaze projected beyond the frame toward 수리영, with the intermittent emergency-light pulses registering across her rigid expression.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 경찰차 (문이 열려 있음) — The vehicle's side and the open door face the camera at a three-quarter angle; used as Its open door creates a hard lateral frame beside 혜수 and establishes her point of arrival; 경찰차 경광등 (번쩍이고 있음) — The camera-facing portions of the light assembly remain visible above the vehicle body; used as The active lights intermittently reshape the contrast across 혜수 and the vehicle.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Flashing police lights create intermittent hard contrast over otherwise subdued nighttime exposure.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S13sh2__bgfirst_bg.png",
  "asset_id": "63c9c1b6-fcab-4d3c-a871-0e76a57032b6",
  "input_asset_ids": [
   "4e40834d-7fd0-4840-861e-263c34f9b9f8",
   "ce6adde7-be5a-4231-9cc2-aae8ac2ca7f2"
  ]
 },
 "S13sh2": {
  "input_fingerprint": "1b85f01f21b71e03",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 번쩍이는 경찰차 경광등 불빛 아래, 차 문을 열고 내린 채 굳은 표정으로 서 있는 혜수.\n\nLOCATION (lock): The outdoor arrival area at the base of the old residential building, beside a police car flashing at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the prescribed chest-height three-quarter position beside the vehicle, hold a static medium frame as 혜수 locks her weight beside the newly opened door, occupying the middle-right while the vehicle body and door define the left side. Keep her gaze projected beyond the frame toward 수리영, with the intermittent emergency-light pulses registering across her rigid expression.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 경찰차 (문이 열려 있음) — The vehicle's side and the open door face the camera at a three-quarter angle; used as Its open door creates a hard lateral frame beside 혜수 and establishes her point of arrival; 경찰차 경광등 (번쩍이고 있음) — The camera-facing portions of the light assembly remain visible above the vehicle body; used as The active lights intermittently reshape the contrast across 혜수 and the vehicle.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Flashing police lights create intermittent hard contrast over otherwise subdued nighttime exposure.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 번쩍이는 경찰차 경광등 불빛 아래, 차 문을 열고 내린 채 굳은 표정으로 서 있는 혜수.\n\nLOCATION (lock): The outdoor arrival area at the base of the old residential building, beside a police car flashing at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the prescribed chest-height three-quarter position beside the vehicle, hold a static medium frame as 혜수 locks her weight beside the newly opened door, occupying the middle-right while the vehicle body and door define the left side. Keep her gaze projected beyond the frame toward 수리영, with the intermittent emergency-light pulses registering across her rigid expression.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 경찰차 (문이 열려 있음) — The vehicle's side and the open door face the camera at a three-quarter angle; used as Its open door creates a hard lateral frame beside 혜수 and establishes her point of arrival; 경찰차 경광등 (번쩍이고 있음) — The camera-facing portions of the light assembly remain visible above the vehicle body; used as The active lights intermittently reshape the contrast across 혜수 and the vehicle.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Flashing police lights create intermittent hard contrast over otherwise subdued nighttime exposure.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 번쩍이는 경찰차 경광등 불빛 아래, 차 문을 열고 내린 채 굳은 표정으로 서 있는 혜수.\n\nLOCATION (lock): The outdoor arrival area at the base of the old residential building, beside a police car flashing at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the prescribed chest-height three-quarter position beside the vehicle, hold a static medium frame as 혜수 locks her weight beside the newly opened door, occupying the middle-right while the vehicle body and door define the left side. Keep her gaze projected beyond the frame toward 수리영, with the intermittent emergency-light pulses registering across her rigid expression.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 경찰차 (문이 열려 있음) — The vehicle's side and the open door face the camera at a three-quarter angle; used as Its open door creates a hard lateral frame beside 혜수 and establishes her point of arrival; 경찰차 경광등 (번쩍이고 있음) — The camera-facing portions of the light assembly remain visible above the vehicle body; used as The active lights intermittently reshape the contrast across 혜수 and the vehicle.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Flashing police lights create intermittent hard contrast over otherwise subdued nighttime exposure.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S13sh2__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S13sh2.png"
    },
    {
     "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:796814>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L04B03.png"
    },
    {
     "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:796814>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "지시된 미디엄 샷 프레이밍과 피사체의 구도를 정확히 구현했으나, 차량이 옥상에 위치하는 물리적으로 불가능한 배경 배치가 발생했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "요구된 프레이밍보다 너무 넓게 촬영되었으며, 프롬프트에 없는 추가 인물(배경의 경찰)이 등장하여 심각한 위반 사항이 되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "지시된 미디엄 샷 프레이밍과 피사체의 구도를 정확히 구현했으나, 차량이 옥상에 위치하는 물리적으로 불가능한 배경 배치가 발생했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "요구된 프레이밍보다 너무 넓게 촬영되었으며, 프롬프트에 없는 추가 인물(배경의 경찰)이 등장하여 심각한 위반 사항이 되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "레퍼런스의 로케이션(옥상 건물)을 정확히 반영했으며, 프롬프트가 지시한 미디엄 샷과 인물 및 차량의 구도를 충실히 따랐습니다."
     },
     {
      "label": "A",
      "score": 2,
      "verdict_ko": "지시되지 않은 인물(배경의 경찰)을 임의로 추가하여 심각한 위반을 범했으며, 지정된 로케이션 레퍼런스를 무시하고 다른 장소로 왜곡했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "레퍼런스의 로케이션(옥상 건물)을 정확히 반영했으며, 프롬프트가 지시한 미디엄 샷과 인물 및 차량의 구도를 충실히 따랐습니다."
     },
     {
      "label": "B",
      "score": 2,
      "verdict_ko": "지시되지 않은 인물(배경의 경찰)을 임의로 추가하여 심각한 위반을 범했으며, 지정된 로케이션 레퍼런스를 무시하고 다른 장소로 왜곡했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 12,
     "B": 5
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 12,
   "B": 5
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 5,
    "verdict_ko": "지시된 미디엄 샷 프레이밍과 피사체의 구도를 정확히 구현했으나, 차량이 옥상에 위치하는 물리적으로 불가능한 배경 배치가 발생했습니다."
   },
   {
    "label": "B",
    "score": 3,
    "verdict_ko": "요구된 프레이밍보다 너무 넓게 촬영되었으며, 프롬프트에 없는 추가 인물(배경의 경찰)이 등장하여 심각한 위반 사항이 되었습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L04B03.png"
   },
   {
    "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:796814>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "원본 배경 이미지에 있던 파란색 물탱크가 건물 옆에서 사라졌습니다.",
     "fix_en": "Restore the large blue water tank next to the small building exactly as it appears in the background reference image."
    },
    {
     "issue_ko": "우측 배경에 있던 콘크리트 계단과 금속 난간이 사다리 형태로 변형되었습니다.",
     "fix_en": "Revert the leaning ladder structure on the right back to the original concrete stairs and metal railing from the background reference image."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Restore the large blue water tank next to the small building exactly as it appears in the background reference image.\n- Revert the leaning ladder structure on the right back to the original concrete stairs and metal railing from the background reference image.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지시된 인물의 외모, 위치, 시선을 정확히 구현하였으며 지정된 원본 배경의 모든 요소(파란 물탱크 등)를 완벽하게 보존했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "인물 구현은 우수하나, 원본 배경에 있던 파란색 물탱크가 사라져 배경 고정 지시를 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지시된 인물의 외모, 위치, 시선을 정확히 구현하였으며 지정된 원본 배경의 모든 요소(파란 물탱크 등)를 완벽하게 보존했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "인물 구현은 우수하나, 원본 배경에 있던 파란색 물탱크가 사라져 배경 고정 지시를 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 원본 배경(물탱크, 계단 난간)을 정확히 유지하며 인물의 배치와 시선을 완벽하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "인물 배치는 적절하나, 절대 변경해서는 안 되는 고정 배경 요소(물탱크 삭제, 난간 변형)를 훼손하여 감점되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 원본 배경(물탱크, 계단 난간)을 정확히 유지하며 인물의 배치와 시선을 완벽하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "인물 배치는 적절하나, 절대 변경해서는 안 되는 고정 배경 요소(물탱크 삭제, 난간 변형)를 훼손하여 감점되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 10,
     "B": 16
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S13sh2__bgfirst_bg.png",
   "bg_asset_id": "63c9c1b6-fcab-4d3c-a871-0e76a57032b6",
   "bg_record_key": "S13sh2::bgfirst_bg",
   "chain_winner": true,
   "authority": "plate"
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L04B01",
    "B": "L04B02",
    "C": "L04B03",
    "D": "L04B04"
   },
   "assigned": "L04B03",
   "choice": "Candidate C",
   "confident": true,
   "reason_ko": "\"경찰차 경광등 불빛 아래\"라는 묘사에 부합하는 경찰차 조명 효과(적색/청색광)가 반영되어 있으며, 다른 서브공간으로 변경할 명확한 근거가 없으므로 현재 할당된 후보를 유지합니다.",
   "kept": "L04B03"
  }
 },
 "S13sh4::variants": {
  "author_fp": "8f6387c2b1c1c393",
  "author": {
   "variants": [
    {
     "approach_ko": "압축된 깊이와 얕은 초점으로 수리영의 절박한 옆얼굴과 움켜쥔 손을 밀도 있게 강조한다.",
     "prompt_en": "Use a compressed, intimate close-up with shallow but controlled depth, keeping Suri-young’s side profile and clenched hands critically sharp while Hyesu remains clearly legible on the slightly deeper plane. Let a police-light flash rake laterally across faces, knuckles, and crushed fabric, leaving the surrounding night subdued and the vehicle reduced to a soft continuity sliver."
    },
    {
     "approach_ko": "손의 물리적 압박을 전경에서 강하게 체감시키고 두 얼굴 사이의 깊이를 선명하게 살리는 긴박한 현장 시점이다.",
     "prompt_en": "Favor an immediate, tactile perspective that makes the two-handed grip the visual anchor in the near foreground while preserving the prescribed side profile and layered separation between both faces. Hold slightly more depth through the pair so facial reactions, fingers, and strained fabric folds read together; catch the intermittent police light as a harder, uneven burst with restrained falloff into the nighttime background."
    }
   ]
  },
  "reused": false
 },
 "S13sh4": {
  "input_fingerprint": "b5822164872330ea",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 혜수의 옷깃을 두 손으로 꽉 쥔 채 절박하게 올려다보는 수리영의 옆얼굴.\n\nLOCATION (lock): The building’s outdoor ground-level approach beside the police vehicle and the entrance to the rooftop stairs. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At shoulder height beside 수리영, finish the prescribed dolly-in on her side profile and the two hands clenched into 혜수's upper clothing, with 수리영 occupying the left two-thirds and 혜수 held close on the right. Their faces remain on different depth planes but both stay legible; compared with the preceding beat, only camera distance tightens.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, foreground, reaches for Hyesu's upper clothing; 혜수 in the middle-right of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 경찰차 (현장에 정차해 있음) — Only part of its side is visible behind the pair; used as A restrained sliver of the police vehicle preserves continuity without distracting from the grip.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Intermittent police-light flashes cut across the pair in tense, subdued nighttime contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the nighttime exterior, flashing police-light colors, building surroundings, and the officer's clothing from the reference. EXCLUDE the open car door and solitary standing pose; show the distressed young woman gripping the officer's lapels.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리); 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 혜수의 옷깃을 두 손으로 꽉 쥔 채 절박하게 올려다보는 수리영의 옆얼굴.\n\nLOCATION (lock): The building’s outdoor ground-level approach beside the police vehicle and the entrance to the rooftop stairs. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At shoulder height beside 수리영, finish the prescribed dolly-in on her side profile and the two hands clenched into 혜수's upper clothing, with 수리영 occupying the left two-thirds and 혜수 held close on the right. Their faces remain on different depth planes but both stay legible; compared with the preceding beat, only camera distance tightens.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, foreground, reaches for Hyesu's upper clothing; 혜수 in the middle-right of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 경찰차 (현장에 정차해 있음) — Only part of its side is visible behind the pair; used as A restrained sliver of the police vehicle preserves continuity without distracting from the grip.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Intermittent police-light flashes cut across the pair in tense, subdued nighttime contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the nighttime exterior, flashing police-light colors, building surroundings, and the officer's clothing from the reference. EXCLUDE the open car door and solitary standing pose; show the distressed young woman gripping the officer's lapels.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리); 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a compressed, intimate close-up with shallow but controlled depth, keeping Suri-young’s side profile and clenched hands critically sharp while Hyesu remains clearly legible on the slightly deeper plane. Let a police-light flash rake laterally across faces, knuckles, and crushed fabric, leaving the surrounding night subdued and the vehicle reduced to a soft continuity sliver.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 혜수의 옷깃을 두 손으로 꽉 쥔 채 절박하게 올려다보는 수리영의 옆얼굴.\n\nLOCATION (lock): The building’s outdoor ground-level approach beside the police vehicle and the entrance to the rooftop stairs. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At shoulder height beside 수리영, finish the prescribed dolly-in on her side profile and the two hands clenched into 혜수's upper clothing, with 수리영 occupying the left two-thirds and 혜수 held close on the right. Their faces remain on different depth planes but both stay legible; compared with the preceding beat, only camera distance tightens.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, foreground, reaches for Hyesu's upper clothing; 혜수 in the middle-right of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 경찰차 (현장에 정차해 있음) — Only part of its side is visible behind the pair; used as A restrained sliver of the police vehicle preserves continuity without distracting from the grip.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Intermittent police-light flashes cut across the pair in tense, subdued nighttime contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the nighttime exterior, flashing police-light colors, building surroundings, and the officer's clothing from the reference. EXCLUDE the open car door and solitary standing pose; show the distressed young woman gripping the officer's lapels.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리); 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFavor an immediate, tactile perspective that makes the two-handed grip the visual anchor in the near foreground while preserving the prescribed side profile and layered separation between both faces. Hold slightly more depth through the pair so facial reactions, fingers, and strained fabric folds read together; catch the intermittent police light as a harder, uneven burst with restrained falloff into the nighttime background.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 3,
   "B": 7
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 3,
    "verdict_ko": "수리영의 팔 하나가 혜수의 가죽 재킷 소매로 렌더링되는 심각한 신체/의상 오류가 발생해 하드 위반입니다."
   },
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "수리영이 혜수의 옷깃을 두 손으로 쥐고 절박하게 올려다보는 측면 샷의 구도와 지정된 배경 요소를 정확하게 구현했습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S13sh2_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   },
   {
    "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:796814>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "이전 샷 레퍼런스에서 혜수의 목과 옷은 깨끗한 상태였으나, 생성된 사진에서는 혜수의 목, 티셔츠, 가죽 재킷 깃에 붉은 자국이 임의로 추가되어 이전 샷의 상태를 그대로 유지하라는 지침을 위반했습니다.",
     "fix_en": "Remove all red smudges and marks from Hye-su's neck, t-shirt, and leather jacket lapel to perfectly match her clean state in the previous shot reference."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all red smudges and marks from Hye-su's neck, t-shirt, and leather jacket lapel to perfectly match her clean state in the previous shot reference.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "수리영이 혜수의 옷깃을 두 손으로 쥐고 올려다보는 측면 구도와 경찰차 배경, 조명 등 모든 요구사항을 정확하게 반영했습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "후보 A와 동일한 이미지로, 제시된 프롬프트와 레퍼런스의 구도, 인물, 배경 조건을 훌륭하게 충족합니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "수리영이 혜수의 옷깃을 두 손으로 쥐고 올려다보는 측면 구도와 경찰차 배경, 조명 등 모든 요구사항을 정확하게 반영했습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "후보 A와 동일한 이미지로, 제시된 프롬프트와 레퍼런스의 구도, 인물, 배경 조건을 훌륭하게 충족합니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 클로즈업 샷 크기와 인물 배치를 충실히 구현했으며, 조명 효과가 자연스럽고 이전 샷의 상태를 비교적 잘 유지하고 있습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "구도와 인물 배치는 훌륭하나, 혜수의 목과 셔츠에 나타난 붉은 얼룩이 조명 반사라기보다 피처럼 보여 이전 샷과의 연속성을 훼손합니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 클로즈업 샷 크기와 인물 배치를 충실히 구현했으며, 조명 효과가 자연스럽고 이전 샷의 상태를 비교적 잘 유지하고 있습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "구도와 인물 배치는 훌륭하나, 혜수의 목과 셔츠에 나타난 붉은 얼룩이 조명 반사라기보다 피처럼 보여 이전 샷과의 연속성을 훼손합니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 14,
     "B": 15
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": false,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S13sh2"
  }
 },
 "S14sh4::bgfirst_bg": {
  "input_fingerprint": "f7762b4e3bb2b54e",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 텅 빈 방 한가운데 서서 핏기가 가신 파리한 얼굴로 허공을 응시하는 수리영.\n\nLOCATION (lock): The unnaturally clean and orderly bedroom inside the rooftop apartment.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level near the room entrance, let the stage-prescribed dolly-in end in a medium view of 수리영 frozen near the center, her body slightly off-center against the conspicuously orderly room. The entrance edges retain a shallow spatial frame around her while the bed, wall, and former curtain-side location remain plainly empty; only the camera distance has changed from the preceding beat.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground; empty orderly room in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 방 입구 (열려 있음) — The camera looks through the entrance into the room, with both side edges partially visible; used as The entrance edges form a restrained frame around 수리영 and expose the room's unchanged layout; 침대 (깨끗하게 정돈되어 있음) — Its side and upper plane are visible behind 수리영; used as The bed anchors the location where the supernatural mark had previously appeared above it; 침대 위 벽 (붉은 표식이 없음); used as The bare area above the bed makes the missing red circle visually explicit; 커튼 (이전 위치에 시신이 없음) — Its room-facing side borders the rear portion of the frame; used as It marks the former location of 강민숙's body at the room edge; 방 바닥 (핏자국 없이 깨끗함); used as The immaculate floor remains visible around 수리영's feet as evidence against her memory.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient illumination and restrained color render the spotless room unnervingly ordinary against 수리영's pallor.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 텅 빈 방 한가운데 서서 핏기가 가신 파리한 얼굴로 허공을 응시하는 수리영.\n\nLOCATION (lock): The unnaturally clean and orderly bedroom inside the rooftop apartment.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level near the room entrance, let the stage-prescribed dolly-in end in a medium view of 수리영 frozen near the center, her body slightly off-center against the conspicuously orderly room. The entrance edges retain a shallow spatial frame around her while the bed, wall, and former curtain-side location remain plainly empty; only the camera distance has changed from the preceding beat.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground; empty orderly room in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 방 입구 (열려 있음) — The camera looks through the entrance into the room, with both side edges partially visible; used as The entrance edges form a restrained frame around 수리영 and expose the room's unchanged layout; 침대 (깨끗하게 정돈되어 있음) — Its side and upper plane are visible behind 수리영; used as The bed anchors the location where the supernatural mark had previously appeared above it; 침대 위 벽 (붉은 표식이 없음); used as The bare area above the bed makes the missing red circle visually explicit; 커튼 (이전 위치에 시신이 없음) — Its room-facing side borders the rear portion of the frame; used as It marks the former location of 강민숙's body at the room edge; 방 바닥 (핏자국 없이 깨끗함); used as The immaculate floor remains visible around 수리영's feet as evidence against her memory.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient illumination and restrained color render the spotless room unnervingly ordinary against 수리영's pallor.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S14sh4__bgfirst_bg.png",
  "asset_id": "375cc179-0218-4075-93f0-9187ad2a4a62",
  "input_asset_ids": [
   "529a3237-a824-419c-9889-9a4d4240a551",
   "162654ce-53bb-4786-8f45-c051f5458517"
  ]
 },
 "S14sh4": {
  "input_fingerprint": "f39ab15b6c01f23f",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 방 한가운데 서서 핏기가 가신 파리한 얼굴로 허공을 응시하는 수리영.\n\nLOCATION (lock): The unnaturally clean and orderly bedroom inside the rooftop apartment. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level near the room entrance, let the stage-prescribed dolly-in end in a medium view of 수리영 frozen near the center, her body slightly off-center against the conspicuously orderly room. The entrance edges retain a shallow spatial frame around her while the bed, wall, and former curtain-side location remain plainly empty; only the camera distance has changed from the preceding beat.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground; empty orderly room in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 방 입구 (열려 있음) — The camera looks through the entrance into the room, with both side edges partially visible; used as The entrance edges form a restrained frame around 수리영 and expose the room's unchanged layout; 침대 (깨끗하게 정돈되어 있음) — Its side and upper plane are visible behind 수리영; used as The bed anchors the location where the supernatural mark had previously appeared above it; 침대 위 벽 (붉은 표식이 없음); used as The bare area above the bed makes the missing red circle visually explicit; 커튼 (이전 위치에 시신이 없음) — Its room-facing side borders the rear portion of the frame; used as It marks the former location of 강민숙's body at the room edge; 방 바닥 (핏자국 없이 깨끗함); used as The immaculate floor remains visible around 수리영's feet as evidence against her memory.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient illumination and restrained color render the spotless room unnervingly ordinary against 수리영's pallor.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung's room is now unnaturally clean: Minsuk's corpse, the blood and footprints, and the red wall circle have all vanished.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 방 한가운데 서서 핏기가 가신 파리한 얼굴로 허공을 응시하는 수리영.\n\nLOCATION (lock): The unnaturally clean and orderly bedroom inside the rooftop apartment. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level near the room entrance, let the stage-prescribed dolly-in end in a medium view of 수리영 frozen near the center, her body slightly off-center against the conspicuously orderly room. The entrance edges retain a shallow spatial frame around her while the bed, wall, and former curtain-side location remain plainly empty; only the camera distance has changed from the preceding beat.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground; empty orderly room in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 방 입구 (열려 있음) — The camera looks through the entrance into the room, with both side edges partially visible; used as The entrance edges form a restrained frame around 수리영 and expose the room's unchanged layout; 침대 (깨끗하게 정돈되어 있음) — Its side and upper plane are visible behind 수리영; used as The bed anchors the location where the supernatural mark had previously appeared above it; 침대 위 벽 (붉은 표식이 없음); used as The bare area above the bed makes the missing red circle visually explicit; 커튼 (이전 위치에 시신이 없음) — Its room-facing side borders the rear portion of the frame; used as It marks the former location of 강민숙's body at the room edge; 방 바닥 (핏자국 없이 깨끗함); used as The immaculate floor remains visible around 수리영's feet as evidence against her memory.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient illumination and restrained color render the spotless room unnervingly ordinary against 수리영's pallor.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung's room is now unnaturally clean: Minsuk's corpse, the blood and footprints, and the red wall circle have all vanished.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 방 한가운데 서서 핏기가 가신 파리한 얼굴로 허공을 응시하는 수리영.\n\nLOCATION (lock): The unnaturally clean and orderly bedroom inside the rooftop apartment. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level near the room entrance, let the stage-prescribed dolly-in end in a medium view of 수리영 frozen near the center, her body slightly off-center against the conspicuously orderly room. The entrance edges retain a shallow spatial frame around her while the bed, wall, and former curtain-side location remain plainly empty; only the camera distance has changed from the preceding beat.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground; empty orderly room in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 방 입구 (열려 있음) — The camera looks through the entrance into the room, with both side edges partially visible; used as The entrance edges form a restrained frame around 수리영 and expose the room's unchanged layout; 침대 (깨끗하게 정돈되어 있음) — Its side and upper plane are visible behind 수리영; used as The bed anchors the location where the supernatural mark had previously appeared above it; 침대 위 벽 (붉은 표식이 없음); used as The bare area above the bed makes the missing red circle visually explicit; 커튼 (이전 위치에 시신이 없음) — Its room-facing side borders the rear portion of the frame; used as It marks the former location of 강민숙's body at the room edge; 방 바닥 (핏자국 없이 깨끗함); used as The immaculate floor remains visible around 수리영's feet as evidence against her memory.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient illumination and restrained color render the spotless room unnervingly ordinary against 수리영's pallor.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung's room is now unnaturally clean: Minsuk's corpse, the blood and footprints, and the red wall circle have all vanished.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S14sh4__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S14sh4.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L05B07.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "요구된 방의 구조(침대 좌측, 커튼 정면)와 캐릭터의 공허한 시선을 레퍼런스에 맞게 잘 구현했으나, 미디엄 샷 대신 전신이 강조된 점이 아쉽습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "미디엄 샷 지시를 어기고 전신을 담았으며, 침대와 창문의 위치가 레퍼런스 공간 구조와 전혀 다르게 왜곡되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "요구된 방의 구조(침대 좌측, 커튼 정면)와 캐릭터의 공허한 시선을 레퍼런스에 맞게 잘 구현했으나, 미디엄 샷 대신 전신이 강조된 점이 아쉽습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "미디엄 샷 지시를 어기고 전신을 담았으며, 침대와 창문의 위치가 레퍼런스 공간 구조와 전혀 다르게 왜곡되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "캐릭터의 외모와 조명, 허공을 응시하는 창백한 표정 연기가 매우 우수하나, 침대 위 벽에 희미한 얼룩이 남아 있어 '표식이 없는 빈 벽' 지시에서 약간의 감점이 있습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "벽은 깨끗하게 묘사되었으나, 침대가 방 한가운데에 어색하게 배치되어 공간감이 떨어지고 캐릭터가 측면을 응시하여 '허공을 응시하는' 정면의 공허한 느낌이 부족합니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "캐릭터의 외모와 조명, 허공을 응시하는 창백한 표정 연기가 매우 우수하나, 침대 위 벽에 희미한 얼룩이 남아 있어 '표식이 없는 빈 벽' 지시에서 약간의 감점이 있습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "벽은 깨끗하게 묘사되었으나, 침대가 방 한가운데에 어색하게 배치되어 공간감이 떨어지고 캐릭터가 측면을 응시하여 '허공을 응시하는' 정면의 공허한 느낌이 부족합니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 9,
     "B": 13
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 9,
   "B": 13
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 6,
    "verdict_ko": "요구된 방의 구조(침대 좌측, 커튼 정면)와 캐릭터의 공허한 시선을 레퍼런스에 맞게 잘 구현했으나, 미디엄 샷 대신 전신이 강조된 점이 아쉽습니다."
   },
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "미디엄 샷 지시를 어기고 전신을 담았으며, 침대와 창문의 위치가 레퍼런스 공간 구조와 전혀 다르게 왜곡되었습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L05B07.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "요청된 미디엄 샷이 아닌 수리영의 전신이 모두 보이는 풀 샷으로 구성되었습니다.",
     "fix_en": "Zoom the camera in to frame Suryoung in a medium shot from the waist up."
    },
    {
     "issue_ko": "비정상적으로 깨끗해야 할 침대 위 벽면에 커다란 아치형 얼룩이 있습니다.",
     "fix_en": "Remove the large arch-shaped stain on the wall above the bed, leaving the surface completely bare and clean."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Zoom the camera in to frame Suryoung in a medium shot from the waist up.\n- Remove the large arch-shaped stain on the wall above the bed, leaving the surface completely bare and clean.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "방 입구의 프레임 연출과 발 주변의 깨끗한 바닥 등 프롬프트가 요구한 핵심 배경 요소들을 정확하게 포함하여 화면을 구성했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "요구된 방 입구의 프레임 효과와 발 주변의 바닥이 화면에서 누락되어 카메라 및 화면 구성 지침을 충실히 따르지 못했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "방 입구의 프레임 연출과 발 주변의 깨끗한 바닥 등 프롬프트가 요구한 핵심 배경 요소들을 정확하게 포함하여 화면을 구성했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "요구된 방 입구의 프레임 효과와 발 주변의 바닥이 화면에서 누락되어 카메라 및 화면 구성 지침을 충실히 따르지 못했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "요청된 미디엄 샷보다 넓은 풀 샷으로 연출되었으나, 핵심 지시사항인 '방 입구 문틀을 통한 카메라 시점'을 정확히 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "인물의 샷 크기는 지정된 스케일에 더 가깝지만, 필수 전경 요소인 '방 입구 문틀 프레임'을 완전히 누락했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "요청된 미디엄 샷보다 넓은 풀 샷으로 연출되었으나, 핵심 지시사항인 '방 입구 문틀을 통한 카메라 시점'을 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "인물의 샷 크기는 지정된 스케일에 더 가깝지만, 필수 전경 요소인 '방 입구 문틀 프레임'을 완전히 누락했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 13,
     "B": 8
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S14sh4__bgfirst_bg.png",
   "bg_asset_id": "375cc179-0218-4075-93f0-9187ad2a4a62",
   "bg_record_key": "S14sh4::bgfirst_bg",
   "chain_winner": false,
   "authority": "plate"
  },
  "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L05B01",
    "B": "L05B02",
    "C": "L05B03",
    "D": "L05B04",
    "E": "L05B05",
    "F": "L05B06",
    "G": "L05B07"
   },
   "assigned": "L05B07",
   "choice": "Candidate G",
   "confident": true,
   "reason_ko": "'텅 빈 방 한가운데 서서'라는 지문에 따라 가구가 비워진 넓은 바닥 공간이 잘 드러나는 현재 배정된 이미지가 가장 적합합니다.",
   "kept": "L05B07"
  }
 },
 "S14sh8::bgfirst_bg": {
  "input_fingerprint": "1fe31a2c8ec665ec",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 수리영의 양어깨를 두 손으로 감싸 쥔 채 멈춘 혜수의 다정한 모습.\n\nLOCATION (lock): The rooftop apartment’s interior living area just outside the bedroom and near the dining table.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level in front of the pair, complete the stage-prescribed advance in a medium-close two-shot, placing 혜수 slightly right of center and 수리영 slightly left as both of 혜수's hands remain fixed around 수리영's shoulders. Keep their upper bodies and the symmetrical contact fully visible while allowing the cleaned dining table to recede behind them; compared with the preceding beat, only camera distance tightens.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground; 혜수 in the middle-right of the frame, midground, reaches for Suriyoung's shoulders.\n- KEY BACKGROUND ELEMENTS: 식탁 (깨끗하며 앞서 보였던 찻잔 두 개가 없음) — The near edge and upper surface angle away behind the two figures; used as Its cleared surface remains softly visible behind the pair, reinforcing the absence of 수리영's claimed evidence.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key, restrained ambient illumination softens 혜수's reassuring gesture without relieving the scene's tense contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 수리영의 양어깨를 두 손으로 감싸 쥔 채 멈춘 혜수의 다정한 모습.\n\nLOCATION (lock): The rooftop apartment’s interior living area just outside the bedroom and near the dining table.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level in front of the pair, complete the stage-prescribed advance in a medium-close two-shot, placing 혜수 slightly right of center and 수리영 slightly left as both of 혜수's hands remain fixed around 수리영's shoulders. Keep their upper bodies and the symmetrical contact fully visible while allowing the cleaned dining table to recede behind them; compared with the preceding beat, only camera distance tightens.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground; 혜수 in the middle-right of the frame, midground, reaches for Suriyoung's shoulders.\n- KEY BACKGROUND ELEMENTS: 식탁 (깨끗하며 앞서 보였던 찻잔 두 개가 없음) — The near edge and upper surface angle away behind the two figures; used as Its cleared surface remains softly visible behind the pair, reinforcing the absence of 수리영's claimed evidence.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key, restrained ambient illumination softens 혜수's reassuring gesture without relieving the scene's tense contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S14sh8__bgfirst_bg.png",
  "asset_id": "3431074c-4188-4eb9-9eaa-15bfba5b57de",
  "input_asset_ids": [
   "5e5e13ad-d585-4219-8aeb-f2ed0c5a6ffd",
   "162654ce-53bb-4786-8f45-c051f5458517"
  ]
 },
 "S14sh8": {
  "input_fingerprint": "7f0c1d0fa864ca15",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 수리영의 양어깨를 두 손으로 감싸 쥔 채 멈춘 혜수의 다정한 모습.\n\nLOCATION (lock): The rooftop apartment’s interior living area just outside the bedroom and near the dining table. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level in front of the pair, complete the stage-prescribed advance in a medium-close two-shot, placing 혜수 slightly right of center and 수리영 slightly left as both of 혜수's hands remain fixed around 수리영's shoulders. Keep their upper bodies and the symmetrical contact fully visible while allowing the cleaned dining table to recede behind them; compared with the preceding beat, only camera distance tightens.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground; 혜수 in the middle-right of the frame, midground, reaches for Suriyoung's shoulders.\n- KEY BACKGROUND ELEMENTS: 식탁 (깨끗하며 앞서 보였던 찻잔 두 개가 없음) — The near edge and upper surface angle away behind the two figures; used as Its cleared surface remains softly visible behind the pair, reinforcing the absence of 수리영's claimed evidence.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key, restrained ambient illumination softens 혜수's reassuring gesture without relieving the scene's tense contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The room remains unnaturally clean and empty of the corpse, blood, and wall mark; Suryoung's backpack and doll are left in slightly altered positions.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리); 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 수리영의 양어깨를 두 손으로 감싸 쥔 채 멈춘 혜수의 다정한 모습.\n\nLOCATION (lock): The rooftop apartment’s interior living area just outside the bedroom and near the dining table. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level in front of the pair, complete the stage-prescribed advance in a medium-close two-shot, placing 혜수 slightly right of center and 수리영 slightly left as both of 혜수's hands remain fixed around 수리영's shoulders. Keep their upper bodies and the symmetrical contact fully visible while allowing the cleaned dining table to recede behind them; compared with the preceding beat, only camera distance tightens.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground; 혜수 in the middle-right of the frame, midground, reaches for Suriyoung's shoulders.\n- KEY BACKGROUND ELEMENTS: 식탁 (깨끗하며 앞서 보였던 찻잔 두 개가 없음) — The near edge and upper surface angle away behind the two figures; used as Its cleared surface remains softly visible behind the pair, reinforcing the absence of 수리영's claimed evidence.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key, restrained ambient illumination softens 혜수's reassuring gesture without relieving the scene's tense contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The room remains unnaturally clean and empty of the corpse, blood, and wall mark; Suryoung's backpack and doll are left in slightly altered positions.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리); 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 수리영의 양어깨를 두 손으로 감싸 쥔 채 멈춘 혜수의 다정한 모습.\n\nLOCATION (lock): The rooftop apartment’s interior living area just outside the bedroom and near the dining table. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level in front of the pair, complete the stage-prescribed advance in a medium-close two-shot, placing 혜수 slightly right of center and 수리영 slightly left as both of 혜수's hands remain fixed around 수리영's shoulders. Keep their upper bodies and the symmetrical contact fully visible while allowing the cleaned dining table to recede behind them; compared with the preceding beat, only camera distance tightens.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground; 혜수 in the middle-right of the frame, midground, reaches for Suriyoung's shoulders.\n- KEY BACKGROUND ELEMENTS: 식탁 (깨끗하며 앞서 보였던 찻잔 두 개가 없음) — The near edge and upper surface angle away behind the two figures; used as Its cleared surface remains softly visible behind the pair, reinforcing the absence of 수리영's claimed evidence.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key, restrained ambient illumination softens 혜수's reassuring gesture without relieving the scene's tense contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The room remains unnaturally clean and empty of the corpse, blood, and wall mark; Suryoung's backpack and doll are left in slightly altered positions.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리); 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S14sh8__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S14sh8.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    },
    {
     "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:796814>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L05B07.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    },
    {
     "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:796814>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 미디엄 샷 스케일(상반신 위주)에 더 가깝게 프레이밍되었으며, 어깨를 감싸쥔 손의 인체 구조와 위치가 훨씬 자연스럽습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "가방과 인형을 보여주기 위해 프레임을 넓혀 샷 스케일 지침을 어겼으며, 어깨에 올린 팔의 위치와 원근감이 해부학적으로 어색합니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 미디엄 샷 스케일(상반신 위주)에 더 가깝게 프레이밍되었으며, 어깨를 감싸쥔 손의 인체 구조와 위치가 훨씬 자연스럽습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "가방과 인형을 보여주기 위해 프레임을 넓혀 샷 스케일 지침을 어겼으며, 어깨에 올린 팔의 위치와 원근감이 해부학적으로 어색합니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "배경 구조(창문 등) 일치도가 높으나, 양손 지시를 어기고 한 손만 올렸으며 손의 해부학적 오류와 소품 누락이 발생함."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "지정된 소품(배낭, 인형)은 포함되었으나 창문 형태가 레퍼런스와 다르고, A와 동일하게 양손 누락 및 손 해부학적 오류가 있음."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "배경 구조(창문 등) 일치도가 높으나, 양손 지시를 어기고 한 손만 올렸으며 손의 해부학적 오류와 소품 누락이 발생함."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "지정된 소품(배낭, 인형)은 포함되었으나 창문 형태가 레퍼런스와 다르고, A와 동일하게 양손 누락 및 손 해부학적 오류가 있음."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 7,
     "B": 11
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 7,
   "B": 11
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "지정된 미디엄 샷 스케일(상반신 위주)에 더 가깝게 프레이밍되었으며, 어깨를 감싸쥔 손의 인체 구조와 위치가 훨씬 자연스럽습니다."
   },
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "가방과 인형을 보여주기 위해 프레임을 넓혀 샷 스케일 지침을 어겼으며, 어깨에 올린 팔의 위치와 원근감이 해부학적으로 어색합니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L05B07.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   },
   {
    "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:796814>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "혜수가 한 손으로만 수리영의 어깨를 잡고 있으며, 측면 구도로 인해 프롬프트가 요구한 '두 손으로 양어깨를 감싸 쥔 대칭적인 접촉'이 보이지 않습니다.",
     "fix_en": "Change the camera angle to the front of the pair and place both of Hyesu's hands symmetrically on both of Suriyoung's shoulders."
    },
    {
     "issue_ko": "식탁이 두 사람의 뒤쪽 배경에 배치되지 않고 프레임 우측 전경에 떨어져 있습니다.",
     "fix_en": "Reposition the camera angle so the dining table recedes into the background directly behind the two characters."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the camera angle to the front of the pair and place both of Hyesu's hands symmetrically on both of Suriyoung's shoulders.\n- Reposition the camera angle so the dining table recedes into the background directly behind the two characters.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "카메라의 정면 앵글과 배경 식탁의 배치는 지시를 잘 따랐으나, 양손이 아닌 한 손만 어깨를 감싸고 있어 아쉽습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "정면이 아닌 측면 프로필 구도로 촬영되었고 식탁의 위치도 어긋났으며, 양손 접촉 조건 역시 충족하지 못했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "카메라의 정면 앵글과 배경 식탁의 배치는 지시를 잘 따랐으나, 양손이 아닌 한 손만 어깨를 감싸고 있어 아쉽습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "정면이 아닌 측면 프로필 구도로 촬영되었고 식탁의 위치도 어긋났으며, 양손 접촉 조건 역시 충족하지 못했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "양어깨를 두 손으로 감싸는 핵심 동작이 한 손으로만 묘사되었고 식탁이 배경 뒤로 물러나지 않았으나, 인물 간의 시선 처리와 원래 공간의 구조를 비교적 자연스럽게 유지했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "마찬가지로 두 손이 아닌 한 손만 어깨에 올려져 있으며, 인물들이 부자연스럽게 카메라를 정면으로 응시하고 식탁의 위치와 스케일이 왜곡되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "양어깨를 두 손으로 감싸는 핵심 동작이 한 손으로만 묘사되었고 식탁이 배경 뒤로 물러나지 않았으나, 인물 간의 시선 처리와 원래 공간의 구조를 비교적 자연스럽게 유지했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "마찬가지로 두 손이 아닌 한 손만 어깨에 올려져 있으며, 인물들이 부자연스럽게 카메라를 정면으로 응시하고 식탁의 위치와 스케일이 왜곡되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 9,
     "B": 10
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": false,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S14sh8__bgfirst_bg.png",
   "bg_asset_id": "3431074c-4188-4eb9-9eaa-15bfba5b57de",
   "bg_record_key": "S14sh8::bgfirst_bg",
   "chain_winner": false,
   "authority": "plate"
  },
  "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L05B01",
    "B": "L05B02",
    "C": "L05B03",
    "D": "L05B04",
    "E": "L05B05",
    "F": "L05B06",
    "G": "L05B07"
   },
   "assigned": "L05B07",
   "choice": "Candidate G",
   "confident": true,
   "reason_ko": "샷 내용(두 인물이 서로 어깨를 감싸 쥐는 모습)만으로는 아파트 내 특정 서브공간을 특정할 수 있는 단서(가구, 문 등)가 없으므로 기존에 할당된 후보를 유지함.",
   "kept": "L05B07"
  }
 },
 "groupbg::해안 버스정류장": {
  "input_fingerprint": "8ab3613ef82f3f8a",
  "meta": {
   "model": "gpt-image-2",
   "size": "1536x864",
   "pack": "9.202607211540",
   "contract": "bgfirst_full_v1",
   "group_sig": {
    "key": "해안 버스정류장",
    "tags": [
     "S15sh1",
     "S15sh3",
     "S15sh5"
    ]
   }
  },
  "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\n\nTHE LOCATION — South Korea, 2026; all people are Korean unless stated: An outdoor bus-stop bench along the coast, facing the dark night sea.\n\nTIME OF DAY (lock): night.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.",
  "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\n\nTHE LOCATION — South Korea, 2026; all people are Korean unless stated: An outdoor bus-stop bench along the coast, facing the dark night sea.\n\nTIME OF DAY (lock): night.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_해안_버스정류장_321da5.png",
  "asset_id": "55645c35-ad76-4074-8b99-11d2a644ceb5",
  "input_asset_ids": [
   "fcb564f4-73b0-429c-bf08-61dedd5c4349"
  ],
  "origin_tag": "S15sh1",
  "place_text": "An outdoor bus-stop bench along the coast, facing the dark night sea."
