{
 "groupbg::dense_forest": {
  "input_fingerprint": "716903dbd5a8c9ea",
  "meta": {
   "model": "gpt-image-2",
   "size": "1536x864",
   "pack": "11.202607220237",
   "contract": "bgfirst_full_v3",
   "group_sig": {
    "key": "dense_forest",
    "tags": [
     "S1sh2",
     "S1sh4",
     "S3sh3"
    ]
   },
   "context_sig": "eca686e671fc8b08"
  },
  "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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — contemporary South Korea, 2026; all people are Korean unless stated: In dense undergrowth along a forest escape path at dusk.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n숲속 살인 현장: 우거진 수풀이 덮여 있는 숲 속 공간. 해질 무렵의 어두운 조명 아래 폴리스 라인이 쳐져 있다. (특징: 우거진 수풀; 해질 무렵의 어두운 조명; 노란색 폴리스 라인)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#1. 숲속 / 해 질 무렵\n- 우거진 수풀을 헤치며 도망치는.\n- 씬1, 2 현장을 감식하는 경찰, 폴리스 라인을 치고 시신을 수거한다.\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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — contemporary South Korea, 2026; all people are Korean unless stated: In dense undergrowth along a forest escape path at dusk.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n숲속 살인 현장: 우거진 수풀이 덮여 있는 숲 속 공간. 해질 무렵의 어두운 조명 아래 폴리스 라인이 쳐져 있다. (특징: 우거진 수풀; 해질 무렵의 어두운 조명; 노란색 폴리스 라인)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#1. 숲속 / 해 질 무렵\n- 우거진 수풀을 헤치며 도망치는.\n- 씬1, 2 현장을 감식하는 경찰, 폴리스 라인을 치고 시신을 수거한다.\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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/groupbg_dense_forest_447a99.png",
  "asset_id": "26d440e5-7e80-41b3-8163-3ff5ff381998",
  "input_asset_ids": [
   "6ddec542-7a13-46fe-9071-61b806506215"
  ],
  "origin_tag": "S1sh2",
  "place_text": "In dense undergrowth along a forest escape path at dusk.",
  "origin_inputs": {
   "place_text": "In dense undergrowth along a forest escape path at dusk.",
   "time_of_day_en": "dusk",
   "conti_asset_id": "6ddec542-7a13-46fe-9071-61b806506215"
  }
 },
 "S1sh2::bgfirst_bg": {
  "input_fingerprint": "f8805bc8cf8a3ac8",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 허벅지에 이빨 자국이 나고 피가 흐르는 채로 우거진 수풀을 헤치며 도망치는 숲속에서 살해된 남자의 잔뜩 겁에 질린 얼굴 클로즈업, 머리카락이 뒤로 흩날리는 순간.\n\nLOCATION (lock): In dense undergrowth along a forest escape path 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: Track backward at the fleeing man's head height from a slight front-side position, tightening until his terrified face occupies most of the center-left frame while vegetation slips past at the edges. Keep the camera near eye level and preserve the backward sweep of his hair along the outer frame, with only fragments of his pitched shoulders visible below.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 숲속에서 살해된 남자 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 우거진 수풀 (Dense and being forced aside by the fleeing man); used as Dense vegetation passes along both frame edges, compressing the escape route around the 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: Fading daylight appropriate to dusk is rendered in restrained, desaturated low-to-mid-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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 허벅지에 이빨 자국이 나고 피가 흐르는 채로 우거진 수풀을 헤치며 도망치는 숲속에서 살해된 남자의 잔뜩 겁에 질린 얼굴 클로즈업, 머리카락이 뒤로 흩날리는 순간.\n\nLOCATION (lock): In dense undergrowth along a forest escape path 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: Track backward at the fleeing man's head height from a slight front-side position, tightening until his terrified face occupies most of the center-left frame while vegetation slips past at the edges. Keep the camera near eye level and preserve the backward sweep of his hair along the outer frame, with only fragments of his pitched shoulders visible below.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 숲속에서 살해된 남자 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 우거진 수풀 (Dense and being forced aside by the fleeing man); used as Dense vegetation passes along both frame edges, compressing the escape route around the 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: Fading daylight appropriate to dusk is rendered in restrained, desaturated low-to-mid-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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S1sh2__bgfirst_bg.png",
  "asset_id": "e1d99de7-1253-4515-a33b-5f1a4e0d915a",
  "input_asset_ids": [
   "6ddec542-7a13-46fe-9071-61b806506215",
   "26d440e5-7e80-41b3-8163-3ff5ff381998"
  ]
 },
 "S1sh2": {
  "input_fingerprint": "79c47488e7d9697d",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 허벅지에 이빨 자국이 나고 피가 흐르는 채로 우거진 수풀을 헤치며 도망치는 숲속에서 살해된 남자의 잔뜩 겁에 질린 얼굴 클로즈업, 머리카락이 뒤로 흩날리는 순간.\n\nLOCATION (lock): In dense undergrowth along a forest escape path 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: Track backward at the fleeing man's head height from a slight front-side position, tightening until his terrified face occupies most of the center-left frame while vegetation slips past at the edges. Keep the camera near eye level and preserve the backward sweep of his hair along the outer frame, with only fragments of his pitched shoulders visible below.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 숲속에서 살해된 남자 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 우거진 수풀 (Dense and being forced aside by the fleeing man); used as Dense vegetation passes along both frame edges, compressing the escape route around the 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: Fading daylight appropriate to dusk is rendered in restrained, desaturated low-to-mid-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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The man is missing one shoe and has a large, actively bleeding bite wound on his thigh; both conditions must continue through the attack.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 숲속에서 살해된 남자 (한국인 남성, 성인, 짧은 검은 머리, 평범한 남성 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 허벅지에 이빨 자국이 나고 피가 흐르는 채로 우거진 수풀을 헤치며 도망치는 숲속에서 살해된 남자의 잔뜩 겁에 질린 얼굴 클로즈업, 머리카락이 뒤로 흩날리는 순간.\n\nLOCATION (lock): In dense undergrowth along a forest escape path 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: Track backward at the fleeing man's head height from a slight front-side position, tightening until his terrified face occupies most of the center-left frame while vegetation slips past at the edges. Keep the camera near eye level and preserve the backward sweep of his hair along the outer frame, with only fragments of his pitched shoulders visible below.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 숲속에서 살해된 남자 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 우거진 수풀 (Dense and being forced aside by the fleeing man); used as Dense vegetation passes along both frame edges, compressing the escape route around the 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: Fading daylight appropriate to dusk is rendered in restrained, desaturated low-to-mid-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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The man is missing one shoe and has a large, actively bleeding bite wound on his thigh; both conditions must continue through the attack.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 숲속에서 살해된 남자 (한국인 남성, 성인, 짧은 검은 머리, 평범한 남성 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 허벅지에 이빨 자국이 나고 피가 흐르는 채로 우거진 수풀을 헤치며 도망치는 숲속에서 살해된 남자의 잔뜩 겁에 질린 얼굴 클로즈업, 머리카락이 뒤로 흩날리는 순간.\n\nLOCATION (lock): In dense undergrowth along a forest escape path 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: Track backward at the fleeing man's head height from a slight front-side position, tightening until his terrified face occupies most of the center-left frame while vegetation slips past at the edges. Keep the camera near eye level and preserve the backward sweep of his hair along the outer frame, with only fragments of his pitched shoulders visible below.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 숲속에서 살해된 남자 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 우거진 수풀 (Dense and being forced aside by the fleeing man); used as Dense vegetation passes along both frame edges, compressing the escape route around the 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: Fading daylight appropriate to dusk is rendered in restrained, desaturated low-to-mid-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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The man is missing one shoe and has a large, actively bleeding bite wound on his thigh; both conditions must continue through the attack.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 숲속에서 살해된 남자 (한국인 남성, 성인, 짧은 검은 머리, 평범한 남성 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S1sh2__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S1sh2.png"
    },
    {
     "label": "CHARACTER REFERENCE — 숲속에서 살해된 남자: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:1309180>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/groupbg_dense_forest_447a99.png"
    },
    {
     "label": "CHARACTER REFERENCE — 숲속에서 살해된 남자: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:1309180>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "어깨 일부만 보여야 한다는 타이트한 클로즈업 지시는 어겼으나, 얼굴의 좌측 중앙 배치 및 수풀이 스쳐 지나가는 동적인 프레이밍 지시를 더 충실히 따랐습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "클로즈업 지시를 어기고 상반신과 손까지 노출되었으며, 얼굴을 좌측 중앙에 배치하라는 구도 지시를 따르지 않고 정중앙에 배치했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "어깨 일부만 보여야 한다는 타이트한 클로즈업 지시는 어겼으나, 얼굴의 좌측 중앙 배치 및 수풀이 스쳐 지나가는 동적인 프레이밍 지시를 더 충실히 따랐습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "클로즈업 지시를 어기고 상반신과 손까지 노출되었으며, 얼굴을 좌측 중앙에 배치하라는 구도 지시를 따르지 않고 정중앙에 배치했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "지정된 얼굴 클로즈업과 어깨선 프레이밍 제한을 어기고 허벅지 상처를 보여주기 위해 프레임을 넓힌 점이 치명적인 감점 요인입니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "어깨 파편만 보이는 타이트한 얼굴 클로즈업 프레이밍, 뒤로 흩날리는 머리카락, 겁에 질린 표정 등 프레이밍과 연출 지시를 정확히 구현했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "지정된 얼굴 클로즈업과 어깨선 프레이밍 제한을 어기고 허벅지 상처를 보여주기 위해 프레임을 넓힌 점이 치명적인 감점 요인입니다."
     },
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "어깨 파편만 보이는 타이트한 얼굴 클로즈업 프레이밍, 뒤로 흩날리는 머리카락, 겁에 질린 표정 등 프레이밍과 연출 지시를 정확히 구현했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 12,
     "B": 9
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": false,
    "policy": 1
   }
  },
  "totals": {
   "A": 12,
   "B": 9
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 5,
    "verdict_ko": "어깨 일부만 보여야 한다는 타이트한 클로즈업 지시는 어겼으나, 얼굴의 좌측 중앙 배치 및 수풀이 스쳐 지나가는 동적인 프레이밍 지시를 더 충실히 따랐습니다."
   },
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "클로즈업 지시를 어기고 상반신과 손까지 노출되었으며, 얼굴을 좌측 중앙에 배치하라는 구도 지시를 따르지 않고 정중앙에 배치했습니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/groupbg_dense_forest_447a99.png"
   },
   {
    "label": "CHARACTER REFERENCE — 숲속에서 살해된 남자: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:1309180>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "캐릭터 레퍼런스 사진에 명확히 보이는 얼굴의 큰 멍과 핏자국(특히 오른쪽 눈가와 뺨)이 누락되었습니다.",
     "fix_en": "Add the prominent bruises and blood stains to the man's face exactly as they appear in the character reference."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Tighten the framing around the man's face so it occupies most of the center-left frame, cropping out the extended hand, arm, and lower torso while retaining only small fragments of his pitched shoulders below.",
     "issue_ko": "프레이밍이 지시보다 넓어 얼굴 외에 상체와 뻗은 팔·손이 크게 보인다. 얼굴이 화면 대부분을 차지하고 아래에는 기울어진 어깨 일부만 보여야 한다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add the prominent bruises and blood stains to the man's face exactly as they appear in the character reference.\n- Tighten the framing around the man's face so it occupies most of the center-left frame, cropping out the extended hand, arm, and lower torso while retaining only small fragments of his pitched shoulders 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": 8,
      "verdict_ko": "제시된 '얼굴 클로즈업' 프레이밍을 정확히 구현하였으며, 흩날리는 머리카락과 레퍼런스의 얼굴 상처까지 충실하게 반영했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "요구된 클로즈업보다 프레임이 너무 넓어(미디엄 샷) 지시사항을 위반했으며, 캐릭터 레퍼런스의 얼굴 상처도 대부분 누락되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "제시된 '얼굴 클로즈업' 프레이밍을 정확히 구현하였으며, 흩날리는 머리카락과 레퍼런스의 얼굴 상처까지 충실하게 반영했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "요구된 클로즈업보다 프레임이 너무 넓어(미디엄 샷) 지시사항을 위반했으며, 캐릭터 레퍼런스의 얼굴 상처도 대부분 누락되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 클로즈업 구도와 스케치의 어깨선 컷을 정확히 준수하여 긴박한 표정을 효과적으로 연출했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "프레임이 넓어져 상체와 팔이 화면에 포함되면서 명시된 클로즈업 및 구도 제한 지시를 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 클로즈업 구도와 스케치의 어깨선 컷을 정확히 준수하여 긴박한 표정을 효과적으로 연출했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "프레임이 넓어져 상체와 팔이 화면에 포함되면서 명시된 클로즈업 및 구도 제한 지시를 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 8,
     "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S1sh2__bgfirst_bg.png",
   "bg_asset_id": "e1d99de7-1253-4515-a33b-5f1a4e0d915a",
   "bg_record_key": "S1sh2::bgfirst_bg",
   "chain_winner": true,
   "authority": "groupbg",
   "group_key": "dense_forest",
   "groupbg_asset_id": "26d440e5-7e80-41b3-8163-3ff5ff381998"
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S1sh4::variants": {
  "author_fp": "22c42563f8d76cf9",
  "author": {
   "variants": [
    {
     "approach_ko": "넓은 화각의 근접 시점과 선택적 초점으로 전경의 다리에서 눌린 얼굴까지 폭력적인 깊이감을 만든다.",
     "prompt_en": "Use a close, wide-angle lens feel at the prescribed near-ground endpoint, making the planted legs loom with tangible weight in the lower-left while the diagonal plane leads sharply to the compressed torso and readable face at middle-right. Keep the face and primary point of bodily pressure crisp, with foreground fabric and edge vegetation falling into restrained softness. Let fading dusk skim laterally across skin, creased clothing, and disturbed ground, with subdued color and dense natural shadow."