 },
 "S15sh1::bgfirst_bg": {
  "input_fingerprint": "a684f9fd486e270a",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 어두운 밤바다를 배경으로 버스정류장 벤치에 멍하니 앉아 있는 수리영의 측면.\n\nLOCATION (lock): An outdoor bus-stop bench along the coast, facing the dark night sea.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold the flow's static side axis at seated shoulder height, framing 수리영 in right-facing profile on one side of the bench with the dark night sea extending beyond her. Use a wide composition that leaves substantial negative space toward the water, keeping her motionless seated figure small enough for the coastal isolation to register.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, looks toward night sea; night sea in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 버스정류장 벤치 (수리영이 앉아 있음) — The bench is seen broadside along the profile camera axis; used as It supports 수리영's seated silhouette and establishes the bus-stop location; 밤바다 (어둡게 보임); used as The sea occupies the distant negative space along 수리영's eyeline.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued nighttime ambient light keeps the palette restrained and the sea-background separation minimal but readable.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 어두운 밤바다를 배경으로 버스정류장 벤치에 멍하니 앉아 있는 수리영의 측면.\n\nLOCATION (lock): An outdoor bus-stop bench along the coast, facing the dark night sea.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold the flow's static side axis at seated shoulder height, framing 수리영 in right-facing profile on one side of the bench with the dark night sea extending beyond her. Use a wide composition that leaves substantial negative space toward the water, keeping her motionless seated figure small enough for the coastal isolation to register.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, looks toward night sea; night sea in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 버스정류장 벤치 (수리영이 앉아 있음) — The bench is seen broadside along the profile camera axis; used as It supports 수리영's seated silhouette and establishes the bus-stop location; 밤바다 (어둡게 보임); used as The sea occupies the distant negative space along 수리영's eyeline.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued nighttime ambient light keeps the palette restrained and the sea-background separation minimal but readable.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S15sh1__bgfirst_bg.png",
  "asset_id": "274b223f-eead-47de-9f72-47b37f074aba",
  "input_asset_ids": [
   "fcb564f4-73b0-429c-bf08-61dedd5c4349",
   "55645c35-ad76-4074-8b99-11d2a644ceb5"
  ]
 },
 "S15sh1": {
  "input_fingerprint": "130808d641f67eb9",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤바다를 배경으로 버스정류장 벤치에 멍하니 앉아 있는 수리영의 측면.\n\nLOCATION (lock): An outdoor bus-stop bench along the coast, facing the dark night sea. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold the flow's static side axis at seated shoulder height, framing 수리영 in right-facing profile on one side of the bench with the dark night sea extending beyond her. Use a wide composition that leaves substantial negative space toward the water, keeping her motionless seated figure small enough for the coastal isolation to register.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, looks toward night sea; night sea in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 버스정류장 벤치 (수리영이 앉아 있음) — The bench is seen broadside along the profile camera axis; used as It supports 수리영's seated silhouette and establishes the bus-stop location; 밤바다 (어둡게 보임); used as The sea occupies the distant negative space along 수리영's eyeline.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued nighttime ambient light keeps the palette restrained and the sea-background separation minimal but readable.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤바다를 배경으로 버스정류장 벤치에 멍하니 앉아 있는 수리영의 측면.\n\nLOCATION (lock): An outdoor bus-stop bench along the coast, facing the dark night sea. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold the flow's static side axis at seated shoulder height, framing 수리영 in right-facing profile on one side of the bench with the dark night sea extending beyond her. Use a wide composition that leaves substantial negative space toward the water, keeping her motionless seated figure small enough for the coastal isolation to register.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, looks toward night sea; night sea in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 버스정류장 벤치 (수리영이 앉아 있음) — The bench is seen broadside along the profile camera axis; used as It supports 수리영's seated silhouette and establishes the bus-stop location; 밤바다 (어둡게 보임); used as The sea occupies the distant negative space along 수리영's eyeline.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued nighttime ambient light keeps the palette restrained and the sea-background separation minimal but readable.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤바다를 배경으로 버스정류장 벤치에 멍하니 앉아 있는 수리영의 측면.\n\nLOCATION (lock): An outdoor bus-stop bench along the coast, facing the dark night sea. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold the flow's static side axis at seated shoulder height, framing 수리영 in right-facing profile on one side of the bench with the dark night sea extending beyond her. Use a wide composition that leaves substantial negative space toward the water, keeping her motionless seated figure small enough for the coastal isolation to register.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, looks toward night sea; night sea in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 버스정류장 벤치 (수리영이 앉아 있음) — The bench is seen broadside along the profile camera axis; used as It supports 수리영's seated silhouette and establishes the bus-stop location; 밤바다 (어둡게 보임); used as The sea occupies the distant negative space along 수리영's eyeline.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued nighttime ambient light keeps the palette restrained and the sea-background separation minimal but readable.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S15sh1__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S15sh1.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_해안_버스정류장_321da5.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "지정된 앵글과 와이드 샷 구도는 잘 따랐으나, 벤치 대비 인물의 비율이 다소 어색하고 배경에 프롬프트에 없는 큰 달이 추가되었습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "우측 측면 앵글과 여백을 살린 와이드 샷 구도를 완벽하게 구현했으며, 인물과 배경의 물리적 비율 및 레퍼런스와의 일치도가 매우 우수합니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "지정된 앵글과 와이드 샷 구도는 잘 따랐으나, 벤치 대비 인물의 비율이 다소 어색하고 배경에 프롬프트에 없는 큰 달이 추가되었습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "우측 측면 앵글과 여백을 살린 와이드 샷 구도를 완벽하게 구현했으며, 인물과 배경의 물리적 비율 및 레퍼런스와의 일치도가 매우 우수합니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 측면 구도와 와이드 샷 크기를 완벽하게 구현했으며, 로케이션과 캐릭터의 외모 및 의상도 레퍼런스와 매우 정확하게 일치합니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "전반적인 프레이밍과 캐릭터 배치는 훌륭하나, 로케이션 레퍼런스 하늘에 없던 달을 임의로 추가하여 장소의 재현도가 약간 떨어집니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 측면 구도와 와이드 샷 크기를 완벽하게 구현했으며, 로케이션과 캐릭터의 외모 및 의상도 레퍼런스와 매우 정확하게 일치합니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "전반적인 프레이밍과 캐릭터 배치는 훌륭하나, 로케이션 레퍼런스 하늘에 없던 달을 임의로 추가하여 장소의 재현도가 약간 떨어집니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 12,
     "B": 14
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 12,
   "B": 14
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 6,
    "verdict_ko": "지정된 앵글과 와이드 샷 구도는 잘 따랐으나, 벤치 대비 인물의 비율이 다소 어색하고 배경에 프롬프트에 없는 큰 달이 추가되었습니다."
   },
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "우측 측면 앵글과 여백을 살린 와이드 샷 구도를 완벽하게 구현했으며, 인물과 배경의 물리적 비율 및 레퍼런스와의 일치도가 매우 우수합니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_해안_버스정류장_321da5.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": []
  },
  "fix_skipped": true,
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S15sh1__bgfirst_bg.png",
   "bg_asset_id": "274b223f-eead-47de-9f72-47b37f074aba",
   "bg_record_key": "S15sh1::bgfirst_bg",
   "chain_winner": false,
   "authority": "groupbg",
   "group_key": "해안 버스정류장",
   "groupbg_asset_id": "55645c35-ad76-4074-8b99-11d2a644ceb5"
  },
  "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S15sh3::variants": {
  "author_fp": "03eca4bd31348ed7",
  "author": {
   "variants": [
    {
     "approach_ko": "도로의 원근과 벤치를 선명하게 살린 깊은 초점의 환경 중심 와이드 숏.",
     "prompt_en": "Use a restrained, rectilinear wide-lens feel with deep focus, letting the road axis create a strong diagonal into the coastal darkness while the angled bench remains a crisp spatial anchor behind her. Keep the waist-height perspective observational and steady, with generous negative space toward her future path. Shape the available nighttime illumination as soft lateral light, preserving facial detail and fabric texture while cleanly outlining both clenched fists against the dim background."
    },
    {
     "approach_ko": "배경을 은은히 압축하고 수리영의 결의를 고립시키는 절제된 장망원풍 전신 숏.",
     "prompt_en": "Use a subtly compressed lens feel from the prescribed farther camera position, isolating her full figure while keeping the bench, road direction, and dark sea clearly readable in layered depth. Hold her near the center with a firm, symmetrical visual weight, allowing the surroundings to recede into gentle softness without losing their identity. Let restrained roadside night light fall more narrowly across her face and torso, with delicate edge separation around her body and fists against the low-luminance coast."
    }
   ]
  },
  "reused": false
 },
 "S15sh3": {
  "input_fingerprint": "0a52211d99e97093",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 벤치에서 몸을 일으킨 채 주먹을 꽉 쥐고 멈춘 수리영의 단호한 전신.\n\nLOCATION (lock): The roadside space beside a coastal bus-stop bench, with the night sea beyond it. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height on the bench's open side, finish the prescribed retreat in a wide full-body view just as 수리영 reaches full height, placing her near center with both clenched fists clearly separated from her silhouette. The bench remains behind her as a scale anchor and the emerging road axis opens beside her future path; compared with the preceding beat, only camera distance expands.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 버스정류장 벤치 (비어 있음) — The bench's front edge and seating plane sit behind 수리영 at a slight angle; used as It records the point from which 수리영 has abruptly risen and prevents the wide frame from becoming spatially ambiguous; 도로 (버스정류장 옆에 이어짐); used as Its axis opens toward the direction 수리영 will run in the following beat; 밤바다 (어둡게 보임); used as The distant water preserves continuity with the earlier seated profile.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Restrained nighttime illumination holds 수리영's resolved silhouette against a low-luminance coastal background.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the dark seaside, bus-stop bench, nighttime lighting, and surrounding roadside setting from the reference. EXCLUDE the seated, vacant pose and the offered drink; show the young woman standing decisively with clenched fists.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 벤치에서 몸을 일으킨 채 주먹을 꽉 쥐고 멈춘 수리영의 단호한 전신.\n\nLOCATION (lock): The roadside space beside a coastal bus-stop bench, with the night sea beyond it. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height on the bench's open side, finish the prescribed retreat in a wide full-body view just as 수리영 reaches full height, placing her near center with both clenched fists clearly separated from her silhouette. The bench remains behind her as a scale anchor and the emerging road axis opens beside her future path; compared with the preceding beat, only camera distance expands.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 버스정류장 벤치 (비어 있음) — The bench's front edge and seating plane sit behind 수리영 at a slight angle; used as It records the point from which 수리영 has abruptly risen and prevents the wide frame from becoming spatially ambiguous; 도로 (버스정류장 옆에 이어짐); used as Its axis opens toward the direction 수리영 will run in the following beat; 밤바다 (어둡게 보임); used as The distant water preserves continuity with the earlier seated profile.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Restrained nighttime illumination holds 수리영's resolved silhouette against a low-luminance coastal background.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the dark seaside, bus-stop bench, nighttime lighting, and surrounding roadside setting from the reference. EXCLUDE the seated, vacant pose and the offered drink; show the young woman standing decisively with clenched fists.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained, rectilinear wide-lens feel with deep focus, letting the road axis create a strong diagonal into the coastal darkness while the angled bench remains a crisp spatial anchor behind her. Keep the waist-height perspective observational and steady, with generous negative space toward her future path. Shape the available nighttime illumination as soft lateral light, preserving facial detail and fabric texture while cleanly outlining both clenched fists against the dim background.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 벤치에서 몸을 일으킨 채 주먹을 꽉 쥐고 멈춘 수리영의 단호한 전신.\n\nLOCATION (lock): The roadside space beside a coastal bus-stop bench, with the night sea beyond it. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height on the bench's open side, finish the prescribed retreat in a wide full-body view just as 수리영 reaches full height, placing her near center with both clenched fists clearly separated from her silhouette. The bench remains behind her as a scale anchor and the emerging road axis opens beside her future path; compared with the preceding beat, only camera distance expands.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 버스정류장 벤치 (비어 있음) — The bench's front edge and seating plane sit behind 수리영 at a slight angle; used as It records the point from which 수리영 has abruptly risen and prevents the wide frame from becoming spatially ambiguous; 도로 (버스정류장 옆에 이어짐); used as Its axis opens toward the direction 수리영 will run in the following beat; 밤바다 (어둡게 보임); used as The distant water preserves continuity with the earlier seated profile.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Restrained nighttime illumination holds 수리영's resolved silhouette against a low-luminance coastal background.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the dark seaside, bus-stop bench, nighttime lighting, and surrounding roadside setting from the reference. EXCLUDE the seated, vacant pose and the offered drink; show the young woman standing decisively with clenched fists.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a subtly compressed lens feel from the prescribed farther camera position, isolating her full figure while keeping the bench, road direction, and dark sea clearly readable in layered depth. Hold her near the center with a firm, symmetrical visual weight, allowing the surroundings to recede into gentle softness without losing their identity. Let restrained roadside night light fall more narrowly across her face and torso, with delicate edge separation around her body and fists against the low-luminance coast.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 7,
   "B": 6
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "지시문대로 주먹을 꽉 쥔 양손이 몸의 실루엣에서 명확히 분리된 전신 구도를 가장 정확히 구현했습니다."
   },
   {
    "label": "B",
    "score": 6,
    "verdict_ko": "배경과 인물의 디테일은 우수하나, 양 주먹이 몸의 실루엣에 다소 붙어 있어 분리해달라는 지시에서 약간 벗어납니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S15sh1_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "수리영이 벤치 옆에 서 있습니다. 지시문은 벤치가 그녀의 뒤에 배치되어야 한다고 명시하고 있습니다.",
     "fix_en": "Move the character to the left so she is standing directly in front of the bench, leaving it visible behind her."
    },
    {
     "issue_ko": "주먹 쥔 양손이 허벅지에 닿아 있어 몸의 실루엣에서 명확히 분리되지 않았습니다.",
     "fix_en": "Angle both arms slightly outward so her clenched fists are completely separated from the silhouette of her body."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Move the character to the left so she is standing directly in front of the bench, leaving it visible behind her.\n- Angle both arms slightly outward so her clenched fists are completely separated from the silhouette of her body.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요구된 풀바디 와이드 샷과 주먹을 쥔 동작을 잘 구현했으며, 레퍼런스의 배경과 벤치 구성을 정확하게 유지했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "레퍼런스에 없는 두 번째 벤치가 인물 뒤에 임의로 생성되는 치명적인 오브젝트 중복 오류가 발생했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요구된 풀바디 와이드 샷과 주먹을 쥔 동작을 잘 구현했으며, 레퍼런스의 배경과 벤치 구성을 정확하게 유지했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "레퍼런스에 없는 두 번째 벤치가 인물 뒤에 임의로 생성되는 치명적인 오브젝트 중복 오류가 발생했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "캐릭터 다리 뒤에 프롬프트와 레퍼런스에 없는 두 번째 벤치가 생성되어 심각한 물리적 오류(Hard Violation)가 발생했습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "레퍼런스의 장소와 벤치를 정확히 유지하면서, 주먹을 쥔 채 서 있는 캐릭터의 단호한 전신 샷을 성공적으로 구현했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "캐릭터 다리 뒤에 프롬프트와 레퍼런스에 없는 두 번째 벤치가 생성되어 심각한 물리적 오류(Hard Violation)가 발생했습니다."
     },
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "레퍼런스의 장소와 벤치를 정확히 유지하면서, 주먹을 쥔 채 서 있는 캐릭터의 단호한 전신 샷을 성공적으로 구현했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 15,
     "B": 6
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S15sh1"
  }
 },
 "S15sh5::variants": {
  "author_fp": "b12b7bdb8c243bec",
  "author": {
   "variants": [
    {
     "approach_ko": "얕은 심도와 압축된 원근으로 혜수의 절박한 외침과 떠난 방향의 빈 공간을 고립시킨다.",
     "prompt_en": "Use a gently compressed lens feel and very shallow depth of field, holding the chest-height medium shot tightly on 혜수’s upper body. Render her cupped hands, open mouth, and fixed off-screen eyeline with crisp tactile detail while the oblique empty bench and narrow road fall into subdued, softly legible layers behind her. Let restrained roadside illumination shape one side of her face, with a delicate falloff into coastal darkness that isolates the urgency of her call."
    },
    {
     "approach_ko": "더 깊은 초점과 열린 공간 구성으로 혜수의 외침을 텅 빈 정류장과 이어지는 도로 속에 놓는다.",
     "prompt_en": "Use a natural, slightly open lens perspective with deeper focus, keeping the chest-height medium framing while allowing the empty bench and receding strip of road to remain clearly readable as spatial anchors. Compose 혜수 middle-center but preserve breathing room along her eyeline toward the frame edge, making the surrounding vacancy carry the dramatic weight. Favor broad, low-contrast nighttime light with restrained highlights on her face and hands, maintaining realistic detail through the dim bus-stop environment without brightening the coastal darkness."
    }
   ]
  },
  "reused": false
 },
 "S15sh5": {
  "input_fingerprint": "ffa4badc09f85df1",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 두 손을 입가에 모은 채, 달려나가는 자세의 수리영을 향해 입을 크게 벌린 혜수의 상체.\n\nLOCATION (lock): The coastal bus-stop area beside the roadway, looking toward the route leading away from the bench. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height on the road axis, complete the pan back toward 혜수 and tighten to her upper body, placing her off-center with both hands cupped around her open mouth. Preserve her eyeline into empty off-screen space where 수리영 has departed; compared with the preceding beat, only the visual attention shifts from the runner to 혜수.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 혜수 in the middle-center of the frame, midground, looks toward departing Suriyoung off-screen.\n- KEY BACKGROUND ELEMENTS: 버스정류장 벤치 (비어 있음) — Its front and seating plane are visible behind 혜수 at an oblique angle; used as The bench remains behind 혜수 as a muted location anchor while the road-side eyeline leads out of frame; 도로 (수리영이 달려간 방향으로 이어짐); used as A narrow portion of the road preserves the direction of 수리영's departure.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued nighttime exposure keeps 혜수 readable while preserving the surrounding coastal darkness and unease.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the same night seaside bus-stop setting, bench placement, and roadside lighting from the reference. EXCLUDE the young woman's stationary foreground pose; show her running away while the officer calls after her, without adding unrelated passersby or vehicles.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 두 손을 입가에 모은 채, 달려나가는 자세의 수리영을 향해 입을 크게 벌린 혜수의 상체.\n\nLOCATION (lock): The coastal bus-stop area beside the roadway, looking toward the route leading away from the bench. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height on the road axis, complete the pan back toward 혜수 and tighten to her upper body, placing her off-center with both hands cupped around her open mouth. Preserve her eyeline into empty off-screen space where 수리영 has departed; compared with the preceding beat, only the visual attention shifts from the runner to 혜수.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 혜수 in the middle-center of the frame, midground, looks toward departing Suriyoung off-screen.\n- KEY BACKGROUND ELEMENTS: 버스정류장 벤치 (비어 있음) — Its front and seating plane are visible behind 혜수 at an oblique angle; used as The bench remains behind 혜수 as a muted location anchor while the road-side eyeline leads out of frame; 도로 (수리영이 달려간 방향으로 이어짐); used as A narrow portion of the road preserves the direction of 수리영's departure.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued nighttime exposure keeps 혜수 readable while preserving the surrounding coastal darkness and unease.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the same night seaside bus-stop setting, bench placement, and roadside lighting from the reference. EXCLUDE the young woman's stationary foreground pose; show her running away while the officer calls after her, without adding unrelated passersby or vehicles.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a gently compressed lens feel and very shallow depth of field, holding the chest-height medium shot tightly on 혜수’s upper body. Render her cupped hands, open mouth, and fixed off-screen eyeline with crisp tactile detail while the oblique empty bench and narrow road fall into subdued, softly legible layers behind her. Let restrained roadside illumination shape one side of her face, with a delicate falloff into coastal darkness that isolates the urgency of her call.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 두 손을 입가에 모은 채, 달려나가는 자세의 수리영을 향해 입을 크게 벌린 혜수의 상체.\n\nLOCATION (lock): The coastal bus-stop area beside the roadway, looking toward the route leading away from the bench. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height on the road axis, complete the pan back toward 혜수 and tighten to her upper body, placing her off-center with both hands cupped around her open mouth. Preserve her eyeline into empty off-screen space where 수리영 has departed; compared with the preceding beat, only the visual attention shifts from the runner to 혜수.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 혜수 in the middle-center of the frame, midground, looks toward departing Suriyoung off-screen.\n- KEY BACKGROUND ELEMENTS: 버스정류장 벤치 (비어 있음) — Its front and seating plane are visible behind 혜수 at an oblique angle; used as The bench remains behind 혜수 as a muted location anchor while the road-side eyeline leads out of frame; 도로 (수리영이 달려간 방향으로 이어짐); used as A narrow portion of the road preserves the direction of 수리영's departure.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued nighttime exposure keeps 혜수 readable while preserving the surrounding coastal darkness and unease.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the same night seaside bus-stop setting, bench placement, and roadside lighting from the reference. EXCLUDE the young woman's stationary foreground pose; show her running away while the officer calls after her, without adding unrelated passersby or vehicles.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a natural, slightly open lens perspective with deeper focus, keeping the chest-height medium framing while allowing the empty bench and receding strip of road to remain clearly readable as spatial anchors. Compose 혜수 middle-center but preserve breathing room along her eyeline toward the frame edge, making the surrounding vacancy carry the dramatic weight. Favor broad, low-contrast nighttime light with restrained highlights on her face and hands, maintaining realistic detail through the dim bus-stop environment without brightening the coastal darkness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 7,
   "B": 5
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "지정된 프레이밍과 포즈를 잘 구현했으며, 시선 방향이 도로 쪽을 향해 프롬프트의 지시와 일치합니다."
   },
   {
    "label": "B",
    "score": 5,
    "verdict_ko": "포즈와 의상은 일치하나, 손의 해부학적 형태가 다소 어색하고 시선이 도로 방향과 맞지 않습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S15sh3_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:796814>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "혜수의 시선이 프레임 밖이 아닌 카메라를 정면으로 향하고 있습니다.",
     "fix_en": "Shift her eyeline and head angle to look off-screen into the distance instead of staring directly at the camera."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Shift her eyeline and head angle to look off-screen into the distance instead of staring directly at the camera.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "화면 밖을 향하는 시선 처리와 상체 미디엄 샷 구도가 프롬프트의 지시를 정확히 충족합니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "지정된 복장과 손동작은 일치하나, 시선이 카메라 정면을 향해 오프스크린 지시를 어겼습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "화면 밖을 향하는 시선 처리와 상체 미디엄 샷 구도가 프롬프트의 지시를 정확히 충족합니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "지정된 복장과 손동작은 일치하나, 시선이 카메라 정면을 향해 오프스크린 지시를 어겼습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "요구된 프레이밍과 포즈를 잘 준수했으며, 수리영이 달려간 도로 방향으로의 시선 처리도 자연스럽습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "포즈와 배경은 지시사항과 일치하나, 시선이 수리영이 달려간 도로 방향이 아닌 정면을 향하고 있어 아쉽습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "요구된 프레이밍과 포즈를 잘 준수했으며, 수리영이 달려간 도로 방향으로의 시선 처리도 자연스럽습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "포즈와 배경은 지시사항과 일치하나, 시선이 수리영이 달려간 도로 방향이 아닌 정면을 향하고 있어 아쉽습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 11,
     "B": 15
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S15sh3"
  }
 },
 "S16sh3::bgfirst_bg": {
  "input_fingerprint": "de0074ad61981423",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 김형사의 눈앞에 특정 번호가 뜬 스마트폰 화면을 바짝 들이민 채 멈춘 혜수의 굳은 옆얼굴.\n\nLOCATION (lock): A nighttime police office workspace beside the on-duty detective’s desk, under the office lighting.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 혜수's shoulder height beside her, complete the prescribed dolly-in with her rigid side profile on the left and the smartphone held between her and 김형사 near the center, its screen angled cleanly toward both him and the camera. 김형사 remains on the right and slightly deeper, close enough to establish the phone as a demand rather than an isolated insert; only camera distance tightens from the preceding beat.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 혜수의 스마트폰 (화면이 켜져 있고 특정 번호가 표시됨) — The front screen faces diagonally toward 김형사 and the camera, displaying the saved number requested for tracing; used as The illuminated content plane forms the central focal bridge between 혜수 and 김형사; 경찰서 사무실 내부 (김형사가 혼자 당직 근무 중); used as The surrounding workspace remains subdued and incomplete at the frame edges, preserving the police-station context.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued office illumination keeps the palette restrained, with the active phone screen providing a localized emphasis rather than broad fill.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 김형사의 눈앞에 특정 번호가 뜬 스마트폰 화면을 바짝 들이민 채 멈춘 혜수의 굳은 옆얼굴.\n\nLOCATION (lock): A nighttime police office workspace beside the on-duty detective’s desk, under the office lighting.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 혜수's shoulder height beside her, complete the prescribed dolly-in with her rigid side profile on the left and the smartphone held between her and 김형사 near the center, its screen angled cleanly toward both him and the camera. 김형사 remains on the right and slightly deeper, close enough to establish the phone as a demand rather than an isolated insert; only camera distance tightens from the preceding beat.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 혜수의 스마트폰 (화면이 켜져 있고 특정 번호가 표시됨) — The front screen faces diagonally toward 김형사 and the camera, displaying the saved number requested for tracing; used as The illuminated content plane forms the central focal bridge between 혜수 and 김형사; 경찰서 사무실 내부 (김형사가 혼자 당직 근무 중); used as The surrounding workspace remains subdued and incomplete at the frame edges, preserving the police-station context.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued office illumination keeps the palette restrained, with the active phone screen providing a localized emphasis rather than broad fill.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S16sh3__bgfirst_bg.png",
  "asset_id": "3a7f9388-b17a-496d-a0a2-0d51278e864e",
  "input_asset_ids": [
   "65d75985-de8b-458a-a431-21c397025b13",
   "2224074a-592d-435b-9bf6-ee9a004f8d3f"
  ]
 },
 "S16sh3": {
  "input_fingerprint": "3f4d684300a75265",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 김형사의 눈앞에 특정 번호가 뜬 스마트폰 화면을 바짝 들이민 채 멈춘 혜수의 굳은 옆얼굴.\n\nLOCATION (lock): A nighttime police office workspace beside the on-duty detective’s desk, under the office lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 혜수's shoulder height beside her, complete the prescribed dolly-in with her rigid side profile on the left and the smartphone held between her and 김형사 near the center, its screen angled cleanly toward both him and the camera. 김형사 remains on the right and slightly deeper, close enough to establish the phone as a demand rather than an isolated insert; only camera distance tightens from the preceding beat.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 혜수의 스마트폰 (화면이 켜져 있고 특정 번호가 표시됨) — The front screen faces diagonally toward 김형사 and the camera, displaying the saved number requested for tracing; used as The illuminated content plane forms the central focal bridge between 혜수 and 김형사; 경찰서 사무실 내부 (김형사가 혼자 당직 근무 중); used as The surrounding workspace remains subdued and incomplete at the frame edges, preserving the police-station context.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued office illumination keeps the palette restrained, with the active phone screen providing a localized emphasis rather than broad fill.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 김형사의 눈앞에 특정 번호가 뜬 스마트폰 화면을 바짝 들이민 채 멈춘 혜수의 굳은 옆얼굴.\n\nLOCATION (lock): A nighttime police office workspace beside the on-duty detective’s desk, under the office lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 혜수's shoulder height beside her, complete the prescribed dolly-in with her rigid side profile on the left and the smartphone held between her and 김형사 near the center, its screen angled cleanly toward both him and the camera. 김형사 remains on the right and slightly deeper, close enough to establish the phone as a demand rather than an isolated insert; only camera distance tightens from the preceding beat.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 혜수의 스마트폰 (화면이 켜져 있고 특정 번호가 표시됨) — The front screen faces diagonally toward 김형사 and the camera, displaying the saved number requested for tracing; used as The illuminated content plane forms the central focal bridge between 혜수 and 김형사; 경찰서 사무실 내부 (김형사가 혼자 당직 근무 중); used as The surrounding workspace remains subdued and incomplete at the frame edges, preserving the police-station context.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued office illumination keeps the palette restrained, with the active phone screen providing a localized emphasis rather than broad fill.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 김형사의 눈앞에 특정 번호가 뜬 스마트폰 화면을 바짝 들이민 채 멈춘 혜수의 굳은 옆얼굴.\n\nLOCATION (lock): A nighttime police office workspace beside the on-duty detective’s desk, under the office lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 혜수's shoulder height beside her, complete the prescribed dolly-in with her rigid side profile on the left and the smartphone held between her and 김형사 near the center, its screen angled cleanly toward both him and the camera. 김형사 remains on the right and slightly deeper, close enough to establish the phone as a demand rather than an isolated insert; only camera distance tightens from the preceding beat.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 혜수의 스마트폰 (화면이 켜져 있고 특정 번호가 표시됨) — The front screen faces diagonally toward 김형사 and the camera, displaying the saved number requested for tracing; used as The illuminated content plane forms the central focal bridge between 혜수 and 김형사; 경찰서 사무실 내부 (김형사가 혼자 당직 근무 중); used as The surrounding workspace remains subdued and incomplete at the frame edges, preserving the police-station context.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued office illumination keeps the palette restrained, with the active phone screen providing a localized emphasis rather than broad fill.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S16sh3__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S16sh3.png"
    },
    {
     "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:796814>"
    },
    {
     "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.",
     "path": "<bytes:802115>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L14B01.png"
    },
    {
     "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:796814>"
    },
    {
     "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.",
     "path": "<bytes:802115>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "지정된 클로즈업을 무시하고 앵글이 넓게 잡혔으며, 스마트폰을 든 팔의 해부학적 구조가 물리적으로 불가능해 보입니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요청된 클로즈업 프레이밍과 스마트폰의 방향을 정확히 구현했고, 레퍼런스의 의상과 장소를 충실히 반영했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "지정된 클로즈업을 무시하고 앵글이 넓게 잡혔으며, 스마트폰을 든 팔의 해부학적 구조가 물리적으로 불가능해 보입니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요청된 클로즈업 프레이밍과 스마트폰의 방향을 정확히 구현했고, 레퍼런스의 의상과 장소를 충실히 반영했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요청된 클로즈업 구도와 인물의 측면 배치, 대각선으로 기울어진 스마트폰 화면 및 레퍼런스 의상(가죽 재킷)을 잘 반영했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "혜수의 레퍼런스 의상이 누락되었고 팔의 해부학적 구조가 어색하며, 스마트폰이 형사를 향하지 않고 정면만을 향하고 있습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요청된 클로즈업 구도와 인물의 측면 배치, 대각선으로 기울어진 스마트폰 화면 및 레퍼런스 의상(가죽 재킷)을 잘 반영했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "혜수의 레퍼런스 의상이 누락되었고 팔의 해부학적 구조가 어색하며, 스마트폰이 형사를 향하지 않고 정면만을 향하고 있습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 6,
     "B": 14
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 6,
   "B": 14
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 3,
    "verdict_ko": "지정된 클로즈업을 무시하고 앵글이 넓게 잡혔으며, 스마트폰을 든 팔의 해부학적 구조가 물리적으로 불가능해 보입니다."
   },
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "요청된 클로즈업 프레이밍과 스마트폰의 방향을 정확히 구현했고, 레퍼런스의 의상과 장소를 충실히 반영했습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L14B01.png"
   },
   {
    "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:796814>"
   },
   {
    "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.",
    "path": "<bytes:802115>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "우측 형사의 유니폼 조끼에 '경찰' 및 영문 텍스트가 포함되어 있어, 모든 텍스트 생성을 금지한 규칙을 위반했습니다.",
     "fix_en": "Remove all text and lettering from the police officer's uniform vest on the right."
    },
    {
     "issue_ko": "스마트폰을 쥐고 있는 팔이 프레임 하단 중앙에서 부자연스럽게 뻗어 나와, 왼쪽의 혜수와 해부학적으로 연결되지 않습니다.",
     "fix_en": "Adjust the angle of the arm holding the phone so it physically connects to Hyesu's shoulder on the left side of the frame."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all text and lettering from the police officer's uniform vest on the right.\n- Adjust the angle of the arm holding the phone so it physically connects to Hyesu's shoulder on the left side of the frame.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "요구된 클로즈업 구도와 인물 배치, 스마트폰의 각도를 충실히 구현했으며 기준 이미지의 디테일도 잘 살렸습니다."
     },
     {
      "label": "B",
      "score": 2,
      "verdict_ko": "화면 좌측 혜수의 어깨 위에 출처를 알 수 없는 손이 얹혀 있는 치명적인 신체/인원 추가 오류가 발생하여 실격입니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "요구된 클로즈업 구도와 인물 배치, 스마트폰의 각도를 충실히 구현했으며 기준 이미지의 디테일도 잘 살렸습니다."
     },
     {
      "label": "B",
      "score": 2,
      "verdict_ko": "화면 좌측 혜수의 어깨 위에 출처를 알 수 없는 손이 얹혀 있는 치명적인 신체/인원 추가 오류가 발생하여 실격입니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "인물의 어깨에 정체불명의 손이 나타나 신체 구조 및 엑스트라 신체 부위 오류(Hard Violation)가 발생하여 실격입니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 클로즈업 샷 사이즈와 혜수의 옆얼굴, 스마트폰의 방향, 배경의 경찰서 디테일이 모두 지시사항에 맞게 정확히 연출되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "인물의 어깨에 정체불명의 손이 나타나 신체 구조 및 엑스트라 신체 부위 오류(Hard Violation)가 발생하여 실격입니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요구된 클로즈업 샷 사이즈와 혜수의 옆얼굴, 스마트폰의 방향, 배경의 경찰서 디테일이 모두 지시사항에 맞게 정확히 연출되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 15,
     "B": 5
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S16sh3__bgfirst_bg.png",
   "bg_asset_id": "3a7f9388-b17a-496d-a0a2-0d51278e864e",
   "bg_record_key": "S16sh3::bgfirst_bg",
   "chain_winner": false,
   "authority": "plate"
  },
  "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L14B01",
    "B": "L14B02",
    "C": "L14B03"
   },
   "assigned": "L14B03",
   "choice": "Candidate C",
   "confident": true,
   "reason_ko": "지문이 인물들의 클로즈업 액션(스마트폰을 들이미는 모습)에만 집중하고 있으며, 특정 서브공간이나 배경 요소를 요구하지 않으므로 기존 할당된 후보를 유지함.",
   "kept": "L14B03"
  }
 },
 "S17sh1::variants": {
  "author_fp": "7aeb05b25746c826",
  "author": {
   "variants": [
    {
     "approach_ko": "수평을 단정히 잡은 절제된 광각 원경으로 건물의 고립감과 넓은 새벽 하늘을 강조한다.",
     "prompt_en": "Use a restrained rectilinear wide-lens feel from a level rooftop-height viewpoint, with clean verticals and strong negative space in the sky. Keep the architecture crisp across the depth, allowing the receding boundary wall to form a quiet lateral lead while the warm opening remains the sole restrained focal accent. Render the dawn ambience softly and naturally, preserving fine wear and texture without stylization."
    },
    {
     "approach_ko": "약한 망원 압축과 낮은 대비의 층위로 옥탑방을 새벽 하늘 아래 고요하고 외롭게 밀착시킨다.",
     "prompt_en": "Photograph from the prescribed rooftop-level distance with a gently compressed lens feel, flattening the architectural planes into calm layers beneath the dominant sky while retaining the required broad face, adjoining side, and edge wall. Favor a slightly denser, more intimate establishing image with subtle atmospheric separation rather than expansive perspective. Let the cool dawn remain low-contrast as the contained yellow glow falls off delicately across the aged exterior surfaces."
    }
   ]
  },
  "reused": false
 },
 "S17sh1": {
  "input_fingerprint": "35af25361f5be6ff",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dawn.\n\nSHOT TEXT (authoritative, Korean): 여명이 밝아오는 푸르스름한 하늘 아래, 노란 불빛이 새어 나오는 낡은 옥탑방 외부 전경.\n\nLOCATION (lock): The rooftop exterior around an old rooftop apartment at dawn, with warm light spilling from inside against the blue sky. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the flow-prescribed rooftop-level exterior position, hold a static wide view that includes the full old rooftop room in the lower portion and a broad field of dawn sky above it. Keep the lit room slightly off-center and leave the boundary wall visible along one edge so the next lateral move toward the cat has a clear spatial destination.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 낡은 옥탑방 외부 (내부 불이 켜져 있음) — A broad exterior face and part of an adjoining side are visible from the distant rooftop-level angle; used as The full exterior is the primary architectural subject and establishes the isolated rooftop location; 옥탑방의 빛이 새어 나오는 개구부 (노란 불빛이 새어 나옴) — The camera sees the yellow interior light through the opening, while interior details remain indistinct; used as The light escaping from the room provides the warm focal point beneath the dawn sky; 여명 하늘 (푸르스름하게 밝아오고 있음); used as It occupies the upper field and establishes the approach of daybreak; 옥상 담장 (옥탑방 외곽을 경계 짓고 있음) — Its upper edge recedes laterally along one side of the frame; used as The wall sits at the edge of the establishing frame as the destination of the next lateral track.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Cool bluish dawn ambience surrounds the rooftop while yellow light from the occupied room creates a restrained warm contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The old rooftop room remains lit as dawn begins.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dawn.\n\nSHOT TEXT (authoritative, Korean): 여명이 밝아오는 푸르스름한 하늘 아래, 노란 불빛이 새어 나오는 낡은 옥탑방 외부 전경.\n\nLOCATION (lock): The rooftop exterior around an old rooftop apartment at dawn, with warm light spilling from inside against the blue sky. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the flow-prescribed rooftop-level exterior position, hold a static wide view that includes the full old rooftop room in the lower portion and a broad field of dawn sky above it. Keep the lit room slightly off-center and leave the boundary wall visible along one edge so the next lateral move toward the cat has a clear spatial destination.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 낡은 옥탑방 외부 (내부 불이 켜져 있음) — A broad exterior face and part of an adjoining side are visible from the distant rooftop-level angle; used as The full exterior is the primary architectural subject and establishes the isolated rooftop location; 옥탑방의 빛이 새어 나오는 개구부 (노란 불빛이 새어 나옴) — The camera sees the yellow interior light through the opening, while interior details remain indistinct; used as The light escaping from the room provides the warm focal point beneath the dawn sky; 여명 하늘 (푸르스름하게 밝아오고 있음); used as It occupies the upper field and establishes the approach of daybreak; 옥상 담장 (옥탑방 외곽을 경계 짓고 있음) — Its upper edge recedes laterally along one side of the frame; used as The wall sits at the edge of the establishing frame as the destination of the next lateral track.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Cool bluish dawn ambience surrounds the rooftop while yellow light from the occupied room creates a restrained warm contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The old rooftop room remains lit as dawn begins.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained rectilinear wide-lens feel from a level rooftop-height viewpoint, with clean verticals and strong negative space in the sky. Keep the architecture crisp across the depth, allowing the receding boundary wall to form a quiet lateral lead while the warm opening remains the sole restrained focal accent. Render the dawn ambience softly and naturally, preserving fine wear and texture without stylization.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dawn.\n\nSHOT TEXT (authoritative, Korean): 여명이 밝아오는 푸르스름한 하늘 아래, 노란 불빛이 새어 나오는 낡은 옥탑방 외부 전경.\n\nLOCATION (lock): The rooftop exterior around an old rooftop apartment at dawn, with warm light spilling from inside against the blue sky. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the flow-prescribed rooftop-level exterior position, hold a static wide view that includes the full old rooftop room in the lower portion and a broad field of dawn sky above it. Keep the lit room slightly off-center and leave the boundary wall visible along one edge so the next lateral move toward the cat has a clear spatial destination.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 낡은 옥탑방 외부 (내부 불이 켜져 있음) — A broad exterior face and part of an adjoining side are visible from the distant rooftop-level angle; used as The full exterior is the primary architectural subject and establishes the isolated rooftop location; 옥탑방의 빛이 새어 나오는 개구부 (노란 불빛이 새어 나옴) — The camera sees the yellow interior light through the opening, while interior details remain indistinct; used as The light escaping from the room provides the warm focal point beneath the dawn sky; 여명 하늘 (푸르스름하게 밝아오고 있음); used as It occupies the upper field and establishes the approach of daybreak; 옥상 담장 (옥탑방 외곽을 경계 짓고 있음) — Its upper edge recedes laterally along one side of the frame; used as The wall sits at the edge of the establishing frame as the destination of the next lateral track.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Cool bluish dawn ambience surrounds the rooftop while yellow light from the occupied room creates a restrained warm contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The old rooftop room remains lit as dawn begins.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from the prescribed rooftop-level distance with a gently compressed lens feel, flattening the architectural planes into calm layers beneath the dominant sky while retaining the required broad face, adjoining side, and edge wall. Favor a slightly denser, more intimate establishing image with subtle atmospheric separation rather than expansive perspective. Let the cool dawn remain low-contrast as the contained yellow glow falls off delicately across the aged exterior surfaces.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 7,
   "B": 3
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "지정된 옥탑방의 구도와 푸르스름한 새벽의 조명 분위기를 잘 구현했으며, 우측 골목길 등 주변 공간의 배치를 원본에 가깝게 유지했습니다."