    },
    {
     "approach_ko": "압축된 망원적 시야와 깊은 초점으로 두 몸과 수풀을 하나의 답답한 압박면처럼 포착한다.",
     "prompt_en": "Use a restrained longer-lens feel while preserving the exact low medium composition, compressing the foreground legs, pinned body, forest floor, and enclosing undergrowth into a tight, oppressive stack of planes. Hold broader depth of field so the attacker's weight-bearing posture and the victim's compressed face register simultaneously, with vegetation clearly hemming the frame rather than dissolving away. Render the dusk as flat, cool, and diffuse beneath the foliage, emphasizing physical contact, fabric tension, and the unforgiving ground without theatrical highlights."
    }
   ]
  },
  "reused": false
 },
 "S1sh4": {
  "input_fingerprint": "7315fae765f382f6",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 정체불명의 남자가 바닥에 쓰러진 숲속에서 살해된 남자의 몸을 체중을 실어 강하게 짓누르고 있는 모습.\n\nLOCATION (lock): On the forest floor amid dense undergrowth at dusk, where the fleeing victim has been brought 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: Dolly in near ground level from beside the fallen bodies, looking diagonally across the unidentified man's lower body toward the victim pinned beneath him. Hold a medium composition in which the attacker's planted legs occupy the lower-left foreground and the victim's compressed torso and face remain readable in the middle-right, emphasizing their unequal vertical weight.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 정체불명의 남자 in the middle-left of the frame, foreground; 숲속에서 살해된 남자 in the middle-right of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 우거진 수풀 (Dense around the fallen pair); used as Vegetation hems in the bodies and partially breaks the frame edges; 숲 바닥 (Supporting both fallen bodies); used as The ground establishes the low impact plane against which the victim is pinned.\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 dusk light remains desaturated and low-to-mid-key, keeping the violence grounded rather than stylized.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The pinned man remains one-shoed, with the 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: 숲속에서 살해된 남자 (한국인 남성, 성인, 짧은 검은 머리, 평범한 남성 얼굴); 정체불명의 남자 (한국인 남성, 성인, 검은 머리, 평범한 남성 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 정체불명의 남자가 바닥에 쓰러진 숲속에서 살해된 남자의 몸을 체중을 실어 강하게 짓누르고 있는 모습.\n\nLOCATION (lock): On the forest floor amid dense undergrowth at dusk, where the fleeing victim has been brought 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: Dolly in near ground level from beside the fallen bodies, looking diagonally across the unidentified man's lower body toward the victim pinned beneath him. Hold a medium composition in which the attacker's planted legs occupy the lower-left foreground and the victim's compressed torso and face remain readable in the middle-right, emphasizing their unequal vertical weight.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 정체불명의 남자 in the middle-left of the frame, foreground; 숲속에서 살해된 남자 in the middle-right of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 우거진 수풀 (Dense around the fallen pair); used as Vegetation hems in the bodies and partially breaks the frame edges; 숲 바닥 (Supporting both fallen bodies); used as The ground establishes the low impact plane against which the victim is pinned.\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 dusk light remains desaturated and low-to-mid-key, keeping the violence grounded rather than stylized.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The pinned man remains one-shoed, with the 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: 숲속에서 살해된 남자 (한국인 남성, 성인, 짧은 검은 머리, 평범한 남성 얼굴); 정체불명의 남자 (한국인 남성, 성인, 검은 머리, 평범한 남성 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a close, wide-angle lens feel at the prescribed near-ground endpoint, making the planted legs loom with tangible weight in the lower-left while the diagonal plane leads sharply to the compressed torso and readable face at middle-right. Keep the face and primary point of bodily pressure crisp, with foreground fabric and edge vegetation falling into restrained softness. Let fading dusk skim laterally across skin, creased clothing, and disturbed ground, with subdued color and dense natural shadow.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 정체불명의 남자가 바닥에 쓰러진 숲속에서 살해된 남자의 몸을 체중을 실어 강하게 짓누르고 있는 모습.\n\nLOCATION (lock): On the forest floor amid dense undergrowth at dusk, where the fleeing victim has been brought 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: Dolly in near ground level from beside the fallen bodies, looking diagonally across the unidentified man's lower body toward the victim pinned beneath him. Hold a medium composition in which the attacker's planted legs occupy the lower-left foreground and the victim's compressed torso and face remain readable in the middle-right, emphasizing their unequal vertical weight.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 정체불명의 남자 in the middle-left of the frame, foreground; 숲속에서 살해된 남자 in the middle-right of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 우거진 수풀 (Dense around the fallen pair); used as Vegetation hems in the bodies and partially breaks the frame edges; 숲 바닥 (Supporting both fallen bodies); used as The ground establishes the low impact plane against which the victim is pinned.\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 dusk light remains desaturated and low-to-mid-key, keeping the violence grounded rather than stylized.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The pinned man remains one-shoed, with the 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: 숲속에서 살해된 남자 (한국인 남성, 성인, 짧은 검은 머리, 평범한 남성 얼굴); 정체불명의 남자 (한국인 남성, 성인, 검은 머리, 평범한 남성 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained longer-lens feel while preserving the exact low medium composition, compressing the foreground legs, pinned body, forest floor, and enclosing undergrowth into a tight, oppressive stack of planes. Hold broader depth of field so the attacker's weight-bearing posture and the victim's compressed face register simultaneously, with vegetation clearly hemming the frame rather than dissolving away. Render the dusk as flat, cool, and diffuse beneath the foliage, emphasizing physical contact, fabric tension, and the unforgiving ground without 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."
  },
  "totals": {
   "B": 7,
   "A": 5
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "지시된 로우 앵글 구도를 정확히 구현했으며, 희생자의 허벅지 교상과 체중을 실어 누르는 동작의 무게감을 사실적으로 잘 표현했습니다."
   },
   {
    "label": "A",
    "score": 5,
    "verdict_ko": "전반적인 구도와 의상은 일치하나, 희생자의 맨발 형태가 해부학적으로 어색하며 짓누르는 자세의 역동성이 상대적으로 떨어집니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S1sh2_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 숲속에서 살해된 남자: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:1309180>"
   },
   {
    "label": "CHARACTER REFERENCE — 정체불명의 남자: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:418704>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "피해자의 다리 상처가 허벅지가 아닌 무릎 아래 정강이 부위에 있습니다.",
     "fix_en": "Move the bloody wound on the victim's leg up from the shin to the thigh."
    },
    {
     "issue_ko": "피해자는 신발을 한 짝만 신고 있어야 하나, 화면에 보이는 양발 모두에 신발을 신고 있습니다.",
     "fix_en": "Remove the shoe from one of the victim's feet so he is visibly one-shoed."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Lower the camera to near forest-floor level beside the bodies and aim diagonally across the attacker's planted legs toward the pinned victim, preserving all staging.",
     "issue_ko": "카메라가 계약보다 높아 두 사람을 위에서 내려다보는 시점이며, 바닥 가까이에서 공격자의 하체 너머로 대각선 조망하는 축이 충분히 구현되지 않았다."
    },
    {
     "fix_en": "Dolly in to a tighter medium framing while keeping the attacker middle-left foreground and the victim's compressed torso and face middle-right midground.",
     "issue_ko": "프레이밍이 요구된 미디엄 숏보다 넓어 두 사람의 다리와 주변 바닥이 과도하게 많이 들어오며, 압축된 상·하체의 긴밀한 화면 구성이 약하다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Move the bloody wound on the victim's leg up from the shin to the thigh.\n- Remove the shoe from one of the victim's feet so he is visibly one-shoed.\n- Lower the camera to near forest-floor level beside the bodies and aim diagonally across the attacker's planted legs toward the pinned victim, preserving all staging.\n- Dolly in to a tighter medium framing while keeping the attacker middle-left foreground and the victim's compressed torso and face middle-right midground.\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": "명시된 카메라 지시대로 가해자의 하반신을 전경 좌측에 배치하고 상반신을 크롭하여, 피해자를 짓누르는 하체의 무게감과 구도를 완벽하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "인물과 배경의 세부 요소는 잘 반영되었으나, 가해자의 상반신이 프레임에 온전히 포함되어 '가해자의 하반신을 가로질러 보는 미디엄 샷'이라는 샷 텍스트의 구도 조건을 충족하지 못했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "명시된 카메라 지시대로 가해자의 하반신을 전경 좌측에 배치하고 상반신을 크롭하여, 피해자를 짓누르는 하체의 무게감과 구도를 완벽하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "인물과 배경의 세부 요소는 잘 반영되었으나, 가해자의 상반신이 프레임에 온전히 포함되어 '가해자의 하반신을 가로질러 보는 미디엄 샷'이라는 샷 텍스트의 구도 조건을 충족하지 못했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "피해자의 한쪽 신발이 벗겨진 상태(one-shoed)와 허벅지의 상처를 지시대로 정확히 구현했으며, 가해자가 체중을 실어 짓누르는 구도 역시 훌륭합니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "가해자의 하체와 피해자의 구도는 좋으나, 피해자가 양쪽 신발을 모두 신고 있어 '신발을 한 짝만 신은 상태'라는 유지 조건(carried state)을 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "피해자의 한쪽 신발이 벗겨진 상태(one-shoed)와 허벅지의 상처를 지시대로 정확히 구현했으며, 가해자가 체중을 실어 짓누르는 구도 역시 훌륭합니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "가해자의 하체와 피해자의 구도는 좋으나, 피해자가 양쪽 신발을 모두 신고 있어 '신발을 한 짝만 신은 상태'라는 유지 조건(carried state)을 위반했습니다."
     }
    ]
   },
   "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": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S1sh2"
  }
 },
 "S2sh1::bgfirst_bg": {
  "input_fingerprint": "606c94e9c3847a77",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 희미한 가로등 불빛 아래, 전봇대에 기댄 뒷골목에서 살해된 여자와 정체불명의 남자가 밀착해 서 있는 전신 구도.\n\nLOCATION (lock): Beside a utility pole in a city back alley at night, under a dim 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: From waist height across the alley, hold a static wide view of the pair pressed tightly against the utility pole, their full bodies clustered around the center of frame. Keep the man slightly left and the woman slightly right, with the pole directly behind their overlap and enough alley space around them to make the intimacy appear threatening rather than romantic.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 정체불명의 남자 in the middle-left of the frame, midground; 뒷골목에서 살해된 여자 in the middle-right of the frame, midground; utility pole in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 전봇대 (Supporting the woman's weight) — Its alley-facing side is visible behind the pair; used as The pole forms the rigid vertical anchor directly behind the intertwined pair; 가로등 (On with faint output) — The fixture is positioned above the pair and directed toward the alley below; used as The streetlight provides the explicitly faint illumination over the pair; 도시 뒷골목 (Night setting); used as Open alley depth surrounds and isolates the central figures.\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 streetlight creates restrained low-key illumination with a desaturated, oppressive night palette.\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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 희미한 가로등 불빛 아래, 전봇대에 기댄 뒷골목에서 살해된 여자와 정체불명의 남자가 밀착해 서 있는 전신 구도.\n\nLOCATION (lock): Beside a utility pole in a city back alley at night, under a dim 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: From waist height across the alley, hold a static wide view of the pair pressed tightly against the utility pole, their full bodies clustered around the center of frame. Keep the man slightly left and the woman slightly right, with the pole directly behind their overlap and enough alley space around them to make the intimacy appear threatening rather than romantic.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 정체불명의 남자 in the middle-left of the frame, midground; 뒷골목에서 살해된 여자 in the middle-right of the frame, midground; utility pole in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 전봇대 (Supporting the woman's weight) — Its alley-facing side is visible behind the pair; used as The pole forms the rigid vertical anchor directly behind the intertwined pair; 가로등 (On with faint output) — The fixture is positioned above the pair and directed toward the alley below; used as The streetlight provides the explicitly faint illumination over the pair; 도시 뒷골목 (Night setting); used as Open alley depth surrounds and isolates the central figures.\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 streetlight creates restrained low-key illumination with a desaturated, oppressive night palette.\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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S2sh1__bgfirst_bg.png",
  "asset_id": "c52b87c6-0b48-4ee5-8d29-ec4fb000f768",
  "input_asset_ids": [
   "a5fe8439-24ea-489a-93f6-7f141004e989",
   "b18407f5-3e3a-45eb-85e0-43ce8eccbd24"
  ]
 },
 "S2sh1": {
  "input_fingerprint": "96d2bace5b7df242",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 희미한 가로등 불빛 아래, 전봇대에 기댄 뒷골목에서 살해된 여자와 정체불명의 남자가 밀착해 서 있는 전신 구도.\n\nLOCATION (lock): Beside a utility pole in a city back alley at night, under a dim streetlamp. The shot takes place here — the attached LOCATION STRUCTURE PHOTOGRAPH is the single authority for this exact place — its fixed structure and permanent site details are LOCKED to it. No separate location photograph exists for this place. Build everything else strictly from the location text above and the shot text; the layout sketch (when attached) governs framing and placement only, and the shot text governs time of day, lighting and action.\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 waist height across the alley, hold a static wide view of the pair pressed tightly against the utility pole, their full bodies clustered around the center of frame. Keep the man slightly left and the woman slightly right, with the pole directly behind their overlap and enough alley space around them to make the intimacy appear threatening rather than romantic.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 정체불명의 남자 in the middle-left of the frame, midground; 뒷골목에서 살해된 여자 in the middle-right of the frame, midground; utility pole in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 전봇대 (Supporting the woman's weight) — Its alley-facing side is visible behind the pair; used as The pole forms the rigid vertical anchor directly behind the intertwined pair; 가로등 (On with faint output) — The fixture is positioned above the pair and directed toward the alley below; used as The streetlight provides the explicitly faint illumination over the pair; 도시 뒷골목 (Night setting); used as Open alley depth surrounds and isolates the central figures.\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 streetlight creates restrained low-key illumination with a desaturated, oppressive night palette.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 뒷골목에서 살해된 여자 (한국인 여성, 성인, 자연스러운 검은 머리, 평범한 여성 얼굴); 정체불명의 남자 (한국인 남성, 성인, 검은 머리, 평범한 남성 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 희미한 가로등 불빛 아래, 전봇대에 기댄 뒷골목에서 살해된 여자와 정체불명의 남자가 밀착해 서 있는 전신 구도.\n\nLOCATION (lock): Beside a utility pole in a city back alley at night, under a dim streetlamp. The shot takes place here — the attached LOCATION STRUCTURE PHOTOGRAPH is the single authority for this exact place — its fixed structure and permanent site details are LOCKED to it. No separate location photograph exists for this place. Build everything else strictly from the location text above and the shot text; the layout sketch (when attached) governs framing and placement only, and the shot text governs time of day, lighting and action.\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 waist height across the alley, hold a static wide view of the pair pressed tightly against the utility pole, their full bodies clustered around the center of frame. Keep the man slightly left and the woman slightly right, with the pole directly behind their overlap and enough alley space around them to make the intimacy appear threatening rather than romantic.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 정체불명의 남자 in the middle-left of the frame, midground; 뒷골목에서 살해된 여자 in the middle-right of the frame, midground; utility pole in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 전봇대 (Supporting the woman's weight) — Its alley-facing side is visible behind the pair; used as The pole forms the rigid vertical anchor directly behind the intertwined pair; 가로등 (On with faint output) — The fixture is positioned above the pair and directed toward the alley below; used as The streetlight provides the explicitly faint illumination over the pair; 도시 뒷골목 (Night setting); used as Open alley depth surrounds and isolates the central figures.\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 streetlight creates restrained low-key illumination with a desaturated, oppressive night palette.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 뒷골목에서 살해된 여자 (한국인 여성, 성인, 자연스러운 검은 머리, 평범한 여성 얼굴); 정체불명의 남자 (한국인 남성, 성인, 검은 머리, 평범한 남성 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 희미한 가로등 불빛 아래, 전봇대에 기댄 뒷골목에서 살해된 여자와 정체불명의 남자가 밀착해 서 있는 전신 구도.\n\nLOCATION (lock): Beside a utility pole in a city back alley at night, under a dim streetlamp. The shot takes place here — the attached LOCATION STRUCTURE PHOTOGRAPH is the single authority for this exact place — its fixed structure and permanent site details are LOCKED to it. No separate location photograph exists for this place. Build everything else strictly from the location text above and the shot text; the layout sketch (when attached) governs framing and placement only, and the shot text governs time of day, lighting and action.\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 waist height across the alley, hold a static wide view of the pair pressed tightly against the utility pole, their full bodies clustered around the center of frame. Keep the man slightly left and the woman slightly right, with the pole directly behind their overlap and enough alley space around them to make the intimacy appear threatening rather than romantic.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 정체불명의 남자 in the middle-left of the frame, midground; 뒷골목에서 살해된 여자 in the middle-right of the frame, midground; utility pole in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 전봇대 (Supporting the woman's weight) — Its alley-facing side is visible behind the pair; used as The pole forms the rigid vertical anchor directly behind the intertwined pair; 가로등 (On with faint output) — The fixture is positioned above the pair and directed toward the alley below; used as The streetlight provides the explicitly faint illumination over the pair; 도시 뒷골목 (Night setting); used as Open alley depth surrounds and isolates the central figures.\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 streetlight creates restrained low-key illumination with a desaturated, oppressive night palette.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 뒷골목에서 살해된 여자 (한국인 여성, 성인, 자연스러운 검은 머리, 평범한 여성 얼굴); 정체불명의 남자 (한국인 남성, 성인, 검은 머리, 평범한 남성 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S2sh1__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S2sh1.png"
    },
    {
     "label": "CHARACTER REFERENCE — 뒷골목에서 살해된 여자: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:774548>"
    },
    {
     "label": "CHARACTER REFERENCE — 정체불명의 남자: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:418704>"
    }
   ],
   "B": [
    {
     "label": "LOCATION STRUCTURE PHOTOGRAPH — the confirmed photograph of this exact place and its fixed structure: it is the SINGLE authority for the location, the structure's shape, proportions, materials, colors, openings and every permanent site detail. Never copy its camera framing, time of day or lighting — the shot text is the authority for those.",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/background_chain/seed_bg_city_back_alley_sel.png"
    },
    {
     "label": "CHARACTER REFERENCE — 뒷골목에서 살해된 여자: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:774548>"
    },
    {
     "label": "CHARACTER REFERENCE — 정체불명의 남자: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:418704>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "살해된 여성의 축 늘어진 자세와 무기력한 상태를 지문에 맞게 잘 표현했으며, 구도와 배경 또한 훌륭하게 구현되었습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "여성이 눈을 뜨고 살아있는 상태로 남자를 바라보고 있어 '살해된 여자'라는 핵심 설정 및 연기 지문을 완전히 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "살해된 여성의 축 늘어진 자세와 무기력한 상태를 지문에 맞게 잘 표현했으며, 구도와 배경 또한 훌륭하게 구현되었습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "여성이 눈을 뜨고 살아있는 상태로 남자를 바라보고 있어 '살해된 여자'라는 핵심 설정 및 연기 지문을 완전히 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "축 늘어진 자세와 표정으로 '살해된 여자'의 상태를 정확하게 연출했으며, 명시된 와이드 앵글과 위치 지정도 잘 구현되었습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "여자가 남자와 시선을 맞추며 살아있는 것처럼 연출되어 '살해된 여자'라는 핵심 설정에 위배됩니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "축 늘어진 자세와 표정으로 '살해된 여자'의 상태를 정확하게 연출했으며, 명시된 와이드 앵글과 위치 지정도 잘 구현되었습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "여자가 남자와 시선을 맞추며 살아있는 것처럼 연출되어 '살해된 여자'라는 핵심 설정에 위배됩니다."
     }
    ]
   },
   "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": "살해된 여성의 축 늘어진 자세와 무기력한 상태를 지문에 맞게 잘 표현했으며, 구도와 배경 또한 훌륭하게 구현되었습니다."
   },
   {
    "label": "B",
    "score": 3,
    "verdict_ko": "여성이 눈을 뜨고 살아있는 상태로 남자를 바라보고 있어 '살해된 여자'라는 핵심 설정 및 연기 지문을 완전히 위반했습니다."
   }
  ],
  "refs": [
   {
    "label": "LOCATION STRUCTURE PHOTOGRAPH — the confirmed photograph of this exact place and its fixed structure: it is the SINGLE authority for the location, the structure's shape, proportions, materials, colors, openings and every permanent site detail. Never copy its camera framing, time of day or lighting — the shot text is the authority for those.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/background_chain/seed_bg_city_back_alley_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 뒷골목에서 살해된 여자: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:774548>"
   },
   {
    "label": "CHARACTER REFERENCE — 정체불명의 남자: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:418704>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "프롬프트에 '살해된 여자'로 명시되어 있으나, 여자가 눈을 뜨고 남자를 쳐다보며 의식이 있는 것처럼 보입니다.",
     "fix_en": "Close the woman's eyes and relax her facial features to make her appear completely lifeless and deceased."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Restore the exact 16:9 framing and camera axis of the locked shot background without shifting, cropping, widening, or rescaling its architecture.",
     "issue_ko": "선택 이미지가 잠긴 배경의 16:9 프레이밍보다 세로가 잘린 넓은 화면으로 바뀌어, 고정 건축물과 가로등의 화면 위치 및 크기가 기준 배경과 다릅니다."
    },
    {
     "fix_en": "Lower the camera to waist height and aim it horizontally across the alley while preserving all subject relationships.",
     "issue_ko": "카메라가 허리 높이의 수평 시점이 아니라 골목을 내려다보는 높은 시점으로 보입니다."
    },
    {
     "fix_en": "Scale both figures down together by about 15 percent and place their full-body cluster in the midground center, keeping the man middle-left, the woman middle-right, and the pole directly behind their overlap.",
     "issue_ko": "두 인물이 레이아웃 스케치보다 화면에서 크게 보여 전경에 가까우며, 하단에 치우쳐 주변 골목 공간이 부족합니다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Close the woman's eyes and relax her facial features to make her appear completely lifeless and deceased.\n- Restore the exact 16:9 framing and camera axis of the locked shot background without shifting, cropping, widening, or rescaling its architecture.\n- Lower the camera to waist height and aim it horizontally across the alley while preserving all subject relationships.\n- Scale both figures down together by about 15 percent and place their full-body cluster in the midground center, keeping the man middle-left, the woman middle-right, and the pole directly behind their overlap.\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": "주어진 배경과 스케치 레이아웃을 정확하게 반영했으며, 인물들의 외모와 복장, 살해된 채 기대어 있는 포즈가 자연스럽고 프롬프트에 매우 충실함."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "프롬프트와 레퍼런스를 잘 따랐으나, 남자의 얼굴 측면과 머리카락 렌더링이 A에 비해 다소 부자연스러움."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "주어진 배경과 스케치 레이아웃을 정확하게 반영했으며, 인물들의 외모와 복장, 살해된 채 기대어 있는 포즈가 자연스럽고 프롬프트에 매우 충실함."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "프롬프트와 레퍼런스를 잘 따랐으나, 남자의 얼굴 측면과 머리카락 렌더링이 A에 비해 다소 부자연스러움."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 원본 배경을 훼손 없이 유지하며, 레퍼런스의 인물 외형과 스케치의 밀착된 포즈를 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "후보 A와 시각적으로 동일한 결과물로, 배경, 구도, 인물 지시사항 등 프롬프트의 모든 조건을 충실히 반영했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 원본 배경을 훼손 없이 유지하며, 레퍼런스의 인물 외형과 스케치의 밀착된 포즈를 정확히 구현했습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "후보 A와 시각적으로 동일한 결과물로, 배경, 구도, 인물 지시사항 등 프롬프트의 모든 조건을 충실히 반영했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 16,
     "B": 15
    },
    "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S2sh1__bgfirst_bg.png",
   "bg_asset_id": "c52b87c6-0b48-4ee5-8d29-ec4fb000f768",
   "bg_record_key": "S2sh1::bgfirst_bg",
   "chain_winner": true,
   "authority": "seed_bg"
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "lane_policy": "ab_select_ready"
 },
 "S2sh3::variants": {
  "author_fp": "d8eda7ffedd13de9",
  "author": {
   "variants": [
    {
     "approach_ko": "접촉면의 물성과 정지된 폭력을 예리하게 포착하고 배경을 얕게 녹이는 임상적 근접 촬영.",
     "prompt_en": "Use a clinically precise, shallow-focus close-up that makes the frozen contact plane the sharpest point in the image. Resolve the pressure of the jaw, embedded teeth, torn skin, and minute surface texture with restrained realism, while the cheek and shoulder contours soften gradually and the utility pole falls into a narrow, indistinct vertical blur. Let the faint streetlight rake laterally across the skin and fabric, creating controlled highlights within dense, desaturated shadow."