   },
   {
    "label": "B",
    "score": 3,
    "verdict_ko": "우측 근경에 레퍼런스에 존재하지 않는 거대한 옥상 구조물을 임의로 생성하여 공간적 맥락을 크게 훼손했습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L04B01.png"
   },
   {
    "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/background_chain/seed_bg_rooftop_villa_sel.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "옥탑방 구조물(창문, 문, 물탱크의 위치와 형태)이 STRUCTURE LOOK 사진이 아닌 LOCATION 사진을 잘못 모방했습니다.",
     "fix_en": "Rebuild the central rooftop structure to exactly match the STRUCTURE LOOK reference, placing the green door on the left-facing wall, the window on the right-facing wall, and moving the blue water tank to the rear right."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Rebuild the central rooftop structure to exactly match the STRUCTURE LOOK reference, placing the green door on the left-facing wall, the window on the right-facing wall, and moving the blue water tank to the rear right.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "레퍼런스의 구조물 형태와 회색 외벽 재질을 정확히 유지했으며, 푸른 새벽 하늘과 따뜻한 실내 조명의 대비를 지시대로 잘 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "파란색 물탱크가 두 개로 복제되었으며, 옥탑방의 외벽 색상과 문/창문의 위치가 구조물 레퍼런스와 다르게 왜곡되는 오류가 발생했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "레퍼런스의 구조물 형태와 회색 외벽 재질을 정확히 유지했으며, 푸른 새벽 하늘과 따뜻한 실내 조명의 대비를 지시대로 잘 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "파란색 물탱크가 두 개로 복제되었으며, 옥탑방의 외벽 색상과 문/창문의 위치가 구조물 레퍼런스와 다르게 왜곡되는 오류가 발생했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "레퍼런스의 구조물 형태, 회색 벽면 재질, 창문과 문의 위치 및 주변 사물을 정확히 재현함."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "구조물 색상이 베이지색으로 변형되었고 문과 창문의 위치가 레퍼런스와 다르며 물탱크가 임의로 추가됨."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "레퍼런스의 구조물 형태, 회색 벽면 재질, 창문과 문의 위치 및 주변 사물을 정확히 재현함."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "구조물 색상이 베이지색으로 변형되었고 문과 창문의 위치가 레퍼런스와 다르며 물탱크가 임의로 추가됨."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 16,
     "B": 7
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "플레이트+seed만 (배경 전용)",
  "share_plan": {
   "ref_plan": "background"
  },
  "lane_policy": "ab_select_bypass:bg_only",
  "plate_select": {
   "candidates": {
    "A": "L04B01",
    "B": "L04B02",
    "C": "L04B03",
    "D": "L04B04"
   },
   "assigned": "L04B01",
   "choice": "A",
   "confident": true,
   "reason_ko": "\"낡은 옥탑방 외부 전경\"을 요구하고 있으므로 옥탑방 전체가 조망되는 현재 할당된 컷이 적합합니다.",
   "chosen": "L04B01"
  }
 },
 "S18sh1::bgfirst_bg": {
  "input_fingerprint": "e3cfeb8d53dd66c5",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 침대 가장자리 바닥에 주저앉아 매트리스에 머리를 기댄 채 눈을 감은 수리영.\n\nLOCATION (lock): The main bedroom inside the rooftop apartment, on the floor beside the bed in early daylight.\n\nTIME OF DAY (lock): early morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From a static knee-height position beside the bed, frame 수리영 in a medium three-quarter view with her collapsed body low in the left half and the bed edge cutting diagonally through the midground. Her head rests against the mattress while the surrounding floor remains visible, holding the composition in an uneasy pause before the camera later rises with her.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: bed and mattress edge (Present beside 수리영) — The side edge faces the camera obliquely and recedes across the frame; used as Diagonal midground anchor supporting 수리영's head and defining the low camera position; bedside floor (Visible); used as Visible lower-frame context around 수리영's seated body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued daytime ambient light with restrained saturation and tense, low-key contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 침대 가장자리 바닥에 주저앉아 매트리스에 머리를 기댄 채 눈을 감은 수리영.\n\nLOCATION (lock): The main bedroom inside the rooftop apartment, on the floor beside the bed in early daylight.\n\nTIME OF DAY (lock): early morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From a static knee-height position beside the bed, frame 수리영 in a medium three-quarter view with her collapsed body low in the left half and the bed edge cutting diagonally through the midground. Her head rests against the mattress while the surrounding floor remains visible, holding the composition in an uneasy pause before the camera later rises with her.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: bed and mattress edge (Present beside 수리영) — The side edge faces the camera obliquely and recedes across the frame; used as Diagonal midground anchor supporting 수리영's head and defining the low camera position; bedside floor (Visible); used as Visible lower-frame context around 수리영's seated body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued daytime ambient light with restrained saturation and tense, low-key contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S18sh1__bgfirst_bg.png",
  "asset_id": "9c7cfa11-f401-41f5-8178-b088451a572e",
  "input_asset_ids": [
   "39bc0136-c92b-4986-9851-d68b72624c9f",
   "d89b9368-6c76-4f92-aa94-cf7a78100776"
  ]
 },
 "S18sh1": {
  "input_fingerprint": "f842d12c864fa387",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 침대 가장자리 바닥에 주저앉아 매트리스에 머리를 기댄 채 눈을 감은 수리영.\n\nLOCATION (lock): The main bedroom inside the rooftop apartment, on the floor beside the bed in early daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From a static knee-height position beside the bed, frame 수리영 in a medium three-quarter view with her collapsed body low in the left half and the bed edge cutting diagonally through the midground. Her head rests against the mattress while the surrounding floor remains visible, holding the composition in an uneasy pause before the camera later rises with her.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: bed and mattress edge (Present beside 수리영) — The side edge faces the camera obliquely and recedes across the frame; used as Diagonal midground anchor supporting 수리영's head and defining the low camera position; bedside floor (Visible); used as Visible lower-frame context around 수리영's seated body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued daytime ambient light with restrained saturation and tense, low-key contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 침대 가장자리 바닥에 주저앉아 매트리스에 머리를 기댄 채 눈을 감은 수리영.\n\nLOCATION (lock): The main bedroom inside the rooftop apartment, on the floor beside the bed in early daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From a static knee-height position beside the bed, frame 수리영 in a medium three-quarter view with her collapsed body low in the left half and the bed edge cutting diagonally through the midground. Her head rests against the mattress while the surrounding floor remains visible, holding the composition in an uneasy pause before the camera later rises with her.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: bed and mattress edge (Present beside 수리영) — The side edge faces the camera obliquely and recedes across the frame; used as Diagonal midground anchor supporting 수리영's head and defining the low camera position; bedside floor (Visible); used as Visible lower-frame context around 수리영's seated body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued daytime ambient light with restrained saturation and tense, low-key contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 침대 가장자리 바닥에 주저앉아 매트리스에 머리를 기댄 채 눈을 감은 수리영.\n\nLOCATION (lock): The main bedroom inside the rooftop apartment, on the floor beside the bed in early daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From a static knee-height position beside the bed, frame 수리영 in a medium three-quarter view with her collapsed body low in the left half and the bed edge cutting diagonally through the midground. Her head rests against the mattress while the surrounding floor remains visible, holding the composition in an uneasy pause before the camera later rises with her.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: bed and mattress edge (Present beside 수리영) — The side edge faces the camera obliquely and recedes across the frame; used as Diagonal midground anchor supporting 수리영's head and defining the low camera position; bedside floor (Visible); used as Visible lower-frame context around 수리영's seated body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued daytime ambient light with restrained saturation and tense, low-key contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S18sh1__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S18sh1.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L05B03.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "매트리스에 머리를 기댄 포즈와 요구된 무릎 높이의 미디엄 샷 프레이밍을 대각선 구도와 함께 정확하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "매트리스가 아닌 팔에 머리를 기대고 있으며, 지정된 미디엄 샷이 아닌 너무 넓은 와이드 샷으로 연출되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "매트리스에 머리를 기댄 포즈와 요구된 무릎 높이의 미디엄 샷 프레이밍을 대각선 구도와 함께 정확하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "매트리스가 아닌 팔에 머리를 기대고 있으며, 지정된 미디엄 샷이 아닌 너무 넓은 와이드 샷으로 연출되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 미디엄 샷 스케일과 대각선으로 가로지르는 침대 모서리 구도를 완벽하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "미디엄 샷 지시를 무시하고 장소 레퍼런스의 와이드 샷 카메라 프레이밍을 그대로 복사하여 크게 감점되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지정된 미디엄 샷 스케일과 대각선으로 가로지르는 침대 모서리 구도를 완벽하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "미디엄 샷 지시를 무시하고 장소 레퍼런스의 와이드 샷 카메라 프레이밍을 그대로 복사하여 크게 감점되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 7,
     "B": 15
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 7,
   "B": 15
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "매트리스에 머리를 기댄 포즈와 요구된 무릎 높이의 미디엄 샷 프레이밍을 대각선 구도와 함께 정확하게 구현했습니다."
   },
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "매트리스가 아닌 팔에 머리를 기대고 있으며, 지정된 미디엄 샷이 아닌 너무 넓은 와이드 샷으로 연출되었습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L05B03.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "캐릭터가 한쪽 발에는 운동화를 신고 있고 다른 쪽 발에는 흰 양말만 신고 있어 레퍼런스와 일치하지 않습니다.",
     "fix_en": "Change the white sock on the front foot into a sneaker that matches the one on the other foot."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the white sock on the front foot into a sneaker that matches the one on the other foot.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "제시된 앵글, 포즈, 공간 및 인물의 외형을 완벽하게 구현하였으며 물리적 오류 없이 깔끔하게 렌더링되었습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "프롬프트의 주요 요구사항은 충족했으나, 바닥에 놓인 신발이 비율상 어색하고 불필요한 복제(오브젝트 생성 오류)로 보여 감점되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "제시된 앵글, 포즈, 공간 및 인물의 외형을 완벽하게 구현하였으며 물리적 오류 없이 깔끔하게 렌더링되었습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "프롬프트의 주요 요구사항은 충족했으나, 바닥에 놓인 신발이 비율상 어색하고 불필요한 복제(오브젝트 생성 오류)로 보여 감점되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "지정된 앵글과 피사체의 자세, 배경의 디테일을 완벽하게 구현했으며 인체 해부학적 오류가 눈에 띄지 않아 사실적인 컷을 완성했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "프롬프트의 구도와 조명 조건은 잘 따랐으나, 바닥에 놓인 오른손의 손가락 형태가 심하게 왜곡되는 치명적인 해부학적 오류가 발생했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지정된 앵글과 피사체의 자세, 배경의 디테일을 완벽하게 구현했으며 인체 해부학적 오류가 눈에 띄지 않아 사실적인 컷을 완성했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "프롬프트의 구도와 조명 조건은 잘 따랐으나, 바닥에 놓인 오른손의 손가락 형태가 심하게 왜곡되는 치명적인 해부학적 오류가 발생했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 16,
     "B": 16
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": false,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S18sh1__bgfirst_bg.png",
   "bg_asset_id": "9c7cfa11-f401-41f5-8178-b088451a572e",
   "bg_record_key": "S18sh1::bgfirst_bg",
   "chain_winner": false,
   "authority": "plate"
  },
  "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L05B01",
    "B": "L05B02",
    "C": "L05B03",
    "D": "L05B04",
    "E": "L05B05",
    "F": "L05B06",
    "G": "L05B07"
   },
   "assigned": "L05B03",
   "choice": "Candidate C",
   "confident": true,
   "reason_ko": "\"침대 가장자리 바닥에 주저앉아 매트리스에 머리를 기댄 채\"라는 묘사를 통해 침대가 있는 메인 침실 공간임을 알 수 있습니다.",
   "kept": "L05B03"
  }
 },
 "S18sh12::variants": {
  "author_fp": "f91a568cda83378d",
  "author": {
   "variants": [
    {
     "approach_ko": "정면의 평면성과 타일 격자를 강조한 냉정한 증거 사진식 접근.",
     "prompt_en": "Use a restrained, low-distortion lens feel and precise frontal geometry, with the tile grid held level and the red circle commanding the center as blunt physical evidence. Keep the marked tile surface and the peripheral basin edge in crisp focus, rendering grout, glaze, wear, and the mark’s contact with the tile with unsentimental clarity under soft, subdued ambient bathroom light."
    },
    {
     "approach_ko": "붉은 원만 날카롭게 붙잡고 주변부를 부드럽게 흘려 altered perception을 강조한 접근.",
     "prompt_en": "Photograph the same locked frontal close-up with a gently compressed, claustrophobic lens feel and selective focus concentrated on the sharply defined red circle. Let the narrow tile margins and basin edge fall subtly softer without losing their identity, while restrained early-morning bathroom ambience creates delicate tonal falloff across the glazed surface and makes the deep red presence feel disturbingly immediate."
    }
   ]
  },
  "reused": false
 },
 "S18sh12": {
  "input_fingerprint": "774c249cd1bd55cb",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 세면대 옆 하얀 타일 벽면에 핏빛의 둥근 붉은 원이 선명하게 나타난 상태.\n\nLOCATION (lock): The small bathroom inside the rooftop apartment, on the tiled wall beside the sink and mirror. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level, place the camera square and perpendicular to the bathroom wall, isolating the deep-red circle near the center while retaining narrow margins of white tile and the edge of the basin for scale. The locked-off, close composition treats the impossible mark as stark visual evidence within 수리영's altered perception.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: white tiled bathroom wall with red circle (The red circle is visibly present) — The marked face is square to the camera, showing a sharply defined round red circle on the tiles; used as Primary focal surface carrying the apparition; washbasin edge (Present beside the wall) — Only the side nearest the wall enters the frame edge; used as Small peripheral scale reference beside the marked wall.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued bathroom ambient light preserves the white-versus-deep-red contrast without stylized illumination.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung now has her doll-adorned backpack with her.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 세면대 옆 하얀 타일 벽면에 핏빛의 둥근 붉은 원이 선명하게 나타난 상태.\n\nLOCATION (lock): The small bathroom inside the rooftop apartment, on the tiled wall beside the sink and mirror. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level, place the camera square and perpendicular to the bathroom wall, isolating the deep-red circle near the center while retaining narrow margins of white tile and the edge of the basin for scale. The locked-off, close composition treats the impossible mark as stark visual evidence within 수리영's altered perception.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: white tiled bathroom wall with red circle (The red circle is visibly present) — The marked face is square to the camera, showing a sharply defined round red circle on the tiles; used as Primary focal surface carrying the apparition; washbasin edge (Present beside the wall) — Only the side nearest the wall enters the frame edge; used as Small peripheral scale reference beside the marked wall.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued bathroom ambient light preserves the white-versus-deep-red contrast without stylized illumination.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung now has her doll-adorned backpack with her.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained, low-distortion lens feel and precise frontal geometry, with the tile grid held level and the red circle commanding the center as blunt physical evidence. Keep the marked tile surface and the peripheral basin edge in crisp focus, rendering grout, glaze, wear, and the mark’s contact with the tile with unsentimental clarity under soft, subdued ambient bathroom light.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 세면대 옆 하얀 타일 벽면에 핏빛의 둥근 붉은 원이 선명하게 나타난 상태.\n\nLOCATION (lock): The small bathroom inside the rooftop apartment, on the tiled wall beside the sink and mirror. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level, place the camera square and perpendicular to the bathroom wall, isolating the deep-red circle near the center while retaining narrow margins of white tile and the edge of the basin for scale. The locked-off, close composition treats the impossible mark as stark visual evidence within 수리영's altered perception.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: white tiled bathroom wall with red circle (The red circle is visibly present) — The marked face is square to the camera, showing a sharply defined round red circle on the tiles; used as Primary focal surface carrying the apparition; washbasin edge (Present beside the wall) — Only the side nearest the wall enters the frame edge; used as Small peripheral scale reference beside the marked wall.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued bathroom ambient light preserves the white-versus-deep-red contrast without stylized illumination.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung now has her doll-adorned backpack with her.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the same locked frontal close-up with a gently compressed, claustrophobic lens feel and selective focus concentrated on the sharply defined red circle. Let the narrow tile margins and basin edge fall subtly softer without losing their identity, while restrained early-morning bathroom ambience creates delicate tonal falloff across the glazed surface and makes the deep red presence feel disturbingly immediate.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 8,
   "B": 5
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 8,
    "verdict_ko": "핏빛 원의 질감이 타일과 줄눈 위에 사실적으로 표현되었으며, 요구된 프레이밍과 피사체의 스케일을 정확히 구현했습니다."
   },
   {
    "label": "B",
    "score": 5,
    "verdict_ko": "원 형태가 줄눈에서 인위적으로 끊어져 있어 그래픽 오버레이처럼 보이며, 물리적 재질감(Material Realism) 지침을 위반했습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L05B01.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "레퍼런스 사진에서 붉은 원은 세면대 바로 위 벽면에 위치하지만, 생성된 이미지에서는 세면대가 원의 오른쪽에 배치되어 공간 구조가 일치하지 않습니다.",
     "fix_en": "Move the edge of the washbasin to the bottom edge of the frame, directly below the red circle."
    },
    {
     "issue_ko": "이미지 상단 중앙에 거울 가장자리가 보이나, 레퍼런스 사진에서 거울은 왼쪽 벽면에 위치하므로 원 위에 있어서는 안 됩니다.",
     "fix_en": "Remove the mirror edge from the top center of the tiled wall."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Move the edge of the washbasin to the bottom edge of the frame, directly below the red circle.\n- Remove the mirror edge from the top center of the tiled wall.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 붉은 원을 중앙에 배치하고 세면대 가장자리만 우측에 노출하여 프레이밍과 공간적 사실성을 모두 정확히 구현함."
     },
     {
      "label": "B",
      "score": 2,
      "verdict_ko": "프레임 하단 중앙과 우측에 세면대가 중복 생성되어 공간 및 구조적 오류(하드 위반)가 발생함."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 붉은 원을 중앙에 배치하고 세면대 가장자리만 우측에 노출하여 프레이밍과 공간적 사실성을 모두 정확히 구현함."
     },
     {
      "label": "B",
      "score": 2,
      "verdict_ko": "프레임 하단 중앙과 우측에 세면대가 중복 생성되어 공간 및 구조적 오류(하드 위반)가 발생함."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "화면에 세면대가 두 개(아래쪽, 오른쪽) 생성되는 심각한 객체 중복 오류가 발생하여 실격입니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 클로즈업 구도를 잘 따랐으며, 화면 가장자리에 세면대 일부만 걸치도록 배치한 지시사항을 정확히 수행했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "화면에 세면대가 두 개(아래쪽, 오른쪽) 생성되는 심각한 객체 중복 오류가 발생하여 실격입니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요구된 클로즈업 구도를 잘 따랐으며, 화면 가장자리에 세면대 일부만 걸치도록 배치한 지시사항을 정확히 수행했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 14,
     "B": 5
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "플레이트만 (배경 전용)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L05B01",
    "B": "L05B02",
    "C": "L05B03",
    "D": "L05B04",
    "E": "L05B05",
    "F": "L05B06",
    "G": "L05B07"
   },
   "assigned": "L05B01",
   "choice": "Candidate A",
   "confident": true,
   "reason_ko": "\"세면대 옆 하얀 타일 벽면\"과 \"붉은 원\"이라는 묘사가 Candidate A의 화장실 내부 모습과 정확히 일치합니다.",
   "kept": "L05B01"
  }
 },
 "S18sh16::variants": {
  "author_fp": "b9c5bd8394cc56c6",
  "author": {
   "variants": [
    {
     "approach_ko": "압축된 망원 느낌과 얕은 초점으로 일그러진 얼굴을 폐쇄적이고 불길하게 고립한다.",
     "prompt_en": "Use a compressed, claustrophobic lens feel from the locked frontal position, holding the distorted face with exact central emphasis while the narrowly visible frame encloses it like a trap. Keep focus critically sharp across the smeared photographic surface, with the thin table reference and outer frame edges falling gently softer; subdued, broad early-morning ambient light and restrained desaturated tones make the warped image feel tactile and physically present."
    },
    {
     "approach_ko": "근접한 정상 원근과 부드러운 측면광으로 액자와 인화면의 물성을 냉정하게 드러낸다.",
     "prompt_en": "Use an intimate close-focus perspective that preserves the photograph as a real object on the table rather than flattening it into a graphic portrait. Maintain the locked face-on symmetry and narrow frame border, with slightly deeper focus carrying crisp detail through the frame material, image-bearing surface, and thin strip of tabletop; let soft directional early-morning ambient light skim gently across the surfaces, revealing minute texture and irregular deformation without any supernatural glow."
    }
   ]
  },
  "reused": false
 },
 "S18sh16": {
  "input_fingerprint": "6f6858b25a3d690d",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 식탁 위 액자 속 강민숙(민숙)의 얼굴 사진이 뭉개진 채 기괴하게 얼룩진 상태 클로즈업.\n\nLOCATION (lock): The dining area of the rooftop apartment, centered on a framed family photograph resting on the table. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold just above table height at the closest point of the frontal push-in, with the distorted face of 강민숙 (사진 왜곡 상태) centered and the photograph frame still narrowly legible around it. Keep the table as a thin lower reference so the face remains an image inside a physical frame rather than becoming a disembodied portrait.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: distorted portrait of 강민숙 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: framed photograph (The photographed face remains visibly distorted) — Its image-bearing front faces the camera directly, showing 강민숙's smiling face smeared and flowing out of shape; used as Primary content plane containing the supernatural facial distortion; dining table surface (The photograph rests on it); used as Thin lower-frame spatial reference beneath the photograph.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient exposure and desaturated contrast keep the distortion tactile without assigning it an artificial light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung still has her doll-adorned backpack; the bathroom circle has already vanished, while the framed photograph's distortion is only momentary and returns to normal.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 식탁 위 액자 속 강민숙(민숙)의 얼굴 사진이 뭉개진 채 기괴하게 얼룩진 상태 클로즈업.\n\nLOCATION (lock): The dining area of the rooftop apartment, centered on a framed family photograph resting on the table. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold just above table height at the closest point of the frontal push-in, with the distorted face of 강민숙 (사진 왜곡 상태) centered and the photograph frame still narrowly legible around it. Keep the table as a thin lower reference so the face remains an image inside a physical frame rather than becoming a disembodied portrait.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: distorted portrait of 강민숙 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: framed photograph (The photographed face remains visibly distorted) — Its image-bearing front faces the camera directly, showing 강민숙's smiling face smeared and flowing out of shape; used as Primary content plane containing the supernatural facial distortion; dining table surface (The photograph rests on it); used as Thin lower-frame spatial reference beneath the photograph.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient exposure and desaturated contrast keep the distortion tactile without assigning it an artificial light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung still has her doll-adorned backpack; the bathroom circle has already vanished, while the framed photograph's distortion is only momentary and returns to normal.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a compressed, claustrophobic lens feel from the locked frontal position, holding the distorted face with exact central emphasis while the narrowly visible frame encloses it like a trap. Keep focus critically sharp across the smeared photographic surface, with the thin table reference and outer frame edges falling gently softer; subdued, broad early-morning ambient light and restrained desaturated tones make the warped image feel tactile and physically present.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 식탁 위 액자 속 강민숙(민숙)의 얼굴 사진이 뭉개진 채 기괴하게 얼룩진 상태 클로즈업.\n\nLOCATION (lock): The dining area of the rooftop apartment, centered on a framed family photograph resting on the table. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold just above table height at the closest point of the frontal push-in, with the distorted face of 강민숙 (사진 왜곡 상태) centered and the photograph frame still narrowly legible around it. Keep the table as a thin lower reference so the face remains an image inside a physical frame rather than becoming a disembodied portrait.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: distorted portrait of 강민숙 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: framed photograph (The photographed face remains visibly distorted) — Its image-bearing front faces the camera directly, showing 강민숙's smiling face smeared and flowing out of shape; used as Primary content plane containing the supernatural facial distortion; dining table surface (The photograph rests on it); used as Thin lower-frame spatial reference beneath the photograph.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient exposure and desaturated contrast keep the distortion tactile without assigning it an artificial light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung still has her doll-adorned backpack; the bathroom circle has already vanished, while the framed photograph's distortion is only momentary and returns to normal.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an intimate close-focus perspective that preserves the photograph as a real object on the table rather than flattening it into a graphic portrait. Maintain the locked face-on symmetry and narrow frame border, with slightly deeper focus carrying crisp detail through the frame material, image-bearing surface, and thin strip of tabletop; let soft directional early-morning ambient light skim gently across the surfaces, revealing minute texture and irregular deformation without any supernatural glow.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 5,
   "B": 7
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 5,
    "verdict_ko": "요구된 '웃는 얼굴' 대신 무표정으로 묘사되었으며, 기준 사진의 공간 구조(화장실과 발코니 문 방향)와 배경이 일치하지 않습니다."
   },
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "기괴하게 얼룩진 웃는 얼굴과 배경의 공간 구조를 정확히 구현했으나, 지시문에서 사라졌어야 할 화장실의 붉은 원이 남아있는 점이 감점 요인입니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L05B01.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "배경 우측 욕실 벽에 붉은색 원 자국이 여전히 희미하게 보입니다 (지시문에서는 이미 사라졌다고 명시됨).",
     "fix_en": "Remove the faint red circular markings from the tiled bathroom wall in the background."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the faint red circular markings from the tiled bathroom wall in the background.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "요구된 클로즈업 구도와 사진 왜곡을 훌륭하게 구현했으며, 화장실의 붉은 원이 사라져야 한다는 세부 지침까지 정확히 준수했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "구도와 피사체의 표현은 우수하나, 배경 화장실에 사라져야 할 붉은 원이 여전히 남아있어 지침을 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "요구된 클로즈업 구도와 사진 왜곡을 훌륭하게 구현했으며, 화장실의 붉은 원이 사라져야 한다는 세부 지침까지 정확히 준수했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "구도와 피사체의 표현은 우수하나, 배경 화장실에 사라져야 할 붉은 원이 여전히 남아있어 지침을 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "프레이밍과 사진의 왜곡 효과가 훌륭하며, 프롬프트의 지시대로 배경 화장실의 붉은 원을 정확히 제거했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "사진의 왜곡과 프레이밍은 좋으나, 프롬프트 지시와 달리 배경 화장실 벽에 붉은 원이 여전히 남아 있어 감점되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "프레이밍과 사진의 왜곡 효과가 훌륭하며, 프롬프트의 지시대로 배경 화장실의 붉은 원을 정확히 제거했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "사진의 왜곡과 프레이밍은 좋으나, 프롬프트 지시와 달리 배경 화장실 벽에 붉은 원이 여전히 남아 있어 감점되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 10,
     "B": 16
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "플레이트만 (배경 전용)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L05B01",
    "B": "L05B02",
    "C": "L05B03",
    "D": "L05B04",
    "E": "L05B05",
    "F": "L05B06",
    "G": "L05B07"
   },
   "assigned": "L05B01",
   "choice": "A",
   "confident": true,
   "reason_ko": "\"식탁 위 액자\"라는 지문에 정확히 부합하게 식탁 위에 가족사진 액자가 놓여 있는 유일한 앵글입니다.",
   "chosen": "L05B01"
  }
 },
 "S19sh1::bgfirst_bg": {
  "input_fingerprint": "4c560c3a1e9810b4",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 이른 아침 한산한 마트 입구 앞, 자전거 핸들을 양손으로 쥔 채 서 있는 수리영의 전신.\n\nLOCATION (lock): The quiet exterior forecourt directly in front of a large supermarket entrance in the early morning.\n\nTIME OF DAY (lock): early morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level directly outside the store entrance, hold a wide frontal view with 수리영 and the bicycle centered, her full body readable behind the handlebars and the entrance opening held in the background. Leave clear space behind her along the entry axis so the eventual backward retreat into the store feels continuous.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: store entrance (Open for entry) — The entrance faces the camera along the same axis as 수리영 and the bicycle; used as Background destination centered behind 수리영 and establishing the next camera path; bicycle (Stationary before being parked) — The handlebar and front of the bicycle face toward the camera; used as Foreground-midground physical anchor held by 수리영.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Restrained early-morning ambient light with muted color and subdued contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 이른 아침 한산한 마트 입구 앞, 자전거 핸들을 양손으로 쥔 채 서 있는 수리영의 전신.\n\nLOCATION (lock): The quiet exterior forecourt directly in front of a large supermarket entrance in the early morning.\n\nTIME OF DAY (lock): early morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level directly outside the store entrance, hold a wide frontal view with 수리영 and the bicycle centered, her full body readable behind the handlebars and the entrance opening held in the background. Leave clear space behind her along the entry axis so the eventual backward retreat into the store feels continuous.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: store entrance (Open for entry) — The entrance faces the camera along the same axis as 수리영 and the bicycle; used as Background destination centered behind 수리영 and establishing the next camera path; bicycle (Stationary before being parked) — The handlebar and front of the bicycle face toward the camera; used as Foreground-midground physical anchor held by 수리영.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Restrained early-morning ambient light with muted color and subdued contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S19sh1__bgfirst_bg.png",
  "asset_id": "6cd940bf-98f9-45e2-bf93-6d3fc8fef3be",
  "input_asset_ids": [
   "dcaee74b-fb53-4e59-9650-f2ba7de921a1",
   "ddd4463c-af4a-48e3-bdea-1f1e845d03e2",
   "40c6cf7b-afda-4940-a200-18e70c018255"
  ]
 },
 "S19sh1": {
  "input_fingerprint": "4f5265a6fcd44088",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 이른 아침 한산한 마트 입구 앞, 자전거 핸들을 양손으로 쥔 채 서 있는 수리영의 전신.\n\nLOCATION (lock): The quiet exterior forecourt directly in front of a large supermarket entrance in the early morning. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level directly outside the store entrance, hold a wide frontal view with 수리영 and the bicycle centered, her full body readable behind the handlebars and the entrance opening held in the background. Leave clear space behind her along the entry axis so the eventual backward retreat into the store feels continuous.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: store entrance (Open for entry) — The entrance faces the camera along the same axis as 수리영 and the bicycle; used as Background destination centered behind 수리영 and establishing the next camera path; bicycle (Stationary before being parked) — The handlebar and front of the bicycle face toward the camera; used as Foreground-midground physical anchor held by 수리영.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Restrained early-morning ambient light with muted color and subdued contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung continues to carry the doll-adorned backpack she took from home.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 이른 아침 한산한 마트 입구 앞, 자전거 핸들을 양손으로 쥔 채 서 있는 수리영의 전신.\n\nLOCATION (lock): The quiet exterior forecourt directly in front of a large supermarket entrance in the early morning. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level directly outside the store entrance, hold a wide frontal view with 수리영 and the bicycle centered, her full body readable behind the handlebars and the entrance opening held in the background. Leave clear space behind her along the entry axis so the eventual backward retreat into the store feels continuous.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: store entrance (Open for entry) — The entrance faces the camera along the same axis as 수리영 and the bicycle; used as Background destination centered behind 수리영 and establishing the next camera path; bicycle (Stationary before being parked) — The handlebar and front of the bicycle face toward the camera; used as Foreground-midground physical anchor held by 수리영.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Restrained early-morning ambient light with muted color and subdued contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung continues to carry the doll-adorned backpack she took from home.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 이른 아침 한산한 마트 입구 앞, 자전거 핸들을 양손으로 쥔 채 서 있는 수리영의 전신.\n\nLOCATION (lock): The quiet exterior forecourt directly in front of a large supermarket entrance in the early morning. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level directly outside the store entrance, hold a wide frontal view with 수리영 and the bicycle centered, her full body readable behind the handlebars and the entrance opening held in the background. Leave clear space behind her along the entry axis so the eventual backward retreat into the store feels continuous.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: store entrance (Open for entry) — The entrance faces the camera along the same axis as 수리영 and the bicycle; used as Background destination centered behind 수리영 and establishing the next camera path; bicycle (Stationary before being parked) — The handlebar and front of the bicycle face toward the camera; used as Foreground-midground physical anchor held by 수리영.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Restrained early-morning ambient light with muted color and subdued contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung continues to carry the doll-adorned backpack she took from home.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S19sh1__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S19sh1.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L10B01.png"
    },
    {
     "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/background_chain/seed_bg_large_market_sel.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "정면을 향한 자전거의 구도와 인형이 달린 백팩의 묘사가 지시사항에 부합하며, 마트 입구 뒤쪽의 후퇴 동선이 더 명확히 확보되었습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "자전거의 각도가 정면을 향하지 않고 프레임 뒤쪽이 부자연스럽게 생략되었으며, 입구 축에 원래 있던 자전거가 남아있어 동선이 다소 가려집니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "정면을 향한 자전거의 구도와 인형이 달린 백팩의 묘사가 지시사항에 부합하며, 마트 입구 뒤쪽의 후퇴 동선이 더 명확히 확보되었습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "자전거의 각도가 정면을 향하지 않고 프레임 뒤쪽이 부자연스럽게 생략되었으며, 입구 축에 원래 있던 자전거가 남아있어 동선이 다소 가려집니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지침에 따라 STRUCTURE LOOK의 평면적인 건물 외관을 정확히 반영했으며, 중앙에 정렬된 피사체와 활짝 열린 입구, 인형이 달린 백팩을 훌륭하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "STRUCTURE LOOK을 우선하지 않고 LOCATION 사진의 모서리 입구를 그대로 사용했으며, 백팩의 인형 장식도 명확하게 보이지 않습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지침에 따라 STRUCTURE LOOK의 평면적인 건물 외관을 정확히 반영했으며, 중앙에 정렬된 피사체와 활짝 열린 입구, 인형이 달린 백팩을 훌륭하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "STRUCTURE LOOK을 우선하지 않고 LOCATION 사진의 모서리 입구를 그대로 사용했으며, 백팩의 인형 장식도 명확하게 보이지 않습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 15,
     "B": 11
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 15,
   "B": 11
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "정면을 향한 자전거의 구도와 인형이 달린 백팩의 묘사가 지시사항에 부합하며, 마트 입구 뒤쪽의 후퇴 동선이 더 명확히 확보되었습니다."
   },
   {
    "label": "B",
    "score": 5,
    "verdict_ko": "자전거의 각도가 정면을 향하지 않고 프레임 뒤쪽이 부자연스럽게 생략되었으며, 입구 축에 원래 있던 자전거가 남아있어 동선이 다소 가려집니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L10B01.png"
   },
   {
    "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/background_chain/seed_bg_large_market_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "자전거의 앞바퀴 중심 축에 페달이 잘못 달려 있습니다.",
     "fix_en": "Remove the pedals protruding from the center axle of the bicycle's front wheel."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the pedals protruding from the center axle of the bicycle's front wheel.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 샷 크기와 배경을 완벽히 유지했으며, 레퍼런스와 흡사한 얼굴 및 인형 달린 가방 등 소품의 형태가 자연스럽습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "스케치와 배경 지침을 충실히 따랐으나, 자전거 핸들부의 구조와 손의 형태가 A에 비해 다소 덜 정교합니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 샷 크기와 배경을 완벽히 유지했으며, 레퍼런스와 흡사한 얼굴 및 인형 달린 가방 등 소품의 형태가 자연스럽습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "스케치와 배경 지침을 충실히 따랐으나, 자전거 핸들부의 구조와 손의 형태가 A에 비해 다소 덜 정교합니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 샷의 구도, 인물의 배치 및 의상, 인형이 달린 가방 등 프롬프트의 요구사항을 물리적 오류 없이 충실히 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "구도와 인물 묘사는 양호하나, 자전거 앞바퀴 축에 페달이 달려 있는 물리적으로 불가능한 구조(치명적 위반)가 발견되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 샷의 구도, 인물의 배치 및 의상, 인형이 달린 가방 등 프롬프트의 요구사항을 물리적 오류 없이 충실히 구현했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "구도와 인물 묘사는 양호하나, 자전거 앞바퀴 축에 페달이 달려 있는 물리적으로 불가능한 구조(치명적 위반)가 발견되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 11,
     "B": 14
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": false,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S19sh1__bgfirst_bg.png",
   "bg_asset_id": "6cd940bf-98f9-45e2-bf93-6d3fc8fef3be",
   "bg_record_key": "S19sh1::bgfirst_bg",
   "chain_winner": true,
   "authority": "plate",
   "seed_attached": true
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "lane_policy": "ab_select_ready",
  "plate_select": {
   "candidates": {
    "A": "L10B01",
    "B": "L10B02"
   },
   "assigned": "L10B01",
   "choice": "Candidate A",
   "confident": true,
   "reason_ko": "\"수리영의 전신\"을 담기 위해서는 인물 전체가 들어갈 수 있는 넓은 화각의 마트 입구 앞 배경(Candidate A)이 필요합니다.",
   "kept": "L10B01"
  }
 },
 "S19sh4::variants": {
  "author_fp": "4f5b8d35b3465653",
  "author": {
   "variants": [
    {
     "approach_ko": "넓은 공간감과 깊은 초점으로 손짓, CCTV, 처마 구조의 감시 동선을 한눈에 읽히게 한다.",
     "prompt_en": "Use a natural wide-normal perspective with controlled deep focus, keeping the pointing gesture, mounted camera, and surrounding storefront architecture crisply legible within one continuous space. Let the awning’s structural lines reinforce the upward diagonal without distortion, while cool early-morning ambience reveals worn surfaces with restrained color and firm, quiet contrast."
    },
    {
     "approach_ko": "망원 압축과 CCTV 중심의 선택적 초점으로 수리영이 감시 장치를 발견한 순간을 긴장감 있게 강조한다.",
     "prompt_en": "Use a restrained long-lens feel to compress the distance between her raised hand and the CCTV while preserving their specified scale and placement. Hold critical focus on the camera housing and pointing fingertip, allowing her back and the farther storefront planes to fall gently softer; shape the subdued early-morning light so the fixture catches a slight metallic edge against the darker awning underside."
    }
   ]
  },
  "reused": false
 },
 "S19sh4": {
  "input_fingerprint": "2202f3f7166f4cd6",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 밖 처마 밑에 매달린 CCTV 카메라를 손가락으로 가리킨 채 멈춘 수리영의 뒷모습.\n\nLOCATION (lock): Outside the supermarket beneath the entrance canopy, directly below a mounted security camera. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From shoulder height behind 수리영 beneath the awning, frame her back in the lower center as her raised hand creates a diagonal toward the CCTV camera above. Keep enough of the store exterior around both anchors to make the camera's coverage position immediately understandable before the planned arc toward her front.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-center of the frame, midground, points to CCTV camera; CCTV camera in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: CCTV camera (Mounted beneath the awning) — Its camera-facing housing and viewing direction are visible beneath the awning, oriented toward the area being checked; used as Upper-frame target of 수리영's pointing gesture; store awning (Present above the store exterior) — Its underside faces the camera and carries the CCTV camera; used as Overhead frame connecting 수리영 and the mounted camera.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural early-morning exterior ambience is rendered with restrained saturation and firm, subdued contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the quiet early-morning supermarket entrance, facade, pavement, and cool daylight from the reference. EXCLUDE the bicycle-centered frontal pose; frame the young woman from behind as she points toward the fixed CCTV camera under the awning.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung still has her doll-adorned backpack with her outside the mart.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 밖 처마 밑에 매달린 CCTV 카메라를 손가락으로 가리킨 채 멈춘 수리영의 뒷모습.\n\nLOCATION (lock): Outside the supermarket beneath the entrance canopy, directly below a mounted security camera. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From shoulder height behind 수리영 beneath the awning, frame her back in the lower center as her raised hand creates a diagonal toward the CCTV camera above. Keep enough of the store exterior around both anchors to make the camera's coverage position immediately understandable before the planned arc toward her front.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-center of the frame, midground, points to CCTV camera; CCTV camera in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: CCTV camera (Mounted beneath the awning) — Its camera-facing housing and viewing direction are visible beneath the awning, oriented toward the area being checked; used as Upper-frame target of 수리영's pointing gesture; store awning (Present above the store exterior) — Its underside faces the camera and carries the CCTV camera; used as Overhead frame connecting 수리영 and the mounted camera.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural early-morning exterior ambience is rendered with restrained saturation and firm, subdued contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the quiet early-morning supermarket entrance, facade, pavement, and cool daylight from the reference. EXCLUDE the bicycle-centered frontal pose; frame the young woman from behind as she points toward the fixed CCTV camera under the awning.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung still has her doll-adorned backpack with her outside the mart.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a natural wide-normal perspective with controlled deep focus, keeping the pointing gesture, mounted camera, and surrounding storefront architecture crisply legible within one continuous space. Let the awning’s structural lines reinforce the upward diagonal without distortion, while cool early-morning ambience reveals worn surfaces with restrained color and firm, quiet contrast.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 밖 처마 밑에 매달린 CCTV 카메라를 손가락으로 가리킨 채 멈춘 수리영의 뒷모습.\n\nLOCATION (lock): Outside the supermarket beneath the entrance canopy, directly below a mounted security camera. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From shoulder height behind 수리영 beneath the awning, frame her back in the lower center as her raised hand creates a diagonal toward the CCTV camera above. Keep enough of the store exterior around both anchors to make the camera's coverage position immediately understandable before the planned arc toward her front.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-center of the frame, midground, points to CCTV camera; CCTV camera in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: CCTV camera (Mounted beneath the awning) — Its camera-facing housing and viewing direction are visible beneath the awning, oriented toward the area being checked; used as Upper-frame target of 수리영's pointing gesture; store awning (Present above the store exterior) — Its underside faces the camera and carries the CCTV camera; used as Overhead frame connecting 수리영 and the mounted camera.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural early-morning exterior ambience is rendered with restrained saturation and firm, subdued contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the quiet early-morning supermarket entrance, facade, pavement, and cool daylight from the reference. EXCLUDE the bicycle-centered frontal pose; frame the young woman from behind as she points toward the fixed CCTV camera under the awning.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung still has her doll-adorned backpack with her outside the mart.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained long-lens feel to compress the distance between her raised hand and the CCTV while preserving their specified scale and placement. Hold critical focus on the camera housing and pointing fingertip, allowing her back and the farther storefront planes to fall gently softer; shape the subdued early-morning light so the fixture catches a slight metallic edge against the darker awning underside.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 8,
   "B": 5
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 8,
    "verdict_ko": "지정된 숄더 뷰와 미디엄 샷 구도를 잘 살렸으며, 레퍼런스의 건축 양식과 인물의 지시 동작을 충실히 구현했습니다."
   },
   {
    "label": "B",
    "score": 5,
    "verdict_ko": "구도와 인물의 포즈는 양호하나, 원본 레퍼런스 건물에는 없는 기둥과 슬레이트 구조물을 우측에 임의로 추가하여 장소의 일관성을 크게 훼손했습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S19sh1_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "기준 이미지에 없는 거대한 콘크리트 캐노피 구조물이 건물 우측에 임의로 추가되어 기존 건물의 외관을 훼손했습니다.",
     "fix_en": "Remove the protruding concrete canopy on the right and mount the CCTV camera directly under the existing gray architectural band above the entrance."
    },
    {
     "issue_ko": "CCTV를 가리키는 오른손의 형태가 뭉개져 손등과 나머지 손가락의 구분이 없는 기형적인 모습입니다.",
     "fix_en": "Redraw the right hand to clearly show a proper anatomical structure with an extended index finger and correctly folded remaining fingers."