    },
    {
     "approach_ko": "남자의 뺨과 여자의 어깨를 어두운 겹으로 묶어 탈출구 없는 압박감을 강조하는 저조도 측면 촬영.",
     "prompt_en": "Shape the close-up as a compressed stack of dark profile planes, with the man’s cheek and the woman’s shoulder enclosing the central bite contact and making the frame feel airless. Hold slightly broader focus across both human contours while keeping the receding utility pole softly recognizable, preserving tactile depth rather than isolating a single detail. Use the weak streetlight as a narrow edge source, allowing most facial and clothing surfaces to sink into low-key, desaturated shadow while a restrained glint defines the locked teeth and damaged skin."
    }
   ]
  },
  "reused": false
 },
 "S2sh3": {
  "input_fingerprint": "8b3259c5754b8021",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 정체불명의 남자의 이빨이 뒷골목에서 살해된 여자의 목덜미 피부에 박힌 채 멈춰있는 측면.\n\nLOCATION (lock): At the utility pole in the dimly lit city back alley at night. 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 neck height and perpendicular to the woman's neck, arrest the lateral track at its closest point so the man's jaw and embedded teeth remain locked in profile. The contact occupies the central portion of the close frame without losing the contour of the woman's shoulder or the man's cheek, while the utility pole recedes as a narrow blurred reference behind them.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 전봇대 (Behind the woman's shoulder) — Only its side edge facing the lateral camera position remains visible; used as A narrow out-of-focus vertical section preserves the pair's position against the pole; 목덜미 피부 (Teeth embedded in it); used as The woman's neck is the contact plane held perpendicular to 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: The faint streetlight maintains low-key, desaturated contrast, with the explicitly red eye detail held as a selective deep-red accent if it enters the edge of frame.\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 dimly lit alley, streetlamp glow, utility pole, and nighttime surfaces. Preserve the woman's position against the pole and the man's close proximity and clothing. Exclude the earlier full-body framing and any street area outside this tight neck-bite composition.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The woman's neck remains caught in the man's teeth and bears the resulting torn bite injury as she loses her strength.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 뒷골목에서 살해된 여자 (한국인 여성, 성인, 자연스러운 검은 머리, 평범한 여성 얼굴); 정체불명의 남자 (한국인 남성, 성인, 검은 머리, 평범한 남성 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 정체불명의 남자의 이빨이 뒷골목에서 살해된 여자의 목덜미 피부에 박힌 채 멈춰있는 측면.\n\nLOCATION (lock): At the utility pole in the dimly lit city back alley at night. 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 neck height and perpendicular to the woman's neck, arrest the lateral track at its closest point so the man's jaw and embedded teeth remain locked in profile. The contact occupies the central portion of the close frame without losing the contour of the woman's shoulder or the man's cheek, while the utility pole recedes as a narrow blurred reference behind them.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 전봇대 (Behind the woman's shoulder) — Only its side edge facing the lateral camera position remains visible; used as A narrow out-of-focus vertical section preserves the pair's position against the pole; 목덜미 피부 (Teeth embedded in it); used as The woman's neck is the contact plane held perpendicular to 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: The faint streetlight maintains low-key, desaturated contrast, with the explicitly red eye detail held as a selective deep-red accent if it enters the edge of frame.\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 dimly lit alley, streetlamp glow, utility pole, and nighttime surfaces. Preserve the woman's position against the pole and the man's close proximity and clothing. Exclude the earlier full-body framing and any street area outside this tight neck-bite composition.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The woman's neck remains caught in the man's teeth and bears the resulting torn bite injury as she loses her strength.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 뒷골목에서 살해된 여자 (한국인 여성, 성인, 자연스러운 검은 머리, 평범한 여성 얼굴); 정체불명의 남자 (한국인 남성, 성인, 검은 머리, 평범한 남성 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a clinically precise, shallow-focus close-up that makes the frozen contact plane the sharpest point in the image. Resolve the pressure of the jaw, embedded teeth, torn skin, and minute surface texture with restrained realism, while the cheek and shoulder contours soften gradually and the utility pole falls into a narrow, indistinct vertical blur. Let the faint streetlight rake laterally across the skin and fabric, creating controlled highlights within dense, desaturated shadow.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 정체불명의 남자의 이빨이 뒷골목에서 살해된 여자의 목덜미 피부에 박힌 채 멈춰있는 측면.\n\nLOCATION (lock): At the utility pole in the dimly lit city back alley at night. 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 neck height and perpendicular to the woman's neck, arrest the lateral track at its closest point so the man's jaw and embedded teeth remain locked in profile. The contact occupies the central portion of the close frame without losing the contour of the woman's shoulder or the man's cheek, while the utility pole recedes as a narrow blurred reference behind them.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 전봇대 (Behind the woman's shoulder) — Only its side edge facing the lateral camera position remains visible; used as A narrow out-of-focus vertical section preserves the pair's position against the pole; 목덜미 피부 (Teeth embedded in it); used as The woman's neck is the contact plane held perpendicular to 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: The faint streetlight maintains low-key, desaturated contrast, with the explicitly red eye detail held as a selective deep-red accent if it enters the edge of frame.\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 dimly lit alley, streetlamp glow, utility pole, and nighttime surfaces. Preserve the woman's position against the pole and the man's close proximity and clothing. Exclude the earlier full-body framing and any street area outside this tight neck-bite composition.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The woman's neck remains caught in the man's teeth and bears the resulting torn bite injury as she loses her strength.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 뒷골목에서 살해된 여자 (한국인 여성, 성인, 자연스러운 검은 머리, 평범한 여성 얼굴); 정체불명의 남자 (한국인 남성, 성인, 검은 머리, 평범한 남성 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nShape the close-up as a compressed stack of dark profile planes, with the man’s cheek and the woman’s shoulder enclosing the central bite contact and making the frame feel airless. Hold slightly broader focus across both human contours while keeping the receding utility pole softly recognizable, preserving tactile depth rather than isolating a single detail. Use the weak streetlight as a narrow edge source, allowing most facial and clothing surfaces to sink into low-key, desaturated shadow while a restrained glint defines the locked teeth and damaged skin.\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": "남자가 여자의 목을 물고 있는 핵심 행동이 누락되었으며, 단지 서로 마주보고 있어 프롬프트의 지시를 위반했습니다."
   },
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "요구된 클로즈업 구도와 비니 등 의상을 잘 유지했으나, 이빨이 피부에 완전히 박혀있기보다 상처 옆에 떠 있는 점이 아쉽습니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S2sh1_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 뒷골목에서 살해된 여자: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:1116195>"
   },
   {
    "label": "CHARACTER REFERENCE — 정체불명의 남자: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:418704>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "남자의 이빨이 여자의 목덜미 피부에 박혀 있지 않고 상처와 떨어져 있습니다.",
     "fix_en": "Close the gap to physically embed the man's teeth directly into the flesh of the woman's neck."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Move the man’s head forward and slightly toward the woman’s neck so his teeth are visibly embedded in the neck wound at the center of the frame, while preserving the same profile camera angle and tight framing.",
     "issue_ko": "남자의 이빨이 여성의 목덜미에 박혀 있어야 하지만, 현재는 턱과 입이 상처에서 떨어져 있어 중앙의 물림 접촉이 성립하지 않는다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Close the gap to physically embed the man's teeth directly into the flesh of the woman's neck.\n- Move the man’s head forward and slightly toward the woman’s neck so his teeth are visibly embedded in the neck wound at the center of the frame, while preserving the same profile camera angle and tight framing.\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": "이빨이 피부에 완전히 박히지 않고 약간 떨어져 있어 아쉬우나, 요구된 이빨 노출과 목덜미의 물어뜯긴 상처를 잘 반영했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "남자가 입을 다물고 있어 이빨이 전혀 보이지 않으며, 핵심인 목덜미의 상처 묘사도 누락되어 프롬프트 지시와 거리가 멉니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "이빨이 피부에 완전히 박히지 않고 약간 떨어져 있어 아쉬우나, 요구된 이빨 노출과 목덜미의 물어뜯긴 상처를 잘 반영했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "남자가 입을 다물고 있어 이빨이 전혀 보이지 않으며, 핵심인 목덜미의 상처 묘사도 누락되어 프롬프트 지시와 거리가 멉니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "남자의 드러난 이빨과 붉은 눈동자, 여자의 목덜미 상처가 지시문과 상당히 유사하게 묘사되어 프롬프트 충실도가 높습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "남자의 입이 완전히 닫혀 있어 '이빨이 피부에 박힌 채 멈춰있는' 핵심 동작과 목의 상처가 전혀 구현되지 않았습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "남자의 드러난 이빨과 붉은 눈동자, 여자의 목덜미 상처가 지시문과 상당히 유사하게 묘사되어 프롬프트 충실도가 높습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "남자의 입이 완전히 닫혀 있어 '이빨이 피부에 박힌 채 멈춰있는' 핵심 동작과 목의 상처가 전혀 구현되지 않았습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 13,
     "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": "S2sh1"
  }
 },
 "S3sh3::bgfirst_bg": {
  "input_fingerprint": "c18eaf3061215607",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 무리 지어 선 한국인 남성 형사들 사이에서 한쪽 팔을 뻗은 채 서 있는 혜수의 모습.\n\nLOCATION (lock): Inside the police-controlled section of the forest crime scene, among investigators conducting the daytime examination.\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: Present the scene as regional-news footage from an eye-level position slightly off one side of the detective group, with subtle screen raster and broadcast compression shaping the image. 혜수 holds the visual center in a medium-wide arrangement, one arm extended across the clustered male detectives while foreground shoulders create a restrained observational frame.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 폴리스 라인 (Installed around the scene) — Its marked outward-facing side is visible across the scene; used as The line identifies and separates the active examination area behind the detectives; 시신 수거 현장 (Police are processing and recovering the bodies); used as Background police activity situates 혜수's command within the ongoing recovery operation; 지역 뉴스 화면 (Showing regional-news footage) — The camera-facing image side displays the field footage of the investigation; used as The visible broadcast image acts as the mediated surface through which the scene is perceived.\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 recorded through a regional-news image is kept desaturated with restrained contrast and mild device-mediated tonal compression.\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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 무리 지어 선 한국인 남성 형사들 사이에서 한쪽 팔을 뻗은 채 서 있는 혜수의 모습.\n\nLOCATION (lock): Inside the police-controlled section of the forest crime scene, among investigators conducting the daytime examination.\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: Present the scene as regional-news footage from an eye-level position slightly off one side of the detective group, with subtle screen raster and broadcast compression shaping the image. 혜수 holds the visual center in a medium-wide arrangement, one arm extended across the clustered male detectives while foreground shoulders create a restrained observational frame.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 폴리스 라인 (Installed around the scene) — Its marked outward-facing side is visible across the scene; used as The line identifies and separates the active examination area behind the detectives; 시신 수거 현장 (Police are processing and recovering the bodies); used as Background police activity situates 혜수's command within the ongoing recovery operation; 지역 뉴스 화면 (Showing regional-news footage) — The camera-facing image side displays the field footage of the investigation; used as The visible broadcast image acts as the mediated surface through which the scene is perceived.\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 recorded through a regional-news image is kept desaturated with restrained contrast and mild device-mediated tonal compression.\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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S3sh3__bgfirst_bg.png",
  "asset_id": "efd53933-330d-480e-801e-5ba25dba6b77",
  "input_asset_ids": [
   "496d7044-23da-49de-84f2-23629fd74510",
   "26d440e5-7e80-41b3-8163-3ff5ff381998"
  ]
 },
 "S3sh3": {
  "input_fingerprint": "55daa89b1a0d3ab5",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 무리 지어 선 한국인 남성 형사들 사이에서 한쪽 팔을 뻗은 채 서 있는 혜수의 모습.\n\nLOCATION (lock): Inside the police-controlled section of the forest crime scene, among investigators conducting the daytime examination. 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: Present the scene as regional-news footage from an eye-level position slightly off one side of the detective group, with subtle screen raster and broadcast compression shaping the image. 혜수 holds the visual center in a medium-wide arrangement, one arm extended across the clustered male detectives while foreground shoulders create a restrained observational frame.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 폴리스 라인 (Installed around the scene) — Its marked outward-facing side is visible across the scene; used as The line identifies and separates the active examination area behind the detectives; 시신 수거 현장 (Police are processing and recovering the bodies); used as Background police activity situates 혜수's command within the ongoing recovery operation; 지역 뉴스 화면 (Showing regional-news footage) — The camera-facing image side displays the field footage of the investigation; used as The visible broadcast image acts as the mediated surface through which the scene is perceived.\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 recorded through a regional-news image is kept desaturated with restrained contrast and mild device-mediated tonal compression.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 중반, 단정한 검은 머리, 성숙한 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 무리 지어 선 한국인 남성 형사들 사이에서 한쪽 팔을 뻗은 채 서 있는 혜수의 모습.\n\nLOCATION (lock): Inside the police-controlled section of the forest crime scene, among investigators conducting the daytime examination. 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: Present the scene as regional-news footage from an eye-level position slightly off one side of the detective group, with subtle screen raster and broadcast compression shaping the image. 혜수 holds the visual center in a medium-wide arrangement, one arm extended across the clustered male detectives while foreground shoulders create a restrained observational frame.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 폴리스 라인 (Installed around the scene) — Its marked outward-facing side is visible across the scene; used as The line identifies and separates the active examination area behind the detectives; 시신 수거 현장 (Police are processing and recovering the bodies); used as Background police activity situates 혜수's command within the ongoing recovery operation; 지역 뉴스 화면 (Showing regional-news footage) — The camera-facing image side displays the field footage of the investigation; used as The visible broadcast image acts as the mediated surface through which the scene is perceived.\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 recorded through a regional-news image is kept desaturated with restrained contrast and mild device-mediated tonal compression.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 중반, 단정한 검은 머리, 성숙한 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 무리 지어 선 한국인 남성 형사들 사이에서 한쪽 팔을 뻗은 채 서 있는 혜수의 모습.\n\nLOCATION (lock): Inside the police-controlled section of the forest crime scene, among investigators conducting the daytime examination. 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: Present the scene as regional-news footage from an eye-level position slightly off one side of the detective group, with subtle screen raster and broadcast compression shaping the image. 혜수 holds the visual center in a medium-wide arrangement, one arm extended across the clustered male detectives while foreground shoulders create a restrained observational frame.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 폴리스 라인 (Installed around the scene) — Its marked outward-facing side is visible across the scene; used as The line identifies and separates the active examination area behind the detectives; 시신 수거 현장 (Police are processing and recovering the bodies); used as Background police activity situates 혜수's command within the ongoing recovery operation; 지역 뉴스 화면 (Showing regional-news footage) — The camera-facing image side displays the field footage of the investigation; used as The visible broadcast image acts as the mediated surface through which the scene is perceived.\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 recorded through a regional-news image is kept desaturated with restrained contrast and mild device-mediated tonal compression.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 중반, 단정한 검은 머리, 성숙한 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S3sh3__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S3sh3.png"
    },
    {
     "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:735029>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/groupbg_dense_forest_447a99.png"
    },
    {
     "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:735029>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "지정된 샷의 구도와 인물 배치를 무난하게 구현했으나, 뉴스 화면 지시를 카메라맨의 직접 등장으로 다소 일차원적으로 해석했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "프롬프트의 배경 지시어인 '시신 수거 현장'이 우측 폴리스 라인 위에 텍스트로 그대로 유출된 치명적인 오류가 있습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "지정된 샷의 구도와 인물 배치를 무난하게 구현했으나, 뉴스 화면 지시를 카메라맨의 직접 등장으로 다소 일차원적으로 해석했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "프롬프트의 배경 지시어인 '시신 수거 현장'이 우측 폴리스 라인 위에 텍스트로 그대로 유출된 치명적인 오류가 있습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "전경의 어깨를 활용한 관찰자적 프레이밍과 혜수의 포즈 지시를 정확히 따랐으며, 지시된 공간적 요소들을 훌륭하게 배치했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "화면 자체를 뉴스 영상처럼 연출하라는 지시와 달리 카메라맨을 프레임 안에 배치하여 의도를 빗나갔습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "전경의 어깨를 활용한 관찰자적 프레이밍과 혜수의 포즈 지시를 정확히 따랐으며, 지시된 공간적 요소들을 훌륭하게 배치했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "화면 자체를 뉴스 영상처럼 연출하라는 지시와 달리 카메라맨을 프레임 안에 배치하여 의도를 빗나갔습니다."
     }
    ]
   },
   "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": 6,
    "verdict_ko": "지정된 샷의 구도와 인물 배치를 무난하게 구현했으나, 뉴스 화면 지시를 카메라맨의 직접 등장으로 다소 일차원적으로 해석했습니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "프롬프트의 배경 지시어인 '시신 수거 현장'이 우측 폴리스 라인 위에 텍스트로 그대로 유출된 치명적인 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/groupbg_dense_forest_447a99.png"
   },
   {
    "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:735029>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "프롬프트의 배경 요소 설명인 '폴리스 라인'과 '시신 수거 현[장]'이라는 단어가 노란색 통제선 테이프 위에 텍스트로 그대로 생성되어 '텍스트 금지' 규칙을 위반했습니다.",
     "fix_en": "Remove all text from the yellow police tape."
    },
    {
     "issue_ko": "지역 뉴스 화면을 통해 매개된 영상을 의도한 프롬프트의 지시와 달리, 화면 래스터(screen raster)나 방송 화질 압축 질감 등의 효과가 전혀 없는 일반적인 고화질 영화 스틸컷처럼 보입니다.",
     "fix_en": "Apply a subtle screen raster and broadcast compression artifacts over the entire image to simulate it being viewed on a regional news screen."
    }
   ]
  },
  "composition_critique": {
   "issues": []
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all text from the yellow police tape.\n- Apply a subtle screen raster and broadcast compression artifacts over the entire image to simulate it being viewed on a regional news 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": "화면에 미세한 래스터(raster) 효과를 적용하여 뉴스 송출 화면이라는 조건을 잘 구현했으며, 텍스트 유출 없이 지시된 구도와 인물의 동작을 안정적으로 표현했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "우측 폴리스 라인에 프롬프트의 배경 요소 지시문인 '시신 수거 현장'이 텍스트('시신 수거 현')로 그대로 유출되는 치명적인 오류가 발생했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "화면에 미세한 래스터(raster) 효과를 적용하여 뉴스 송출 화면이라는 조건을 잘 구현했으며, 텍스트 유출 없이 지시된 구도와 인물의 동작을 안정적으로 표현했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "우측 폴리스 라인에 프롬프트의 배경 요소 지시문인 '시신 수거 현장'이 텍스트('시신 수거 현')로 그대로 유출되는 치명적인 오류가 발생했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요구된 샷의 구도(전경의 어깨, 미디엄 와이드), 혜수의 동작 및 배경의 수사 활동이 프롬프트와 잘 일치합니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "프롬프트의 배경 요소 지시어인 '시신 수거 현장'이 우측 폴리스 라인 테이프 위에 텍스트로 누출되는 치명적인 위반(Hard Violation)이 발생했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 샷의 구도(전경의 어깨, 미디엄 와이드), 혜수의 동작 및 배경의 수사 활동이 프롬프트와 잘 일치합니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "프롬프트의 배경 요소 지시어인 '시신 수거 현장'이 우측 폴리스 라인 테이프 위에 텍스트로 누출되는 치명적인 위반(Hard Violation)이 발생했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 6,
     "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S3sh3__bgfirst_bg.png",
   "bg_asset_id": "efd53933-330d-480e-801e-5ba25dba6b77",
   "bg_record_key": "S3sh3::bgfirst_bg",
   "chain_winner": false,
   "authority": "groupbg",
   "group_key": "dense_forest",
   "groupbg_asset_id": "26d440e5-7e80-41b3-8163-3ff5ff381998"
  },
  "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S4sh1::variants": {
  "author_fp": "36e57363b2e1a7b1",
  "author": {
   "variants": [
    {
     "approach_ko": "수직선을 곧게 유지한 깊은 초점의 건축적 부감으로 옥탑방과 빌라의 공간 관계를 선명하게 드러낸다.",
     "prompt_en": "Use a rectilinear, deep-focus overhead view with disciplined architectural precision. Keep the roof edges parallel to the frame, the rooftop room clearly resolved at the center, and the surrounding upper building mass and outer district legible in layered detail. Render the daylight neutrally, emphasizing authentic weathering, roof textures, and grounded spatial scale."
    },
    {
     "approach_ko": "완만한 망원 압축과 중심 선택성을 활용해 옥탑방을 항공 접근의 도착점처럼 고립시킨다.",
     "prompt_en": "Photograph the direct overhead wide view with a gently compressed lens feel, making the rooftop room register as the visual endpoint while the surrounding villa roof forms a dense enclosing field. Hold the center with crisp material detail and allow a subtle, natural reduction of definition toward the outer district, without losing its location context. Keep contrast restrained and let tonal separation rather than exaggerated color guide the eye inward."
    }
   ]
  },
  "reused": false
 },
 "S4sh1": {
  "input_fingerprint": "d74aa64d32f0c975",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 낡은 다세대 빌라 건물의 옥탑방이 내려다보이는 주간의 부감 구도.\n\nLOCATION (lock): Above the rooftop of an aging low-rise residential building, overlooking the exterior of its rooftop room in daylight. 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: Complete the descending crane move in a direct overhead wide shot, settling with the rooftop room near the center and the upper portion of the old multi-unit villa surrounding it. Keep the building's roof geometry square to the frame so the rooftop room reads as the destination of the preceding aerial approach rather than as an abstract shape.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: rooftop room in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 옥탑방 (Part of the old villa) — Its top and perimeter face the directly overhead camera; used as The rooftop room is the central destination of the overhead approach; 낡은 다세대 빌라 (Old multi-unit residential building) — Its roof and upper outline face the camera from directly below; used as The upper building mass provides scale and spatial context around the rooftop room; 인천 변두리 (Daytime exterior); used as The surrounding district remains visible at the outer frame as location 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: Daylight appropriate to the stated daytime exterior is rendered with restrained saturation and grounded 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 낡은 다세대 빌라 건물의 옥탑방이 내려다보이는 주간의 부감 구도.\n\nLOCATION (lock): Above the rooftop of an aging low-rise residential building, overlooking the exterior of its rooftop room in daylight. 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: Complete the descending crane move in a direct overhead wide shot, settling with the rooftop room near the center and the upper portion of the old multi-unit villa surrounding it. Keep the building's roof geometry square to the frame so the rooftop room reads as the destination of the preceding aerial approach rather than as an abstract shape.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: rooftop room in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 옥탑방 (Part of the old villa) — Its top and perimeter face the directly overhead camera; used as The rooftop room is the central destination of the overhead approach; 낡은 다세대 빌라 (Old multi-unit residential building) — Its roof and upper outline face the camera from directly below; used as The upper building mass provides scale and spatial context around the rooftop room; 인천 변두리 (Daytime exterior); used as The surrounding district remains visible at the outer frame as location 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: Daylight appropriate to the stated daytime exterior is rendered with restrained saturation and grounded 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 rectilinear, deep-focus overhead view with disciplined architectural precision. Keep the roof edges parallel to the frame, the rooftop room clearly resolved at the center, and the surrounding upper building mass and outer district legible in layered detail. Render the daylight neutrally, emphasizing authentic weathering, roof textures, and grounded spatial scale.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 낡은 다세대 빌라 건물의 옥탑방이 내려다보이는 주간의 부감 구도.\n\nLOCATION (lock): Above the rooftop of an aging low-rise residential building, overlooking the exterior of its rooftop room in daylight. 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: Complete the descending crane move in a direct overhead wide shot, settling with the rooftop room near the center and the upper portion of the old multi-unit villa surrounding it. Keep the building's roof geometry square to the frame so the rooftop room reads as the destination of the preceding aerial approach rather than as an abstract shape.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: rooftop room in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 옥탑방 (Part of the old villa) — Its top and perimeter face the directly overhead camera; used as The rooftop room is the central destination of the overhead approach; 낡은 다세대 빌라 (Old multi-unit residential building) — Its roof and upper outline face the camera from directly below; used as The upper building mass provides scale and spatial context around the rooftop room; 인천 변두리 (Daytime exterior); used as The surrounding district remains visible at the outer frame as location 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: Daylight appropriate to the stated daytime exterior is rendered with restrained saturation and grounded 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 direct overhead wide view with a gently compressed lens feel, making the rooftop room register as the visual endpoint while the surrounding villa roof forms a dense enclosing field. Hold the center with crisp material detail and allow a subtle, natural reduction of definition toward the outer district, without losing its location context. Keep contrast restrained and let tonal separation rather than exaggerated color guide the eye inward.\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": "지시된 '건물 외관(Structure Look)' 레퍼런스의 구조를 완전히 무시하고 '위치(Location)' 사진의 구도와 구조를 그대로 복사하여 큰 감점 요소가 되었습니다."