    },
    {
     "issue_ko": "기준 이미지에서 가방의 앞쪽 어깨끈에 달려있던 인형들이 가방 뒷면 본체 우측에 달려 있습니다.",
     "fix_en": "Move the hanging dolls from the back of the backpack to the front left shoulder strap."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the protruding concrete canopy on the right and mount the CCTV camera directly under the existing gray architectural band above the entrance.\n- Redraw the right hand to clearly show a proper anatomical structure with an extended index finger and correctly folded remaining fingers.\n- Move the hanging dolls from the back of the backpack to the front left shoulder strap.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "인물의 포즈와 샷 크기가 훌륭하며, CCTV가 처마 밑(underside)에 매달려 있다는 지시사항을 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "전반적인 구도와 포즈는 좋으나, CCTV가 처마 아래가 아닌 수직 외벽에 부착되어 위치 관계 지시를 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "인물의 포즈와 샷 크기가 훌륭하며, CCTV가 처마 밑(underside)에 매달려 있다는 지시사항을 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "전반적인 구도와 포즈는 좋으나, CCTV가 처마 아래가 아닌 수직 외벽에 부착되어 위치 관계 지시를 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "레퍼런스의 건축 구조와 형태를 그대로 유지하면서 인물의 뒷모습, 지시하는 동작, 가방의 인형, 그리고 앵글을 충실하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "처마 아래를 묘사하기 위해 레퍼런스 이미지에는 존재하지 않는 거대한 콘크리트 구조물을 화면 우측 상단에 임의로 생성하여 장소 일관성 조건을 심각하게 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "레퍼런스의 건축 구조와 형태를 그대로 유지하면서 인물의 뒷모습, 지시하는 동작, 가방의 인형, 그리고 앵글을 충실하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "처마 아래를 묘사하기 위해 레퍼런스 이미지에는 존재하지 않는 거대한 콘크리트 구조물을 화면 우측 상단에 임의로 생성하여 장소 일관성 조건을 심각하게 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 10,
     "B": 12
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": false,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S19sh1"
  },
  "lane_policy": "ab_select_bypass:prev"
 },
 "S20sh3::bgfirst_bg": {
  "input_fingerprint": "a02122215d68fea6",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 흐릿한 모니터 화면 불빛에 비친 수리영의 불안하게 확장된 눈동자 클로즈업.\n\nLOCATION (lock): A supermarket administration and security office, immediately behind the desk holding the CCTV monitor. The dim footage glows from the screen.\n\nTIME OF DAY (lock): early morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Continue the dolly directly along 수리영's eyeline until only her widened eyes and the bridge of her nose remain sharp, centered in a level extreme close view. The monitor image stays indistinct, registered only as unstable luminance across her eyes so the frame isolates her escalating anxiety rather than the footage.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영's widened eyes in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: CCTV monitor display (On, replaying footage with an unusually dark section) — The active image-bearing face is directed toward 수리영 and remains outside focus from the camera's position; used as Off-frame source of the indistinct luminance reflected across 수리영's eyes.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Indistinct monitor illumination grazes her widened eyes within a dim, desaturated, high-tension exposure.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 흐릿한 모니터 화면 불빛에 비친 수리영의 불안하게 확장된 눈동자 클로즈업.\n\nLOCATION (lock): A supermarket administration and security office, immediately behind the desk holding the CCTV monitor. The dim footage glows from the screen.\n\nTIME OF DAY (lock): early morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Continue the dolly directly along 수리영's eyeline until only her widened eyes and the bridge of her nose remain sharp, centered in a level extreme close view. The monitor image stays indistinct, registered only as unstable luminance across her eyes so the frame isolates her escalating anxiety rather than the footage.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영's widened eyes in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: CCTV monitor display (On, replaying footage with an unusually dark section) — The active image-bearing face is directed toward 수리영 and remains outside focus from the camera's position; used as Off-frame source of the indistinct luminance reflected across 수리영's eyes.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Indistinct monitor illumination grazes her widened eyes within a dim, desaturated, high-tension exposure.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S20sh3__bgfirst_bg.png",
  "asset_id": "8e2c4490-836f-40bb-8809-579b6d240c03",
  "input_asset_ids": [
   "1784ef17-cda3-46f8-9961-6d0ec3dd9ae5",
   "62240680-2dc0-4031-84cf-68bca60caafe"
  ]
 },
 "S20sh3": {
  "input_fingerprint": "6845aa554aa3b901",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 흐릿한 모니터 화면 불빛에 비친 수리영의 불안하게 확장된 눈동자 클로즈업.\n\nLOCATION (lock): A supermarket administration and security office, immediately behind the desk holding the CCTV monitor. The dim footage glows from the screen. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Continue the dolly directly along 수리영's eyeline until only her widened eyes and the bridge of her nose remain sharp, centered in a level extreme close view. The monitor image stays indistinct, registered only as unstable luminance across her eyes so the frame isolates her escalating anxiety rather than the footage.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영's widened eyes in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: CCTV monitor display (On, replaying footage with an unusually dark section) — The active image-bearing face is directed toward 수리영 and remains outside focus from the camera's position; used as Off-frame source of the indistinct luminance reflected across 수리영's eyes.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Indistinct monitor illumination grazes her widened eyes within a dim, desaturated, high-tension exposure.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung continues to have the doll-adorned backpack with her while watching the CCTV footage.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 흐릿한 모니터 화면 불빛에 비친 수리영의 불안하게 확장된 눈동자 클로즈업.\n\nLOCATION (lock): A supermarket administration and security office, immediately behind the desk holding the CCTV monitor. The dim footage glows from the screen. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Continue the dolly directly along 수리영's eyeline until only her widened eyes and the bridge of her nose remain sharp, centered in a level extreme close view. The monitor image stays indistinct, registered only as unstable luminance across her eyes so the frame isolates her escalating anxiety rather than the footage.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영's widened eyes in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: CCTV monitor display (On, replaying footage with an unusually dark section) — The active image-bearing face is directed toward 수리영 and remains outside focus from the camera's position; used as Off-frame source of the indistinct luminance reflected across 수리영's eyes.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Indistinct monitor illumination grazes her widened eyes within a dim, desaturated, high-tension exposure.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung continues to have the doll-adorned backpack with her while watching the CCTV footage.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 흐릿한 모니터 화면 불빛에 비친 수리영의 불안하게 확장된 눈동자 클로즈업.\n\nLOCATION (lock): A supermarket administration and security office, immediately behind the desk holding the CCTV monitor. The dim footage glows from the screen. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Continue the dolly directly along 수리영's eyeline until only her widened eyes and the bridge of her nose remain sharp, centered in a level extreme close view. The monitor image stays indistinct, registered only as unstable luminance across her eyes so the frame isolates her escalating anxiety rather than the footage.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영's widened eyes in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: CCTV monitor display (On, replaying footage with an unusually dark section) — The active image-bearing face is directed toward 수리영 and remains outside focus from the camera's position; used as Off-frame source of the indistinct luminance reflected across 수리영's eyes.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Indistinct monitor illumination grazes her widened eyes within a dim, desaturated, high-tension exposure.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung continues to have the doll-adorned backpack with her while watching the CCTV footage.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S20sh3__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S20sh3.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L11B01.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "지시된 '익스트림 클로즈업' 프레이밍(눈과 콧대만 강조)과 어둡고 긴장감 넘치는 조명 분위기를 매우 정확하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "눈이 확장된 연기와 모니터 반사는 좋으나 프레이밍이 덜 타이트하여 얼굴의 넓은 면적과 배경이 과도하게 노출되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "지시된 '익스트림 클로즈업' 프레이밍(눈과 콧대만 강조)과 어둡고 긴장감 넘치는 조명 분위기를 매우 정확하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "눈이 확장된 연기와 모니터 반사는 좋으나 프레이밍이 덜 타이트하여 얼굴의 넓은 면적과 배경이 과도하게 노출되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "지시된 익스트림 클로즈업 앵글을 정확히 구현하여 눈동자와 콧등만 프레임에 남기고, 요구된 모니터 불빛의 반사광을 훌륭하게 표현했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "요구된 익스트림 클로즈업보다 샷 크기가 넓어 지시와 달리 뒷배경과 얼굴의 다른 부위가 너무 많이 노출되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "지시된 익스트림 클로즈업 앵글을 정확히 구현하여 눈동자와 콧등만 프레임에 남기고, 요구된 모니터 불빛의 반사광을 훌륭하게 표현했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "요구된 익스트림 클로즈업보다 샷 크기가 넓어 지시와 달리 뒷배경과 얼굴의 다른 부위가 너무 많이 노출되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 13,
     "B": 20
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 13,
   "B": 20
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 10,
    "verdict_ko": "지시된 '익스트림 클로즈업' 프레이밍(눈과 콧대만 강조)과 어둡고 긴장감 넘치는 조명 분위기를 매우 정확하게 구현했습니다."
   },
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "눈이 확장된 연기와 모니터 반사는 좋으나 프레이밍이 덜 타이트하여 얼굴의 넓은 면적과 배경이 과도하게 노출되었습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L11B01.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "눈동자에 비친 반사광이 방이나 창문의 구체적인 형태를 보일 정도로 너무 선명합니다. 프롬프트는 화면이 형태가 불분명한 불안정한 빛(indistinct, unstable luminance)으로만 표현되어야 한다고 지시했습니다.",
     "fix_en": "Blur the sharp reflections in the eyes into an indistinct, hazy glow of light without any recognizable shapes or details."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Blur the sharp reflections in the eyes into an indistinct, hazy glow of light without any recognizable shapes or details.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "정상적인 사람의 동공과 홍채 구조를 유지한 채 모니터 불빛이 반사되는 불안한 눈동자를 익스트림 클로즈업으로 정확하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 2,
      "verdict_ko": "눈동자를 비정상적으로 하얗게 변형시키지 말라는 해부학적 제약 지시를 정면으로 위반하여 눈을 기괴하게 렌더링했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "정상적인 사람의 동공과 홍채 구조를 유지한 채 모니터 불빛이 반사되는 불안한 눈동자를 익스트림 클로즈업으로 정확하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 2,
      "verdict_ko": "눈동자를 비정상적으로 하얗게 변형시키지 말라는 해부학적 제약 지시를 정면으로 위반하여 눈을 기괴하게 렌더링했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 0,
      "verdict_ko": "눈동자가 하얗게 변색되어 해부학적으로 비정상적이므로 'EXPRESSIONS ARE ACTED, NEVER ANATOMICAL' 규칙을 심각하게 위반했습니다."
     },
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "요구된 익스트림 클로즈업 앵글을 완벽하게 구현했으며, 자연스러운 눈동자 안에 모니터 불빛이 비치는 모습을 지시사항에 맞게 잘 표현했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 0,
      "verdict_ko": "눈동자가 하얗게 변색되어 해부학적으로 비정상적이므로 'EXPRESSIONS ARE ACTED, NEVER ANATOMICAL' 규칙을 심각하게 위반했습니다."
     },
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "요구된 익스트림 클로즈업 앵글을 완벽하게 구현했으며, 자연스러운 눈동자 안에 모니터 불빛이 비치는 모습을 지시사항에 맞게 잘 표현했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 20,
     "B": 2
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S20sh3__bgfirst_bg.png",
   "bg_asset_id": "8e2c4490-836f-40bb-8809-579b6d240c03",
   "bg_record_key": "S20sh3::bgfirst_bg",
   "chain_winner": false,
   "authority": "plate"
  },
  "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S21sh3::bgfirst_bg": {
  "input_fingerprint": "843f432e7b665936",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 양손으로 현장 사진을 꽉 쥔 채 공포에 질려 입을 크게 벌린 수리영의 얼굴 정면.\n\nLOCATION (lock): Inside the police office at a detective’s desk, where enlarged crime-scene photographs are spread across the work surface.\n\nTIME OF DAY (lock): late morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level directly across the desk, finish the frontal push-in with 수리영's terrified face occupying most of the frame and both hands crushing the photographs along the lower edge. Preserve a narrow strip of desk and overlapping image corners as context, while her open mouth and rigid grip form the emotional center.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: crime-scene photographs (Spread across the desk and gripped in both hands) — Their image-bearing faces angle upward toward 수리영 and partly toward the camera, revealing enlarged images of severe wounds; used as Lower-foreground trigger for 수리영's reaction, partially visible between her hands; desk surface (Covered with scattered enlarged photographs); used as Narrow lower-frame support for the photographs and hands.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued indoor daytime ambience uses desaturated contrast to keep her face and the photographic evidence stark but grounded.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 양손으로 현장 사진을 꽉 쥔 채 공포에 질려 입을 크게 벌린 수리영의 얼굴 정면.\n\nLOCATION (lock): Inside the police office at a detective’s desk, where enlarged crime-scene photographs are spread across the work surface.\n\nTIME OF DAY (lock): late morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level directly across the desk, finish the frontal push-in with 수리영's terrified face occupying most of the frame and both hands crushing the photographs along the lower edge. Preserve a narrow strip of desk and overlapping image corners as context, while her open mouth and rigid grip form the emotional center.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: crime-scene photographs (Spread across the desk and gripped in both hands) — Their image-bearing faces angle upward toward 수리영 and partly toward the camera, revealing enlarged images of severe wounds; used as Lower-foreground trigger for 수리영's reaction, partially visible between her hands; desk surface (Covered with scattered enlarged photographs); used as Narrow lower-frame support for the photographs and hands.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued indoor daytime ambience uses desaturated contrast to keep her face and the photographic evidence stark but grounded.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S21sh3__bgfirst_bg.png",
  "asset_id": "b06e598e-2e0e-4bc8-9770-832b71ab941a",
  "input_asset_ids": [
   "0146f448-4d04-4242-85bf-d5600807cd1f",
   "2224074a-592d-435b-9bf6-ee9a004f8d3f"
  ]
 },
 "S21sh3": {
  "input_fingerprint": "44df550a78b0cf3c",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 양손으로 현장 사진을 꽉 쥔 채 공포에 질려 입을 크게 벌린 수리영의 얼굴 정면.\n\nLOCATION (lock): Inside the police office at a detective’s desk, where enlarged crime-scene photographs are spread across the work surface. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level directly across the desk, finish the frontal push-in with 수리영's terrified face occupying most of the frame and both hands crushing the photographs along the lower edge. Preserve a narrow strip of desk and overlapping image corners as context, while her open mouth and rigid grip form the emotional center.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: crime-scene photographs (Spread across the desk and gripped in both hands) — Their image-bearing faces angle upward toward 수리영 and partly toward the camera, revealing enlarged images of severe wounds; used as Lower-foreground trigger for 수리영's reaction, partially visible between her hands; desk surface (Covered with scattered enlarged photographs); used as Narrow lower-frame support for the photographs and hands.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued indoor daytime ambience uses desaturated contrast to keep her face and the photographic evidence stark but grounded.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung still has her doll-adorned backpack with her as she examines the crime-scene photographs.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 양손으로 현장 사진을 꽉 쥔 채 공포에 질려 입을 크게 벌린 수리영의 얼굴 정면.\n\nLOCATION (lock): Inside the police office at a detective’s desk, where enlarged crime-scene photographs are spread across the work surface. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level directly across the desk, finish the frontal push-in with 수리영's terrified face occupying most of the frame and both hands crushing the photographs along the lower edge. Preserve a narrow strip of desk and overlapping image corners as context, while her open mouth and rigid grip form the emotional center.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: crime-scene photographs (Spread across the desk and gripped in both hands) — Their image-bearing faces angle upward toward 수리영 and partly toward the camera, revealing enlarged images of severe wounds; used as Lower-foreground trigger for 수리영's reaction, partially visible between her hands; desk surface (Covered with scattered enlarged photographs); used as Narrow lower-frame support for the photographs and hands.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued indoor daytime ambience uses desaturated contrast to keep her face and the photographic evidence stark but grounded.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung still has her doll-adorned backpack with her as she examines the crime-scene photographs.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 양손으로 현장 사진을 꽉 쥔 채 공포에 질려 입을 크게 벌린 수리영의 얼굴 정면.\n\nLOCATION (lock): Inside the police office at a detective’s desk, where enlarged crime-scene photographs are spread across the work surface. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level directly across the desk, finish the frontal push-in with 수리영's terrified face occupying most of the frame and both hands crushing the photographs along the lower edge. Preserve a narrow strip of desk and overlapping image corners as context, while her open mouth and rigid grip form the emotional center.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: crime-scene photographs (Spread across the desk and gripped in both hands) — Their image-bearing faces angle upward toward 수리영 and partly toward the camera, revealing enlarged images of severe wounds; used as Lower-foreground trigger for 수리영's reaction, partially visible between her hands; desk surface (Covered with scattered enlarged photographs); used as Narrow lower-frame support for the photographs and hands.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued indoor daytime ambience uses desaturated contrast to keep her face and the photographic evidence stark but grounded.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung still has her doll-adorned backpack with her as she examines the crime-scene photographs.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S21sh3__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S21sh3.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L14B01.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 샷 크기, 표정, 배경(위치 잠금), 그리고 인형이 달린 백팩까지 모든 요구 사항을 충실히 반영한 훌륭한 결과물입니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "표정과 행동은 묘사되었으나, 제공된 배경 레퍼런스와 일치하지 않으며 필수 소품인 인형 백팩이 누락되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 샷 크기, 표정, 배경(위치 잠금), 그리고 인형이 달린 백팩까지 모든 요구 사항을 충실히 반영한 훌륭한 결과물입니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "표정과 행동은 묘사되었으나, 제공된 배경 레퍼런스와 일치하지 않으며 필수 소품인 인형 백팩이 누락되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 클로즈업 샷 크기, 프레임 대부분을 차지하는 얼굴, 사진을 책상 위에서 꽉 쥐는 동작 등 프레이밍과 연출을 매우 충실히 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "배경과 가방 소품은 잘 반영되었으나, 샷이 너무 넓게 잡혀 지시된 클로즈업 프레이밍과 피사체 비율 조건을 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 클로즈업 샷 크기, 프레임 대부분을 차지하는 얼굴, 사진을 책상 위에서 꽉 쥐는 동작 등 프레이밍과 연출을 매우 충실히 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "배경과 가방 소품은 잘 반영되었으나, 샷이 너무 넓게 잡혀 지시된 클로즈업 프레이밍과 피사체 비율 조건을 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 12,
     "B": 11
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": false,
    "policy": 1
   }
  },
  "totals": {
   "A": 12,
   "B": 11
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 8,
    "verdict_ko": "지정된 샷 크기, 표정, 배경(위치 잠금), 그리고 인형이 달린 백팩까지 모든 요구 사항을 충실히 반영한 훌륭한 결과물입니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "표정과 행동은 묘사되었으나, 제공된 배경 레퍼런스와 일치하지 않으며 필수 소품인 인형 백팩이 누락되었습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L14B01.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "레이아웃 스케치의 시선 화살표는 아래를 향해 사진을 보게 되어 있으나, 인물이 카메라 정면을 응시하고 있습니다.",
     "fix_en": "Point the character's eyes downward so she is looking at the photographs, matching the layout sketch's gaze arrows."
    },
    {
     "issue_ko": "인물이 들고 있는 사진이 본인의 눈을 향하지 않고 카메라 쪽으로 완전히 바깥을 향해 들려 있습니다.",
     "fix_en": "Rotate the held photographs so their image-bearing sides face inward toward the character's eyes, keeping them angled upward enough to remain partly visible to the camera."
    },
    {
     "issue_ko": "사진을 꽉 쥐고 구기는(crushing) 모습이 아니라 손가락으로 종이 끝을 가볍게 잡고 있습니다.",
     "fix_en": "Modify the hands to tightly clench and physically crumple the edges of the held photographs."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Point the character's eyes downward so she is looking at the photographs, matching the layout sketch's gaze arrows.\n- Rotate the held photographs so their image-bearing sides face inward toward the character's eyes, keeping them angled upward enough to remain partly visible to the camera.\n- Modify the hands to tightly clench and physically crumple the edges of the held photographs.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "표정과 구도는 좋으나, 레이아웃 스케치의 시선 방향(사진을 내려다보는 시선)을 무시하고 카메라를 정면으로 응시하여 감점되었습니다."
     },
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지시된 클로즈업 프레임, 인물의 착장 및 백팩 디테일을 완벽하게 구현했으며, 특히 스케치에 제시된 시선 방향(사진을 향한 시선)을 정확히 연출하여 가장 우수합니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "표정과 구도는 좋으나, 레이아웃 스케치의 시선 방향(사진을 내려다보는 시선)을 무시하고 카메라를 정면으로 응시하여 감점되었습니다."
     },
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지시된 클로즈업 프레임, 인물의 착장 및 백팩 디테일을 완벽하게 구현했으며, 특히 스케치에 제시된 시선 방향(사진을 향한 시선)을 정확히 연출하여 가장 우수합니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "레이아웃 스케치에 지시된 시선 방향(아래의 사진)을 정확히 따랐으며, 겁에 질린 표정과 소품 묘사가 우수합니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "스케치의 시선 방향 지시를 무시하고 카메라 정면을 응시하여 연출 의도를 벗어났습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "레이아웃 스케치에 지시된 시선 방향(아래의 사진)을 정확히 따랐으며, 겁에 질린 표정과 소품 묘사가 우수합니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "스케치의 시선 방향 지시를 무시하고 카메라 정면을 응시하여 연출 의도를 벗어났습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 11,
     "B": 17
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S21sh3__bgfirst_bg.png",
   "bg_asset_id": "b06e598e-2e0e-4bc8-9770-832b71ab941a",
   "bg_record_key": "S21sh3::bgfirst_bg",
   "chain_winner": true,
   "authority": "plate"
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L14B01",
    "B": "L14B02",
    "C": "L14B03"
   },
   "assigned": "L14B01",
   "choice": "Candidate A",
   "confident": true,
   "reason_ko": "지문에서 인물이 현장 사진을 들고 있다고 묘사하고 있으며, 현재 할당된 후보 이미지의 책상 위에 현장 사진들이 놓여 있으므로 해당 작업 공간이 가장 적합합니다.",
   "kept": "L14B01"
  }
 },
 "S21sh6::variants": {
  "author_fp": "9b10ff0dbd39f3f2",
  "author": {
   "variants": [
    {
     "approach_ko": "정면 대칭과 균일한 초점으로 수사 증거처럼 냉정하고 판독성 높게 포착한다.",
     "prompt_en": "Use a dead-on, monitor-height macro perspective with strict bilateral balance, making the active display feel clinical and evidentiary. Keep the full information surface uniformly sharp and geometrically undistorted, with only the slim housing edge anchoring it as a physical object. Render the restrained screen glow cleanly, allowing the warm point to command the otherwise cool, desaturated field."
    },
    {
     "approach_ko": "화면 가장자리의 미세한 광학 감쇠와 물성을 살려 발견 순간의 긴장감을 촉각적으로 강조한다.",
     "prompt_en": "Photograph from an intimate monitor-height close-focus position with a subtly compressed lens feel, preserving the nearly square-on display while letting the thin housing boundary soften gently toward the corners. Hold every essential screen element crisp and aligned across the central plane, but use delicate luminance falloff and faint natural surface reflections to give the monitor tangible glass depth. Let the isolated warm accent cut through the muted sea map with quiet, ominous tension."
    }
   ]
  },
  "reused": false
 },
 "S21sh6": {
  "input_fingerprint": "a80d8a20db4c67e6",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 바다 위 붉은 점과 좌표 글씨가 떠 있는 모니터 화면 클로즈업.\n\nLOCATION (lock): A detective’s workstation inside the police office, focused on the computer monitor displaying a location result. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Complete the advance at monitor height until the active display becomes the central viewing field, with a narrow trace of the monitor boundary preserving its physical presence. Hold the sea map, red location point, coordinate line, timestamp, source, and offline status in crisp alignment while surrounding office detail falls away.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: monitor location-result display in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: monitor location-result display (On and displaying 강민숙's phone-location result) — The active front face is nearly square to the camera, showing a sea map with a red point, coordinates 37.484400 N / 126.072486 E, timestamp 2026-02-18 11:21:08, GPS/Network source, and offline status; used as Primary information surface revealing the phone's last recorded location; monitor housing (Stationary on the desk) — Only a narrow front-facing border remains visible around the active display; used as Thin peripheral boundary confirming that the information is being viewed on an office monitor.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The active monitor provides restrained screen luminance, with the red point as the sole sharp warm accent against a desaturated display.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 바다 위 붉은 점과 좌표 글씨가 떠 있는 모니터 화면 클로즈업.\n\nLOCATION (lock): A detective’s workstation inside the police office, focused on the computer monitor displaying a location result. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Complete the advance at monitor height until the active display becomes the central viewing field, with a narrow trace of the monitor boundary preserving its physical presence. Hold the sea map, red location point, coordinate line, timestamp, source, and offline status in crisp alignment while surrounding office detail falls away.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: monitor location-result display in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: monitor location-result display (On and displaying 강민숙's phone-location result) — The active front face is nearly square to the camera, showing a sea map with a red point, coordinates 37.484400 N / 126.072486 E, timestamp 2026-02-18 11:21:08, GPS/Network source, and offline status; used as Primary information surface revealing the phone's last recorded location; monitor housing (Stationary on the desk) — Only a narrow front-facing border remains visible around the active display; used as Thin peripheral boundary confirming that the information is being viewed on an office monitor.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The active monitor provides restrained screen luminance, with the red point as the sole sharp warm accent against a desaturated display.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a dead-on, monitor-height macro perspective with strict bilateral balance, making the active display feel clinical and evidentiary. Keep the full information surface uniformly sharp and geometrically undistorted, with only the slim housing edge anchoring it as a physical object. Render the restrained screen glow cleanly, allowing the warm point to command the otherwise cool, desaturated field.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 바다 위 붉은 점과 좌표 글씨가 떠 있는 모니터 화면 클로즈업.\n\nLOCATION (lock): A detective’s workstation inside the police office, focused on the computer monitor displaying a location result. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Complete the advance at monitor height until the active display becomes the central viewing field, with a narrow trace of the monitor boundary preserving its physical presence. Hold the sea map, red location point, coordinate line, timestamp, source, and offline status in crisp alignment while surrounding office detail falls away.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: monitor location-result display in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: monitor location-result display (On and displaying 강민숙's phone-location result) — The active front face is nearly square to the camera, showing a sea map with a red point, coordinates 37.484400 N / 126.072486 E, timestamp 2026-02-18 11:21:08, GPS/Network source, and offline status; used as Primary information surface revealing the phone's last recorded location; monitor housing (Stationary on the desk) — Only a narrow front-facing border remains visible around the active display; used as Thin peripheral boundary confirming that the information is being viewed on an office monitor.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The active monitor provides restrained screen luminance, with the red point as the sole sharp warm accent against a desaturated display.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from an intimate monitor-height close-focus position with a subtly compressed lens feel, preserving the nearly square-on display while letting the thin housing boundary soften gently toward the corners. Hold every essential screen element crisp and aligned across the central plane, but use delicate luminance falloff and faint natural surface reflections to give the monitor tangible glass depth. Let the isolated warm accent cut through the muted sea map with quiet, ominous tension.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 9,
   "B": 4
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 9,
    "verdict_ko": "지시된 모든 텍스트(좌표, 타임스탬프 등)를 실제 모니터 UI처럼 자연스럽게 구현했으며, 프레이밍과 화면 비율이 완벽합니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "화면 중앙에 거대한 자막 형식의 텍스트가 삽입되어 그래픽 오버레이나 캡션을 금지한 지시사항을 위반하고 사실성을 떨어뜨립니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S21sh3_sel.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "화면 우측 하단 시스템 트레이의 날짜가 프롬프트에 명시된 2026년이 아닌 2020년으로 잘못 표시되어 있습니다.",
     "fix_en": "Change the date in the bottom right system tray from '2020' to '2026' to match the specified year."
    },
    {
     "issue_ko": "창 제목('Policess Location'), 지도 지명('South Koas') 및 화면 우측 하단에 철자가 틀린 영어와 의미 없는 텍스트가 생성되었습니다.",
     "fix_en": "Remove the misspelled English window title 'Policess Location', the map label 'South Koas', and the gibberish text at the bottom right."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the date in the bottom right system tray from '2020' to '2026' to match the specified year.\n- Remove the misspelled English window title 'Policess Location', the map label 'South Koas', and the gibberish text at the bottom right.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "요구된 모니터 클로즈업 구도를 정확히 구현했으며, 지시된 모든 텍스트(좌표, 시간, 상태 등)가 시각적 오류 없이 선명하게 렌더링되었습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "전반적인 구도와 화면의 디테일은 훌륭하지만, 지정된 좌표 텍스트의 'E' 글자 부분에 그래픽이 깨지는 아티팩트가 발생해 감점 요인이 되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "요구된 모니터 클로즈업 구도를 정확히 구현했으며, 지시된 모든 텍스트(좌표, 시간, 상태 등)가 시각적 오류 없이 선명하게 렌더링되었습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "전반적인 구도와 화면의 디테일은 훌륭하지만, 지정된 좌표 텍스트의 'E' 글자 부분에 그래픽이 깨지는 아티팩트가 발생해 감점 요인이 되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "지정된 텍스트와 좌표, UI 요소들이 정확하게 구현되었으며, 모니터 화면의 클로즈업 프레이밍도 지시사항을 잘 따랐습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "프레이밍과 주요 텍스트는 훌륭하지만, 지도 우측 하단에 불필요한 의미 불명의 텍스트(hhq dsa...)가 생성되는 오류가 있습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지정된 텍스트와 좌표, UI 요소들이 정확하게 구현되었으며, 모니터 화면의 클로즈업 프레이밍도 지시사항을 잘 따랐습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "프레이밍과 주요 텍스트는 훌륭하지만, 지도 우측 하단에 불필요한 의미 불명의 텍스트(hhq dsa...)가 생성되는 오류가 있습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 16,
     "B": 17
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": false,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev만 (배경 전용·공유 계획)",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S21sh3"
  },
  "lane_policy": "share_plan_prev_bgonly",
  "plate_select": {
   "candidates": {
    "A": "L14B01",
    "B": "L14B02",
    "C": "L14B03"
   },
   "assigned": "L14B01",
   "choice": "Candidate A",
   "confident": true,
   "reason_ko": "지문에서 '모니터 화면 클로즈업'과 '바다 위 붉은 점'을 요구하고 있으므로, 우측 책상 위 켜진 모니터에 지도가 명확히 표시된 Candidate A가 가장 적합합니다.",
   "kept": "L14B01"
  }
 },
 "S21sh9::variants": {
  "author_fp": "ecb3d10c06f26c9e",
  "author": {
   "variants": [
    {
     "approach_ko": "눈높이의 압축된 원근과 얕은 초점으로 수리영의 멈칫한 표정과 문턱의 긴장을 강조한다.",
     "prompt_en": "Photograph at eye level with a restrained, slightly compressed lens feel, holding the prescribed medium framing while isolating her profile in crisp focus against a gently softened office depth. Let the oblique glass edge create a firm threshold line beside her, with subdued late-morning ambient light giving delicate definition to her cheek, hand, and the real reflections in the glass. Keep the open left side quiet and low-contrast so her backward gaze carries the tension across the emptiness."
    },
    {
     "approach_ko": "가까운 공간감과 깊은 초점으로 유리문, 수리영, 빈 사무실을 하나의 불안한 시선 축에 묶는다.",
     "prompt_en": "Use a natural, spatially open lens feel from a subtly lower-than-eye-level position, preserving the exact medium scale and right-weighted layout while allowing the doorway and office depth to remain clearly legible. Favor deeper focus and layered planes: her hand and profile, the oblique glass surface, then the empty workspace along the path of her glance. Render the neutral late-morning interior illumination broadly and evenly, with restrained desaturated color and faint glass reflections that never obscure the physical space."
    }
   ]
  },
  "reused": false
 },
 "S21sh9": {
  "input_fingerprint": "61a78682062da6ca",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 사무실 유리문을 한 손으로 반쯤 밀어 연 채 뒤를 향해 고개를 돌린 수리영의 측면.\n\nLOCATION (lock): The interior doorway of the police office, at the glass door leading out of the workspace. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From inside the office near the doorway, shift far enough to 수리영's side for a medium-close profile while keeping her hand on the half-open glass door in the same frame. Place her toward the right and let her backward glance travel across the open left side toward the empty office, freezing the instant before she slips out.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-right of the frame, midground, looks toward empty office; empty office in the middle-left of the frame, background.\n- KEY BACKGROUND ELEMENTS: office glass door (Partly open) — The door is seen obliquely with the empty office visible through and beyond it; used as Side-frame threshold held by 수리영 and used to preserve her profile; office interior (Empty after the detectives leave); used as Background destination of 수리영's backward check.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Neutral indoor daytime ambience remains subdued and desaturated, emphasizing the tension of the empty office.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung leaves with her doll-adorned backpack and the folded note bearing Minsuk's last GPS coordinates in her pocket.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 사무실 유리문을 한 손으로 반쯤 밀어 연 채 뒤를 향해 고개를 돌린 수리영의 측면.\n\nLOCATION (lock): The interior doorway of the police office, at the glass door leading out of the workspace. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From inside the office near the doorway, shift far enough to 수리영's side for a medium-close profile while keeping her hand on the half-open glass door in the same frame. Place her toward the right and let her backward glance travel across the open left side toward the empty office, freezing the instant before she slips out.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-right of the frame, midground, looks toward empty office; empty office in the middle-left of the frame, background.\n- KEY BACKGROUND ELEMENTS: office glass door (Partly open) — The door is seen obliquely with the empty office visible through and beyond it; used as Side-frame threshold held by 수리영 and used to preserve her profile; office interior (Empty after the detectives leave); used as Background destination of 수리영's backward check.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Neutral indoor daytime ambience remains subdued and desaturated, emphasizing the tension of the empty office.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung leaves with her doll-adorned backpack and the folded note bearing Minsuk's last GPS coordinates in her pocket.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph at eye level with a restrained, slightly compressed lens feel, holding the prescribed medium framing while isolating her profile in crisp focus against a gently softened office depth. Let the oblique glass edge create a firm threshold line beside her, with subdued late-morning ambient light giving delicate definition to her cheek, hand, and the real reflections in the glass. Keep the open left side quiet and low-contrast so her backward gaze carries the tension across the emptiness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 사무실 유리문을 한 손으로 반쯤 밀어 연 채 뒤를 향해 고개를 돌린 수리영의 측면.\n\nLOCATION (lock): The interior doorway of the police office, at the glass door leading out of the workspace. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From inside the office near the doorway, shift far enough to 수리영's side for a medium-close profile while keeping her hand on the half-open glass door in the same frame. Place her toward the right and let her backward glance travel across the open left side toward the empty office, freezing the instant before she slips out.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-right of the frame, midground, looks toward empty office; empty office in the middle-left of the frame, background.\n- KEY BACKGROUND ELEMENTS: office glass door (Partly open) — The door is seen obliquely with the empty office visible through and beyond it; used as Side-frame threshold held by 수리영 and used to preserve her profile; office interior (Empty after the detectives leave); used as Background destination of 수리영's backward check.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Neutral indoor daytime ambience remains subdued and desaturated, emphasizing the tension of the empty office.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung leaves with her doll-adorned backpack and the folded note bearing Minsuk's last GPS coordinates in her pocket.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a natural, spatially open lens feel from a subtly lower-than-eye-level position, preserving the exact medium scale and right-weighted layout while allowing the doorway and office depth to remain clearly legible. Favor deeper focus and layered planes: her hand and profile, the oblique glass surface, then the empty workspace along the path of her glance. Render the neutral late-morning interior illumination broadly and evenly, with restrained desaturated color and faint glass reflections that never obscure the physical space.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 7,
   "B": 4
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "지정된 앵글과 인물의 측면 시선을 정확히 묘사했으며, 손의 해부학과 유리문 반사가 사실적입니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "손잡이를 잡은 손의 심각한 해부학적 왜곡과 유리에 비친 정체불명의 인물(불가능한 반사)로 인해 치명적인 오류가 발생했습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S21sh6_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "수리영의 몸이 시선과 같은 방향(왼쪽)을 향하고 있어, 사무실을 나서며 뒤를 돌아보는 자세가 아닙니다.",
     "fix_en": "Turn Suryoung's body to face right toward the exit so she is looking backward over her shoulder to the left."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Turn Suryoung's body to face right toward the exit so she is looking backward over her shoulder to the left.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "지시된 구도와 조명, 캐릭터의 외양 및 소지품을 완벽하게 구현했으며, 유리문을 잡은 손의 해부학적 형태도 자연스럽고 정확합니다."
     },
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "지시된 구도와 텅 빈 사무실의 분위기를 훌륭하게 연출했으나, 유리문을 잡고 있는 손의 검지와 중지 사이에 살이 붙어 있는 해부학적 오류가 존재합니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "지시된 구도와 조명, 캐릭터의 외양 및 소지품을 완벽하게 구현했으며, 유리문을 잡은 손의 해부학적 형태도 자연스럽고 정확합니다."
     },
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "지시된 구도와 텅 빈 사무실의 분위기를 훌륭하게 연출했으나, 유리문을 잡고 있는 손의 검지와 중지 사이에 살이 붙어 있는 해부학적 오류가 존재합니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "유리문을 밀며 뒤를 돌아보는 수리영의 측면 미디엄 샷이 정확히 연출되었으며, 인상착의와 가방의 인형 등 모든 설정이 완벽하게 구현되었습니다."
     },
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "A 후보와 동일한 이미지로, 샷 텍스트의 구도, 인물의 행동 및 복장 등 모든 프롬프트 요구사항을 완벽하게 충족하고 있습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "유리문을 밀며 뒤를 돌아보는 수리영의 측면 미디엄 샷이 정확히 연출되었으며, 인상착의와 가방의 인형 등 모든 설정이 완벽하게 구현되었습니다."
     },
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "A 후보와 동일한 이미지로, 샷 텍스트의 구도, 인물의 행동 및 복장 등 모든 프롬프트 요구사항을 완벽하게 충족하고 있습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 19,
     "B": 20
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S21sh6"
  }
 },
 "S22sh1::bgfirst_bg": {
  "input_fingerprint": "97d17e68dd6caad5",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 경찰서 밖 도로, 자전거 안장에 앉아 페달에 발을 올린 채 비장한 표정을 지은 수리영의 정면.\n\nLOCATION (lock): The road directly outside the police station, beside the area where the bicycle was parked.\n\nTIME OF DAY (lock): late morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Center the camera directly in front of 수리영 at her seated eye level, holding a wide frontal view that includes her determined face, hands on the handlebars, and both feet set on the pedals. Keep the road extending behind the camera axis and the police-station exterior behind her so the frame is poised for the backward tracking move as she accelerates.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: bicycle (Stationary at the instant before departure) — The front wheel and handlebars face directly toward the camera along the intended travel axis; used as Full-body movement platform and central compositional anchor; police-station exterior (Visible behind the bicycle) — Its outward-facing side sits behind 수리영; used as Background origin establishing where 수리영 is leaving from; road outside the police station (Clear in the direction of departure); used as Forward travel corridor aligned with the lens axis.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural daytime exterior ambience is restrained and desaturated, with tense contrast supporting her determined expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 경찰서 밖 도로, 자전거 안장에 앉아 페달에 발을 올린 채 비장한 표정을 지은 수리영의 정면.\n\nLOCATION (lock): The road directly outside the police station, beside the area where the bicycle was parked.\n\nTIME OF DAY (lock): late morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Center the camera directly in front of 수리영 at her seated eye level, holding a wide frontal view that includes her determined face, hands on the handlebars, and both feet set on the pedals. Keep the road extending behind the camera axis and the police-station exterior behind her so the frame is poised for the backward tracking move as she accelerates.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: bicycle (Stationary at the instant before departure) — The front wheel and handlebars face directly toward the camera along the intended travel axis; used as Full-body movement platform and central compositional anchor; police-station exterior (Visible behind the bicycle) — Its outward-facing side sits behind 수리영; used as Background origin establishing where 수리영 is leaving from; road outside the police station (Clear in the direction of departure); used as Forward travel corridor aligned with the lens axis.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural daytime exterior ambience is restrained and desaturated, with tense contrast supporting her determined expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S22sh1__bgfirst_bg.png",
  "asset_id": "7357b195-146e-4543-82dc-74c27eddf15d",
  "input_asset_ids": [
   "ef30bdc8-6f7c-408d-b0bc-d950fb0f64de",
   "a7666712-e63c-4b80-ab78-3d70775c5c5b"
  ]
 },
 "S22sh1": {
  "input_fingerprint": "5fde9633b8ddea2a",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 경찰서 밖 도로, 자전거 안장에 앉아 페달에 발을 올린 채 비장한 표정을 지은 수리영의 정면.\n\nLOCATION (lock): The road directly outside the police station, beside the area where the bicycle was parked. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Center the camera directly in front of 수리영 at her seated eye level, holding a wide frontal view that includes her determined face, hands on the handlebars, and both feet set on the pedals. Keep the road extending behind the camera axis and the police-station exterior behind her so the frame is poised for the backward tracking move as she accelerates.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: bicycle (Stationary at the instant before departure) — The front wheel and handlebars face directly toward the camera along the intended travel axis; used as Full-body movement platform and central compositional anchor; police-station exterior (Visible behind the bicycle) — Its outward-facing side sits behind 수리영; used as Background origin establishing where 수리영 is leaving from; road outside the police station (Clear in the direction of departure); used as Forward travel corridor aligned with the lens axis.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural daytime exterior ambience is restrained and desaturated, with tense contrast supporting her determined expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung continues to carry her backpack and the folded coordinate note in her pocket as she rides away.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 경찰서 밖 도로, 자전거 안장에 앉아 페달에 발을 올린 채 비장한 표정을 지은 수리영의 정면.\n\nLOCATION (lock): The road directly outside the police station, beside the area where the bicycle was parked. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Center the camera directly in front of 수리영 at her seated eye level, holding a wide frontal view that includes her determined face, hands on the handlebars, and both feet set on the pedals. Keep the road extending behind the camera axis and the police-station exterior behind her so the frame is poised for the backward tracking move as she accelerates.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: bicycle (Stationary at the instant before departure) — The front wheel and handlebars face directly toward the camera along the intended travel axis; used as Full-body movement platform and central compositional anchor; police-station exterior (Visible behind the bicycle) — Its outward-facing side sits behind 수리영; used as Background origin establishing where 수리영 is leaving from; road outside the police station (Clear in the direction of departure); used as Forward travel corridor aligned with the lens axis.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural daytime exterior ambience is restrained and desaturated, with tense contrast supporting her determined expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung continues to carry her backpack and the folded coordinate note in her pocket as she rides away.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 경찰서 밖 도로, 자전거 안장에 앉아 페달에 발을 올린 채 비장한 표정을 지은 수리영의 정면.\n\nLOCATION (lock): The road directly outside the police station, beside the area where the bicycle was parked. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Center the camera directly in front of 수리영 at her seated eye level, holding a wide frontal view that includes her determined face, hands on the handlebars, and both feet set on the pedals. Keep the road extending behind the camera axis and the police-station exterior behind her so the frame is poised for the backward tracking move as she accelerates.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: bicycle (Stationary at the instant before departure) — The front wheel and handlebars face directly toward the camera along the intended travel axis; used as Full-body movement platform and central compositional anchor; police-station exterior (Visible behind the bicycle) — Its outward-facing side sits behind 수리영; used as Background origin establishing where 수리영 is leaving from; road outside the police station (Clear in the direction of departure); used as Forward travel corridor aligned with the lens axis.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural daytime exterior ambience is restrained and desaturated, with tense contrast supporting her determined expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung continues to carry her backpack and the folded coordinate note in her pocket as she rides away.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S22sh1__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S22sh1.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L15B01.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 정면 프레임과 자세를 잘 구현했으며, 레퍼런스의 배경 건축물(경찰서 및 주변)을 매우 정확하게 재현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "인물의 구도는 양호하나, 지정된 경찰서 배경의 건축 구조(입구 형태 등)가 레퍼런스와 전혀 다르게 변형되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 정면 프레임과 자세를 잘 구현했으며, 레퍼런스의 배경 건축물(경찰서 및 주변)을 매우 정확하게 재현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "인물의 구도는 양호하나, 지정된 경찰서 배경의 건축 구조(입구 형태 등)가 레퍼런스와 전혀 다르게 변형되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 구도와 인물의 자세를 완벽히 구현했으며, 특히 레퍼런스 이미지의 장소(경찰서와 우측 붉은 벽돌 건물 등)를 매우 정확하게 재현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "구도와 인물의 자세는 대체로 맞으나, 배경 우측의 건물 형태와 구조가 레퍼런스 장소와 완전히 다르게 왜곡되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요구된 구도와 인물의 자세를 완벽히 구현했으며, 특히 레퍼런스 이미지의 장소(경찰서와 우측 붉은 벽돌 건물 등)를 매우 정확하게 재현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "구도와 인물의 자세는 대체로 맞으나, 배경 우측의 건물 형태와 구조가 레퍼런스 장소와 완전히 다르게 왜곡되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 14,
     "B": 7
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 14,
   "B": 7
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "지정된 정면 프레임과 자세를 잘 구현했으며, 레퍼런스의 배경 건축물(경찰서 및 주변)을 매우 정확하게 재현했습니다."