   },
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "요구된 부감 구도를 잘 구현했으며, 물탱크나 계단 등 '건물 외관' 레퍼런스의 주요 형태적 특징을 프레임 내에 충실히 반영하여 우수합니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L03B02.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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/background_chain/seed_bg_rooftop_villa_sel.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "카메라 앵글이 수직으로 완전히 내려다보는 직각 부감(direct overhead) 구도가 아니며 건물의 측면 벽들이 다수 노출되어 있습니다.",
     "fix_en": "Change the camera angle to a direct 90-degree overhead shot pointing straight down at the roof, so only the top surfaces are visible."
    },
    {
     "issue_ko": "건물 구조 레퍼런스(STRUCTURE LOOK) 사진에는 옥탑방 문이 유리창 없는 형태이나, 생성된 이미지에는 문 상단에 창문이 있습니다.",
     "fix_en": "Replace the window on the green door of the rooftop room with a solid green panel to match the structure reference."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Move the camera to a true vertical bird’s-eye position directly above the villa, keeping the roof geometry square to the frame and the rooftop room centered while preserving the wide framing.",
     "issue_ko": "카메라가 정수직 부감이 아니라 옥탑방 정면과 건물 측면이 크게 보이는 사선 부감이며, 지붕 윤곽도 프레임에 직각으로 정렬되지 않았다."
    }
   ]
  },
  "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 a direct 90-degree overhead shot pointing straight down at the roof, so only the top surfaces are visible.\n- Replace the window on the green door of the rooftop room with a solid green panel to match the structure reference.\n- Move the camera to a true vertical bird’s-eye position directly above the villa, keeping the roof geometry square to the frame and the rooftop room centered while preserving the wide framing.\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 '수직 상단(direct overhead)' 카메라 구도를 정확히 구현하여 옥탑방 지붕을 정중앙에 성공적으로 배치했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "카메라 앵글이 수직 상단(direct overhead)이 아닌 사선 부감으로 연출되어 명시된 촬영 구도 지시를 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 '수직 상단(direct overhead)' 카메라 구도를 정확히 구현하여 옥탑방 지붕을 정중앙에 성공적으로 배치했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "카메라 앵글이 수직 상단(direct overhead)이 아닌 사선 부감으로 연출되어 명시된 촬영 구도 지시를 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요청된 수직 부감(Direct overhead) 카메라 축과 프레임에 수평/수직을 맞춘 지붕 기하학을 정확하게 구현하여 최우선 지침인 프레이밍 요건을 충실히 따랐습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "옥탑방의 구조적 특징은 비교적 잘 나타냈으나, 필수 지침인 '완전한 수직 부감(Direct overhead)' 대신 사선 앵글을 사용하여 프레이밍 우선순위를 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요청된 수직 부감(Direct overhead) 카메라 축과 프레임에 수평/수직을 맞춘 지붕 기하학을 정확하게 구현하여 최우선 지침인 프레이밍 요건을 충실히 따랐습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "옥탑방의 구조적 특징은 비교적 잘 나타냈으나, 필수 지침인 '완전한 수직 부감(Direct overhead)' 대신 사선 앵글을 사용하여 프레이밍 우선순위를 위반했습니다."
     }
    ]
   },
   "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": "플레이트+seed만 (배경 전용)",
  "share_plan": {
   "ref_plan": "background"
  },
  "lane_policy": "ab_select_bypass:bg_only",
  "plate_select": {
   "candidates": {
    "A": "L03B01",
    "B": "L03B02",
    "C": "L03B03"
   },
   "assigned": "L03B02",
   "choice": "Candidate B",
   "confident": true,
   "reason_ko": "지문에서 요구하는 '주간의 부감 구도'에 정확히 부합하며, 다른 후보들과 비교해 서브공간을 명확히 구분할 추가 단서가 없으므로 현재 배정된 후보를 유지합니다.",
   "kept": "L03B02"
  }
 },
 "S5sh7::bgfirst_bg": {
  "input_fingerprint": "0d07934ee9bd3730",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 컵 테두리에 입술을 댄 수리영의 얼굴이 미간을 찌푸린 채 멈춘 모습.\n\nLOCATION (lock): At the dining table in the rooftop apartment's small living area, reached by daylight from a small window.\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: Settle the table-side dolly-in at seated eye level, framing 수리영's face tightly in three-quarter view as her lips remain against the cup rim. Her furrowed brow occupies the upper center and the cup stays below it at a natural scale, with only a narrow suggestion of table and room left around the reaction.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 컵 (Held with its rim against her lips) — Its open top angles toward 수리영 while the side remains visible to camera; used as The rim crosses the lower portion of 수리영's face and anchors her reaction to the drink; 식탁 (수리영 is seated beside it) — Its upper face is visible along the lower frame edge; used as A small portion of the table remains beneath the cup as spatial context; 거실의 작은 창 (Admitting a narrow shaft of sunlight) — Sunlight passes through it into the room; the view beyond is not specified; used as The window supplies the explicitly described narrow sunlight from the 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: Narrow sunlight from the small living-room window shapes a restrained, desaturated close-up with low-to-mid-key interior 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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 컵 테두리에 입술을 댄 수리영의 얼굴이 미간을 찌푸린 채 멈춘 모습.\n\nLOCATION (lock): At the dining table in the rooftop apartment's small living area, reached by daylight from a small window.\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: Settle the table-side dolly-in at seated eye level, framing 수리영's face tightly in three-quarter view as her lips remain against the cup rim. Her furrowed brow occupies the upper center and the cup stays below it at a natural scale, with only a narrow suggestion of table and room left around the reaction.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 컵 (Held with its rim against her lips) — Its open top angles toward 수리영 while the side remains visible to camera; used as The rim crosses the lower portion of 수리영's face and anchors her reaction to the drink; 식탁 (수리영 is seated beside it) — Its upper face is visible along the lower frame edge; used as A small portion of the table remains beneath the cup as spatial context; 거실의 작은 창 (Admitting a narrow shaft of sunlight) — Sunlight passes through it into the room; the view beyond is not specified; used as The window supplies the explicitly described narrow sunlight from the 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: Narrow sunlight from the small living-room window shapes a restrained, desaturated close-up with low-to-mid-key interior 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S5sh7__bgfirst_bg.png",
  "asset_id": "23bac759-7ef6-4f36-a057-17fa9f46b361",
  "input_asset_ids": [
   "09d78c37-97b5-4a7e-b840-28b106ed23e6",
   "56a91bbe-83f6-4863-8952-ce28537e5796"
  ]
 },
 "S5sh7": {
  "input_fingerprint": "9ee03d749621b428",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 컵 테두리에 입술을 댄 수리영의 얼굴이 미간을 찌푸린 채 멈춘 모습.\n\nLOCATION (lock): At the dining table in the rooftop apartment's small living area, reached by daylight from a small window. 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: Settle the table-side dolly-in at seated eye level, framing 수리영's face tightly in three-quarter view as her lips remain against the cup rim. Her furrowed brow occupies the upper center and the cup stays below it at a natural scale, with only a narrow suggestion of table and room left around the reaction.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 컵 (Held with its rim against her lips) — Its open top angles toward 수리영 while the side remains visible to camera; used as The rim crosses the lower portion of 수리영's face and anchors her reaction to the drink; 식탁 (수리영 is seated beside it) — Its upper face is visible along the lower frame edge; used as A small portion of the table remains beneath the cup as spatial context; 거실의 작은 창 (Admitting a narrow shaft of sunlight) — Sunlight passes through it into the room; the view beyond is not specified; used as The window supplies the explicitly described narrow sunlight from the 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: Narrow sunlight from the small living-room window shapes a restrained, desaturated close-up with low-to-mid-key interior 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk already bears the small circular mark on her wrist, a long-standing feature present throughout this scene.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 컵 테두리에 입술을 댄 수리영의 얼굴이 미간을 찌푸린 채 멈춘 모습.\n\nLOCATION (lock): At the dining table in the rooftop apartment's small living area, reached by daylight from a small window. 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: Settle the table-side dolly-in at seated eye level, framing 수리영's face tightly in three-quarter view as her lips remain against the cup rim. Her furrowed brow occupies the upper center and the cup stays below it at a natural scale, with only a narrow suggestion of table and room left around the reaction.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 컵 (Held with its rim against her lips) — Its open top angles toward 수리영 while the side remains visible to camera; used as The rim crosses the lower portion of 수리영's face and anchors her reaction to the drink; 식탁 (수리영 is seated beside it) — Its upper face is visible along the lower frame edge; used as A small portion of the table remains beneath the cup as spatial context; 거실의 작은 창 (Admitting a narrow shaft of sunlight) — Sunlight passes through it into the room; the view beyond is not specified; used as The window supplies the explicitly described narrow sunlight from the 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: Narrow sunlight from the small living-room window shapes a restrained, desaturated close-up with low-to-mid-key interior 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk already bears the small circular mark on her wrist, a long-standing feature present throughout this scene.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 컵 테두리에 입술을 댄 수리영의 얼굴이 미간을 찌푸린 채 멈춘 모습.\n\nLOCATION (lock): At the dining table in the rooftop apartment's small living area, reached by daylight from a small window. 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: Settle the table-side dolly-in at seated eye level, framing 수리영's face tightly in three-quarter view as her lips remain against the cup rim. Her furrowed brow occupies the upper center and the cup stays below it at a natural scale, with only a narrow suggestion of table and room left around the reaction.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 컵 (Held with its rim against her lips) — Its open top angles toward 수리영 while the side remains visible to camera; used as The rim crosses the lower portion of 수리영's face and anchors her reaction to the drink; 식탁 (수리영 is seated beside it) — Its upper face is visible along the lower frame edge; used as A small portion of the table remains beneath the cup as spatial context; 거실의 작은 창 (Admitting a narrow shaft of sunlight) — Sunlight passes through it into the room; the view beyond is not specified; used as The window supplies the explicitly described narrow sunlight from the 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: Narrow sunlight from the small living-room window shapes a restrained, desaturated close-up with low-to-mid-key interior 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk already bears the small circular mark on her wrist, a long-standing feature present throughout this scene.\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 초반, 긴 검은 머리, 타원형의 젊은 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S5sh7__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S5sh7.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L04B04.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 클로즈업 샷 크기와 미간을 찌푸린 표정을 타이트하게 잘 구현했으며, 손목의 둥근 표식과 좁은 햇빛의 묘사도 훌륭합니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "식탁은 잘 보이나 프레이밍이 요구된 클로즈업보다 넓으며, 명시된 손목의 동그라미 표식이 누락되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 클로즈업 샷 크기와 미간을 찌푸린 표정을 타이트하게 잘 구현했으며, 손목의 둥근 표식과 좁은 햇빛의 묘사도 훌륭합니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "식탁은 잘 보이나 프레이밍이 요구된 클로즈업보다 넓으며, 명시된 손목의 동그라미 표식이 누락되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "화면 하단에 지정된 식탁이 포함되었으며, 레퍼런스의 의상(후드티)과 미간을 찌푸린 표정을 안정적으로 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "클로즈업 샷에 더 가깝지만 프레임 하단에 명시된 식탁이 완전히 누락되었고, 기준 이미지의 겉옷을 입고 있지 않습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "화면 하단에 지정된 식탁이 포함되었으며, 레퍼런스의 의상(후드티)과 미간을 찌푸린 표정을 안정적으로 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "클로즈업 샷에 더 가깝지만 프레임 하단에 명시된 식탁이 완전히 누락되었고, 기준 이미지의 겉옷을 입고 있지 않습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 11,
     "B": 11
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": false,
    "policy": 1
   }
  },
  "totals": {
   "A": 11,
   "B": 11
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "지시된 클로즈업 샷 크기와 미간을 찌푸린 표정을 타이트하게 잘 구현했으며, 손목의 둥근 표식과 좁은 햇빛의 묘사도 훌륭합니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "식탁은 잘 보이나 프레이밍이 요구된 클로즈업보다 넓으며, 명시된 손목의 동그라미 표식이 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L04B04.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "입술이 컵 테두리에 닿아 있지 않고 컵이 입 아래로 내려가 있습니다.",
     "fix_en": "Raise the cup so its rim presses directly against her lips."
    },
    {
     "issue_ko": "화면 하단 가장자리에 보여야 할 식탁 윗면이 프레임에서 잘려 보이지 않습니다.",
     "fix_en": "Reveal a small strip of the table's upper surface along the bottom edge of the frame."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Reframe slightly downward or pull back just enough to reveal a narrow strip of the dining tabletop along the lower frame edge beneath the cup, preserving the seated eye-level close-up and all other staging.",
     "issue_ko": "하단 프레임 가장자리에 보여야 할 식탁 상판이 전혀 보이지 않아 컵 아래의 공간적 맥락이 빠져 있다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Raise the cup so its rim presses directly against her lips.\n- Reveal a small strip of the table's upper surface along the bottom edge of the frame.\n- Reframe slightly downward or pull back just enough to reveal a narrow strip of the dining tabletop along the lower frame edge beneath the cup, preserving the seated eye-level close-up and all other staging.\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": "배경 보존, 스케치의 레이아웃, 인물의 미간을 찌푸린 표정 및 손목의 동그라미 표식까지 모든 지침을 해부학적 오류 없이 충실하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "컵을 쥔 손의 겉면에 5개의 손가락(엄지 제외)이 묘사되는 치명적인 해부학적 오류(하드 위반)가 발생했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "배경 보존, 스케치의 레이아웃, 인물의 미간을 찌푸린 표정 및 손목의 동그라미 표식까지 모든 지침을 해부학적 오류 없이 충실하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "컵을 쥔 손의 겉면에 5개의 손가락(엄지 제외)이 묘사되는 치명적인 해부학적 오류(하드 위반)가 발생했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지시된 프레이밍, 캐릭터의 표정과 자세, 배경 요소 및 손목의 원형 마크까지 스케치와 텍스트의 요구사항을 충실히 반영했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "컵을 쥔 손의 손가락이 5개로 묘사되어 해부학적으로 불가능한 신체 구조를 보여주는 치명적인 오류가 있습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지시된 프레이밍, 캐릭터의 표정과 자세, 배경 요소 및 손목의 원형 마크까지 스케치와 텍스트의 요구사항을 충실히 반영했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "컵을 쥔 손의 손가락이 5개로 묘사되어 해부학적으로 불가능한 신체 구조를 보여주는 치명적인 오류가 있습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 11,
     "B": 10
    },
    "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S5sh7__bgfirst_bg.png",
   "bg_asset_id": "23bac759-7ef6-4f36-a057-17fa9f46b361",
   "bg_record_key": "S5sh7::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",
    "E": "L04B05",
    "F": "L04B06",
    "G": "L04B07"
   },
   "assigned": "L04B04",
   "choice": "Candidate D",
   "confident": true,
   "reason_ko": "\"컵 테두리에 입술을 댄\"이라는 행동으로 보아 컵을 사용하는 식탁 및 주방 공간이 가장 적합하므로 기존 할당을 유지함.",
   "kept": "L04B04"
  }
 },
 "S5sh9::variants": {
  "author_fp": "9069607937eec024",
  "author": {
   "variants": [
    {
     "approach_ko": "얕은 심도로 감긴 눈과 어깨 접촉의 촉각적 친밀감을 강조한 관찰적 클로즈업.",
     "prompt_en": "Use an intimate, gently compressed lens feel with very shallow depth of field, placing precise focus on Suri-young’s closed lashes and the point where her nose meets Minsuk’s shoulder. Let Minsuk’s turned profile fall into delicate softness while remaining clearly readable, and reduce the entrance behind them to muted tonal layers. Shape the soft window light across skin, hair, and creased fabric with restrained contrast, making the contact feel tactile, private, and suspended."
    },
    {
     "approach_ko": "두 옆얼굴과 현관의 경계선을 또렷한 평면 구성으로 묶어 이별의 긴장을 강조한 클로즈업.",
     "prompt_en": "Photograph the opposing profiles with a more neutral lens feel and moderately deep focus, holding both faces and the shoulder contact with comparable clarity. Organize the entrance’s existing lines and subdued surfaces as a firm planar boundary behind them, creating quiet visual pressure around the farewell. Keep the daylight broad and even with restrained color, subtle shadow separation, and an unsentimental observational stillness."
    }
   ]
  },
  "reused": false
 },
 "S5sh9": {
  "input_fingerprint": "86e997fdd5c9bf84",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 수리영이 민숙(강민숙)의 어깨 쪽에 코를 댄 채 눈을 감은 측면.\n\nLOCATION (lock): In the interior entryway of the rooftop apartment, beside the spot where shoes are put on. 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 beside the entrance, hold a tightened lateral two-shot that keeps both women in opposing profiles. 수리영's closed eyes and nose-to-shoulder contact sit just left of center while 강민숙's shoulder and softly turned face occupy the right, leaving their lower bodies outside the close frame.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, foreground; 강민숙 in the middle-right of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 현관 (Location of their farewell interaction); used as The entrance area provides a subdued spatial boundary behind the shoulder-level contact.\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: Interior ambient light and the sunlight already established through the small window are kept soft, restrained, and desaturated around the intimate contact.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The small circular mark remains on Minsuk's wrist while Suri-young leans in toward her 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: 수리영 (한국인 여성, 20대 초반, 긴 검은 머리, 타원형의 젊은 얼굴); 강민숙 (한국인 여성, 40대 초반, 자연스러운 검은 머리, 부드러운 중년 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 수리영이 민숙(강민숙)의 어깨 쪽에 코를 댄 채 눈을 감은 측면.\n\nLOCATION (lock): In the interior entryway of the rooftop apartment, beside the spot where shoes are put on. 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 beside the entrance, hold a tightened lateral two-shot that keeps both women in opposing profiles. 수리영's closed eyes and nose-to-shoulder contact sit just left of center while 강민숙's shoulder and softly turned face occupy the right, leaving their lower bodies outside the close frame.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, foreground; 강민숙 in the middle-right of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 현관 (Location of their farewell interaction); used as The entrance area provides a subdued spatial boundary behind the shoulder-level contact.\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: Interior ambient light and the sunlight already established through the small window are kept soft, restrained, and desaturated around the intimate contact.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The small circular mark remains on Minsuk's wrist while Suri-young leans in toward her 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: 수리영 (한국인 여성, 20대 초반, 긴 검은 머리, 타원형의 젊은 얼굴); 강민숙 (한국인 여성, 40대 초반, 자연스러운 검은 머리, 부드러운 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an intimate, gently compressed lens feel with very shallow depth of field, placing precise focus on Suri-young’s closed lashes and the point where her nose meets Minsuk’s shoulder. Let Minsuk’s turned profile fall into delicate softness while remaining clearly readable, and reduce the entrance behind them to muted tonal layers. Shape the soft window light across skin, hair, and creased fabric with restrained contrast, making the contact feel tactile, private, and suspended.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 수리영이 민숙(강민숙)의 어깨 쪽에 코를 댄 채 눈을 감은 측면.\n\nLOCATION (lock): In the interior entryway of the rooftop apartment, beside the spot where shoes are put on. 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 beside the entrance, hold a tightened lateral two-shot that keeps both women in opposing profiles. 수리영's closed eyes and nose-to-shoulder contact sit just left of center while 강민숙's shoulder and softly turned face occupy the right, leaving their lower bodies outside the close frame.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, foreground; 강민숙 in the middle-right of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 현관 (Location of their farewell interaction); used as The entrance area provides a subdued spatial boundary behind the shoulder-level contact.\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: Interior ambient light and the sunlight already established through the small window are kept soft, restrained, and desaturated around the intimate contact.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The small circular mark remains on Minsuk's wrist while Suri-young leans in toward her 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: 수리영 (한국인 여성, 20대 초반, 긴 검은 머리, 타원형의 젊은 얼굴); 강민숙 (한국인 여성, 40대 초반, 자연스러운 검은 머리, 부드러운 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the opposing profiles with a more neutral lens feel and moderately deep focus, holding both faces and the shoulder contact with comparable clarity. Organize the entrance’s existing lines and subdued surfaces as a firm planar boundary behind them, creating quiet visual pressure around the farewell. Keep the daylight broad and even with restrained color, subtle shadow separation, and an unsentimental observational 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."
  },
  "totals": {
   "B": 7,
   "A": 4
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "지정된 샷 크기와 두 인물의 구도(코를 어깨에 댄 측면)를 충실히 구현했으며, 지침대로 강민숙의 손목에 원형 자국을 정확히 배치했습니다."
   },
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "두 인물의 구도는 양호하나, 강민숙의 손목에 있어야 할 원형 자국을 수리영의 손목에 배치하여 명시된 유지 상태(Carried State) 지침을 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S5sh7_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   },
   {
    "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:397486>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "배경이 이전 장면 스틸에 고정된 꽃무늬 벽지와 주방 카운터가 아닌, 신발장이 있는 평범한 복도로 변경되었습니다.",
     "fix_en": "Replace the plain hallway and shoe rack in the background with the floral wallpaper and kitchen counter established in the previous shot still."
    },
    {
     "issue_ko": "수리영이 이전 장면 스틸에서 입고 있던 연회색 티셔츠 대신 캐릭터 레퍼런스의 짙은 회색 후드티를 입고 있습니다.",
     "fix_en": "Change Suri-young's dark grey hoodie to the light grey t-shirt she wears in the previous shot still."
    },
    {
     "issue_ko": "강민숙이 캐릭터 레퍼런스에 있는 초록색 모자와 베이지색 티셔츠를 착용하지 않았습니다.",
     "fix_en": "Dress Minsuk in the green cap and beige t-shirt exactly as shown in her character reference."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Reframe slightly to the right so the nose-to-shoulder contact falls just left of frame center while keeping Suri-young middle-left and Minsuk middle-right at the same close-up scale.",
     "issue_ko": "코와 어깨의 접촉점이 프레임 중앙 바로 왼쪽이 아니라 오른쪽 3분의 1 부근에 치우쳐 있다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Replace the plain hallway and shoe rack in the background with the floral wallpaper and kitchen counter established in the previous shot still.\n- Change Suri-young's dark grey hoodie to the light grey t-shirt she wears in the previous shot still.\n- Dress Minsuk in the green cap and beige t-shirt exactly as shown in her character reference.\n- Reframe slightly to the right so the nose-to-shoulder contact falls just left of frame center while keeping Suri-young middle-left and Minsuk middle-right at the same close-up scale.\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": "이전 샷의 배경(꽃무늬 벽지, 주방)과 수리영의 의상 연속성을 완벽히 유지했으며, 강민숙의 레퍼런스 의상(베이지색 티셔츠, 모자)도 정확하게 반영했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "이전 샷과 일치해야 하는 배경과 수리영의 의상 연속성을 위반했으며, 강민숙의 캐릭터 레퍼런스 의상(모자 등) 또한 누락되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "이전 샷의 배경(꽃무늬 벽지, 주방)과 수리영의 의상 연속성을 완벽히 유지했으며, 강민숙의 레퍼런스 의상(베이지색 티셔츠, 모자)도 정확하게 반영했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "이전 샷과 일치해야 하는 배경과 수리영의 의상 연속성을 위반했으며, 강민숙의 캐릭터 레퍼런스 의상(모자 등) 또한 누락되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "이전 샷에서 확립된 배경(꽃무늬 벽지, 주방)과 수리영의 의상을 정확히 유지했으며, 요구된 프레이밍과 밀착된 포즈를 충실히 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "이전 샷의 배경을 무시하고 임의의 신발장 배경을 생성했으며, 수리영의 확립된 의상(회색 티셔츠)도 반영하지 않아 일관성이 부족합니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "이전 샷에서 확립된 배경(꽃무늬 벽지, 주방)과 수리영의 의상을 정확히 유지했으며, 요구된 프레이밍과 밀착된 포즈를 충실히 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "이전 샷의 배경을 무시하고 임의의 신발장 배경을 생성했으며, 수리영의 확립된 의상(회색 티셔츠)도 반영하지 않아 일관성이 부족합니다."
     }
    ]
   },
   "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": "S5sh7"
  }
 },
 "S5sh10::variants": {
  "author_fp": "bc9e49cab329875b",
  "author": {
   "variants": [
    {
     "approach_ko": "평면적이고 증거 사진처럼 절제된 망원 매크로 감각으로 표식의 선명함을 냉정하게 강조한다.",
     "prompt_en": "Use a compressed, observational macro feel that renders the wrist plane nearly flat and undistorted. Hold precise focus across the mark and adjacent skin texture, letting the entrance dissolve into a smooth, unobtrusive blur. Keep exposure neutral and restrained, with clean local contrast making the mark feel discovered rather than dramatized."
    },
    {
     "approach_ko": "피부의 곡면과 빛의 미세한 흐름을 살린 친밀하고 불길한 촉각적 클로즈업으로 연출한다.",
     "prompt_en": "Use an intimate close-focus perspective with gentle dimensionality along the wrist’s natural curvature. Let the established ambient light skim subtly across the skin, revealing pores, fine tonal shifts, and the mark’s physical integration with the surface; allow focus to fall away delicately beyond the locked center. Preserve soft entrance shapes as faint spatial context, creating a quiet, bodily sense of unease."