   },
   {
    "label": "B",
    "score": 3,
    "verdict_ko": "인물의 구도는 양호하나, 지정된 경찰서 배경의 건축 구조(입구 형태 등)가 레퍼런스와 전혀 다르게 변형되었습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L15B01.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "청바지 왼쪽 주머니 바깥에 지시사항에 없는 사각형 신분증/태그가 매달려 있습니다.",
     "fix_en": "Remove the rectangular badge hanging outside her left front jeans pocket."
    },
    {
     "issue_ko": "자전거 앞바퀴에 스포크(바큇살)가 없고 내부가 둥글고 까만 기둥처럼 막혀 있습니다.",
     "fix_en": "Render clear metal spokes and an open space inside the rim of the bicycle's front wheel."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the rectangular badge hanging outside her left front jeans pocket.\n- Render clear metal spokes and an open space inside the rim of the bicycle's front wheel.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 앵글과 스케치 구도를 정확히 반영했으며, 인물의 자세와 자전거의 물리적 형태가 사실적으로 구현되었습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "자전거 앞바퀴의 바큇살(스포크)이 생략되고 검은 막대처럼 렌더링되어 물리적으로 불가능한 구조를 보입니다(치명적 위반)."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 앵글과 스케치 구도를 정확히 반영했으며, 인물의 자세와 자전거의 물리적 형태가 사실적으로 구현되었습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "자전거 앞바퀴의 바큇살(스포크)이 생략되고 검은 막대처럼 렌더링되어 물리적으로 불가능한 구조를 보입니다(치명적 위반)."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "스케치와 배경은 잘 따랐으나, 자전거 앞바퀴 포크의 한쪽이 누락되는 구조적 오류가 있고 주머니의 쪽지가 묘사되지 않았습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 구도와 배경을 정확히 유지하면서 주머니에 꽂힌 쪽지 디테일까지 잘 구현했고 자전거의 형태도 상대적으로 더 온전합니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "스케치와 배경은 잘 따랐으나, 자전거 앞바퀴 포크의 한쪽이 누락되는 구조적 오류가 있고 주머니의 쪽지가 묘사되지 않았습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 구도와 배경을 정확히 유지하면서 주머니에 꽂힌 쪽지 디테일까지 잘 구현했고 자전거의 형태도 상대적으로 더 온전합니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 10,
     "B": 12
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": false,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S22sh1__bgfirst_bg.png",
   "bg_asset_id": "7357b195-146e-4543-82dc-74c27eddf15d",
   "bg_record_key": "S22sh1::bgfirst_bg",
   "chain_winner": true,
   "authority": "plate"
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "groupbg::피시방 좌석 구역": {
  "input_fingerprint": "3588b46fb7d03062",
  "meta": {
   "model": "gpt-image-2",
   "size": "1536x864",
   "pack": "9.202607211540",
   "contract": "bgfirst_full_v1",
   "group_sig": {
    "key": "피시방 좌석 구역",
    "tags": [
     "S23sh1",
     "S23sh2"
    ]
   }
  },
  "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\n\nTHE LOCATION — South Korea, 2026; all people are Korean unless stated: A computer station inside an internet café, lit by the monitor and surrounding neon lighting.\n\nTIME OF DAY (lock): late morning.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.",
  "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\n\nTHE LOCATION — South Korea, 2026; all people are Korean unless stated: A computer station inside an internet café, lit by the monitor and surrounding neon lighting.\n\nTIME OF DAY (lock): late morning.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_피시방_좌석_구역_27315d.png",
  "asset_id": "011e0e25-aac7-4ce7-be5e-c104c923f3bb",
  "input_asset_ids": [
   "b0f1fb60-0604-41f0-83ff-8fce0a1dde8f"
  ],
  "origin_tag": "S23sh1",
  "place_text": "A computer station inside an internet café, lit by the monitor and surrounding neon lighting."
 },
 "S23sh1::bgfirst_bg": {
  "input_fingerprint": "95a2b0da00c7ff16",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): PC방의 화려한 네온 조명 아래, 모니터 화면을 매섭게 응시하며 마우스를 쥔 수리영의 상체.\n\nLOCATION (lock): A computer station inside an internet café, lit by the monitor and surrounding neon lighting.\n\nTIME OF DAY (lock): late morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level, the camera completes its frontal three-quarter dolly toward 수리영 and holds at a firm medium distance, with her upper body placed just left of center and the monitor occupying the right midground. Her face and mouse hand remain simultaneously legible, while the workstation edges create a restrained frame around her fixed concentration.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: PC monitor (On) — The active display face is angled toward both 수리영 and the camera, showing the location and coordinate interface; used as Midground visual counterweight beside 수리영 and the target of her stare; PC café neon lighting (On); used as Provides the surrounding practical illumination and identifies the PC café setting; Computer workstation (In use) — Its user-facing side angles across the frame toward 수리영; used as Forms the lower and lateral boundaries of the workstation composition.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Colorful PC café neon is held at subdued exposure against a restrained, desaturated image, sharpening the tension around 수리영's face.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): PC방의 화려한 네온 조명 아래, 모니터 화면을 매섭게 응시하며 마우스를 쥔 수리영의 상체.\n\nLOCATION (lock): A computer station inside an internet café, lit by the monitor and surrounding neon lighting.\n\nTIME OF DAY (lock): late morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level, the camera completes its frontal three-quarter dolly toward 수리영 and holds at a firm medium distance, with her upper body placed just left of center and the monitor occupying the right midground. Her face and mouse hand remain simultaneously legible, while the workstation edges create a restrained frame around her fixed concentration.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: PC monitor (On) — The active display face is angled toward both 수리영 and the camera, showing the location and coordinate interface; used as Midground visual counterweight beside 수리영 and the target of her stare; PC café neon lighting (On); used as Provides the surrounding practical illumination and identifies the PC café setting; Computer workstation (In use) — Its user-facing side angles across the frame toward 수리영; used as Forms the lower and lateral boundaries of the workstation composition.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Colorful PC café neon is held at subdued exposure against a restrained, desaturated image, sharpening the tension around 수리영's face.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S23sh1__bgfirst_bg.png",
  "asset_id": "0de4218e-dbc0-4178-9013-1f01fdc90ca8",
  "input_asset_ids": [
   "b0f1fb60-0604-41f0-83ff-8fce0a1dde8f",
   "011e0e25-aac7-4ce7-be5e-c104c923f3bb"
  ]
 },
 "S23sh1": {
  "input_fingerprint": "3337291806a267a4",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): PC방의 화려한 네온 조명 아래, 모니터 화면을 매섭게 응시하며 마우스를 쥔 수리영의 상체.\n\nLOCATION (lock): A computer station inside an internet café, lit by the monitor and surrounding neon lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level, the camera completes its frontal three-quarter dolly toward 수리영 and holds at a firm medium distance, with her upper body placed just left of center and the monitor occupying the right midground. Her face and mouse hand remain simultaneously legible, while the workstation edges create a restrained frame around her fixed concentration.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: PC monitor (On) — The active display face is angled toward both 수리영 and the camera, showing the location and coordinate interface; used as Midground visual counterweight beside 수리영 and the target of her stare; PC café neon lighting (On); used as Provides the surrounding practical illumination and identifies the PC café setting; Computer workstation (In use) — Its user-facing side angles across the frame toward 수리영; used as Forms the lower and lateral boundaries of the workstation composition.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Colorful PC café neon is held at subdued exposure against a restrained, desaturated image, sharpening the tension around 수리영's face.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung has her backpack and the coordinate note with her while locating the position on the computer.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): PC방의 화려한 네온 조명 아래, 모니터 화면을 매섭게 응시하며 마우스를 쥔 수리영의 상체.\n\nLOCATION (lock): A computer station inside an internet café, lit by the monitor and surrounding neon lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level, the camera completes its frontal three-quarter dolly toward 수리영 and holds at a firm medium distance, with her upper body placed just left of center and the monitor occupying the right midground. Her face and mouse hand remain simultaneously legible, while the workstation edges create a restrained frame around her fixed concentration.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: PC monitor (On) — The active display face is angled toward both 수리영 and the camera, showing the location and coordinate interface; used as Midground visual counterweight beside 수리영 and the target of her stare; PC café neon lighting (On); used as Provides the surrounding practical illumination and identifies the PC café setting; Computer workstation (In use) — Its user-facing side angles across the frame toward 수리영; used as Forms the lower and lateral boundaries of the workstation composition.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Colorful PC café neon is held at subdued exposure against a restrained, desaturated image, sharpening the tension around 수리영's face.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung has her backpack and the coordinate note with her while locating the position on the computer.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): PC방의 화려한 네온 조명 아래, 모니터 화면을 매섭게 응시하며 마우스를 쥔 수리영의 상체.\n\nLOCATION (lock): A computer station inside an internet café, lit by the monitor and surrounding neon lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level, the camera completes its frontal three-quarter dolly toward 수리영 and holds at a firm medium distance, with her upper body placed just left of center and the monitor occupying the right midground. Her face and mouse hand remain simultaneously legible, while the workstation edges create a restrained frame around her fixed concentration.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: PC monitor (On) — The active display face is angled toward both 수리영 and the camera, showing the location and coordinate interface; used as Midground visual counterweight beside 수리영 and the target of her stare; PC café neon lighting (On); used as Provides the surrounding practical illumination and identifies the PC café setting; Computer workstation (In use) — Its user-facing side angles across the frame toward 수리영; used as Forms the lower and lateral boundaries of the workstation composition.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Colorful PC café neon is held at subdued exposure against a restrained, desaturated image, sharpening the tension around 수리영's face.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung has her backpack and the coordinate note with her while locating the position on the computer.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S23sh1__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S23sh1.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_피시방_좌석_구역_27315d.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 앵글과 레퍼런스 장소를 정확히 구현했으나, 소지품 조건인 좌표 메모지가 누락되었습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "메모지 등 소지품 조건은 지켰으나, 지문에 없는 인물들이 배경에 추가되었고 레퍼런스와 다른 철제 의자가 등장하는 치명적 오류가 있습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 앵글과 레퍼런스 장소를 정확히 구현했으나, 소지품 조건인 좌표 메모지가 누락되었습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "메모지 등 소지품 조건은 지켰으나, 지문에 없는 인물들이 배경에 추가되었고 레퍼런스와 다른 철제 의자가 등장하는 치명적 오류가 있습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 인물만 정확히 등장하며, 요구된 미디엄 샷 구도와 조명, 피사체의 배치 및 동작을 가장 충실하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "프롬프트에 명시되지 않은 배경 인물들을 임의로 추가하여 '발명된 인물'에 해당하는 치명적인 지시 위반이 발생했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 인물만 정확히 등장하며, 요구된 미디엄 샷 구도와 조명, 피사체의 배치 및 동작을 가장 충실하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "프롬프트에 명시되지 않은 배경 인물들을 임의로 추가하여 '발명된 인물'에 해당하는 치명적인 지시 위반이 발생했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 7,
     "B": 14
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 7,
   "B": 14
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "지정된 앵글과 레퍼런스 장소를 정확히 구현했으나, 소지품 조건인 좌표 메모지가 누락되었습니다."
   },
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "메모지 등 소지품 조건은 지켰으나, 지문에 없는 인물들이 배경에 추가되었고 레퍼런스와 다른 철제 의자가 등장하는 치명적 오류가 있습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_피시방_좌석_구역_27315d.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "설정상 수리영이 가지고 있어야 할 좌표가 적힌 쪽지(coordinate note)가 프레임 안에 보이지 않습니다.",
     "fix_en": "Add the coordinate note on the desk near the keyboard or resting in her free hand."
    },
    {
     "issue_ko": "화면에 가장 가까운 왼쪽 어깨에서 이어진 팔 끝에 달린 손(마우스를 쥔 손)이 오른손으로 묘사되어 엄지손가락이 카메라 쪽을 향하는 해부학적 오류가 있습니다.",
     "fix_en": "Correct the hand holding the mouse to an anatomically correct left hand, with the thumb resting on the far side of the mouse."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add the coordinate note on the desk near the keyboard or resting in her free hand.\n- Correct the hand holding the mouse to an anatomically correct left hand, with the thumb resting on the far side of the mouse.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "요구된 촬영 구도, 인물의 형태와 조명 등은 잘 구현되었으나, 지시된 좌표 노트 소품이 프레임 내에 보이지 않습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 카메라 구도, 조명, 인물의 외양을 정확히 반영했으며, 요구된 배낭과 좌표 노트 소품까지 모두 충실히 포함했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "요구된 촬영 구도, 인물의 형태와 조명 등은 잘 구현되었으나, 지시된 좌표 노트 소품이 프레임 내에 보이지 않습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 카메라 구도, 조명, 인물의 외양을 정확히 반영했으며, 요구된 배낭과 좌표 노트 소품까지 모두 충실히 포함했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지시된 화각과 인물 배치를 정확히 구현했으며, 배낭과 함께 마우스패드 위에 좌표 메모지를 배치하여 소품 요구사항을 충실히 반영했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "전반적인 구도와 조명, 인물의 외양은 기준에 부합하나, 요구된 '좌표 메모지' 소품이 프레임 내에 누락되어 감점되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지시된 화각과 인물 배치를 정확히 구현했으며, 배낭과 함께 마우스패드 위에 좌표 메모지를 배치하여 소품 요구사항을 충실히 반영했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "전반적인 구도와 조명, 인물의 외양은 기준에 부합하나, 요구된 '좌표 메모지' 소품이 프레임 내에 누락되어 감점되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 12,
     "B": 15
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S23sh1__bgfirst_bg.png",
   "bg_asset_id": "0de4218e-dbc0-4178-9013-1f01fdc90ca8",
   "bg_record_key": "S23sh1::bgfirst_bg",
   "chain_winner": false,
   "authority": "groupbg",
   "group_key": "피시방 좌석 구역",
   "groupbg_asset_id": "011e0e25-aac7-4ce7-be5e-c104c923f3bb"
  },
  "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S23sh2::variants": {
  "author_fp": "5ffde61cca6a4b99",
  "author": {
   "variants": [
    {
     "approach_ko": "정면의 기계적 대칭과 정지된 종이의 긴장감을 강조한 절제된 인서트.",
     "prompt_en": "Use a precise, near-orthographic lens feel with rigid frontal geometry, making the output slot and sheet edges form a controlled, almost mechanical symmetry. Keep the exposed map crisply resolved while focus drops away immediately behind the output area; subdued, even illumination and restrained neutral tones turn the halted paper into a tense visual interruption."
    },
    {
     "approach_ko": "종이 표면을 스치는 빛과 얕은 심도로 물성과 멈춤의 순간을 촉각적으로 포착한 인서트.",
     "prompt_en": "Give the close insert a tactile, intimate lens feel, with razor-thin focus favoring the coastal contours and the leading edge of the suspended sheet. Let the existing café light skim gently across the paper fibers and printer surfaces, producing soft directional modeling and a rapid falloff into blur, so the physical weight and arrested motion of the page dominate the moment."
    }
   ]
  },
  "reused": false
 },
 "S23sh2": {
  "input_fingerprint": "31d172911be2af19",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 프린터 기기에서 절반쯤 빠져나온 흑백 해안가 지도의 멈춘 상태.\n\nLOCATION (lock): The printer station inside the internet café, where a coastal map is being produced. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At the printer output-slot height, the camera holds close and square to the direction of the emerging sheet, isolating the half-printed coastal map in the central lower portion while retaining the printer body and work surface as scale references. The page remains below forty percent of the frame, with shallow focus suppressing everything beyond the output area.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: Printed coastal map (Half-emerged and momentarily still) — The black-and-white printed face is visible to camera, with the coastal geography readable on the exposed half; used as Primary focal detail suspended across the central lower frame; Printer (Printing) — The output-slot side faces the camera along the paper's travel axis; used as Surrounds the page and establishes its physical emergence point; Work surface (Visible); used as Provides a narrow environmental strip beneath the suspended page.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient PC café illumination keeps the monochrome page crisp while the surrounding workstation falls into restrained contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The printed coastal maps are being produced for Suryoung and will remain folded inside her doll-adorned backpack.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 프린터 기기에서 절반쯤 빠져나온 흑백 해안가 지도의 멈춘 상태.\n\nLOCATION (lock): The printer station inside the internet café, where a coastal map is being produced. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At the printer output-slot height, the camera holds close and square to the direction of the emerging sheet, isolating the half-printed coastal map in the central lower portion while retaining the printer body and work surface as scale references. The page remains below forty percent of the frame, with shallow focus suppressing everything beyond the output area.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: Printed coastal map (Half-emerged and momentarily still) — The black-and-white printed face is visible to camera, with the coastal geography readable on the exposed half; used as Primary focal detail suspended across the central lower frame; Printer (Printing) — The output-slot side faces the camera along the paper's travel axis; used as Surrounds the page and establishes its physical emergence point; Work surface (Visible); used as Provides a narrow environmental strip beneath the suspended page.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient PC café illumination keeps the monochrome page crisp while the surrounding workstation falls into restrained contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The printed coastal maps are being produced for Suryoung and will remain folded inside her doll-adorned backpack.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a precise, near-orthographic lens feel with rigid frontal geometry, making the output slot and sheet edges form a controlled, almost mechanical symmetry. Keep the exposed map crisply resolved while focus drops away immediately behind the output area; subdued, even illumination and restrained neutral tones turn the halted paper into a tense visual interruption.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 프린터 기기에서 절반쯤 빠져나온 흑백 해안가 지도의 멈춘 상태.\n\nLOCATION (lock): The printer station inside the internet café, where a coastal map is being produced. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At the printer output-slot height, the camera holds close and square to the direction of the emerging sheet, isolating the half-printed coastal map in the central lower portion while retaining the printer body and work surface as scale references. The page remains below forty percent of the frame, with shallow focus suppressing everything beyond the output area.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: Printed coastal map (Half-emerged and momentarily still) — The black-and-white printed face is visible to camera, with the coastal geography readable on the exposed half; used as Primary focal detail suspended across the central lower frame; Printer (Printing) — The output-slot side faces the camera along the paper's travel axis; used as Surrounds the page and establishes its physical emergence point; Work surface (Visible); used as Provides a narrow environmental strip beneath the suspended page.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient PC café illumination keeps the monochrome page crisp while the surrounding workstation falls into restrained contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The printed coastal maps are being produced for Suryoung and will remain folded inside her doll-adorned backpack.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nGive the close insert a tactile, intimate lens feel, with razor-thin focus favoring the coastal contours and the leading edge of the suspended sheet. Let the existing café light skim gently across the paper fibers and printer surfaces, producing soft directional modeling and a rapid falloff into blur, so the physical weight and arrested motion of the page dominate the moment.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 7,
   "B": 4
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "지시된 정면 구도를 정확히 구현하였으며, 출력물과 프린터, 작업대의 배치와 비율이 프레이밍 요구사항에 잘 부합합니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "카메라가 정면 구도(square)를 벗어나 측면에서 촬영되었으며, 출력물이 프레임의 40% 이하를 차지하라는 지시를 어겼습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S23sh1_sel.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "출력 중인 지도 표면과 왼쪽 배경의 모니터 화면에 텍스트가 포함되어 있습니다.",
     "fix_en": "Remove all text, labels, and annotations from the printed map and the blurred monitor screen on the left."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all text, labels, and annotations from the printed map and the blurred monitor screen on the left.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "프롬프트의 지시대로 텍스트나 주석을 전혀 포함하지 않고 인쇄 중인 지도를 정확한 구도와 얕은 심도로 완벽하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "구도와 조명은 훌륭하지만 지도 위에 텍스트 형태의 주석이 나타나 '텍스트 및 주석 금지' 지시를 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "프롬프트의 지시대로 텍스트나 주석을 전혀 포함하지 않고 인쇄 중인 지도를 정확한 구도와 얕은 심도로 완벽하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "구도와 조명은 훌륭하지만 지도 위에 텍스트 형태의 주석이 나타나 '텍스트 및 주석 금지' 지시를 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "지시된 프레이밍과 프린터에서 절반쯤 빠져나온 해안가 지도의 상태를 정확히 연출했으며, 텍스트 금지 규칙을 완벽하게 준수했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "구도와 흑백 지도의 상태는 지시사항에 잘 부합하나, 지도 표면에 작은 텍스트(주석)들이 다수 포함되어 '텍스트 금지' 규칙을 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지시된 프레이밍과 프린터에서 절반쯤 빠져나온 해안가 지도의 상태를 정확히 연출했으며, 텍스트 금지 규칙을 완벽하게 준수했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "구도와 흑백 지도의 상태는 지시사항에 잘 부합하나, 지도 표면에 작은 텍스트(주석)들이 다수 포함되어 '텍스트 금지' 규칙을 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 13,
     "B": 19
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev만 (배경 전용·공유 계획)",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S23sh1"
  },
  "lane_policy": "share_plan_prev_bgonly"
 },
 "S24sh2::bgfirst_bg": {
  "input_fingerprint": "4d950d691ab18e5b",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 젖은 머리카락이 이마에 달라붙은 채 빗물 섞인 눈을 부릅뜬 수리영의 얼굴 정면.\n\nLOCATION (lock): An exposed coastal road under a violent rain shower and dark storm clouds.\n\nTIME OF DAY (lock): afternoon, overcast with a thunderstorm.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The camera tracks backward at 수리영's eye level directly ahead of her, holding her rain-struck face in a close frontal composition while the bicycle's forward motion registers only at the lower margins. Her eyes occupy the upper central band, and her wet hair and tightened expression carry the frame without introducing a second subject.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground, moves toward forward route along the coastal road; coastal road behind her in the lower-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: Coastal road (Under heavy rain) — Its route recedes behind 수리영 along the central movement axis; used as Thin, softened background trace that preserves her forward travel direction; Heavy rain (Falling); used as Crosses her face and separates her eyes, hair, and skin into tactile details; Lightning (Intermittent); used as Briefly modulates exposure across her face without changing the camera axis.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Storm-darkened natural light with intermittent lightning keeps the palette desaturated and the rain-covered face sharply tense.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 젖은 머리카락이 이마에 달라붙은 채 빗물 섞인 눈을 부릅뜬 수리영의 얼굴 정면.\n\nLOCATION (lock): An exposed coastal road under a violent rain shower and dark storm clouds.\n\nTIME OF DAY (lock): afternoon, overcast with a thunderstorm.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The camera tracks backward at 수리영's eye level directly ahead of her, holding her rain-struck face in a close frontal composition while the bicycle's forward motion registers only at the lower margins. Her eyes occupy the upper central band, and her wet hair and tightened expression carry the frame without introducing a second subject.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground, moves toward forward route along the coastal road; coastal road behind her in the lower-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: Coastal road (Under heavy rain) — Its route recedes behind 수리영 along the central movement axis; used as Thin, softened background trace that preserves her forward travel direction; Heavy rain (Falling); used as Crosses her face and separates her eyes, hair, and skin into tactile details; Lightning (Intermittent); used as Briefly modulates exposure across her face without changing the camera axis.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Storm-darkened natural light with intermittent lightning keeps the palette desaturated and the rain-covered face sharply tense.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S24sh2__bgfirst_bg.png",
  "asset_id": "913bbf7c-2518-4ede-ac6b-708248a072dd",
  "input_asset_ids": [
   "765dcf4a-32de-4475-b995-11a12d93a39b",
   "519694bd-31b4-4c58-9c2d-45ff2fb4767c"
  ]
 },
 "S24sh2": {
  "input_fingerprint": "04bab81cf46ff09c",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): afternoon, overcast with a thunderstorm.\n\nSHOT TEXT (authoritative, Korean): 젖은 머리카락이 이마에 달라붙은 채 빗물 섞인 눈을 부릅뜬 수리영의 얼굴 정면.\n\nLOCATION (lock): An exposed coastal road under a violent rain shower and dark storm clouds. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The camera tracks backward at 수리영's eye level directly ahead of her, holding her rain-struck face in a close frontal composition while the bicycle's forward motion registers only at the lower margins. Her eyes occupy the upper central band, and her wet hair and tightened expression carry the frame without introducing a second subject.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground, moves toward forward route along the coastal road; coastal road behind her in the lower-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: Coastal road (Under heavy rain) — Its route recedes behind 수리영 along the central movement axis; used as Thin, softened background trace that preserves her forward travel direction; Heavy rain (Falling); used as Crosses her face and separates her eyes, hair, and skin into tactile details; Lightning (Intermittent); used as Briefly modulates exposure across her face without changing the camera axis.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Storm-darkened natural light with intermittent lightning keeps the palette desaturated and the rain-covered face sharply tense.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung continues through the rain with the folded printed maps inside her backpack.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): afternoon, overcast with a thunderstorm.\n\nSHOT TEXT (authoritative, Korean): 젖은 머리카락이 이마에 달라붙은 채 빗물 섞인 눈을 부릅뜬 수리영의 얼굴 정면.\n\nLOCATION (lock): An exposed coastal road under a violent rain shower and dark storm clouds. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The camera tracks backward at 수리영's eye level directly ahead of her, holding her rain-struck face in a close frontal composition while the bicycle's forward motion registers only at the lower margins. Her eyes occupy the upper central band, and her wet hair and tightened expression carry the frame without introducing a second subject.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground, moves toward forward route along the coastal road; coastal road behind her in the lower-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: Coastal road (Under heavy rain) — Its route recedes behind 수리영 along the central movement axis; used as Thin, softened background trace that preserves her forward travel direction; Heavy rain (Falling); used as Crosses her face and separates her eyes, hair, and skin into tactile details; Lightning (Intermittent); used as Briefly modulates exposure across her face without changing the camera axis.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Storm-darkened natural light with intermittent lightning keeps the palette desaturated and the rain-covered face sharply tense.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung continues through the rain with the folded printed maps inside her backpack.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): afternoon, overcast with a thunderstorm.\n\nSHOT TEXT (authoritative, Korean): 젖은 머리카락이 이마에 달라붙은 채 빗물 섞인 눈을 부릅뜬 수리영의 얼굴 정면.\n\nLOCATION (lock): An exposed coastal road under a violent rain shower and dark storm clouds. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The camera tracks backward at 수리영's eye level directly ahead of her, holding her rain-struck face in a close frontal composition while the bicycle's forward motion registers only at the lower margins. Her eyes occupy the upper central band, and her wet hair and tightened expression carry the frame without introducing a second subject.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground, moves toward forward route along the coastal road; coastal road behind her in the lower-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: Coastal road (Under heavy rain) — Its route recedes behind 수리영 along the central movement axis; used as Thin, softened background trace that preserves her forward travel direction; Heavy rain (Falling); used as Crosses her face and separates her eyes, hair, and skin into tactile details; Lightning (Intermittent); used as Briefly modulates exposure across her face without changing the camera axis.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Storm-darkened natural light with intermittent lightning keeps the palette desaturated and the rain-covered face sharply tense.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung continues through the rain with the folded printed maps inside her backpack.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S24sh2__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S24sh2.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_해안_도로_8394b3.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "레퍼런스의 베이지색 재킷을 정확히 착용하고 있으며, 왼쪽 바다와 오른쪽 언덕이라는 배경의 지리적 구조를 충실히 구현하여 프롬프트와 높은 일치도를 보입니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "인물의 재킷 색상이 검은색으로 임의 변경되었고, 바다와 언덕의 위치가 레퍼런스와 반대로 뒤집혀 로케이션 지침을 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "레퍼런스의 베이지색 재킷을 정확히 착용하고 있으며, 왼쪽 바다와 오른쪽 언덕이라는 배경의 지리적 구조를 충실히 구현하여 프롬프트와 높은 일치도를 보입니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "인물의 재킷 색상이 검은색으로 임의 변경되었고, 바다와 언덕의 위치가 레퍼런스와 반대로 뒤집혀 로케이션 지침을 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "레퍼런스와 일치하는 의상(베이지색 재킷)과 해안 도로의 디테일(철책)을 정확히 구현했으며, 요구된 정면 클로즈업 구도와 표정을 충실히 따랐습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "클로즈업 구도와 비 오는 분위기는 묘사되었으나, 레퍼런스와 다른 검은색 의상을 입고 있으며 해안 도로의 형태(가드레일)가 일치하지 않습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "레퍼런스와 일치하는 의상(베이지색 재킷)과 해안 도로의 디테일(철책)을 정확히 구현했으며, 요구된 정면 클로즈업 구도와 표정을 충실히 따랐습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "클로즈업 구도와 비 오는 분위기는 묘사되었으나, 레퍼런스와 다른 검은색 의상을 입고 있으며 해안 도로의 형태(가드레일)가 일치하지 않습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 15,
     "B": 8
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 15,
   "B": 8
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "레퍼런스의 베이지색 재킷을 정확히 착용하고 있으며, 왼쪽 바다와 오른쪽 언덕이라는 배경의 지리적 구조를 충실히 구현하여 프롬프트와 높은 일치도를 보입니다."
   },
   {
    "label": "B",
    "score": 3,
    "verdict_ko": "인물의 재킷 색상이 검은색으로 임의 변경되었고, 바다와 언덕의 위치가 레퍼런스와 반대로 뒤집혀 로케이션 지침을 위반했습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_해안_도로_8394b3.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "프롬프트에는 지도가 배낭 '안에(inside)' 있다고 명시되어 있으나, 배낭 밖으로 지도가 튀어나와 있습니다.",
     "fix_en": "Remove the visible folded maps sticking out of the backpack on the right side of the frame."
    },
    {
     "issue_ko": "원본 배경 이미지를 그대로 유지해야 하나, 산 위의 원래 번개가 사라지고 좌측 바다 위에 새로운 번개가 임의로 배치되었습니다.",
     "fix_en": "Remove the lightning bolt over the ocean and restore the exact original lightning strike over the distant hills from the reference background."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the visible folded maps sticking out of the backpack on the right side of the frame.\n- Remove the lightning bolt over the ocean and restore the exact original lightning strike over the distant hills from the reference background.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "인물의 강렬한 표정과 프레이밍 지시를 훌륭히 소화했으며 스케치의 배경선을 배제했으나, 원본 배경의 번개 위치를 우측으로 임의 변경한 점이 아쉽습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "배낭의 지도는 묘사되었으나, 명시적인 금지 지시를 어기고 레이아웃 스케치에 있는 왼쪽 번개 선을 그대로 렌더링하여 배경 일관성을 훼손했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "인물의 강렬한 표정과 프레이밍 지시를 훌륭히 소화했으며 스케치의 배경선을 배제했으나, 원본 배경의 번개 위치를 우측으로 임의 변경한 점이 아쉽습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "배낭의 지도는 묘사되었으나, 명시적인 금지 지시를 어기고 레이아웃 스케치에 있는 왼쪽 번개 선을 그대로 렌더링하여 배경 일관성을 훼손했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 원본 배경(우측 번개)을 훼손 없이 정확히 유지했으며, 지도가 배낭 안쪽에 있다는 설정에 맞게 화면에 노출하지 않아 모든 지침을 훌륭하게 준수했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "절대 변경해서는 안 되는 배경 요소(번개)의 위치를 임의로 왼쪽으로 이동시켰으며, 배낭 '안'에 있어야 할 지도를 밖으로 어색하게 노출시켜 주요 지침들을 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 원본 배경(우측 번개)을 훼손 없이 정확히 유지했으며, 지도가 배낭 안쪽에 있다는 설정에 맞게 화면에 노출하지 않아 모든 지침을 훌륭하게 준수했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "절대 변경해서는 안 되는 배경 요소(번개)의 위치를 임의로 왼쪽으로 이동시켰으며, 배낭 '안'에 있어야 할 지도를 밖으로 어색하게 노출시켜 주요 지침들을 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 9,
     "B": 14
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S24sh2__bgfirst_bg.png",
   "bg_asset_id": "913bbf7c-2518-4ede-ac6b-708248a072dd",
   "bg_record_key": "S24sh2::bgfirst_bg",
   "chain_winner": true,
   "authority": "groupbg",
   "group_key": "해안 도로",
   "groupbg_asset_id": "519694bd-31b4-4c58-9c2d-45ff2fb4767c"
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S25sh4::variants": {
  "author_fp": "f9b0f494c999d351",
  "author": {
   "variants": [
    {
     "approach_ko": "정면 대칭과 깊은 초점으로 훼손된 방을 차갑고 증거물처럼 기록하는 접근.",
     "prompt_en": "Use a restrained rectilinear wide-lens feel with strict near-symmetry, keeping the doorway edges, marked wall, painted circle, and pinned photograph clearly resolved in deep focus. Let the subdued dusk ambience fall evenly across the physical wall texture, preserving muted shadows and an austere, evidentiary stillness; render the dense red paint and stained, worn photograph with tactile surface detail."
    },
    {
     "approach_ko": "전경 문틀을 부드럽게 흐리고 표식 벽면을 압축해 위협적인 응시점으로 만드는 접근.",
     "prompt_en": "Use a subtly compressed perspective and selective depth, allowing the close doorway edges to fall softly out of focus while the marked wall and central photograph remain sharply fixed beyond them. Keep the exposure low and the dusk ambience directional enough to let the room recede into gentle shadow, making the red circle feel heavy and confrontational without losing the real brush buildup, wall grain, pinning, and age of the photograph."
    }
   ]
  },
  "reused": false
 },
 "S25sh4": {
  "input_fingerprint": "53570aafb11a0b41",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 수리영의 방 하얀 벽면 한가운데 붉은색으로 짙게 칠해진 커다란 원형 표식과 중앙에 꽂힌 낡은 사진.\n\nLOCATION (lock): The white wall of a bedroom inside the ransacked rooftop apartment, marked by a red circle with a photograph pinned at its center. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height just outside the open bedroom, the pan settles into a near-frontal wide hold across the doorway, aligning the large red circle and its centered photograph as one accusatory graphic. The marked wall occupies the central background, while the doorway edges remain visible on both sides to preserve the violated domestic space.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: red circular wall mark with centered photograph in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: Bedroom wall with red circle (Marked with a red circle) — The marked face is held nearly frontal to camera, displaying the large, heavily painted red circle; used as Primary graphic focal plane in the center of the composition; Old photograph (Fixed to the wall and blood-stained) — Its blood-stained image face points toward camera from the exact center of the circle; used as Small central anchor within the red circle; Bedroom doorway (Open) — The opening faces the camera, with both doorway edges visible; used as Frames the marked wall and establishes the camera as observing from outside the room.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk-appropriate ambient light is kept subdued and desaturated, allowing the wall's red mark and the blood on the photograph to form the only forceful color contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains ransacked, with footprints, disturbed household goods, an open living-room window, and overturned dining chairs. Suryoung still carries the folded maps in her backpack.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 수리영의 방 하얀 벽면 한가운데 붉은색으로 짙게 칠해진 커다란 원형 표식과 중앙에 꽂힌 낡은 사진.\n\nLOCATION (lock): The white wall of a bedroom inside the ransacked rooftop apartment, marked by a red circle with a photograph pinned at its center. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height just outside the open bedroom, the pan settles into a near-frontal wide hold across the doorway, aligning the large red circle and its centered photograph as one accusatory graphic. The marked wall occupies the central background, while the doorway edges remain visible on both sides to preserve the violated domestic space.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: red circular wall mark with centered photograph in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: Bedroom wall with red circle (Marked with a red circle) — The marked face is held nearly frontal to camera, displaying the large, heavily painted red circle; used as Primary graphic focal plane in the center of the composition; Old photograph (Fixed to the wall and blood-stained) — Its blood-stained image face points toward camera from the exact center of the circle; used as Small central anchor within the red circle; Bedroom doorway (Open) — The opening faces the camera, with both doorway edges visible; used as Frames the marked wall and establishes the camera as observing from outside the room.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk-appropriate ambient light is kept subdued and desaturated, allowing the wall's red mark and the blood on the photograph to form the only forceful color contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains ransacked, with footprints, disturbed household goods, an open living-room window, and overturned dining chairs. Suryoung still carries the folded maps in her backpack.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained rectilinear wide-lens feel with strict near-symmetry, keeping the doorway edges, marked wall, painted circle, and pinned photograph clearly resolved in deep focus. Let the subdued dusk ambience fall evenly across the physical wall texture, preserving muted shadows and an austere, evidentiary stillness; render the dense red paint and stained, worn photograph with tactile surface detail.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 수리영의 방 하얀 벽면 한가운데 붉은색으로 짙게 칠해진 커다란 원형 표식과 중앙에 꽂힌 낡은 사진.\n\nLOCATION (lock): The white wall of a bedroom inside the ransacked rooftop apartment, marked by a red circle with a photograph pinned at its center. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height just outside the open bedroom, the pan settles into a near-frontal wide hold across the doorway, aligning the large red circle and its centered photograph as one accusatory graphic. The marked wall occupies the central background, while the doorway edges remain visible on both sides to preserve the violated domestic space.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: red circular wall mark with centered photograph in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: Bedroom wall with red circle (Marked with a red circle) — The marked face is held nearly frontal to camera, displaying the large, heavily painted red circle; used as Primary graphic focal plane in the center of the composition; Old photograph (Fixed to the wall and blood-stained) — Its blood-stained image face points toward camera from the exact center of the circle; used as Small central anchor within the red circle; Bedroom doorway (Open) — The opening faces the camera, with both doorway edges visible; used as Frames the marked wall and establishes the camera as observing from outside the room.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk-appropriate ambient light is kept subdued and desaturated, allowing the wall's red mark and the blood on the photograph to form the only forceful color contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains ransacked, with footprints, disturbed household goods, an open living-room window, and overturned dining chairs. Suryoung still carries the folded maps in her backpack.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a subtly compressed perspective and selective depth, allowing the close doorway edges to fall softly out of focus while the marked wall and central photograph remain sharply fixed beyond them. Keep the exposure low and the dusk ambience directional enough to let the room recede into gentle shadow, making the red circle feel heavy and confrontational without losing the real brush buildup, wall grain, pinning, and age of the photograph.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "B": 7,
   "A": 3
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "프레이밍과 조명 지침을 잘 따랐으며, 레퍼런스의 침실 구조(우측 침대)와 어질러진 방의 상태를 충실히 반영했습니다."
   },
   {
    "label": "A",
    "score": 3,
    "verdict_ko": "문틀 너머로 보이는 방 내부에 레퍼런스에 없는 대칭형의 두 개의 문을 임의로 생성하여 공간 구조를 크게 왜곡했습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L05B07.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "레퍼런스 사진에 따르면 침실 문 안쪽 정면 벽에 창문과 커튼이 있어야 하지만, 생성된 이미지에서는 창문이 생략되고 빈 하얀 벽으로 잘못 묘사되었습니다.",
     "fix_en": "Add the window with curtains to the wall facing the doorway inside the bedroom, matching the fixed architecture of the reference location."
    },
    {
     "issue_ko": "모든 인물은 한국인이어야 한다는 지침이 있으나, 벽에 꽂힌 낡은 사진 속 인물들이 백인의 모습을 하고 있습니다.",
     "fix_en": "Change the people depicted in the old photograph to have clearly Korean features and appropriate attire."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add the window with curtains to the wall facing the doorway inside the bedroom, matching the fixed architecture of the reference location.\n- Change the people depicted in the old photograph to have clearly Korean features and appropriate attire.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지정된 구도와 조명 조건을 잘 충족하며, 사진 속 인물들이 한국인이라는 지침을 정확히 반영했습니다."
     },
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "구도와 프레이밍은 훌륭하나, 중앙 사진 속 인물이 한국인이라는 전제 조건(서양인에 가깝게 묘사됨)을 완벽히 충족하지 못했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지정된 구도와 조명 조건을 잘 충족하며, 사진 속 인물들이 한국인이라는 지침을 정확히 반영했습니다."
     },
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "구도와 프레이밍은 훌륭하나, 중앙 사진 속 인물이 한국인이라는 전제 조건(서양인에 가깝게 묘사됨)을 완벽히 충족하지 못했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "문틀을 통해 방 안의 붉은 원과 중앙의 피 묻은 사진을 지침대로 정확히 프레이밍했으며, 사진 속 인물들을 한국인으로 적절히 묘사했습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "구도와 프레이밍, 붉은 원의 묘사는 훌륭하나 사진 속에 등장하는 인물이 한국인이라는 설정이 명확하게 반영되지 않았습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "문틀을 통해 방 안의 붉은 원과 중앙의 피 묻은 사진을 지침대로 정확히 프레이밍했으며, 사진 속 인물들을 한국인으로 적절히 묘사했습니다."
     },
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "구도와 프레이밍, 붉은 원의 묘사는 훌륭하나 사진 속에 등장하는 인물이 한국인이라는 설정이 명확하게 반영되지 않았습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 16,
     "B": 18
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "플레이트만 (배경 전용)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L05B01",
    "B": "L05B02",
    "C": "L05B03",
    "D": "L05B04",
    "E": "L05B05",
    "F": "L05B06",
    "G": "L05B07"
   },
   "assigned": "L05B04",
   "choice": "Candidate A",
   "confident": true,
   "reason_ko": "원문에 묘사된 '붉은색으로 짙게 칠해진 커다란 원형 표식'이 화장실 하얀 벽면에 명확하게 존재함.",
   "kept": "L05B04"
  }
 },
 "S25sh6::variants": {
  "author_fp": "ddf29d7bc42a61aa",
  "author": {
   "variants": [
    {
     "approach_ko": "얕은 심도로 사진의 피와 낡은 유제 질감을 전면에 세우고, 인화된 두 소녀와 건물로 초점을 정밀하게 모으는 불길한 매크로 시선.",
     "prompt_en": "Use an intimate macro-like insert with the photograph held perfectly frontal and centered, its worn paper edges and blood-darkened emulsion rendered with tactile, shallow-depth detail. Let focus resolve decisively across the printed restaurant, the two seated girls, and the uncanny circular rooftop sign, while the narrow red-marked wall fragments fall softly away at the perimeter. Shape the subdued dusk light obliquely across the photograph so dried blood catches a restrained gloss and the aged image retains muted tonal depth."
    },
    {
     "approach_ko": "깊은 초점과 정면 대칭 구도로 사진 전체를 차갑고 증거물처럼 읽히게 하며, 지붕 위 원형 표식의 십자가 같은 실루엣을 중심축에 강조한다.",
     "prompt_en": "Photograph the insert with a restrained, nearly forensic frontal clarity and a deeper plane of focus, keeping the entire old photograph legible as one physical piece rather than isolating a single stain or face. Build a rigid central composition through the restaurant facade and the circular rooftop sign above it, allowing the two seated girls to anchor the lower portion of the printed image and intensify the sign’s cross-like silhouette. Keep the dusk illumination broad and low-contrast, with the bloodstains and peripheral fragments of the red-marked wall providing the only dense, ominous color accents."
    }
   ]
  },
  "reused": false
 },
 "S25sh6": {
  "input_fingerprint": "3ddda57c1010a648",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 십자가처럼 기괴한 둥근 원이 달린 건물과 두 명의 한국인 어린 여자아이가 찍힌 낡고 피 묻은 사진 클로즈업.\n\nLOCATION (lock): Inside the rooftop apartment’s bedroom, focused tightly on an old bloodstained photograph held after being removed from the wall. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The camera completes the frontal dolly toward the photograph, isolating its image at the center while leaving thin fragments of the red-marked wall around it as physical context. The photograph stays within a realistic portion of the frame as focus resolves the rural restaurant, the roof-mounted circular sign, and the two seated girls.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: blood-stained photograph in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: Old photograph (Old and blood-stained) — The blood-stained image face is presented frontally, showing an old rural restaurant, a circular roof sign, and two seated girls; used as Primary isolated image detail at the center of the frame; Rural restaurant in photograph (Depicted in an old photograph) — Its front facade faces the original photograph camera; used as Visible within the photograph as the architectural backdrop behind the girls; Circular roof sign in photograph (Installed on the roof) — Its circular face is visible above the building roof; used as Creates the uncanny cross-like silhouette above the photographed building; Red-marked wall (Marked with a red circle) — Only peripheral sections of the wall face are visible around the photograph, carrying fragments of the red circle; used as Narrow peripheral context confirming that the photograph remains attached to the marked wall.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambient light preserves the photograph's aged tonal separation while the red mark and blood retain a deep, ominous contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The ransacked condition of the room remains, and the old photograph—bloodstained and showing two girls before the building with a circular rooftop sign—has been removed from the red wall circle into Suryoung's possession.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 십자가처럼 기괴한 둥근 원이 달린 건물과 두 명의 한국인 어린 여자아이가 찍힌 낡고 피 묻은 사진 클로즈업.\n\nLOCATION (lock): Inside the rooftop apartment’s bedroom, focused tightly on an old bloodstained photograph held after being removed from the wall. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The camera completes the frontal dolly toward the photograph, isolating its image at the center while leaving thin fragments of the red-marked wall around it as physical context. The photograph stays within a realistic portion of the frame as focus resolves the rural restaurant, the roof-mounted circular sign, and the two seated girls.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: blood-stained photograph in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: Old photograph (Old and blood-stained) — The blood-stained image face is presented frontally, showing an old rural restaurant, a circular roof sign, and two seated girls; used as Primary isolated image detail at the center of the frame; Rural restaurant in photograph (Depicted in an old photograph) — Its front facade faces the original photograph camera; used as Visible within the photograph as the architectural backdrop behind the girls; Circular roof sign in photograph (Installed on the roof) — Its circular face is visible above the building roof; used as Creates the uncanny cross-like silhouette above the photographed building; Red-marked wall (Marked with a red circle) — Only peripheral sections of the wall face are visible around the photograph, carrying fragments of the red circle; used as Narrow peripheral context confirming that the photograph remains attached to the marked wall.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambient light preserves the photograph's aged tonal separation while the red mark and blood retain a deep, ominous contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The ransacked condition of the room remains, and the old photograph—bloodstained and showing two girls before the building with a circular rooftop sign—has been removed from the red wall circle into Suryoung's possession.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an intimate macro-like insert with the photograph held perfectly frontal and centered, its worn paper edges and blood-darkened emulsion rendered with tactile, shallow-depth detail. Let focus resolve decisively across the printed restaurant, the two seated girls, and the uncanny circular rooftop sign, while the narrow red-marked wall fragments fall softly away at the perimeter. Shape the subdued dusk light obliquely across the photograph so dried blood catches a restrained gloss and the aged image retains muted tonal depth.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 십자가처럼 기괴한 둥근 원이 달린 건물과 두 명의 한국인 어린 여자아이가 찍힌 낡고 피 묻은 사진 클로즈업.\n\nLOCATION (lock): Inside the rooftop apartment’s bedroom, focused tightly on an old bloodstained photograph held after being removed from the wall. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The camera completes the frontal dolly toward the photograph, isolating its image at the center while leaving thin fragments of the red-marked wall around it as physical context. The photograph stays within a realistic portion of the frame as focus resolves the rural restaurant, the roof-mounted circular sign, and the two seated girls.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: blood-stained photograph in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: Old photograph (Old and blood-stained) — The blood-stained image face is presented frontally, showing an old rural restaurant, a circular roof sign, and two seated girls; used as Primary isolated image detail at the center of the frame; Rural restaurant in photograph (Depicted in an old photograph) — Its front facade faces the original photograph camera; used as Visible within the photograph as the architectural backdrop behind the girls; Circular roof sign in photograph (Installed on the roof) — Its circular face is visible above the building roof; used as Creates the uncanny cross-like silhouette above the photographed building; Red-marked wall (Marked with a red circle) — Only peripheral sections of the wall face are visible around the photograph, carrying fragments of the red circle; used as Narrow peripheral context confirming that the photograph remains attached to the marked wall.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambient light preserves the photograph's aged tonal separation while the red mark and blood retain a deep, ominous contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The ransacked condition of the room remains, and the old photograph—bloodstained and showing two girls before the building with a circular rooftop sign—has been removed from the red wall circle into Suryoung's possession.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the insert with a restrained, nearly forensic frontal clarity and a deeper plane of focus, keeping the entire old photograph legible as one physical piece rather than isolating a single stain or face. Build a rigid central composition through the restaurant facade and the circular rooftop sign above it, allowing the two seated girls to anchor the lower portion of the printed image and intensify the sign’s cross-like silhouette. Keep the dusk illumination broad and low-contrast, with the bloodstains and peripheral fragments of the red-marked wall providing the only dense, ominous color accents.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 7,
   "B": 3
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "벽면의 질감과 붉은 원의 배치가 참조 이미지와 일치하며 십자가 형태의 간판을 잘 구현했으나, 사진에 두 명의 여자아이 외에 성인이 추가로 등장한 점이 감점 요소입니다."
   },
   {
    "label": "B",
    "score": 3,
    "verdict_ko": "오른쪽 붉은 원이 허공에 떠 있는 물리적으로 불가능한 공간감이 연출되었고, 사진이 벽에 밀착되지 않아 프롬프트를 크게 위반했습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S25sh4_sel.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "사진 속 건물 간판에 텍스트가 포함되어 있습니다.",
     "fix_en": "Remove all text and characters from the building's signs."
    },
    {
     "issue_ko": "사진 속에 두 명의 여자아이 외에 성인 여성이 포함되어 있습니다.",
     "fix_en": "Remove the adult woman from the photograph so that only the two girls remain."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all text and characters from the building's signs.\n- Remove the adult woman from the photograph so that only the two girls remain.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "텍스트 생성 금지 규칙을 위반하여 사진 속 간판에 임의의 문자가 생성되었으며, 두 명의 소녀가 아닌 세 명의 인물이 등장해 지시사항과 어긋납니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "글자가 포함되지 않아야 한다는 규칙을 잘 준수하였고, 십자가 형태의 기괴한 간판과 두 명의 소녀가 있는 낡은 사진을 정확히 구현했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "텍스트 생성 금지 규칙을 위반하여 사진 속 간판에 임의의 문자가 생성되었으며, 두 명의 소녀가 아닌 세 명의 인물이 등장해 지시사항과 어긋납니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "글자가 포함되지 않아야 한다는 규칙을 잘 준수하였고, 십자가 형태의 기괴한 간판과 두 명의 소녀가 있는 낡은 사진을 정확히 구현했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "텍스트 생성 금지 규칙을 엄격히 준수했으며, 지시문대로 사진 속에 정확히 두 명의 소녀와 기괴한 십자가 모양 간판을 잘 구현함."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "사진 속 간판에 텍스트가 생성되어 '텍스트 금지' 규칙을 위반했으며, 지시된 두 명의 소녀 외에 다른 인물이 추가됨."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "텍스트 생성 금지 규칙을 엄격히 준수했으며, 지시문대로 사진 속에 정확히 두 명의 소녀와 기괴한 십자가 모양 간판을 잘 구현함."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "사진 속 간판에 텍스트가 생성되어 '텍스트 금지' 규칙을 위반했으며, 지시된 두 명의 소녀 외에 다른 인물이 추가됨."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 6,
     "B": 15
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev만 (배경 전용·공유 계획)",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S25sh4"
  },
  "lane_policy": "share_plan_prev_bgonly",
  "plate_select": {
   "candidates": {
    "A": "L05B01",
    "B": "L05B02",
    "C": "L05B03",
    "D": "L05B04",
    "E": "L05B05",
    "F": "L05B06",
    "G": "L05B07"
   },
   "assigned": "L05B04",
   "choice": "Candidate C",
   "confident": true,
   "reason_ko": "로케이션 설명에서 숏의 공간을 옥탑방의 '침실 내부(Inside the rooftop apartment’s bedroom)'로 명확히 지정하고 있으므로, 침대가 놓인 침실 안을 보여주는 후보를 선택해야 합니다.",
   "kept": "L05B04"
  }
 },
 "S25sh8::bgfirst_bg": {
  "input_fingerprint": "49898fb419d4a3ec",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 백팩을 메고 한 손에 구겨진 사진을 꽉 쥔 채 열린 현관문 턱을 한 발로 디딘 자세로 굳은 수리영.\n\nLOCATION (lock): The open front-door threshold of the rooftop apartment, between the ransacked interior and the exterior rooftop.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level inside the living area, the parallel track settles roughly opposite the open entrance as 수리영 freezes with one foot planted on the threshold. She occupies the middle-right of the wide frame with the backpack close to her body and the crumpled photograph held visibly but naturally small in one fist, while the disturbed room remains behind her to the left.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 paused at threshold in the middle-right of the frame, midground; open entrance in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: Entrance door (Open) — The open doorway faces diagonally toward the interior camera; used as Defines the exit line directly beside 수리영's paused body; Living area (Ransacked) — The room opens away from the entrance into the left background; used as Extends behind 수리영 and preserves evidence of the intrusion; Dining chairs (Overturned) — Their overturned sides and legs angle toward the camera from the background; used as Angular background obstacles that break the otherwise direct route to the exit; Living-room window (Open) — The open window provides a view beyond the room through its opening; used as Secondary opening visible within the disturbed living area.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambient light maintains a desaturated interior, with the crumpled blood-stained photograph providing a restrained deep-red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 백팩을 메고 한 손에 구겨진 사진을 꽉 쥔 채 열린 현관문 턱을 한 발로 디딘 자세로 굳은 수리영.\n\nLOCATION (lock): The open front-door threshold of the rooftop apartment, between the ransacked interior and the exterior rooftop.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level inside the living area, the parallel track settles roughly opposite the open entrance as 수리영 freezes with one foot planted on the threshold. She occupies the middle-right of the wide frame with the backpack close to her body and the crumpled photograph held visibly but naturally small in one fist, while the disturbed room remains behind her to the left.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 paused at threshold in the middle-right of the frame, midground; open entrance in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: Entrance door (Open) — The open doorway faces diagonally toward the interior camera; used as Defines the exit line directly beside 수리영's paused body; Living area (Ransacked) — The room opens away from the entrance into the left background; used as Extends behind 수리영 and preserves evidence of the intrusion; Dining chairs (Overturned) — Their overturned sides and legs angle toward the camera from the background; used as Angular background obstacles that break the otherwise direct route to the exit; Living-room window (Open) — The open window provides a view beyond the room through its opening; used as Secondary opening visible within the disturbed living area.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambient light maintains a desaturated interior, with the crumpled blood-stained photograph providing a restrained deep-red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S25sh8__bgfirst_bg.png",
  "asset_id": "e910f597-0726-4f25-a790-aa1dc8dfd395",
  "input_asset_ids": [
   "1b5e5766-4ded-4622-9921-39b79419e057",
   "cdeff3a4-6071-4fe9-924a-4a305ce19b8b"
  ]
 },
 "S25sh8": {
  "input_fingerprint": "11a22575835e62c9",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 백팩을 메고 한 손에 구겨진 사진을 꽉 쥔 채 열린 현관문 턱을 한 발로 디딘 자세로 굳은 수리영.\n\nLOCATION (lock): The open front-door threshold of the rooftop apartment, between the ransacked interior and the exterior rooftop. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level inside the living area, the parallel track settles roughly opposite the open entrance as 수리영 freezes with one foot planted on the threshold. She occupies the middle-right of the wide frame with the backpack close to her body and the crumpled photograph held visibly but naturally small in one fist, while the disturbed room remains behind her to the left.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 paused at threshold in the middle-right of the frame, midground; open entrance in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: Entrance door (Open) — The open doorway faces diagonally toward the interior camera; used as Defines the exit line directly beside 수리영's paused body; Living area (Ransacked) — The room opens away from the entrance into the left background; used as Extends behind 수리영 and preserves evidence of the intrusion; Dining chairs (Overturned) — Their overturned sides and legs angle toward the camera from the background; used as Angular background obstacles that break the otherwise direct route to the exit; Living-room window (Open) — The open window provides a view beyond the room through its opening; used as Secondary opening visible within the disturbed living area.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambient light maintains a desaturated interior, with the crumpled blood-stained photograph providing a restrained deep-red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung leaves wearing the backpack that contains the printed maps and keeps the crumpled, bloodstained old photograph tightly in one hand. The room remains ransacked behind her, with the red wall circle left in place without the photograph.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 백팩을 메고 한 손에 구겨진 사진을 꽉 쥔 채 열린 현관문 턱을 한 발로 디딘 자세로 굳은 수리영.\n\nLOCATION (lock): The open front-door threshold of the rooftop apartment, between the ransacked interior and the exterior rooftop. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level inside the living area, the parallel track settles roughly opposite the open entrance as 수리영 freezes with one foot planted on the threshold. She occupies the middle-right of the wide frame with the backpack close to her body and the crumpled photograph held visibly but naturally small in one fist, while the disturbed room remains behind her to the left.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 paused at threshold in the middle-right of the frame, midground; open entrance in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: Entrance door (Open) — The open doorway faces diagonally toward the interior camera; used as Defines the exit line directly beside 수리영's paused body; Living area (Ransacked) — The room opens away from the entrance into the left background; used as Extends behind 수리영 and preserves evidence of the intrusion; Dining chairs (Overturned) — Their overturned sides and legs angle toward the camera from the background; used as Angular background obstacles that break the otherwise direct route to the exit; Living-room window (Open) — The open window provides a view beyond the room through its opening; used as Secondary opening visible within the disturbed living area.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambient light maintains a desaturated interior, with the crumpled blood-stained photograph providing a restrained deep-red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung leaves wearing the backpack that contains the printed maps and keeps the crumpled, bloodstained old photograph tightly in one hand. The room remains ransacked behind her, with the red wall circle left in place without the photograph.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 백팩을 메고 한 손에 구겨진 사진을 꽉 쥔 채 열린 현관문 턱을 한 발로 디딘 자세로 굳은 수리영.\n\nLOCATION (lock): The open front-door threshold of the rooftop apartment, between the ransacked interior and the exterior rooftop. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level inside the living area, the parallel track settles roughly opposite the open entrance as 수리영 freezes with one foot planted on the threshold. She occupies the middle-right of the wide frame with the backpack close to her body and the crumpled photograph held visibly but naturally small in one fist, while the disturbed room remains behind her to the left.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 paused at threshold in the middle-right of the frame, midground; open entrance in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: Entrance door (Open) — The open doorway faces diagonally toward the interior camera; used as Defines the exit line directly beside 수리영's paused body; Living area (Ransacked) — The room opens away from the entrance into the left background; used as Extends behind 수리영 and preserves evidence of the intrusion; Dining chairs (Overturned) — Their overturned sides and legs angle toward the camera from the background; used as Angular background obstacles that break the otherwise direct route to the exit; Living-room window (Open) — The open window provides a view beyond the room through its opening; used as Secondary opening visible within the disturbed living area.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambient light maintains a desaturated interior, with the crumpled blood-stained photograph providing a restrained deep-red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung leaves wearing the backpack that contains the printed maps and keeps the crumpled, bloodstained old photograph tightly in one hand. The room remains ransacked behind her, with the red wall circle left in place without the photograph.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S25sh8__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S25sh8.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    },
    {
     "label": "PROP REFERENCE — 두 어린 여자아이와 원형 장식 건물이 찍힌 낡고 피 묻은 흑백사진: the exact object appearing in this shot; match its look, material and wear exactly.",
     "path": "<bytes:1268182>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L05B05.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    },
    {
     "label": "PROP REFERENCE — 두 어린 여자아이와 원형 장식 건물이 찍힌 낡고 피 묻은 흑백사진: the exact object appearing in this shot; match its look, material and wear exactly.",
     "path": "<bytes:1268182>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "백팩의 어깨끈이 전혀 없어 물리적으로 불가능한 형태이며, 손에 쥔 사진이 참조 이미지와 다르고 피 묻은 질감도 없습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 와이드 샷 구도와 인물의 문턱 디딤 자세를 정확히 구현했으며, 가방 끈과 참조 사진의 디테일도 잘 살렸습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "백팩의 어깨끈이 전혀 없어 물리적으로 불가능한 형태이며, 손에 쥔 사진이 참조 이미지와 다르고 피 묻은 질감도 없습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 와이드 샷 구도와 인물의 문턱 디딤 자세를 정확히 구현했으며, 가방 끈과 참조 사진의 디테일도 잘 살렸습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "한 발을 현관문 턱에 디딘 구체적인 지문과 나가는 방향의 자세를 정확하게 연출했으나 사진의 핏자국 묘사는 다소 부족함."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "문턱에 발을 디딘 자세가 구현되지 않았고, 동선과 무관하게 카메라를 향해 정면으로 서 있어 연출 의도와 어긋남."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "한 발을 현관문 턱에 디딘 구체적인 지문과 나가는 방향의 자세를 정확하게 연출했으나 사진의 핏자국 묘사는 다소 부족함."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "문턱에 발을 디딘 자세가 구현되지 않았고, 동선과 무관하게 카메라를 향해 정면으로 서 있어 연출 의도와 어긋남."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 11,
     "B": 12
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": false,
    "policy": 1
   }
  },
  "totals": {
   "A": 11,
   "B": 12
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "백팩의 어깨끈이 전혀 없어 물리적으로 불가능한 형태이며, 손에 쥔 사진이 참조 이미지와 다르고 피 묻은 질감도 없습니다."
   },
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "지정된 와이드 샷 구도와 인물의 문턱 디딤 자세를 정확히 구현했으며, 가방 끈과 참조 사진의 디테일도 잘 살렸습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L05B05.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   },
   {
    "label": "PROP REFERENCE — 두 어린 여자아이와 원형 장식 건물이 찍힌 낡고 피 묻은 흑백사진: the exact object appearing in this shot; match its look, material and wear exactly.",
    "path": "<bytes:1268182>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "수리영이 사진을 주먹에 꽉 쥐지 않고, 평평한 상태의 사진을 느슨하게 잡고 있습니다.",
     "fix_en": "Crumple the photograph into a small shape and make her grip it tightly within a closed fist."
    },
    {
     "issue_ko": "손에 든 사진에 프롬프트가 요구한 붉은 핏자국이 없습니다.",
     "fix_en": "Add deep-red bloodstains to the photograph in her hand."
    },
    {
     "issue_ko": "레퍼런스 사진의 평평한 바닥과 달리, 현관문 바닥에 존재하지 않는 깊은 단차(도랑)가 잘못 생성되었습니다.",
     "fix_en": "Flatten the floor at the doorway to match the level, continuous threshold shown in the location reference."
    },
    {
     "issue_ko": "레퍼런스의 고정된 건축 구조(바깥으로 열리는 현관문)와 반대로 문이 방 안쪽으로 열려 있습니다.",
     "fix_en": "Change the door to swing outwards onto the exterior rooftop instead of inwards into the room."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Crumple the photograph into a small shape and make her grip it tightly within a closed fist.\n- Add deep-red bloodstains to the photograph in her hand.\n- Flatten the floor at the doorway to match the level, continuous threshold shown in the location reference.\n- Change the door to swing outwards onto the exterior rooftop instead of inwards into the room.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "주먹에 꽉 쥔 구겨진 사진과 핏자국의 붉은색 포인트를 정확하고 자연스럽게 연출하여 프롬프트의 지시를 훌륭하게 완수했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "배경과 인물의 구도는 매우 좋으나, 사진을 주먹에 쥐고 구겨진 형태가 아닌 평평하게 들고 있어 소품 묘사 지시를 어겼습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "주먹에 꽉 쥔 구겨진 사진과 핏자국의 붉은색 포인트를 정확하고 자연스럽게 연출하여 프롬프트의 지시를 훌륭하게 완수했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "배경과 인물의 구도는 매우 좋으나, 사진을 주먹에 쥐고 구겨진 형태가 아닌 평평하게 들고 있어 소품 묘사 지시를 어겼습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "구겨진 사진을 주먹에 꽉 쥔 디테일과 붉은색 포인트가 지시문과 정확히 일치하며 훌륭한 구도를 보여줍니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "사진이 전혀 구겨지지 않은 채 평평하게 들려 있어 '구겨진 사진을 꽉 쥔'이라는 핵심 소품 상태를 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "구겨진 사진을 주먹에 꽉 쥔 디테일과 붉은색 포인트가 지시문과 정확히 일치하며 훌륭한 구도를 보여줍니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "사진이 전혀 구겨지지 않은 채 평평하게 들려 있어 '구겨진 사진을 꽉 쥔'이라는 핵심 소품 상태를 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 10,
     "B": 15
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S25sh8__bgfirst_bg.png",
   "bg_asset_id": "e910f597-0726-4f25-a790-aa1dc8dfd395",
   "bg_record_key": "S25sh8::bgfirst_bg",
   "chain_winner": false,
   "authority": "plate"
  },
  "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L05B01",
    "B": "L05B02",
    "C": "L05B03",
    "D": "L05B04",
    "E": "L05B05",
    "F": "L05B06",
    "G": "L05B07"
   },
   "assigned": "L05B05",
   "choice": "Candidate E",
   "confident": true,
   "reason_ko": "열린 현관문 턱에 서 있는 수리영을 묘사하고 있으므로, 열린 현관문이 직접적으로 보이는 서브공간이 필요합니다.",
   "kept": "L05B05"
  }
 },
 "groupbg::안개 낀 포구": {
  "input_fingerprint": "2567761c46a63136",
  "meta": {
   "model": "gpt-image-2",
   "size": "1536x864",
   "pack": "9.202607211540",
   "contract": "bgfirst_full_v1",
   "group_sig": {
    "key": "안개 낀 포구",
    "tags": [
     "S26sh5",
     "S26sh6",
     "S26sh9"
    ]
   }
  },
  "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\n\nTHE LOCATION — South Korea, 2026; all people are Korean unless stated: The fogbound dockside beside a small fishing boat, at a mooring post where its heavy line is being secured.\n\nTIME OF DAY (lock): dusk, foggy after rain.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.",
  "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\n\nTHE LOCATION — South Korea, 2026; all people are Korean unless stated: The fogbound dockside beside a small fishing boat, at a mooring post where its heavy line is being secured.\n\nTIME OF DAY (lock): dusk, foggy after rain.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_안개_낀_포구_a58666.png",
  "asset_id": "39344657-a18d-4e06-a017-7d12d95059e2",
  "input_asset_ids": [
   "10e234f1-0012-42a4-befe-0d07de9912fd"
  ],
  "origin_tag": "S26sh5",
  "place_text": "The fogbound dockside beside a small fishing boat, at a mooring post where its heavy line is being secured."
 },
 "S26sh5::bgfirst_bg": {
  "input_fingerprint": "5a577fa6750a0de8",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 선착장 기둥에 굵은 밧줄을 반쯤 감은 채 양손으로 팽팽하게 잡아당기고 있는 인우의 상체.\n\nLOCATION (lock): The fogbound dockside beside a small fishing boat, at a mooring post where its heavy line is being secured.\n\nTIME OF DAY (lock): dusk, foggy after rain.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height beside the docking point, the camera completes its dolly toward 인우 and holds a medium three-quarter view of his upper body and both hands. He is placed slightly off center with the half-wrapped rope crossing the lower frame toward the mooring post, preserving realistic scale and the force of his pull.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: Mooring post (Rope half-wrapped around it) — Its side faces the camera while the rope turns around it; used as Anchors the rope action beside 인우 in the lower side of the frame; Thick mooring rope (Pulled taut); used as Creates a taut diagonal between 인우's hands and the mooring post; Dense fog (Dense); used as Softens the harbor depth behind the close physical action; Harbor (Empty); used as Provides the isolated harbor context behind the docking point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Diffuse dusk light filtered by dense fog keeps the image low in saturation and contrast while preserving tactile definition in 인우's hands and the taut rope.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 선착장 기둥에 굵은 밧줄을 반쯤 감은 채 양손으로 팽팽하게 잡아당기고 있는 인우의 상체.\n\nLOCATION (lock): The fogbound dockside beside a small fishing boat, at a mooring post where its heavy line is being secured.\n\nTIME OF DAY (lock): dusk, foggy after rain.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height beside the docking point, the camera completes its dolly toward 인우 and holds a medium three-quarter view of his upper body and both hands. He is placed slightly off center with the half-wrapped rope crossing the lower frame toward the mooring post, preserving realistic scale and the force of his pull.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: Mooring post (Rope half-wrapped around it) — Its side faces the camera while the rope turns around it; used as Anchors the rope action beside 인우 in the lower side of the frame; Thick mooring rope (Pulled taut); used as Creates a taut diagonal between 인우's hands and the mooring post; Dense fog (Dense); used as Softens the harbor depth behind the close physical action; Harbor (Empty); used as Provides the isolated harbor context behind the docking point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Diffuse dusk light filtered by dense fog keeps the image low in saturation and contrast while preserving tactile definition in 인우's hands and the taut rope.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S26sh5__bgfirst_bg.png",
  "asset_id": "6ea8d46c-33ce-4eb5-8aad-34aabfb49b2f",
  "input_asset_ids": [
   "10e234f1-0012-42a4-befe-0d07de9912fd",
   "39344657-a18d-4e06-a017-7d12d95059e2"
  ]
 },
 "S26sh5": {
  "input_fingerprint": "046dc0ac88ffc947",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy after rain.\n\nSHOT TEXT (authoritative, Korean): 선착장 기둥에 굵은 밧줄을 반쯤 감은 채 양손으로 팽팽하게 잡아당기고 있는 인우의 상체.\n\nLOCATION (lock): The fogbound dockside beside a small fishing boat, at a mooring post where its heavy line is being secured. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height beside the docking point, the camera completes its dolly toward 인우 and holds a medium three-quarter view of his upper body and both hands. He is placed slightly off center with the half-wrapped rope crossing the lower frame toward the mooring post, preserving realistic scale and the force of his pull.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: Mooring post (Rope half-wrapped around it) — Its side faces the camera while the rope turns around it; used as Anchors the rope action beside 인우 in the lower side of the frame; Thick mooring rope (Pulled taut); used as Creates a taut diagonal between 인우's hands and the mooring post; Dense fog (Dense); used as Softens the harbor depth behind the close physical action; Harbor (Empty); used as Provides the isolated harbor context behind the docking point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Diffuse dusk light filtered by dense fog keeps the image low in saturation and contrast while preserving tactile definition in 인우's hands and the taut rope.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung still has her backpack, the folded maps inside it, and the bloodstained old photograph in her possession while Inwoo secures the boat.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 남성, 20대 중반의 젊은 얼굴, 반듯한 얼굴선, 짧은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy after rain.\n\nSHOT TEXT (authoritative, Korean): 선착장 기둥에 굵은 밧줄을 반쯤 감은 채 양손으로 팽팽하게 잡아당기고 있는 인우의 상체.\n\nLOCATION (lock): The fogbound dockside beside a small fishing boat, at a mooring post where its heavy line is being secured. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height beside the docking point, the camera completes its dolly toward 인우 and holds a medium three-quarter view of his upper body and both hands. He is placed slightly off center with the half-wrapped rope crossing the lower frame toward the mooring post, preserving realistic scale and the force of his pull.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: Mooring post (Rope half-wrapped around it) — Its side faces the camera while the rope turns around it; used as Anchors the rope action beside 인우 in the lower side of the frame; Thick mooring rope (Pulled taut); used as Creates a taut diagonal between 인우's hands and the mooring post; Dense fog (Dense); used as Softens the harbor depth behind the close physical action; Harbor (Empty); used as Provides the isolated harbor context behind the docking point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Diffuse dusk light filtered by dense fog keeps the image low in saturation and contrast while preserving tactile definition in 인우's hands and the taut rope.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung still has her backpack, the folded maps inside it, and the bloodstained old photograph in her possession while Inwoo secures the boat.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 남성, 20대 중반의 젊은 얼굴, 반듯한 얼굴선, 짧은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy after rain.\n\nSHOT TEXT (authoritative, Korean): 선착장 기둥에 굵은 밧줄을 반쯤 감은 채 양손으로 팽팽하게 잡아당기고 있는 인우의 상체.\n\nLOCATION (lock): The fogbound dockside beside a small fishing boat, at a mooring post where its heavy line is being secured. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height beside the docking point, the camera completes its dolly toward 인우 and holds a medium three-quarter view of his upper body and both hands. He is placed slightly off center with the half-wrapped rope crossing the lower frame toward the mooring post, preserving realistic scale and the force of his pull.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: Mooring post (Rope half-wrapped around it) — Its side faces the camera while the rope turns around it; used as Anchors the rope action beside 인우 in the lower side of the frame; Thick mooring rope (Pulled taut); used as Creates a taut diagonal between 인우's hands and the mooring post; Dense fog (Dense); used as Softens the harbor depth behind the close physical action; Harbor (Empty); used as Provides the isolated harbor context behind the docking point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Diffuse dusk light filtered by dense fog keeps the image low in saturation and contrast while preserving tactile definition in 인우's hands and the taut rope.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung still has her backpack, the folded maps inside it, and the bloodstained old photograph in her possession while Inwoo secures the boat.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 남성, 20대 중반의 젊은 얼굴, 반듯한 얼굴선, 짧은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S26sh5__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S26sh5.png"
    },
    {
     "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:761032>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_안개_낀_포구_a58666.png"
    },
    {
     "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:761032>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "레퍼런스의 공간 배치(좌측 기둥, 우측 배)를 정확히 유지하며 밧줄을 당기는 지문과 구도를 충실히 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "원래 좌측에 있는 기둥을 우측으로 임의로 이동시켜 로케이션의 공간적 진실성을 훼손하는 구조적 오류를 범했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "레퍼런스의 공간 배치(좌측 기둥, 우측 배)를 정확히 유지하며 밧줄을 당기는 지문과 구도를 충실히 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "원래 좌측에 있는 기둥을 우측으로 임의로 이동시켜 로케이션의 공간적 진실성을 훼손하는 구조적 오류를 범했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "레퍼런스 이미지의 공간 구조(왼쪽 기둥, 오른쪽 배)를 정확히 유지하며 프롬프트의 행동 지시를 훌륭하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "밧줄을 당기는 동작은 표현되었으나, 기둥이 우측에 배치되어 지정된 로케이션의 공간 진실성에 위배됩니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "레퍼런스 이미지의 공간 구조(왼쪽 기둥, 오른쪽 배)를 정확히 유지하며 프롬프트의 행동 지시를 훌륭하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "밧줄을 당기는 동작은 표현되었으나, 기둥이 우측에 배치되어 지정된 로케이션의 공간 진실성에 위배됩니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 15,
     "B": 9
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 15,
   "B": 9
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "레퍼런스의 공간 배치(좌측 기둥, 우측 배)를 정확히 유지하며 밧줄을 당기는 지문과 구도를 충실히 구현했습니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "원래 좌측에 있는 기둥을 우측으로 임의로 이동시켜 로케이션의 공간적 진실성을 훼손하는 구조적 오류를 범했습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/groupbg_안개_낀_포구_a58666.png"
   },
   {
    "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:761032>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "인우의 몸 방향과 시선이 레이아웃 스케치와 완전히 다릅니다 (스케치는 등을 보이며 오른쪽을 향하고 있으나, 생성된 이미지는 정면을 향하며 왼쪽을 봄).",
     "fix_en": "Turn Inwoo's body so his back and right shoulder face the camera, and turn his head to look far to the right, exactly matching the layout sketch."
    },
    {
     "issue_ko": "인우의 왼손(화면 오른쪽)에서 이어진 밧줄이 허공에서 비현실적인 매듭을 만들며 다른 밧줄과 엉켜 있습니다.",
     "fix_en": "Remove the floating twisted loop of rope on the right and draw it as a single, continuous taut line extending outward from his hand."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Turn Inwoo's body so his back and right shoulder face the camera, and turn his head to look far to the right, exactly matching the layout sketch.\n- Remove the floating twisted loop of rope on the right and draw it as a single, continuous taut line extending outward from his hand.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "레이아웃 스케치의 카메라 방향(가슴이 보이는 앞모습)을 잘 따랐으며, 양손으로 밧줄을 팽팽하게 당기는 행동을 정확하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "스케치와 달리 캐릭터가 등을 보이고 있으며, 양손으로 밧줄을 당긴다는 프롬프트의 핵심 행동이 제대로 묘사되지 않았습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "레이아웃 스케치의 카메라 방향(가슴이 보이는 앞모습)을 잘 따랐으며, 양손으로 밧줄을 팽팽하게 당기는 행동을 정확하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "스케치와 달리 캐릭터가 등을 보이고 있으며, 양손으로 밧줄을 당긴다는 프롬프트의 핵심 행동이 제대로 묘사되지 않았습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "스케치와 반대로 몸의 방향이 뒤집혀 있으며, 밧줄을 양손으로 당기는 동작이 제대로 구현되지 않은 치명적인 오류가 있습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "레이아웃 스케치의 포즈와 시선을 정확히 따랐으며, 기둥에 감긴 밧줄을 양손으로 팽팽하게 당기는 지시를 잘 구현했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "스케치와 반대로 몸의 방향이 뒤집혀 있으며, 밧줄을 양손으로 당기는 동작이 제대로 구현되지 않은 치명적인 오류가 있습니다."
     },
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "레이아웃 스케치의 포즈와 시선을 정확히 따랐으며, 기둥에 감긴 밧줄을 양손으로 팽팽하게 당기는 지시를 잘 구현했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 15,
     "B": 6
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S26sh5__bgfirst_bg.png",
   "bg_asset_id": "6ea8d46c-33ce-4eb5-8aad-34aabfb49b2f",
   "bg_record_key": "S26sh5::bgfirst_bg",
   "chain_winner": true,
   "authority": "groupbg",
   "group_key": "안개 낀 포구",
   "groupbg_asset_id": "39344657-a18d-4e06-a017-7d12d95059e2"
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S26sh6::variants": {
  "author_fp": "58af22c4cd6035b5",
  "author": {
   "variants": [
    {
     "approach_ko": "얕은 심도로 수리영의 절박한 옆얼굴을 선명하게 붙잡고 인우와 안개 낀 부두를 부드럽게 분리한다.",
     "prompt_en": "Use an intimate, gently compressed medium profile with shallow depth of field, holding Suryoung’s forward-leaning face and half-open mouth in crisp focus while Inwoo remains softly but recognizably present across her sightline. Let the lower-center docking edge lead through the narrow focus falloff between them, with diffuse fog-muted dusk light shaping subtle moisture and restrained highlights on real surfaces."
    },
    {
     "approach_ko": "두 인물과 그 사이의 부두 공간을 함께 읽히는 깊은 초점으로 담아 물리적 거리와 협상의 긴장을 강조한다.",
     "prompt_en": "Photograph the medium profile with a more observational lens feel and extended depth, keeping both figures and the working docking edge clearly legible as a layered foreground-to-midground arrangement. Emphasize the open physical interval along their negotiation axis, using the angled boat side and dense fog as quiet structural planes; keep the diffuse dusk illumination even and subdued, with muted color and natural wet-surface response."
    }
   ]
  },
  "reused": false
 },
 "S26sh6": {
  "input_fingerprint": "27fec0aacb7b17b7",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy after rain.\n\nSHOT TEXT (authoritative, Korean): 인우를 향해 상체를 앞으로 기울이고 입을 반쯤 벌린 간절한 표정의 수리영 측면.\n\nLOCATION (lock): The edge of a foggy harbor pier beside the moored fishing boat. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At shoulder height beside 수리영, the lateral track tightens into a medium profile composition with her forward-leaning face in the left foreground and 인우 held across her sightline in the right midground. Her half-open mouth remains clear while the docking area between them preserves their physical separation and negotiation axis.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, foreground, looks toward 인우 across the docking point; 인우 in the middle-right of the frame, midground, looks toward 수리영 in the foreground; shared docking point in the lower-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: Docking point (In use) — Its working edge runs between the foreground and midground figures; used as Shared spatial anchor separating and connecting the two speakers; Dense fog (Dense); used as Obscures distant harbor detail and isolates the exchange; Boat (Moored) — Its dock-facing side angles behind 인우; used as Places the boat beside the negotiation without overpowering either face.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Diffuse dusk light through dense fog creates restrained luminance and muted color, keeping the emotional contrast concentrated between 수리영 and 인우.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the dense harbor fog, dock surfaces, mooring post, rope, and muted dusk lighting from the reference. EXCLUDE the rope-pulling action as the main focus; frame the young woman leaning toward the boatman with an urgent expression.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung continues to carry her backpack, folded maps, and the old bloodstained photograph as she pleads with Inwoo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy after rain.\n\nSHOT TEXT (authoritative, Korean): 인우를 향해 상체를 앞으로 기울이고 입을 반쯤 벌린 간절한 표정의 수리영 측면.\n\nLOCATION (lock): The edge of a foggy harbor pier beside the moored fishing boat. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At shoulder height beside 수리영, the lateral track tightens into a medium profile composition with her forward-leaning face in the left foreground and 인우 held across her sightline in the right midground. Her half-open mouth remains clear while the docking area between them preserves their physical separation and negotiation axis.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, foreground, looks toward 인우 across the docking point; 인우 in the middle-right of the frame, midground, looks toward 수리영 in the foreground; shared docking point in the lower-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: Docking point (In use) — Its working edge runs between the foreground and midground figures; used as Shared spatial anchor separating and connecting the two speakers; Dense fog (Dense); used as Obscures distant harbor detail and isolates the exchange; Boat (Moored) — Its dock-facing side angles behind 인우; used as Places the boat beside the negotiation without overpowering either face.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Diffuse dusk light through dense fog creates restrained luminance and muted color, keeping the emotional contrast concentrated between 수리영 and 인우.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the dense harbor fog, dock surfaces, mooring post, rope, and muted dusk lighting from the reference. EXCLUDE the rope-pulling action as the main focus; frame the young woman leaning toward the boatman with an urgent expression.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung continues to carry her backpack, folded maps, and the old bloodstained photograph as she pleads with Inwoo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an intimate, gently compressed medium profile with shallow depth of field, holding Suryoung’s forward-leaning face and half-open mouth in crisp focus while Inwoo remains softly but recognizably present across her sightline. Let the lower-center docking edge lead through the narrow focus falloff between them, with diffuse fog-muted dusk light shaping subtle moisture and restrained highlights on real surfaces.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy after rain.\n\nSHOT TEXT (authoritative, Korean): 인우를 향해 상체를 앞으로 기울이고 입을 반쯤 벌린 간절한 표정의 수리영 측면.\n\nLOCATION (lock): The edge of a foggy harbor pier beside the moored fishing boat. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At shoulder height beside 수리영, the lateral track tightens into a medium profile composition with her forward-leaning face in the left foreground and 인우 held across her sightline in the right midground. Her half-open mouth remains clear while the docking area between them preserves their physical separation and negotiation axis.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, foreground, looks toward 인우 across the docking point; 인우 in the middle-right of the frame, midground, looks toward 수리영 in the foreground; shared docking point in the lower-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: Docking point (In use) — Its working edge runs between the foreground and midground figures; used as Shared spatial anchor separating and connecting the two speakers; Dense fog (Dense); used as Obscures distant harbor detail and isolates the exchange; Boat (Moored) — Its dock-facing side angles behind 인우; used as Places the boat beside the negotiation without overpowering either face.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Diffuse dusk light through dense fog creates restrained luminance and muted color, keeping the emotional contrast concentrated between 수리영 and 인우.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the dense harbor fog, dock surfaces, mooring post, rope, and muted dusk lighting from the reference. EXCLUDE the rope-pulling action as the main focus; frame the young woman leaning toward the boatman with an urgent expression.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung continues to carry her backpack, folded maps, and the old bloodstained photograph as she pleads with Inwoo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the medium profile with a more observational lens feel and extended depth, keeping both figures and the working docking edge clearly legible as a layered foreground-to-midground arrangement. Emphasize the open physical interval along their negotiation axis, using the angled boat side and dense fog as quiet structural planes; keep the diffuse dusk illumination even and subdued, with muted color and natural wet-surface response.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 8,
   "B": 5
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 8,
    "verdict_ko": "요구된 미디엄 샷 프레이밍과 상체를 기울인 수리영의 간절한 표정을 훌륭하게 구현했으며, 두 인물의 화면 내 거리감과 소품 상태도 정확합니다."
   },
   {
    "label": "B",
    "score": 5,
    "verdict_ko": "수리영의 포즈와 소품은 지시를 따랐으나, 인우가 전신이 보이는 풀 샷 수준으로 멀리 배치되어 미디엄 샷 프레이밍 지시를 어겼습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S26sh5_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "수리영이 들고 있는 지도 표면에 숫자와 텍스트(우측 하단 등)가 생성되어 '텍스트 금지' 규칙을 위반했습니다.",
     "fix_en": "Remove all visible text, numbers, and characters from the surfaces of the folded map."
    },
    {
     "issue_ko": "지문에서 카메라를 향해 보여주라는 지시가 없었음에도, 지도와 사진의 앞면(내용)이 카메라 렌즈 쪽을 향하도록 부자연스럽게 노출되어 소품 방향 규칙을 위반했습니다.",
     "fix_en": "Tilt the photograph and maps inward so their functional printed sides face toward Suryoung's body or downward, hiding their contents from the camera."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all visible text, numbers, and characters from the surfaces of the folded map.\n- Tilt the photograph and maps inward so their functional printed sides face toward Suryoung's body or downward, hiding their contents from the camera.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 인물 구도와 표정을 잘 구현했으며, 물리적 시점에 맞는 자연스러운 소품 방향을 보여줍니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "지도에 텍스트가 노출되는 금지 사항을 위반했으며, 사진이 카메라를 향해 부자연스럽게 돌아가 있습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 인물 구도와 표정을 잘 구현했으며, 물리적 시점에 맞는 자연스러운 소품 방향을 보여줍니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "지도에 텍스트가 노출되는 금지 사항을 위반했으며, 사진이 카메라를 향해 부자연스럽게 돌아가 있습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 샷 크기와 구도를 완벽히 구현했으며, 수리영이 들고 있는 피 묻은 사진과 지도까지 정확하게 표현했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "전반적인 구도와 인물의 자세는 요구사항을 따랐으나, 손에 든 피 묻은 사진과 지도의 디테일이 생략되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 샷 크기와 구도를 완벽히 구현했으며, 수리영이 들고 있는 피 묻은 사진과 지도까지 정확하게 표현했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "전반적인 구도와 인물의 자세는 요구사항을 따랐으나, 손에 든 피 묻은 사진과 지도의 디테일이 생략되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 11,
     "B": 12
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": false,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S26sh5"
  }
 },
 "S26sh9::variants": {
  "author_fp": "afc65802fe95f9a2",
  "author": {
   "variants": [
    {
     "approach_ko": "얕은 심도로 얼굴과 휴대폰에 시선을 고정하는 압축된 심리적 클로즈 관찰.",
     "prompt_en": "Use a restrained long-lens feel with shallow depth of field, holding her lowered eyes, mouth, and smartphone in one crisp plane while the harbor falls into a soft fog-muted wash. Preserve quiet negative space around her off-center placement, with gentle diffuse modeling across her face and a compressed background that intensifies the privacy of the moment."
    },
    {
     "approach_ko": "갑판과 안개 낀 바다의 층위를 살리는 깊은 초점의 담담한 환경적 관찰.",
     "prompt_en": "Use a natural perspective and moderately deep focus for an observational, grounded image, keeping the receding deck plane legible beneath and behind her while the sea remains only faintly discernible through the fog. Organize the frame in clear foreground, subject, and background layers, with subdued dusk light revealing damp, worn surfaces and her braced posture without breaking the scene’s muted restraint."
    }
   ]
  },
  "reused": false
 },
 "S26sh9": {
  "input_fingerprint": "98f7cc7728becd93",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy after rain.\n\nSHOT TEXT (authoritative, Korean): 어선 갑판 위에 서서 스마트폰을 입가에 바짝 댄 채 눈을 내리깐 수리영의 바스트 샷.\n\nLOCATION (lock): The open deck of a small fishing boat while it remains at the foggy harbor before departure. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height on the boat deck and slightly below 수리영's eye line, the camera holds a stable bust shot with her placed just off center. Her raised smartphone stays close to her mouth in the lower central portion, while her lowered eyes and braced posture separate the private message from the fog-muted harbor behind her.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: Boat deck (Supporting 수리영) — The deck plane recedes behind and beneath 수리영; used as Provides a narrow lower-frame base for 수리영's braced posture; Dense fog (Dense); used as Suppresses distant detail around her isolated message; Sea (Viewed through fog); used as Barely readable background context beyond the deck.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Diffuse dusk illumination through the fog remains desaturated and subdued, holding gentle facial contrast around 수리영's lowered eyes and phone.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): On the boat, Suryoung still has the backpack containing the maps and retains the old bloodstained photograph.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy after rain.\n\nSHOT TEXT (authoritative, Korean): 어선 갑판 위에 서서 스마트폰을 입가에 바짝 댄 채 눈을 내리깐 수리영의 바스트 샷.\n\nLOCATION (lock): The open deck of a small fishing boat while it remains at the foggy harbor before departure. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height on the boat deck and slightly below 수리영's eye line, the camera holds a stable bust shot with her placed just off center. Her raised smartphone stays close to her mouth in the lower central portion, while her lowered eyes and braced posture separate the private message from the fog-muted harbor behind her.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: Boat deck (Supporting 수리영) — The deck plane recedes behind and beneath 수리영; used as Provides a narrow lower-frame base for 수리영's braced posture; Dense fog (Dense); used as Suppresses distant detail around her isolated message; Sea (Viewed through fog); used as Barely readable background context beyond the deck.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Diffuse dusk illumination through the fog remains desaturated and subdued, holding gentle facial contrast around 수리영's lowered eyes and phone.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): On the boat, Suryoung still has the backpack containing the maps and retains the old bloodstained photograph.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained long-lens feel with shallow depth of field, holding her lowered eyes, mouth, and smartphone in one crisp plane while the harbor falls into a soft fog-muted wash. Preserve quiet negative space around her off-center placement, with gentle diffuse modeling across her face and a compressed background that intensifies the privacy of the moment.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy after rain.\n\nSHOT TEXT (authoritative, Korean): 어선 갑판 위에 서서 스마트폰을 입가에 바짝 댄 채 눈을 내리깐 수리영의 바스트 샷.\n\nLOCATION (lock): The open deck of a small fishing boat while it remains at the foggy harbor before departure. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height on the boat deck and slightly below 수리영's eye line, the camera holds a stable bust shot with her placed just off center. Her raised smartphone stays close to her mouth in the lower central portion, while her lowered eyes and braced posture separate the private message from the fog-muted harbor behind her.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: Boat deck (Supporting 수리영) — The deck plane recedes behind and beneath 수리영; used as Provides a narrow lower-frame base for 수리영's braced posture; Dense fog (Dense); used as Suppresses distant detail around her isolated message; Sea (Viewed through fog); used as Barely readable background context beyond the deck.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Diffuse dusk illumination through the fog remains desaturated and subdued, holding gentle facial contrast around 수리영's lowered eyes and phone.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): On the boat, Suryoung still has the backpack containing the maps and retains the old bloodstained photograph.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a natural perspective and moderately deep focus for an observational, grounded image, keeping the receding deck plane legible beneath and behind her while the sea remains only faintly discernible through the fog. Organize the frame in clear foreground, subject, and background layers, with subdued dusk light revealing damp, worn surfaces and her braced posture without breaking the scene’s muted restraint.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "B": 9,
   "A": 7
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 9,
    "verdict_ko": "지시된 바스트 샷 규격과 어선 갑판의 배경을 정확히 구현했고, 스마트폰의 기능적 면(화면)이 캐릭터를 향하는 올바른 방향성과 배낭에 담긴 지도 디테일이 매우 뛰어납니다."
   },
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "프레이밍과 전반적인 분위기는 지시문과 잘 맞으나, 스마트폰 화면이 위로 향해 있어 사용 방향이 다소 어색하며 배낭 안의 지도에 대한 묘사가 누락되었습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S26sh6_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   },
   {
    "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.",
    "path": "<bytes:802115>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "수리영이 서 있는 바닥이 지시된 어선의 갑판이 아니라 콘크리트 부두로 묘사되어 있습니다.",
     "fix_en": "Change the concrete pier ground beneath and behind her into the surface of a fishing boat deck."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the concrete pier ground beneath and behind her into the surface of a fishing boat deck.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 어선 갑판 위라는 장소를 정확히 구현했으며, 프레이밍과 피사체의 포즈, 소지품 모두 지시사항을 훌륭하게 따랐습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "어선 갑판 위라는 명확한 장소 지시를 무시하고 인물을 선착장(부두)에 배치하여 감점되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 어선 갑판 위라는 장소를 정확히 구현했으며, 프레이밍과 피사체의 포즈, 소지품 모두 지시사항을 훌륭하게 따랐습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "어선 갑판 위라는 명확한 장소 지시를 무시하고 인물을 선착장(부두)에 배치하여 감점되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "지시된 바스트 샷 앵글, 조명, 안개 낀 분위기를 완벽하게 구현했으며, 뒤로 멀어지는 갑판의 깊이감이 매우 자연스럽습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "레퍼런스 이미지의 배경을 그대로 잘라 붙인 듯한 콜라주 현상이 발생하여 원근감과 공간적 일관성이 크게 훼손되었습니다 (하드 위반)."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지시된 바스트 샷 앵글, 조명, 안개 낀 분위기를 완벽하게 구현했으며, 뒤로 멀어지는 갑판의 깊이감이 매우 자연스럽습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "레퍼런스 이미지의 배경을 그대로 잘라 붙인 듯한 콜라주 현상이 발생하여 원근감과 공간적 일관성이 크게 훼손되었습니다 (하드 위반)."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 7,
     "B": 16
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S26sh6"
  }
 },
 "S27sh1::variants": {
  "author_fp": "b1ae73472333ef1d",
  "author": {
   "variants": [
    {
     "approach_ko": "정면에 가까운 깊은 초점의 와이드 숏으로, 방의 혼란스러운 선들이 중앙의 혜수에게 모이게 한다.",
     "prompt_en": "Use a restrained, nearly frontal eye-level wide shot with deep spatial clarity, letting the disordered room’s lines and overturned chairs converge visually on 혜수 near the center. Keep her three-quarter stance and grave face crisply legible while the entrance door remains a small, distant endpoint beyond her. Shape the subdued dusk ambience with soft lateral falloff, holding muted color and dense, realistic shadow detail."
    },
    {
     "approach_ko": "전경의 흐릿한 어수선함 너머로 혜수를 포착해, 주변 공간이 그녀를 압박하는 듯한 층위감과 고립감을 만든다.",
     "prompt_en": "Photograph the same eye-level wide composition through a close layer of softly defocused disturbed belongings at the room boundary, creating strong foreground, middle-ground, and background separation without obscuring 혜수. Hold her near center in a rigid three-quarter listening pause, isolated in the clearest plane, with overturned chairs forming uneasy shapes around her and the moving entrance door receding beyond. Let fading dusk light skim her face unevenly, preserving a tense, desaturated naturalism."
    }
   ]
  },
  "reused": false
 },
 "S27sh1": {
  "input_fingerprint": "591eecfe0df1ab03",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 어수선한 옥탑방 거실 한가운데 서서 스마트폰을 귀에 댄 채 심각한 표정으로 굳어 있는 혜수.\n\nLOCATION (lock): The disordered living room inside the rooftop apartment near the rattling front door, in the fading evening light. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level near the room boundary, the camera advances into a wide composition with 혜수 held near the center of the disordered living area, the smartphone pressed to her ear. Her body is caught in a rigid listening pause at a three-quarter angle, while the room's scattered forms remain visible around her without competing with her grave expression.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: Rooftop-room living area (Disordered) — The room opens from the camera boundary toward 혜수 at its center; used as Surrounds 혜수 with evidence of the disturbed domestic space; Dining chairs (Overturned) — Their overturned sides are visible behind and beside 혜수; used as Provides an unstable background shape within the living area; Entrance door (Moving in the wind) — Its interior-facing side is visible beyond 혜수; used as Creates the distant audible endpoint of the scene without taking focus from 혜수.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk-appropriate ambient light is subdued and desaturated, leaving 혜수's grave face in tense contrast against the disordered room.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains ransacked, with footprints, disturbed belongings, an open window, and overturned dining chairs; the red circle remains on Suryoung's wall after the photograph was removed.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 어수선한 옥탑방 거실 한가운데 서서 스마트폰을 귀에 댄 채 심각한 표정으로 굳어 있는 혜수.\n\nLOCATION (lock): The disordered living room inside the rooftop apartment near the rattling front door, in the fading evening light. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level near the room boundary, the camera advances into a wide composition with 혜수 held near the center of the disordered living area, the smartphone pressed to her ear. Her body is caught in a rigid listening pause at a three-quarter angle, while the room's scattered forms remain visible around her without competing with her grave expression.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: Rooftop-room living area (Disordered) — The room opens from the camera boundary toward 혜수 at its center; used as Surrounds 혜수 with evidence of the disturbed domestic space; Dining chairs (Overturned) — Their overturned sides are visible behind and beside 혜수; used as Provides an unstable background shape within the living area; Entrance door (Moving in the wind) — Its interior-facing side is visible beyond 혜수; used as Creates the distant audible endpoint of the scene without taking focus from 혜수.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk-appropriate ambient light is subdued and desaturated, leaving 혜수's grave face in tense contrast against the disordered room.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains ransacked, with footprints, disturbed belongings, an open window, and overturned dining chairs; the red circle remains on Suryoung's wall after the photograph was removed.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained, nearly frontal eye-level wide shot with deep spatial clarity, letting the disordered room’s lines and overturned chairs converge visually on 혜수 near the center. Keep her three-quarter stance and grave face crisply legible while the entrance door remains a small, distant endpoint beyond her. Shape the subdued dusk ambience with soft lateral falloff, holding muted color and dense, realistic shadow detail.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 어수선한 옥탑방 거실 한가운데 서서 스마트폰을 귀에 댄 채 심각한 표정으로 굳어 있는 혜수.\n\nLOCATION (lock): The disordered living room inside the rooftop apartment near the rattling front door, in the fading evening light. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level near the room boundary, the camera advances into a wide composition with 혜수 held near the center of the disordered living area, the smartphone pressed to her ear. Her body is caught in a rigid listening pause at a three-quarter angle, while the room's scattered forms remain visible around her without competing with her grave expression.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: Rooftop-room living area (Disordered) — The room opens from the camera boundary toward 혜수 at its center; used as Surrounds 혜수 with evidence of the disturbed domestic space; Dining chairs (Overturned) — Their overturned sides are visible behind and beside 혜수; used as Provides an unstable background shape within the living area; Entrance door (Moving in the wind) — Its interior-facing side is visible beyond 혜수; used as Creates the distant audible endpoint of the scene without taking focus from 혜수.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk-appropriate ambient light is subdued and desaturated, leaving 혜수's grave face in tense contrast against the disordered room.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains ransacked, with footprints, disturbed belongings, an open window, and overturned dining chairs; the red circle remains on Suryoung's wall after the photograph was removed.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반의 얼굴, 단정한 얼굴선, 정리된 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the same eye-level wide composition through a close layer of softly defocused disturbed belongings at the room boundary, creating strong foreground, middle-ground, and background separation without obscuring 혜수. Hold her near center in a rigid three-quarter listening pause, isolated in the clearest plane, with overturned chairs forming uneasy shapes around her and the moving entrance door receding beyond. Let fading dusk light skim her face unevenly, preserving a tense, desaturated naturalism.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 7,
   "B": 3
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "지정된 앵글과 어수선한 방의 구조를 충실히 구현했으나, 요구된 현관문 짝이 누락되었고 인물이 3/4 각도가 아닌 정면을 향하고 있습니다."
   },
   {
    "label": "B",
    "score": 3,
    "verdict_ko": "참조 이미지에 없는 거대한 책상과 소품들을 전경에 임의로 추가하여 공간을 왜곡했으며, 뒤집힌 의자들이 서로 물리적으로 겹쳐 있는 치명적 오류가 있습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S25sh8_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:796814>"
   },
   {
    "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.",
    "path": "<bytes:802115>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "혜수의 몸이 4분의 3 측면(three-quarter angle)이 아닌 완전히 정면을 향하고 있습니다.",
     "fix_en": "Turn Hye-soo's body to a three-quarter angle while keeping her rigid and holding the phone to her ear."
    },
    {
     "issue_ko": "프레임 오른쪽의 출입구에 있어야 할 현관문(Entrance door)이 완전히 사라져 있습니다.",
     "fix_en": "Restore the heavy grey metal entrance door to the right doorway, showing its interior-facing side open inward."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Turn Hye-soo's body to a three-quarter angle while keeping her rigid and holding the phone to her ear.\n- Restore the heavy grey metal entrance door to the right doorway, showing its interior-facing side open inward.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "피사체가 정면을 향해 있어 3/4 측면 각도 지시에 어긋나며, 현관문 자체가 완전히 사라져 배경 연속성 오류가 큽니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "요구된 3/4 측면 각도와 심각한 표정을 잘 구현했으며, 어수선한 실내와 현관문의 상태 등 공간의 연속성을 정확히 유지했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "피사체가 정면을 향해 있어 3/4 측면 각도 지시에 어긋나며, 현관문 자체가 완전히 사라져 배경 연속성 오류가 큽니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "요구된 3/4 측면 각도와 심각한 표정을 잘 구현했으며, 어수선한 실내와 현관문의 상태 등 공간의 연속성을 정확히 유지했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 3/4 측면 각도와 어수선한 방의 상태, 그리고 열려 있는 현관문의 디테일 등을 모두 훌륭하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "캐릭터가 정면을 향하고 있어 3/4 각도 지시를 어겼으며, 우측 프레임에 있어야 할 현관문 자체가 사라지는 오류가 있습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 3/4 측면 각도와 어수선한 방의 상태, 그리고 열려 있는 현관문의 디테일 등을 모두 훌륭하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "캐릭터가 정면을 향하고 있어 3/4 각도 지시를 어겼으며, 우측 프레임에 있어야 할 현관문 자체가 사라지는 오류가 있습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 8,
     "B": 15
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S25sh8"
  }
 },
 "S27sh3::variants": {
  "author_fp": "362b1632cccc08cd",
  "author": {
   "variants": [
    {
     "approach_ko": "얕은 심도와 정면 대칭 구도로 표정의 미세한 긴장을 고립시키는 압축된 인물 클로즈업.",
     "prompt_en": "Use a compressed, nearly symmetrical portrait perspective, holding his face dead center with the folded forearms forming a firm visual base along the lower edge. Keep the depth of field very shallow so the room falls into subdued, desaturated softness, while restrained dusk illumination models the facial planes with firm contrast and preserves every slight tension around the raised mouth corner."
    },
    {
     "approach_ko": "가까운 거리의 자연스러운 원근과 비교적 깊은 초점으로 황폐한 공간의 압박감을 얼굴 뒤에 남기는 클로즈업.",
     "prompt_en": "Photograph from close range with a natural, slightly broader perspective, still eye-level and frontal, letting the face dominate while the folded forearms feel immediate at the bottom of frame. Maintain moderately deep focus so a narrow trace of the disordered room and entrance area remains physically legible behind him without competing for attention; shape the dusk ambience from the side for a restrained falloff across his displeased expression and worn surroundings."
    }
   ]
  },
  "reused": false
 },
 "S27sh3": {
  "input_fingerprint": "4dd9950546298b79",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 팔짱을 낀 채 입꼬리를 비틀어 올리고 못마땅한 표정을 지은 김형사의 얼굴 정면.\n\nLOCATION (lock): The bleak, disordered living room inside the rooftop apartment, close to the wind-rattled entrance door. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level directly in front of 김형사, hold a static close frame in which his face occupies most of the image while his folded forearms remain visible along the lower edge. Keep the composition nearly frontal and tense, with only minimal room context, preparing for the subsequent level pan toward the entrance.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 김형사 in the middle-center of the frame, foreground.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient illumination consistent with sunset supports a restrained, desaturated image and firm facial contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the disordered rooftop-home living room, dusk light, wind-disturbed doorway, and desolate atmosphere from the reference. EXCLUDE the woman using the smartphone; focus on the displeased male detective with folded arms and omit any restored furniture or cleaned surfaces.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The same ransacked state of the rooftop room persists around Hyesu and Detective Kim, including the footprints, open window, overturned chairs, and the now-empty red wall circle.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김형사 (한국인 남성, 30대의 얼굴, 평범한 얼굴선, 짧은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 팔짱을 낀 채 입꼬리를 비틀어 올리고 못마땅한 표정을 지은 김형사의 얼굴 정면.\n\nLOCATION (lock): The bleak, disordered living room inside the rooftop apartment, close to the wind-rattled entrance door. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level directly in front of 김형사, hold a static close frame in which his face occupies most of the image while his folded forearms remain visible along the lower edge. Keep the composition nearly frontal and tense, with only minimal room context, preparing for the subsequent level pan toward the entrance.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 김형사 in the middle-center of the frame, foreground.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient illumination consistent with sunset supports a restrained, desaturated image and firm facial contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the disordered rooftop-home living room, dusk light, wind-disturbed doorway, and desolate atmosphere from the reference. EXCLUDE the woman using the smartphone; focus on the displeased male detective with folded arms and omit any restored furniture or cleaned surfaces.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The same ransacked state of the rooftop room persists around Hyesu and Detective Kim, including the footprints, open window, overturned chairs, and the now-empty red wall circle.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김형사 (한국인 남성, 30대의 얼굴, 평범한 얼굴선, 짧은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a compressed, nearly symmetrical portrait perspective, holding his face dead center with the folded forearms forming a firm visual base along the lower edge. Keep the depth of field very shallow so the room falls into subdued, desaturated softness, while restrained dusk illumination models the facial planes with firm contrast and preserves every slight tension around the raised mouth corner.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 팔짱을 낀 채 입꼬리를 비틀어 올리고 못마땅한 표정을 지은 김형사의 얼굴 정면.\n\nLOCATION (lock): The bleak, disordered living room inside the rooftop apartment, close to the wind-rattled entrance door. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level directly in front of 김형사, hold a static close frame in which his face occupies most of the image while his folded forearms remain visible along the lower edge. Keep the composition nearly frontal and tense, with only minimal room context, preparing for the subsequent level pan toward the entrance.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 김형사 in the middle-center of the frame, foreground.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient illumination consistent with sunset supports a restrained, desaturated image and firm facial contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the disordered rooftop-home living room, dusk light, wind-disturbed doorway, and desolate atmosphere from the reference. EXCLUDE the woman using the smartphone; focus on the displeased male detective with folded arms and omit any restored furniture or cleaned surfaces.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The same ransacked state of the rooftop room persists around Hyesu and Detective Kim, including the footprints, open window, overturned chairs, and the now-empty red wall circle.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김형사 (한국인 남성, 30대의 얼굴, 평범한 얼굴선, 짧은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from close range with a natural, slightly broader perspective, still eye-level and frontal, letting the face dominate while the folded forearms feel immediate at the bottom of frame. Maintain moderately deep focus so a narrow trace of the disordered room and entrance area remains physically legible behind him without competing for attention; shape the dusk ambience from the side for a restrained falloff across his displeased expression and worn surroundings.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 7,
   "B": 4
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "클로즈업 프레이밍과 레퍼런스의 배경 구조(주방, 철문)는 잘 일치하나, 입꼬리를 올린 구체적인 표정과 바닥의 어질러진 흔적들이 누락되었습니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "입꼬리를 비틀어 올린 표정은 정확히 연출했으나, 레퍼런스와 전혀 다른 구조의 방(왼쪽 창문, 오른쪽 나무문)을 생성하여 장소 일관성을 심각하게 위반했습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S27sh1_sel.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "지시된 '입꼬리를 비틀어 올린' 표정이 반영되지 않고, 입술을 아래로 꾹 다문 채 찡그리고만 있습니다.",
     "fix_en": "Twist one corner of his mouth upwards into a tense smirk or sneer."
    },
    {
     "issue_ko": "배경의 바닥이 깨끗하게 치워져 있어 이전 샷의 어질러진 상태(진흙 발자국, 흩어진 서류, 쓰러진 의자들)가 유지되지 않았습니다.",
     "fix_en": "Restore the scattered papers, overturned chairs, and muddy footprints on the floor in the background to match the reference's ransacked state."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Twist one corner of his mouth upwards into a tense smirk or sneer.\n- Restore the scattered papers, overturned chairs, and muddy footprints on the floor in the background to match the reference's ransacked state.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "요구된 '입꼬리를 비틀어 올린' 표정이 아니며, 이전 샷의 어질러진 방바닥(서류, 쓰러진 의자, 발자국) 상태가 유지되지 않고 깨끗하게 지워져 지시를 위반했습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "입꼬리를 비틀어 올린 불만스러운 표정을 정확히 묘사했으며, 배경의 어질러진 바닥과 빈 벽 등 이전 샷의 상태를 충실하게 유지했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "요구된 '입꼬리를 비틀어 올린' 표정이 아니며, 이전 샷의 어질러진 방바닥(서류, 쓰러진 의자, 발자국) 상태가 유지되지 않고 깨끗하게 지워져 지시를 위반했습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "입꼬리를 비틀어 올린 불만스러운 표정을 정확히 묘사했으며, 배경의 어질러진 바닥과 빈 벽 등 이전 샷의 상태를 충실하게 유지했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "입꼬리를 비틀어 올린 디테일한 표정과 이전 샷의 어질러진 방 상태(종이, 발자국 등)를 매우 정확하게 유지하여 지시사항을 훌륭히 충족했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "입꼬리가 올라간 표정 묘사가 누락되었으며, 바닥의 난장판이 모두 깨끗하게 정리된 상태로 묘사되어 공간의 연속성 유지에 완전히 실패했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "입꼬리를 비틀어 올린 디테일한 표정과 이전 샷의 어질러진 방 상태(종이, 발자국 등)를 매우 정확하게 유지하여 지시사항을 훌륭히 충족했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "입꼬리가 올라간 표정 묘사가 누락되었으며, 바닥의 난장판이 모두 깨끗하게 정리된 상태로 묘사되어 공간의 연속성 유지에 완전히 실패했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 6,
     "B": 15
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S27sh1"
  }
 },
 "S28sh5::bgfirst_bg": {
  "input_fingerprint": "02e897f41fe0d37a",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 사진을 마주하고 두 눈이 커진 채 어깨를 흠칫 굳힌 인우의 상체.\n\nLOCATION (lock): The exposed deck of a small fishing boat underway at sea, beside the cabin entrance at dusk.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From close beside the axis connecting 인우 and the photograph, finish the pan on his three-quarter upper body and hold as his widened eyes drop toward the image and his shoulders lock. Keep the photograph small in the lower foreground and 인우 in the center-right midground, preserving enough deck context to ground his reaction.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: photograph (Held out before 인우) — Its image-bearing face is angled partly toward the camera and partly toward 인우, showing the photographed scene rather than its reverse; used as A small lower-foreground counterpoint connecting the reaction to its cause; fishing-boat deck (Visible around the seated position); used as Provides limited spatial context behind 인우 without competing with his reaction.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Fading ambient light appropriate to sunset is kept subdued and desaturated, with tense contrast concentrated on 인우's expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 사진을 마주하고 두 눈이 커진 채 어깨를 흠칫 굳힌 인우의 상체.\n\nLOCATION (lock): The exposed deck of a small fishing boat underway at sea, beside the cabin entrance at dusk.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From close beside the axis connecting 인우 and the photograph, finish the pan on his three-quarter upper body and hold as his widened eyes drop toward the image and his shoulders lock. Keep the photograph small in the lower foreground and 인우 in the center-right midground, preserving enough deck context to ground his reaction.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: photograph (Held out before 인우) — Its image-bearing face is angled partly toward the camera and partly toward 인우, showing the photographed scene rather than its reverse; used as A small lower-foreground counterpoint connecting the reaction to its cause; fishing-boat deck (Visible around the seated position); used as Provides limited spatial context behind 인우 without competing with his reaction.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Fading ambient light appropriate to sunset is kept subdued and desaturated, with tense contrast concentrated on 인우's expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S28sh5__bgfirst_bg.png",
  "asset_id": "f3df93d4-a97b-404f-aa10-9513e4d2c384",
  "input_asset_ids": [
   "930adb1e-0dd8-4620-9301-a22d75b1807d",
   "404d0c08-b5aa-4652-917d-02ffd1e347b3"
  ]
 },
 "S28sh5": {
  "input_fingerprint": "40faf1f063593f8b",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 사진을 마주하고 두 눈이 커진 채 어깨를 흠칫 굳힌 인우의 상체.\n\nLOCATION (lock): The exposed deck of a small fishing boat underway at sea, beside the cabin entrance at dusk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From close beside the axis connecting 인우 and the photograph, finish the pan on his three-quarter upper body and hold as his widened eyes drop toward the image and his shoulders lock. Keep the photograph small in the lower foreground and 인우 in the center-right midground, preserving enough deck context to ground his reaction.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: photograph (Held out before 인우) — Its image-bearing face is angled partly toward the camera and partly toward 인우, showing the photographed scene rather than its reverse; used as A small lower-foreground counterpoint connecting the reaction to its cause; fishing-boat deck (Visible around the seated position); used as Provides limited spatial context behind 인우 without competing with his reaction.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Fading ambient light appropriate to sunset is kept subdued and desaturated, with tense contrast concentrated on 인우's expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung has brought the folded printed maps and the bloodstained old photograph aboard, while her doll-adorned backpack remains with her. The wheelhouse walls already bear the numerous circular markings revealed moments later.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 남성, 20대 중반의 젊은 얼굴, 반듯한 얼굴선, 짧은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 사진을 마주하고 두 눈이 커진 채 어깨를 흠칫 굳힌 인우의 상체.\n\nLOCATION (lock): The exposed deck of a small fishing boat underway at sea, beside the cabin entrance at dusk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From close beside the axis connecting 인우 and the photograph, finish the pan on his three-quarter upper body and hold as his widened eyes drop toward the image and his shoulders lock. Keep the photograph small in the lower foreground and 인우 in the center-right midground, preserving enough deck context to ground his reaction.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: photograph (Held out before 인우) — Its image-bearing face is angled partly toward the camera and partly toward 인우, showing the photographed scene rather than its reverse; used as A small lower-foreground counterpoint connecting the reaction to its cause; fishing-boat deck (Visible around the seated position); used as Provides limited spatial context behind 인우 without competing with his reaction.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Fading ambient light appropriate to sunset is kept subdued and desaturated, with tense contrast concentrated on 인우's expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung has brought the folded printed maps and the bloodstained old photograph aboard, while her doll-adorned backpack remains with her. The wheelhouse walls already bear the numerous circular markings revealed moments later.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 남성, 20대 중반의 젊은 얼굴, 반듯한 얼굴선, 짧은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 사진을 마주하고 두 눈이 커진 채 어깨를 흠칫 굳힌 인우의 상체.\n\nLOCATION (lock): The exposed deck of a small fishing boat underway at sea, beside the cabin entrance at dusk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From close beside the axis connecting 인우 and the photograph, finish the pan on his three-quarter upper body and hold as his widened eyes drop toward the image and his shoulders lock. Keep the photograph small in the lower foreground and 인우 in the center-right midground, preserving enough deck context to ground his reaction.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: photograph (Held out before 인우) — Its image-bearing face is angled partly toward the camera and partly toward 인우, showing the photographed scene rather than its reverse; used as A small lower-foreground counterpoint connecting the reaction to its cause; fishing-boat deck (Visible around the seated position); used as Provides limited spatial context behind 인우 without competing with his reaction.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Fading ambient light appropriate to sunset is kept subdued and desaturated, with tense contrast concentrated on 인우's expression.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung has brought the folded printed maps and the bloodstained old photograph aboard, while her doll-adorned backpack remains with her. The wheelhouse walls already bear the numerous circular markings revealed moments later.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 남성, 20대 중반의 젊은 얼굴, 반듯한 얼굴선, 짧은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S28sh5__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S28sh5.png"
    },
    {
     "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:761032>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L18B02.png"
    },
    {
     "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:761032>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "레퍼런스 배경의 올바른 공간 구조(우측 선실)를 유지하고, 우측 중앙에 인물을 배치하라는 지시와 소품 디테일들을 훌륭히 구현했으나 샷 크기가 약간 넓습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "미디엄 샷의 크기와 인물의 표정 연기는 좋으나, 배경인 배의 좌우 구조가 거울처럼 완전히 뒤집히는 결정적인 위치 오류가 발생했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "레퍼런스 배경의 올바른 공간 구조(우측 선실)를 유지하고, 우측 중앙에 인물을 배치하라는 지시와 소품 디테일들을 훌륭히 구현했으나 샷 크기가 약간 넓습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "미디엄 샷의 크기와 인물의 표정 연기는 좋으나, 배경인 배의 좌우 구조가 거울처럼 완전히 뒤집히는 결정적인 위치 오류가 발생했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "정확한 미디엄 샷과 상체 프레이밍을 유지하며 인물의 시선과 긴장된 표정을 지시문대로 완벽히 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "상체만 보여야 하는 미디엄 샷 지시를 어기고 프레임을 넓혀 주변 소품을 보여주는 데 치중하여 감점되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "정확한 미디엄 샷과 상체 프레이밍을 유지하며 인물의 시선과 긴장된 표정을 지시문대로 완벽히 구현했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "상체만 보여야 하는 미디엄 샷 지시를 어기고 프레임을 넓혀 주변 소품을 보여주는 데 치중하여 감점되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 10,
     "B": 11
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": false,
    "policy": 1
   }
  },
  "totals": {
   "A": 10,
   "B": 11
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "레퍼런스 배경의 올바른 공간 구조(우측 선실)를 유지하고, 우측 중앙에 인물을 배치하라는 지시와 소품 디테일들을 훌륭히 구현했으나 샷 크기가 약간 넓습니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "미디엄 샷의 크기와 인물의 표정 연기는 좋으나, 배경인 배의 좌우 구조가 거울처럼 완전히 뒤집히는 결정적인 위치 오류가 발생했습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L18B02.png"
   },
   {
    "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:761032>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "샷 텍스트에 명시되지 않은 타인의 신체 일부(좌측 하단의 어깨와 사진을 들고 있는 손)가 화면에 등장했습니다.",
     "fix_en": "Remove the shoulder and hand holding the photograph from the bottom left foreground, ensuring no body parts of anyone other than In-woo are visible."
    },
    {
     "issue_ko": "조타실 외벽에 수많은 원형 표시들이 있어야 한다는 지시가 누락되었습니다.",
     "fix_en": "Add numerous circular markings onto the white exterior wall of the wheelhouse behind In-woo."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the shoulder and hand holding the photograph from the bottom left foreground, ensuring no body parts of anyone other than In-woo are visible.\n- Add numerous circular markings onto the white exterior wall of the wheelhouse behind In-woo.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "인물의 표정과 구도 모두 훌륭하며, 특히 조타실 외벽에 명시된 수많은 원형 표식들을 정확히 반영하여 프롬프트의 세부 지시를 완벽히 이행했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "구도와 인물의 감정 표현은 뛰어나지만, 프롬프트에 명시된 조타실 벽면의 원형 표식 디테일이 누락되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "인물의 표정과 구도 모두 훌륭하며, 특히 조타실 외벽에 명시된 수많은 원형 표식들을 정확히 반영하여 프롬프트의 세부 지시를 완벽히 이행했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "구도와 인물의 감정 표현은 뛰어나지만, 프롬프트에 명시된 조타실 벽면의 원형 표식 디테일이 누락되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "인물, 촬영 구도, 조명을 매우 충실히 구현했으며, 특히 프롬프트가 요구한 조타실 벽면의 수많은 원형 표시들을 정확하게 반영했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "인물의 표정과 전체적인 구도는 훌륭하나, 조타실 벽면에 명시된 수많은 원형 표시들이 묘사되지 않아 세부 지시 사항을 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "인물, 촬영 구도, 조명을 매우 충실히 구현했으며, 특히 프롬프트가 요구한 조타실 벽면의 수많은 원형 표시들을 정확하게 반영했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "인물의 표정과 전체적인 구도는 훌륭하나, 조타실 벽면에 명시된 수많은 원형 표시들이 묘사되지 않아 세부 지시 사항을 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 12,
     "B": 16
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S28sh5__bgfirst_bg.png",
   "bg_asset_id": "f3df93d4-a97b-404f-aa10-9513e4d2c384",
   "bg_record_key": "S28sh5::bgfirst_bg",
   "chain_winner": false,
   "authority": "plate"
  },
  "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L18B01",
    "B": "L18B02"
   },
   "assigned": "L18B02",
   "choice": "Candidate B",
   "confident": true,
   "reason_ko": "지문에 명시된 '사진'이 갑판 위 초록색 상자에 놓여 있는 것을 Candidate B에서 확인할 수 있습니다.",
   "kept": "L18B02"
  }
 },
 "S28sh9::variants": {
  "author_fp": "26a44d9bbe1033ef",
  "author": {
   "variants": [
    {
     "approach_ko": "정면 대칭과 깊은 초점으로 문틀과 양팔의 수평선을 냉정하고 완강한 장벽처럼 강조한다.",
     "prompt_en": "Use a restrained, natural-perspective lens feel and rigorous head-on symmetry, with the doorway’s verticals and Suryoung’s outstretched arms forming a stark, immovable crossbar composition. Hold both depth planes in firm focus so her planted back and Inwoo beyond remain spatially distinct, while soft desaturated dusk light outlines the threshold against the dimmer wheelhouse interior."
    },
    {
     "approach_ko": "망원 압축과 선택적 초점으로 수리영의 등이 인우를 짓누르는 듯한 밀도 높은 대치로 포착한다.",
     "prompt_en": "Use a subtly compressed lens feel and shallow, controlled focus centered on Suryoung’s back and rigidly extended arms, letting Inwoo beyond remain recognizable but slightly softened by depth. Keep the interior low-key and tactile, with subdued dusk illumination grazing worn surfaces and a narrow edge of light defining her human outline without turning her into a silhouette."
    }
   ]
  },
  "reused": false
 },
 "S28sh9": {
  "input_fingerprint": "2177f06bf78da2ad",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 조타실 문턱에 서서 양팔을 벌린 채 인우의 진입을 가로막고 버티는 수리영의 정지 구도.\n\nLOCATION (lock): The fishing boat’s open deck directly at the wheelhouse doorway. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From several steps inside the wheelhouse at chest height, frame the open doorway head-on with 수리영 centered across it, her spread arms forming a hard horizontal barrier. Her back occupies the midground while 인우 is held beyond her in the doorway, compressing both figures along the same axis without changing the static camera position.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground; 인우 in the middle-center of the frame, background; wheelhouse doorway in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: wheelhouse doorway (Open and physically blocked by 수리영) — The opening faces the camera from inside the wheelhouse, with 인우 visible beyond 수리영; used as Frames 수리영's outstretched arms and separates the wheelhouse interior from 인우's approach.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient illumination consistent with sunset keeps the doorway confrontation desaturated and sharply separated in depth.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the weathered boat deck, dusk marine light, cabin exterior, and rolling-sea atmosphere from the reference. EXCLUDE the close reaction to the photograph; show the young woman planted in the wheelhouse doorway with both arms spread to block entry.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The wheelhouse walls already carry numerous strange circular markings, including a red circle. Suryoung still possesses her backpack, maps, and old photograph as she blocks Inwoo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 조타실 문턱에 서서 양팔을 벌린 채 인우의 진입을 가로막고 버티는 수리영의 정지 구도.\n\nLOCATION (lock): The fishing boat’s open deck directly at the wheelhouse doorway. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From several steps inside the wheelhouse at chest height, frame the open doorway head-on with 수리영 centered across it, her spread arms forming a hard horizontal barrier. Her back occupies the midground while 인우 is held beyond her in the doorway, compressing both figures along the same axis without changing the static camera position.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground; 인우 in the middle-center of the frame, background; wheelhouse doorway in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: wheelhouse doorway (Open and physically blocked by 수리영) — The opening faces the camera from inside the wheelhouse, with 인우 visible beyond 수리영; used as Frames 수리영's outstretched arms and separates the wheelhouse interior from 인우's approach.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient illumination consistent with sunset keeps the doorway confrontation desaturated and sharply separated in depth.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the weathered boat deck, dusk marine light, cabin exterior, and rolling-sea atmosphere from the reference. EXCLUDE the close reaction to the photograph; show the young woman planted in the wheelhouse doorway with both arms spread to block entry.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The wheelhouse walls already carry numerous strange circular markings, including a red circle. Suryoung still possesses her backpack, maps, and old photograph as she blocks Inwoo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained, natural-perspective lens feel and rigorous head-on symmetry, with the doorway’s verticals and Suryoung’s outstretched arms forming a stark, immovable crossbar composition. Hold both depth planes in firm focus so her planted back and Inwoo beyond remain spatially distinct, while soft desaturated dusk light outlines the threshold against the dimmer wheelhouse interior.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 조타실 문턱에 서서 양팔을 벌린 채 인우의 진입을 가로막고 버티는 수리영의 정지 구도.\n\nLOCATION (lock): The fishing boat’s open deck directly at the wheelhouse doorway. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From several steps inside the wheelhouse at chest height, frame the open doorway head-on with 수리영 centered across it, her spread arms forming a hard horizontal barrier. Her back occupies the midground while 인우 is held beyond her in the doorway, compressing both figures along the same axis without changing the static camera position.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground; 인우 in the middle-center of the frame, background; wheelhouse doorway in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: wheelhouse doorway (Open and physically blocked by 수리영) — The opening faces the camera from inside the wheelhouse, with 인우 visible beyond 수리영; used as Frames 수리영's outstretched arms and separates the wheelhouse interior from 인우's approach.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient illumination consistent with sunset keeps the doorway confrontation desaturated and sharply separated in depth.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the weathered boat deck, dusk marine light, cabin exterior, and rolling-sea atmosphere from the reference. EXCLUDE the close reaction to the photograph; show the young woman planted in the wheelhouse doorway with both arms spread to block entry.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The wheelhouse walls already carry numerous strange circular markings, including a red circle. Suryoung still possesses her backpack, maps, and old photograph as she blocks Inwoo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a subtly compressed lens feel and shallow, controlled focus centered on Suryoung’s back and rigidly extended arms, letting Inwoo beyond remain recognizable but slightly softened by depth. Keep the interior low-key and tactile, with subdued dusk illumination grazing worn surfaces and a narrow edge of light defining her human outline without turning her into a silhouette.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 7,
   "B": 4
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "조타실 내부에서 밖을 향하는 지정된 카메라 앵글을 잘 구현했으나, 배낭 등의 소지품이 누락되었습니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "배낭 등 소지품 묘사는 좋으나, 카메라가 조타실 외부에 위치해 핵심 프레이밍 지시를 어겼습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S28sh5_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "수리영이 메고 있어야 할 배낭과 들고 있어야 할 지도, 낡은 사진이 보이지 않습니다.",
     "fix_en": "Add a backpack onto Suryoung's back and place maps and an old photograph in her hands."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add a backpack onto Suryoung's back and place maps and an old photograph in her hands.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "배낭과 들고 있어야 할 소품들이 누락되었으나, 구도와 조명, 인물의 배치가 자연스럽게 구현되어 전반적인 사실성을 유지했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "요구된 소품(사진, 지도, 배낭)을 포함하려 했으나, 손에 든 물건들이 조명과 원근감이 무시된 채 화면 위에 2D 스티커처럼 합성되어 '재질 현실성(Material Realism)' 지침을 심각하게 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "배낭과 들고 있어야 할 소품들이 누락되었으나, 구도와 조명, 인물의 배치가 자연스럽게 구현되어 전반적인 사실성을 유지했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "요구된 소품(사진, 지도, 배낭)을 포함하려 했으나, 손에 든 물건들이 조명과 원근감이 무시된 채 화면 위에 2D 스티커처럼 합성되어 '재질 현실성(Material Realism)' 지침을 심각하게 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "지도와 사진이 물리적 소품이 아닌 화면 위에 평면적으로 합성된 스티커(콜라주/그래픽 오버레이) 형태이므로 치명적인 위반(Priority 1)입니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "지시된 앵글과 두 인물의 차단 구도는 정확히 구현되었으나, 수리영의 소지품(배낭, 지도, 사진)이 모두 누락되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "지도와 사진이 물리적 소품이 아닌 화면 위에 평면적으로 합성된 스티커(콜라주/그래픽 오버레이) 형태이므로 치명적인 위반(Priority 1)입니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "지시된 앵글과 두 인물의 차단 구도는 정확히 구현되었으나, 수리영의 소지품(배낭, 지도, 사진)이 모두 누락되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 12,
     "B": 6
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S28sh5"
  },
  "lane_policy": "ab_select_bypass:prev"
 },
 "S29sh4::bgfirst_bg": {
  "input_fingerprint": "95732b0cc7def495",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 낡은 사진 속 한국인 여자아이의 얼굴 위를 꾹 짚은 수리영의 떨리는 손가락 클로즈업.\n\nLOCATION (lock): Inside the cramped, worn wheelhouse of the small fishing boat, over the old photograph being shown between the two occupants.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Directly above the photograph at hand height, track into an isolated overhead detail of 수리영's trembling fingertip pressing the photographed girl's face. The finger and face share the center of the frame, while a narrow margin of the photograph remains visible to preserve its physical scale and worn condition.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: worn photograph (Worn and being pressed firmly) — The image-bearing face is fully presented to the overhead camera, with the Korean girl's face visible directly beneath the fingertip; used as Primary image surface beneath 수리영's fingertip.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient light appropriate to the sunset interior preserves tactile detail in the finger and photograph without introducing a new light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\n\nSHOT TEXT this background must serve (Korean): 낡은 사진 속 한국인 여자아이의 얼굴 위를 꾹 짚은 수리영의 떨리는 손가락 클로즈업.\n\nLOCATION (lock): Inside the cramped, worn wheelhouse of the small fishing boat, over the old photograph being shown between the two occupants.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Directly above the photograph at hand height, track into an isolated overhead detail of 수리영's trembling fingertip pressing the photographed girl's face. The finger and face share the center of the frame, while a narrow margin of the photograph remains visible to preserve its physical scale and worn condition.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: worn photograph (Worn and being pressed firmly) — The image-bearing face is fully presented to the overhead camera, with the Korean girl's face visible directly beneath the fingertip; used as Primary image surface beneath 수리영's fingertip.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient light appropriate to the sunset interior preserves tactile detail in the finger and photograph without introducing a new light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.",
  "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S29sh4__bgfirst_bg.png",
  "asset_id": "3b5aee04-ede0-41c4-9e54-d9bdb380ae57",
  "input_asset_ids": [
   "4c198da0-39f3-4a8c-9f2d-28c4e6b781a0",
   "5ff25a59-e3dd-468c-ab19-7cc37a54fc37"
  ]
 },
 "S29sh4": {
  "input_fingerprint": "2ed8844ad78e8ff8",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 낡은 사진 속 한국인 여자아이의 얼굴 위를 꾹 짚은 수리영의 떨리는 손가락 클로즈업.\n\nLOCATION (lock): Inside the cramped, worn wheelhouse of the small fishing boat, over the old photograph being shown between the two occupants. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Directly above the photograph at hand height, track into an isolated overhead detail of 수리영's trembling fingertip pressing the photographed girl's face. The finger and face share the center of the frame, while a narrow margin of the photograph remains visible to preserve its physical scale and worn condition.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: worn photograph (Worn and being pressed firmly) — The image-bearing face is fully presented to the overhead camera, with the Korean girl's face visible directly beneath the fingertip; used as Primary image surface beneath 수리영's fingertip.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient light appropriate to the sunset interior preserves tactile detail in the finger and photograph without introducing a new light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung retains the backpack and maps, and she holds the same bloodstained old photograph while pointing to the girl identified as Minsuk. The wheelhouse's many circular wall markings remain in place.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 낡은 사진 속 한국인 여자아이의 얼굴 위를 꾹 짚은 수리영의 떨리는 손가락 클로즈업.\n\nLOCATION (lock): Inside the cramped, worn wheelhouse of the small fishing boat, over the old photograph being shown between the two occupants. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Directly above the photograph at hand height, track into an isolated overhead detail of 수리영's trembling fingertip pressing the photographed girl's face. The finger and face share the center of the frame, while a narrow margin of the photograph remains visible to preserve its physical scale and worn condition.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: worn photograph (Worn and being pressed firmly) — The image-bearing face is fully presented to the overhead camera, with the Korean girl's face visible directly beneath the fingertip; used as Primary image surface beneath 수리영's fingertip.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient light appropriate to the sunset interior preserves tactile detail in the finger and photograph without introducing a new light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung retains the backpack and maps, and she holds the same bloodstained old photograph while pointing to the girl identified as Minsuk. The wheelhouse's many circular wall markings remain in place.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 낡은 사진 속 한국인 여자아이의 얼굴 위를 꾹 짚은 수리영의 떨리는 손가락 클로즈업.\n\nLOCATION (lock): Inside the cramped, worn wheelhouse of the small fishing boat, over the old photograph being shown between the two occupants. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Directly above the photograph at hand height, track into an isolated overhead detail of 수리영's trembling fingertip pressing the photographed girl's face. The finger and face share the center of the frame, while a narrow margin of the photograph remains visible to preserve its physical scale and worn condition.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: worn photograph (Worn and being pressed firmly) — The image-bearing face is fully presented to the overhead camera, with the Korean girl's face visible directly beneath the fingertip; used as Primary image surface beneath 수리영's fingertip.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient light appropriate to the sunset interior preserves tactile detail in the finger and photograph without introducing a new light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suryoung retains the backpack and maps, and she holds the same bloodstained old photograph while pointing to the girl identified as Minsuk. The wheelhouse's many circular wall markings remain in place.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반의 젊은 얼굴, 타원형 얼굴, 묶은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere."
  },
  "roll_refs": {
   "A": [
    {
     "label": "SHOT BACKGROUND",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S29sh4__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/conti/conti_S29sh4.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    },
    {
     "label": "PROP REFERENCE — 두 어린 여자아이와 원형 장식 건물이 찍힌 낡고 피 묻은 흑백사진: the exact object appearing in this shot; match its look, material and wear exactly.",
     "path": "<bytes:1268182>"
    }
   ],
   "B": [
    {
     "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L19B01.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:884717>"
    },
    {
     "label": "PROP REFERENCE — 두 어린 여자아이와 원형 장식 건물이 찍힌 낡고 피 묻은 흑백사진: the exact object appearing in this shot; match its look, material and wear exactly.",
     "path": "<bytes:1268182>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "지정된 사진 소품(두 소녀) 대신 전혀 다른 이미지를 생성했으며, 선실 벽의 원형 마커가 사진 안으로 유출되는 심각한 오류가 발생했습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "레퍼런스 사진 소품을 정확히 재현하고 피 묻은 질감을 살렸으며, 손가락이 얼굴을 짚고 있는 디테일한 클로즈업 구도를 충실히 따랐습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "지정된 사진 소품(두 소녀) 대신 전혀 다른 이미지를 생성했으며, 선실 벽의 원형 마커가 사진 안으로 유출되는 심각한 오류가 발생했습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "레퍼런스 사진 소품을 정확히 재현하고 피 묻은 질감을 살렸으며, 손가락이 얼굴을 짚고 있는 디테일한 클로즈업 구도를 충실히 따랐습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 소품 레퍼런스(두 소녀가 있는 흑백사진)를 정확히 반영하고 혈흔과 손가락의 짚는 동작을 클로즈업 구도로 잘 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "주어진 소품 레퍼런스와 전혀 다른 이미지(한 소녀의 얼굴만 크게 확대된 사진)를 생성하여 프롬프트의 소품 지침을 크게 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 소품 레퍼런스(두 소녀가 있는 흑백사진)를 정확히 반영하고 혈흔과 손가락의 짚는 동작을 클로즈업 구도로 잘 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "주어진 소품 레퍼런스와 전혀 다른 이미지(한 소녀의 얼굴만 크게 확대된 사진)를 생성하여 프롬프트의 소품 지침을 크게 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 7,
     "B": 15
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 7,
   "B": 15
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 3,
    "verdict_ko": "지정된 사진 소품(두 소녀) 대신 전혀 다른 이미지를 생성했으며, 선실 벽의 원형 마커가 사진 안으로 유출되는 심각한 오류가 발생했습니다."
   },
   {
    "label": "B",
    "score": 8,
    "verdict_ko": "레퍼런스 사진 소품을 정확히 재현하고 피 묻은 질감을 살렸으며, 손가락이 얼굴을 짚고 있는 디테일한 클로즈업 구도를 충실히 따랐습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/episodes/8a0a9553-3523-4c59-b883-177c901cb630/images/background_chain/L19B01.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:884717>"
   },
   {
    "label": "PROP REFERENCE — 두 어린 여자아이와 원형 장식 건물이 찍힌 낡고 피 묻은 흑백사진: the exact object appearing in this shot; match its look, material and wear exactly.",
    "path": "<bytes:1268182>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "사진 속 건물 지붕 꼭대기에 있어야 할 원형 십자가 상징물이 누락되었습니다.",
     "fix_en": "Restore the large circular symbol on the peak of the building's roof inside the photograph exactly as it appears in the prop reference."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Restore the large circular symbol on the peak of the building's roof inside the photograph exactly as it appears in the prop reference.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 클로즈업 앵글과 낡은 사진의 디테일을 잘 살렸으며, 불필요한 시각적 왜곡이 적어 프롬프트에 충실합니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "구도와 앵글은 좋으나 사진 속 건물 지붕에 레퍼런스에 없는 임의의 문양이 생성되어 감점되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 클로즈업 앵글과 낡은 사진의 디테일을 잘 살렸으며, 불필요한 시각적 왜곡이 적어 프롬프트에 충실합니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "구도와 앵글은 좋으나 사진 속 건물 지붕에 레퍼런스에 없는 임의의 문양이 생성되어 감점되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 인서트 클로즈업 구도와 손가락 위치를 정확히 연출했으며, 소품의 물리적 질감을 잘 살렸습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "사진 상단에 원본 소품에는 없는 평면적인 워터마크(그래픽 오버레이)가 삽입되어 사실성 지침을 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요구된 인서트 클로즈업 구도와 손가락 위치를 정확히 연출했으며, 소품의 물리적 질감을 잘 살렸습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "사진 상단에 원본 소품에는 없는 평면적인 워터마크(그래픽 오버레이)가 삽입되어 사실성 지침을 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 14,
     "B": 9
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S29sh4__bgfirst_bg.png",
   "bg_asset_id": "3b5aee04-ede0-41c4-9e54-d9bdb380ae57",
   "bg_record_key": "S29sh4::bgfirst_bg",
   "chain_winner": false,
   "authority": "plate"
  },
  "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S29sh7::variants": {
  "author_fp": "873440de9a1595ec",
  "author": {
   "variants": [
    {
     "approach_ko": "정면에 가까운 절제된 와이드와 깊은 초점으로 붉은 원들의 반복과 낡은 벽의 물성을 냉정하게 드러낸다.",
     "prompt_en": "Use a restrained, near-frontal wide composition with the camera held level and the marked surface carrying most of the frame. Keep the full visible wall and receding edge planes in deep, even focus, letting the repeated circles accumulate with an unnerving forensic clarity. Soft, subdued dusk ambience reveals worn texture and layered pigment without theatrical highlights, with the red remaining the only saturated accent."
    },
    {
     "approach_ko": "넓은 원근감과 가장자리의 급격한 후퇴, 비스듬히 스치는 황혼빛으로 조타실의 압박감과 원 형상의 불길함을 강조한다.",
     "prompt_en": "Favor an assertive wide-angle sense of depth: hold the revealed wall square enough for every repeated form to register, while the adjoining planes taper sharply at the edges and make the room feel compressed around it. Let dim dusk illumination rake obliquely across the surface, catching uneven wear and the physical buildup of overlapping marks before falling quickly into desaturated edge shadow. Preserve the broad reveal while creating a subtle visual pull toward the densest layering of circles."
    }
   ]
  },
  "reused": false
 },
 "S29sh7": {
  "input_fingerprint": "d56152d102d5fc33",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 좁고 낡은 조타실 내부 벽면 가득 기괴하게 덧그려진 수많은 붉은 원 형상들의 섬뜩한 전경.\n\nLOCATION (lock): The cramped interior of the fishing boat’s old wheelhouse, with numerous strange red circles covering its walls. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At standing eye level from just beyond 인우's former shoulder line, complete the track on a wide view of the wall's full visible span. Hold the circle-covered face of the wall square enough to read the repeated forms, while slight perspective convergence preserves the cramped wheelhouse geometry.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: wheelhouse wall covered with red circles (Covered with many red circles) — The camera-facing wall surface bears numerous strange, overlapping red circular forms across its visible span; used as The central reveal, presented across the broad background as a repeated occult pattern; narrow wheelhouse interior (Narrow and old) — The wall planes recede away from the revealed marked surface at the frame edges; used as Provides the confined boundaries around the marked wall.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued sunset-appropriate ambient illumination holds the interior in desaturated shadow while the documented red circles provide the sole deep-red emphasis.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The wheelhouse walls remain covered with numerous strange circles, including the newly noticed red one; Suryoung still has the old photograph, maps, and backpack with her.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 좁고 낡은 조타실 내부 벽면 가득 기괴하게 덧그려진 수많은 붉은 원 형상들의 섬뜩한 전경.\n\nLOCATION (lock): The cramped interior of the fishing boat’s old wheelhouse, with numerous strange red circles covering its walls. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At standing eye level from just beyond 인우's former shoulder line, complete the track on a wide view of the wall's full visible span. Hold the circle-covered face of the wall square enough to read the repeated forms, while slight perspective convergence preserves the cramped wheelhouse geometry.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: wheelhouse wall covered with red circles (Covered with many red circles) — The camera-facing wall surface bears numerous strange, overlapping red circular forms across its visible span; used as The central reveal, presented across the broad background as a repeated occult pattern; narrow wheelhouse interior (Narrow and old) — The wall planes recede away from the revealed marked surface at the frame edges; used as Provides the confined boundaries around the marked wall.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued sunset-appropriate ambient illumination holds the interior in desaturated shadow while the documented red circles provide the sole deep-red emphasis.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The wheelhouse walls remain covered with numerous strange circles, including the newly noticed red one; Suryoung still has the old photograph, maps, and backpack with her.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained, near-frontal wide composition with the camera held level and the marked surface carrying most of the frame. Keep the full visible wall and receding edge planes in deep, even focus, letting the repeated circles accumulate with an unnerving forensic clarity. Soft, subdued dusk ambience reveals worn texture and layered pigment without theatrical highlights, with the red remaining the only saturated accent.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 좁고 낡은 조타실 내부 벽면 가득 기괴하게 덧그려진 수많은 붉은 원 형상들의 섬뜩한 전경.\n\nLOCATION (lock): The cramped interior of the fishing boat’s old wheelhouse, with numerous strange red circles covering its walls. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At standing eye level from just beyond 인우's former shoulder line, complete the track on a wide view of the wall's full visible span. Hold the circle-covered face of the wall square enough to read the repeated forms, while slight perspective convergence preserves the cramped wheelhouse geometry.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: wheelhouse wall covered with red circles (Covered with many red circles) — The camera-facing wall surface bears numerous strange, overlapping red circular forms across its visible span; used as The central reveal, presented across the broad background as a repeated occult pattern; narrow wheelhouse interior (Narrow and old) — The wall planes recede away from the revealed marked surface at the frame edges; used as Provides the confined boundaries around the marked wall.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued sunset-appropriate ambient illumination holds the interior in desaturated shadow while the documented red circles provide the sole deep-red emphasis.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The wheelhouse walls remain covered with numerous strange circles, including the newly noticed red one; Suryoung still has the old photograph, maps, and backpack with her.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFavor an assertive wide-angle sense of depth: hold the revealed wall square enough for every repeated form to register, while the adjoining planes taper sharply at the edges and make the room feel compressed around it. Let dim dusk illumination rake obliquely across the surface, catching uneven wear and the physical buildup of overlapping marks before falling quickly into desaturated edge shadow. Preserve the broad reveal while creating a subtle visual pull toward the densest layering of circles.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 4,
   "B": 8
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "지시되지 않은 기괴한 문자/문양이 원 안에 임의로 추가되었으며, 조타실의 공간감이 부족합니다."
   },
   {
    "label": "B",
    "score": 8,
    "verdict_ko": "좁은 조타실의 공간감과 해질녘의 조명을 잘 살렸으며, 요구된 붉은 원 형상들을 정확하게 묘사했습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S29sh4_sel.png"
   }
  ],
  "critique": {
   "issues": []
  },
  "fix_skipped": true,
  "ref_mode": "prev만 (배경 전용·공유 계획)",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S29sh4"
  },
  "lane_policy": "share_plan_prev_bgonly"
 },
 "S29sh10::variants": {
  "author_fp": "8a550d5f91582bff",
  "author": {
   "variants": [
    {
     "approach_ko": "얕은 심도와 압축된 원근으로 인우의 미세한 냉기와 시선에 집중하는 밀도 높은 클로즈업.",
     "prompt_en": "Use a compressed, intimate close-up with very shallow depth, holding Inwoo’s expressionless face in crisp focus while the lower foreground and marked wall dissolve softly. Let the subdued dusk ambience skim across his slightly lowered face, preserving natural skin texture and tiny tensions around the eyelids and mouth; the restrained red arcs behind him remain muted, out-of-focus traces of threat."
    },
    {
     "approach_ko": "근접한 광각 원근과 깊이감으로 아래에서 올려다보는 권력 관계와 비좁은 선실의 압박을 강조한 클로즈업.",
     "prompt_en": "Use a close, subtly wide perspective that makes the low upward viewpoint and vertical power relation feel immediate without distorting either face. Keep more depth legible through the frame: the unstable lower foreground falls away toward Inwoo in the upper center, while a narrow plane of the weathered, red-marked wall remains recognizable behind him. Shape the dusk light with restrained directional contrast, leaving his downward gaze cool, still, and severe."
    }
   ]
  },
  "reused": false
 },
 "S29sh10": {
  "input_fingerprint": "6e2c44e0fd613060",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 고개를 살짝 숙여 수리영을 무표정하고 차갑게 내려다보는 인우의 서늘한 시선.\n\nLOCATION (lock): Inside the narrow wheelhouse of the fishing boat, amid the walls marked with red circles. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Dolly in from near 수리영's lowered eye level with a slight upward lens axis, placing 인우's expressionless face in the upper center as his eyes cut downward. Keep 수리영 partial and unsteady in the lower foreground, allowing his face to remain the close-frame priority while their vertical power relation stays unmistakable.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 인우 in the upper-center of the frame, midground, looks toward 수리영 below him; 수리영 in the lower-left of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: wall with red circles (Partially visible and covered with red circles) — A limited portion of the camera-facing marked wall remains visible behind 인우, carrying partial red circular forms; used as Falls soft behind 인우, retaining a trace of the preceding reveal without distracting from his face.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient light appropriate to sunset keeps 인우's face coolly desaturated in tone, with any visible red wall markings held as a restrained threat.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the cramped, weathered wheelhouse, dim dusk illumination, and the many red circular markings covering the interior walls from the reference. EXCLUDE the empty wide-room emphasis; frame the boatman looking coldly downward and do not add the photograph or pointing hand.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The many circular markings remain across the wheelhouse walls. Suryoung still carries her backpack, maps, and old photograph as she becomes disoriented before collapsing onto Inwoo's shoulder.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 남성, 20대 중반의 젊은 얼굴, 반듯한 얼굴선, 짧은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 고개를 살짝 숙여 수리영을 무표정하고 차갑게 내려다보는 인우의 서늘한 시선.\n\nLOCATION (lock): Inside the narrow wheelhouse of the fishing boat, amid the walls marked with red circles. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Dolly in from near 수리영's lowered eye level with a slight upward lens axis, placing 인우's expressionless face in the upper center as his eyes cut downward. Keep 수리영 partial and unsteady in the lower foreground, allowing his face to remain the close-frame priority while their vertical power relation stays unmistakable.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 인우 in the upper-center of the frame, midground, looks toward 수리영 below him; 수리영 in the lower-left of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: wall with red circles (Partially visible and covered with red circles) — A limited portion of the camera-facing marked wall remains visible behind 인우, carrying partial red circular forms; used as Falls soft behind 인우, retaining a trace of the preceding reveal without distracting from his face.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient light appropriate to sunset keeps 인우's face coolly desaturated in tone, with any visible red wall markings held as a restrained threat.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the cramped, weathered wheelhouse, dim dusk illumination, and the many red circular markings covering the interior walls from the reference. EXCLUDE the empty wide-room emphasis; frame the boatman looking coldly downward and do not add the photograph or pointing hand.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The many circular markings remain across the wheelhouse walls. Suryoung still carries her backpack, maps, and old photograph as she becomes disoriented before collapsing onto Inwoo's shoulder.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 남성, 20대 중반의 젊은 얼굴, 반듯한 얼굴선, 짧은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a compressed, intimate close-up with very shallow depth, holding Inwoo’s expressionless face in crisp focus while the lower foreground and marked wall dissolve softly. Let the subdued dusk ambience skim across his slightly lowered face, preserving natural skin texture and tiny tensions around the eyelids and mouth; the restrained red arcs behind him remain muted, out-of-focus traces of threat.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 고개를 살짝 숙여 수리영을 무표정하고 차갑게 내려다보는 인우의 서늘한 시선.\n\nLOCATION (lock): Inside the narrow wheelhouse of the fishing boat, amid the walls marked with red circles. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Dolly in from near 수리영's lowered eye level with a slight upward lens axis, placing 인우's expressionless face in the upper center as his eyes cut downward. Keep 수리영 partial and unsteady in the lower foreground, allowing his face to remain the close-frame priority while their vertical power relation stays unmistakable.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 인우 in the upper-center of the frame, midground, looks toward 수리영 below him; 수리영 in the lower-left of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: wall with red circles (Partially visible and covered with red circles) — A limited portion of the camera-facing marked wall remains visible behind 인우, carrying partial red circular forms; used as Falls soft behind 인우, retaining a trace of the preceding reveal without distracting from his face.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued ambient light appropriate to sunset keeps 인우's face coolly desaturated in tone, with any visible red wall markings held as a restrained threat.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): TAKE the cramped, weathered wheelhouse, dim dusk illumination, and the many red circular markings covering the interior walls from the reference. EXCLUDE the empty wide-room emphasis; frame the boatman looking coldly downward and do not add the photograph or pointing hand.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The many circular markings remain across the wheelhouse walls. Suryoung still carries her backpack, maps, and old photograph as she becomes disoriented before collapsing onto Inwoo's shoulder.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 남성, 20대 중반의 젊은 얼굴, 반듯한 얼굴선, 짧은 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a close, subtly wide perspective that makes the low upward viewpoint and vertical power relation feel immediate without distorting either face. Keep more depth legible through the frame: the unstable lower foreground falls away toward Inwoo in the upper center, while a narrow plane of the weathered, red-marked wall remains recognizable behind him. Shape the dusk light with restrained directional contrast, leaving his downward gaze cool, still, and severe.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 8,
   "B": 4
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 8,
    "verdict_ko": "프레임 레이아웃과 붉은 원이 그려진 카메라를 향한 배경 벽을 지시대로 정확히 구현했으며, 인물의 시선과 조명 분위기도 훌륭합니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "인물의 외모는 참조와 가깝지만, 인물 뒤로 붉은 원이 있는 벽이 배치되지 않아 배경 설정에 어긋나며 소품(사진)이 부자연스럽게 카메라를 향하고 있습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S29sh7_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:761032>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "샷 크기가 클로즈업이 아니며, 지시사항에서 배제하라고 한 넓은 선실 내부(양옆 창문 등)가 모두 보입니다.",
     "fix_en": "Tighten the framing to a compressed close-up of Inwoo's face, cropping out the wide room and side windows."
    },
    {
     "issue_ko": "배경의 빨간 원형 무늬 벽에 초점이 선명하게 맞아, 매우 얕은 피사계 심도로 배경을 부드럽게 흐리라는 지시를 위반했습니다.",
     "fix_en": "Apply a very shallow depth of field to make the background wall softly out of focus, keeping crisp focus only on Inwoo's face."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Tighten the framing to a compressed close-up of Inwoo's face, cropping out the wide room and side windows.\n- Apply a very shallow depth of field to make the background wall softly out of focus, keeping crisp focus only on Inwoo's face.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "지시된 클로즈업과 얕은 심도를 무시하고 너무 넓은 화각으로 배경을 과도하게 보여주어 프레이밍 규칙을 위반했습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 압축된 클로즈업 화각과 얕은 심도를 정확히 구현했으며, 인물들의 위치와 시선 처리도 지시문과 완벽하게 일치합니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "지시된 클로즈업과 얕은 심도를 무시하고 너무 넓은 화각으로 배경을 과도하게 보여주어 프레이밍 규칙을 위반했습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 압축된 클로즈업 화각과 얕은 심도를 정확히 구현했으며, 인물들의 위치와 시선 처리도 지시문과 완벽하게 일치합니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "클로즈업 샷 크기와 얕은 피사계 심도 등 프레이밍 지시사항을 정확하게 따르며 인물의 표정을 잘 포착했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "지정된 클로즈업을 무시하고 넓은 화각으로 배경을 과도하게 노출하여 프레이밍 지시를 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "클로즈업 샷 크기와 얕은 피사계 심도 등 프레이밍 지시사항을 정확하게 따르며 인물의 표정을 잘 포착했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "지정된 클로즈업을 무시하고 넓은 화각으로 배경을 과도하게 노출하여 프레이밍 지시를 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 9,
     "B": 15
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S29sh7"
  }
 },
 "S30sh2::variants": {
  "author_fp": "77a3d3caf8acc750",
  "author": {
   "variants": [
    {
     "approach_ko": "해수면 바로 위의 정면 대칭 구도로 안개 장벽의 수직적 규모와 압도감을 강조한다.",
     "prompt_en": "Use an expansive wide-lens feel with severe frontal symmetry, holding the camera just above the sea and letting the upward framing make the fog bank dominate almost the entire image. Keep the narrow golden water strip visually compressed along the bottom edge while dense, physically textured gray vapor rises as a monumental continuous field. Render the restrained sunset reflection with soft specular detail and preserve deep, subdued tonal separation within the fog."
    },
    {
     "approach_ko": "압축된 원근과 절제된 명암층으로 바다와 하늘이 하나의 잿빛 덩어리로 합쳐지는 불길함을 강조한다.",
     "prompt_en": "Use a compressed lens feel from the same low, head-on vantage, flattening the fog bank into an impenetrable gray plane with no readable horizon line. Build the composition through broad horizontal tonal layers: a very narrow, dim golden reflection below and an immense neutral-dark fog mass above, with subtle density variation providing depth rather than a visible boundary. Keep the image austere and low in contrast, allowing only the muted warmth on the sea to interrupt the enveloping gray."
    }
   ]
  },
  "reused": false
 },
 "S30sh2": {
  "input_fingerprint": "1235d5c5c5b9ffe2",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog.\n\nSHOT TEXT (authoritative, Korean): 바다와 하늘의 경계를 완전히 지워버리듯 수평선을 가로막고 솟아오른 짙은 잿빛 안개 장벽의 압도적인 전경.\n\nLOCATION (lock): The open sea at dusk before a dense gray wall of fog that erases the visible horizon. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From a low, wide vantage just above sea level, complete the upward tilt with the dense fog bank extending across the horizon head-on and erasing the division between sea and sky. Exclude most nearby water, retaining only a narrow lower band of golden reflection beneath the immense fog field before the camera begins its forward approach.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: dense dark-gray fog bank (Dense and extending across the full horizon); used as Primary environmental subject spanning the horizon and obscuring the sea-sky boundary; sea (Visible mainly as a narrow foreground band) — The visible surface reflects the golden light of the setting sun beneath the fog bank; used as A narrow lower strip that carries the sunset reflection and establishes the camera's low height; horizon (Obscured by the fog bank); used as Its disappearance is the visual measure of the fog bank's scale.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Golden sunset reflection is restrained against the deep gray fog, with subdued luminance and ominous warm-versus-neutral contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog.\n\nSHOT TEXT (authoritative, Korean): 바다와 하늘의 경계를 완전히 지워버리듯 수평선을 가로막고 솟아오른 짙은 잿빛 안개 장벽의 압도적인 전경.\n\nLOCATION (lock): The open sea at dusk before a dense gray wall of fog that erases the visible horizon. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From a low, wide vantage just above sea level, complete the upward tilt with the dense fog bank extending across the horizon head-on and erasing the division between sea and sky. Exclude most nearby water, retaining only a narrow lower band of golden reflection beneath the immense fog field before the camera begins its forward approach.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: dense dark-gray fog bank (Dense and extending across the full horizon); used as Primary environmental subject spanning the horizon and obscuring the sea-sky boundary; sea (Visible mainly as a narrow foreground band) — The visible surface reflects the golden light of the setting sun beneath the fog bank; used as A narrow lower strip that carries the sunset reflection and establishes the camera's low height; horizon (Obscured by the fog bank); used as Its disappearance is the visual measure of the fog bank's scale.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Golden sunset reflection is restrained against the deep gray fog, with subdued luminance and ominous warm-versus-neutral contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an expansive wide-lens feel with severe frontal symmetry, holding the camera just above the sea and letting the upward framing make the fog bank dominate almost the entire image. Keep the narrow golden water strip visually compressed along the bottom edge while dense, physically textured gray vapor rises as a monumental continuous field. Render the restrained sunset reflection with soft specular detail and preserve deep, subdued tonal separation within the fog.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog.\n\nSHOT TEXT (authoritative, Korean): 바다와 하늘의 경계를 완전히 지워버리듯 수평선을 가로막고 솟아오른 짙은 잿빛 안개 장벽의 압도적인 전경.\n\nLOCATION (lock): The open sea at dusk before a dense gray wall of fog that erases the visible horizon. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From a low, wide vantage just above sea level, complete the upward tilt with the dense fog bank extending across the horizon head-on and erasing the division between sea and sky. Exclude most nearby water, retaining only a narrow lower band of golden reflection beneath the immense fog field before the camera begins its forward approach.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: dense dark-gray fog bank (Dense and extending across the full horizon); used as Primary environmental subject spanning the horizon and obscuring the sea-sky boundary; sea (Visible mainly as a narrow foreground band) — The visible surface reflects the golden light of the setting sun beneath the fog bank; used as A narrow lower strip that carries the sunset reflection and establishes the camera's low height; horizon (Obscured by the fog bank); used as Its disappearance is the visual measure of the fog bank's scale.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Golden sunset reflection is restrained against the deep gray fog, with subdued luminance and ominous warm-versus-neutral contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a compressed lens feel from the same low, head-on vantage, flattening the fog bank into an impenetrable gray plane with no readable horizon line. Build the composition through broad horizontal tonal layers: a very narrow, dim golden reflection below and an immense neutral-dark fog mass above, with subtle density variation providing depth rather than a visible boundary. Keep the image austere and low in contrast, allowing only the muted warmth on the sea to interrupt the enveloping gray.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "A": 8,
   "B": 5
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 8,
    "verdict_ko": "수평선을 가로막은 거대한 잿빛 안개 장벽과 좁은 바다 표면의 황금빛 반사가 지시문에 정확히 부합합니다."
   },
   {
    "label": "B",
    "score": 5,
    "verdict_ko": "안개가 평범한 흐린 하늘처럼 보여 '솟아오른 장벽'이라는 압도적인 느낌과 형태를 살리지 못했습니다."
   }
  ],
  "refs": [],
  "critique": {
   "issues": []
  },
  "fix_skipped": true,
  "ref_mode": "플레이트만 (배경 전용)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S30sh3::variants": {
  "author_fp": "0e07c1bbee3dbcae",
  "author": {
   "variants": [
    {
     "approach_ko": "수면에 밀착한 광각 시점으로 갈라지는 물살의 촉감과 뱃머리의 전진감을 강하게 포착한다.",
     "prompt_en": "Use an immersive wide-angle perspective with the camera skimming just above the sea, making the V-shaped division of the water the strongest graphic path through the lower frame while the three-quarter bow remains naturally scaled in the lower right. Hold broad depth so the worn hull, glossy disturbed water, and soft wall of fog remain physically legible together; let the muted dusk reflection rake faintly across the ripples before dissolving into low-contrast mist."
    },
    {
     "approach_ko": "완만한 장초점 압축으로 어선과 안개 띠 사이의 거리를 눌러, 배가 안개에 삼켜지는 긴장을 강조한다.",
     "prompt_en": "Photograph the wide shot with a restrained long-lens feel, compressing the visible distance between the advancing bow and the dense fog so the mist seems to close around its route. Keep the camera low and tracking backward as specified, but favor layered tonal recession: crisp pressure and texture where the bow parts the sea, then rapidly softened hull edges and subdued dusk color toward the fog, creating a quiet sense of the boat being progressively erased."
    }
   ]
  },
  "reused": false
 },
 "S30sh3": {
  "input_fingerprint": "17dc9e3a1b8e2292",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog.\n\nSHOT TEXT (authoritative, Korean): 두꺼운 안개 띠 속으로 낡은 어선이 진입하며 뱃머리를 따라 물살이 V자로 갈라지는 찰나의 모습.\n\nLOCATION (lock): The open sea at the leading edge of a dense fog bank, where a small fishing boat is entering the mist. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track backward just above sea level from ahead and slightly beside the bow, briefly matching the boat's speed as the lens lowers toward the instant the water divides into a V. Keep the bow in the lower-right portion at a natural scale while the open V-shaped wake leads toward the lower center and the dense fog occupies the route ahead.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: old fishing boat (Entering the dense fog) — Its bow presents a three-quarter front face to the camera as it advances past the lens axis; used as Moving foreground subject whose bow creates the V-shaped division in the water; sea around the bow (Parting into a V-shape); used as Forms the V-shaped lines around the bow and provides visible forward motion; dense fog band (Thick around the boat's line of entry); used as The destination of the boat and the background that progressively removes its outline.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The remaining golden sunset reflection is muted by the dense fog, maintaining low luminance and restrained desaturated tension.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The same old fishing boat carrying Inwoo and the unconscious Suryoung continues forward into the dense fog bank.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.",
  "roll_prompts": {
   "A": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog.\n\nSHOT TEXT (authoritative, Korean): 두꺼운 안개 띠 속으로 낡은 어선이 진입하며 뱃머리를 따라 물살이 V자로 갈라지는 찰나의 모습.\n\nLOCATION (lock): The open sea at the leading edge of a dense fog bank, where a small fishing boat is entering the mist. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track backward just above sea level from ahead and slightly beside the bow, briefly matching the boat's speed as the lens lowers toward the instant the water divides into a V. Keep the bow in the lower-right portion at a natural scale while the open V-shaped wake leads toward the lower center and the dense fog occupies the route ahead.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: old fishing boat (Entering the dense fog) — Its bow presents a three-quarter front face to the camera as it advances past the lens axis; used as Moving foreground subject whose bow creates the V-shaped division in the water; sea around the bow (Parting into a V-shape); used as Forms the V-shaped lines around the bow and provides visible forward motion; dense fog band (Thick around the boat's line of entry); used as The destination of the boat and the background that progressively removes its outline.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The remaining golden sunset reflection is muted by the dense fog, maintaining low luminance and restrained desaturated tension.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The same old fishing boat carrying Inwoo and the unconscious Suryoung continues forward into the dense fog bank.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an immersive wide-angle perspective with the camera skimming just above the sea, making the V-shaped division of the water the strongest graphic path through the lower frame while the three-quarter bow remains naturally scaled in the lower right. Hold broad depth so the worn hull, glossy disturbed water, and soft wall of fog remain physically legible together; let the muted dusk reflection rake faintly across the ripples before dissolving into low-contrast mist.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog.\n\nSHOT TEXT (authoritative, Korean): 두꺼운 안개 띠 속으로 낡은 어선이 진입하며 뱃머리를 따라 물살이 V자로 갈라지는 찰나의 모습.\n\nLOCATION (lock): The open sea at the leading edge of a dense fog bank, where a small fishing boat is entering the mist. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track backward just above sea level from ahead and slightly beside the bow, briefly matching the boat's speed as the lens lowers toward the instant the water divides into a V. Keep the bow in the lower-right portion at a natural scale while the open V-shaped wake leads toward the lower center and the dense fog occupies the route ahead.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: old fishing boat (Entering the dense fog) — Its bow presents a three-quarter front face to the camera as it advances past the lens axis; used as Moving foreground subject whose bow creates the V-shaped division in the water; sea around the bow (Parting into a V-shape); used as Forms the V-shaped lines around the bow and provides visible forward motion; dense fog band (Thick around the boat's line of entry); used as The destination of the boat and the background that progressively removes its outline.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The remaining golden sunset reflection is muted by the dense fog, maintaining low luminance and restrained desaturated tension.\n- MATERIAL REALISM: every object and surface must read as a real,\n  physical material — correct texture, weight, wear and light response\n  (metal reflects, fabric drapes and creases, liquid is glossy and\n  pools, painted or drawn marks sit ON a surface and follow its\n  curvature and lighting). Nothing may look like a flat sticker, a\n  doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The same old fishing boat carrying Inwoo and the unconscious Suryoung continues forward into the dense fog bank.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the wide shot with a restrained long-lens feel, compressing the visible distance between the advancing bow and the dense fog so the mist seems to close around its route. Keep the camera low and tracking backward as specified, but favor layered tonal recession: crisp pressure and texture where the bow parts the sea, then rapidly softened hull edges and subdued dusk color toward the fog, creating a quiet sense of the boat being progressively erased.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins."
  },
  "totals": {
   "B": 7,
   "A": 4
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "카메라가 뱃머리 앞 측면에서 후퇴하며 V자 물살과 3/4 정면 구도를 지침대로 정확히 구현했습니다."
   },
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "지시된 3/4 정면 앵글이 아닌 선박의 측면을 보여주어 카메라 배치 지침을 따르지 않았습니다."
   }
  ],
  "refs": [
   {
    "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/705ead84-e08d-44fd-a244-0a671a7a4f7e/images/8a0a9553-3523-4c59-b883-177c901cb630/scene/recipe/S30sh2_sel.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "어선이 짙은 안개 속으로 진입해야 한다는 설정과 달리, 안개가 배 뒤편에만 있고 배의 진행 방향(왼쪽 및 앞쪽)은 맑은 상태입니다.",
     "fix_en": "Fill the left side and foreground with dense fog occupying the boat's forward path, enveloping the advancing bow to show it entering the mist."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Fill the left side and foreground with dense fog occupying the boat's forward path, enveloping the advancing bow to show it entering the mist.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.",
  "fix_rejudge": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 구도(우측 하단의 뱃머리와 중앙으로 향하는 V자 물살)와 레퍼런스의 황혼 조명 및 안개 띠를 매우 충실하게 구현함."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "V자 물살의 형태와 구도가 지시사항에 덜 부합하며, 레퍼런스의 특징적인 조명과 안개 구조가 다소 약함."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 구도(우측 하단의 뱃머리와 중앙으로 향하는 V자 물살)와 레퍼런스의 황혼 조명 및 안개 띠를 매우 충실하게 구현함."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "V자 물살의 형태와 구도가 지시사항에 덜 부합하며, 레퍼런스의 특징적인 조명과 안개 구조가 다소 약함."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "레퍼런스의 황혼 무드와 안개 질감을 완벽하게 유지하면서, 뱃머리와 V자 물살의 구도를 정확히 구현함."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "구도와 피사체의 형태는 지시사항을 잘 따랐으나, 레퍼런스의 황금빛 노을 반사가 부족하고 톤이 다소 밝음."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "레퍼런스의 황혼 무드와 안개 질감을 완벽하게 유지하면서, 뱃머리와 V자 물살의 구도를 정확히 구현함."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "구도와 피사체의 형태는 지시사항을 잘 따랐으나, 레퍼런스의 황금빛 노을 반사가 부족하고 톤이 다소 밝음."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 16,
     "B": 12
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev만 (배경 전용·공유 계획)",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S30sh2"
  },
  "lane_policy": "share_plan_prev_bgonly"
 }
}