    }
   ]
  },
  "reused": false
 },
 "S5sh10": {
  "input_fingerprint": "bd8c0242a79ff0aa",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 작은 원형 표식이 선명하게 드러난 민숙(강민숙)의 손목 클로즈업.\n\nLOCATION (lock): In the rooftop apartment's interior entryway, near the door where the daughter is preparing to leave. 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: Finish the downward tilt very close beside 강민숙 at wrist height, perpendicular to the exposed side of her wrist, then lock the frame. Center the small circular mark within the wrist detail while retaining a slim margin of forearm and the softly indistinct entrance area so the mark remains naturally scaled.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 강민숙의 손목 (Small circular mark clearly visible) — The marked side is perpendicular to and fully visible from the camera position; used as The visible wrist surface provides the physical plane for the revealed mark; 작은 원형 표식 (Unusually distinct today); used as The mark is held at the locked focal center of the insert; 현관 (Out of focus behind the wrist); used as The entrance remains only as soft peripheral context after the tilt.\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: Established interior ambient light reveals the circular mark clearly while maintaining restrained saturation and grounded 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk's long-standing circular wrist mark is unusually distinct today.\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 초반, 자연스러운 검은 머리, 부드러운 중년 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 작은 원형 표식이 선명하게 드러난 민숙(강민숙)의 손목 클로즈업.\n\nLOCATION (lock): In the rooftop apartment's interior entryway, near the door where the daughter is preparing to leave. 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: Finish the downward tilt very close beside 강민숙 at wrist height, perpendicular to the exposed side of her wrist, then lock the frame. Center the small circular mark within the wrist detail while retaining a slim margin of forearm and the softly indistinct entrance area so the mark remains naturally scaled.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 강민숙의 손목 (Small circular mark clearly visible) — The marked side is perpendicular to and fully visible from the camera position; used as The visible wrist surface provides the physical plane for the revealed mark; 작은 원형 표식 (Unusually distinct today); used as The mark is held at the locked focal center of the insert; 현관 (Out of focus behind the wrist); used as The entrance remains only as soft peripheral context after the tilt.\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: Established interior ambient light reveals the circular mark clearly while maintaining restrained saturation and grounded 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk's long-standing circular wrist mark is unusually distinct today.\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 초반, 자연스러운 검은 머리, 부드러운 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a compressed, observational macro feel that renders the wrist plane nearly flat and undistorted. Hold precise focus across the mark and adjacent skin texture, letting the entrance dissolve into a smooth, unobtrusive blur. Keep exposure neutral and restrained, with clean local contrast making the mark feel discovered rather than dramatized.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 작은 원형 표식이 선명하게 드러난 민숙(강민숙)의 손목 클로즈업.\n\nLOCATION (lock): In the rooftop apartment's interior entryway, near the door where the daughter is preparing to leave. 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: Finish the downward tilt very close beside 강민숙 at wrist height, perpendicular to the exposed side of her wrist, then lock the frame. Center the small circular mark within the wrist detail while retaining a slim margin of forearm and the softly indistinct entrance area so the mark remains naturally scaled.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 강민숙의 손목 (Small circular mark clearly visible) — The marked side is perpendicular to and fully visible from the camera position; used as The visible wrist surface provides the physical plane for the revealed mark; 작은 원형 표식 (Unusually distinct today); used as The mark is held at the locked focal center of the insert; 현관 (Out of focus behind the wrist); used as The entrance remains only as soft peripheral context after the tilt.\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: Established interior ambient light reveals the circular mark clearly while maintaining restrained saturation and grounded 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk's long-standing circular wrist mark is unusually distinct today.\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 초반, 자연스러운 검은 머리, 부드러운 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an intimate close-focus perspective with gentle dimensionality along the wrist’s natural curvature. Let the established ambient light skim subtly across the skin, revealing pores, fine tonal shifts, and the mark’s physical integration with the surface; allow focus to fall away delicately beyond the locked center. Preserve soft entrance shapes as faint spatial context, creating a quiet, bodily sense of 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."
  },
  "totals": {
   "B": 7,
   "A": 6
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "지정된 인서트 클로즈업 앵글과 피사체의 수직 구도를 정확하게 구현했으며, 배경의 아웃포커싱도 적절함."
   },
   {
    "label": "A",
    "score": 6,
    "verdict_ko": "손목 표식과 배경 등 주요 요소는 잘 표현되었으나, 원형 표식이 프레임 중앙에 덜 맞춰져 있고 구도가 약간 틀어짐."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S5sh9_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:397486>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "이전 샷에서 손을 가슴 부근에 얹고 있던 자세와 달리, 손목이 위를 향해 수직으로 들려 있습니다.",
     "fix_en": "Reposition the wrist to match the exact bent-arm pose established in the previous shot."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Reframe slightly to the left so the circular wrist mark sits exactly at the center of the frame, preserving the same close-up scale and background context.",
     "issue_ko": "원형 표식이 프레임의 고정 중심보다 약간 왼쪽에 치우쳐 있다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Reposition the wrist to match the exact bent-arm pose established in the previous shot.\n- Reframe slightly to the left so the circular wrist mark sits exactly at the center of the frame, preserving the same close-up scale and background context.\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 손목 클로즈업 앵글과 원형 표식을 정확히 구현했으며, 배경과 의상도 이전 샷과 완벽히 일치합니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "후보 A와 동일한 이미지로, 프레이밍, 원형 표식, 배경 등 모든 요구 사항을 충실히 반영했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 손목 클로즈업 앵글과 원형 표식을 정확히 구현했으며, 배경과 의상도 이전 샷과 완벽히 일치합니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "후보 A와 동일한 이미지로, 프레이밍, 원형 표식, 배경 등 모든 요구 사항을 충실히 반영했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 샷 크기와 피사체 위치, 피부의 질감 및 원형 표식을 매우 충실하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "A와 동일한 수준으로 프레이밍과 디테일 등 프롬프트의 지시사항을 잘 따랐습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지정된 샷 크기와 피사체 위치, 피부의 질감 및 원형 표식을 매우 충실하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "A와 동일한 수준으로 프레이밍과 디테일 등 프롬프트의 지시사항을 잘 따랐습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 15,
     "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": "S5sh9"
  }
 },
 "groupbg::coastal_road": {
  "input_fingerprint": "bf73645c3ed57022",
  "meta": {
   "model": "gpt-image-2",
   "size": "1536x864",
   "pack": "11.202607220237",
   "contract": "bgfirst_full_v3",
   "group_sig": {
    "key": "coastal_road",
    "tags": [
     "S24sh1",
     "S6sh1",
     "S9sh1"
    ]
   },
   "context_sig": "a53a3df113c95bda"
  },
  "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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — contemporary South Korea, 2026; all people are Korean unless stated: On the seaward lane of a coastal road in daylight.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n해안가 도로: 바다와 인접해 있는 아스팔트 도로. 자전거가 달릴 수 있는 탁 트인 길이다. (특징: 아스팔트 포장도로; 바다를 향한 전망; 해안선)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#6. 해안가 도로 / D\n- 해안도로를 자전거를 타고 바삐 집으로 달려가는 수리영.\n- 자전거를 타고 빗속을 달리는 수리영.\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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — contemporary South Korea, 2026; all people are Korean unless stated: On the seaward lane of a coastal road in daylight.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n해안가 도로: 바다와 인접해 있는 아스팔트 도로. 자전거가 달릴 수 있는 탁 트인 길이다. (특징: 아스팔트 포장도로; 바다를 향한 전망; 해안선)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#6. 해안가 도로 / D\n- 해안도로를 자전거를 타고 바삐 집으로 달려가는 수리영.\n- 자전거를 타고 빗속을 달리는 수리영.\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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/groupbg_coastal_road_16062d.png",
  "asset_id": "20148f1f-f264-4442-8f6d-57b8ec8f2c2e",
  "input_asset_ids": [
   "b915b446-b697-4a6c-a6c9-5eb337cb7047"
  ],
  "origin_tag": "S6sh1",
  "place_text": "On the seaward lane of a coastal road in daylight.",
  "origin_inputs": {
   "place_text": "On the seaward lane of a coastal road in daylight.",
   "time_of_day_en": "day",
   "conti_asset_id": "b915b446-b697-4a6c-a6c9-5eb337cb7047"
  }
 },
 "S6sh1::bgfirst_bg": {
  "input_fingerprint": "ece004005dcaa5d7",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 해안가 도로에서 한쪽 페달을 밟아 내린 자세로 상체를 기울인 수리영의 측면.\n\nLOCATION (lock): On the seaward lane of a coastal road 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: Track parallel at torso height from a direct side position, keeping 수리영 and the bicycle in a wide profile as she rides toward screen right. Place her leaned upper body in the middle-left and the lowered pedal and bent leg in the lower center, preserving enough road ahead on the right for her continued movement.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, moves toward open road at screen right.\n- KEY BACKGROUND ELEMENTS: 해안가 도로 (수리영 is riding along it); used as The road forms the horizontal travel axis and leaves visible space ahead of the bicycle; 바다 (Beside the road); used as The sea establishes the coastal setting behind the parallel side view; 자전거 (Moving forward under pedal power) — Its full side profile faces the tracking camera; used as The bicycle carries the downward pedal action central to the shot.\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 coastal road is rendered in a restrained, desaturated palette while preserving 수리영's bright, energetic presence.\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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 해안가 도로에서 한쪽 페달을 밟아 내린 자세로 상체를 기울인 수리영의 측면.\n\nLOCATION (lock): On the seaward lane of a coastal road 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: Track parallel at torso height from a direct side position, keeping 수리영 and the bicycle in a wide profile as she rides toward screen right. Place her leaned upper body in the middle-left and the lowered pedal and bent leg in the lower center, preserving enough road ahead on the right for her continued movement.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, moves toward open road at screen right.\n- KEY BACKGROUND ELEMENTS: 해안가 도로 (수리영 is riding along it); used as The road forms the horizontal travel axis and leaves visible space ahead of the bicycle; 바다 (Beside the road); used as The sea establishes the coastal setting behind the parallel side view; 자전거 (Moving forward under pedal power) — Its full side profile faces the tracking camera; used as The bicycle carries the downward pedal action central to the shot.\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 coastal road is rendered in a restrained, desaturated palette while preserving 수리영's bright, energetic presence.\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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S6sh1__bgfirst_bg.png",
  "asset_id": "f77c449e-22dd-46c9-ba52-b5d9566b2c99",
  "input_asset_ids": [
   "b915b446-b697-4a6c-a6c9-5eb337cb7047",
   "20148f1f-f264-4442-8f6d-57b8ec8f2c2e"
  ]
 },
 "S6sh1": {
  "input_fingerprint": "83e54d25c7cfe1c9",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 해안가 도로에서 한쪽 페달을 밟아 내린 자세로 상체를 기울인 수리영의 측면.\n\nLOCATION (lock): On the seaward lane of a coastal road 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: Track parallel at torso height from a direct side position, keeping 수리영 and the bicycle in a wide profile as she rides toward screen right. Place her leaned upper body in the middle-left and the lowered pedal and bent leg in the lower center, preserving enough road ahead on the right for her continued movement.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, moves toward open road at screen right.\n- KEY BACKGROUND ELEMENTS: 해안가 도로 (수리영 is riding along it); used as The road forms the horizontal travel axis and leaves visible space ahead of the bicycle; 바다 (Beside the road); used as The sea establishes the coastal setting behind the parallel side view; 자전거 (Moving forward under pedal power) — Its full side profile faces the tracking camera; used as The bicycle carries the downward pedal action central to the shot.\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 coastal road is rendered in a restrained, desaturated palette while preserving 수리영's bright, energetic presence.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young is riding her bicycle with her earphones in.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 해안가 도로에서 한쪽 페달을 밟아 내린 자세로 상체를 기울인 수리영의 측면.\n\nLOCATION (lock): On the seaward lane of a coastal road 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: Track parallel at torso height from a direct side position, keeping 수리영 and the bicycle in a wide profile as she rides toward screen right. Place her leaned upper body in the middle-left and the lowered pedal and bent leg in the lower center, preserving enough road ahead on the right for her continued movement.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, moves toward open road at screen right.\n- KEY BACKGROUND ELEMENTS: 해안가 도로 (수리영 is riding along it); used as The road forms the horizontal travel axis and leaves visible space ahead of the bicycle; 바다 (Beside the road); used as The sea establishes the coastal setting behind the parallel side view; 자전거 (Moving forward under pedal power) — Its full side profile faces the tracking camera; used as The bicycle carries the downward pedal action central to the shot.\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 coastal road is rendered in a restrained, desaturated palette while preserving 수리영's bright, energetic presence.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young is riding her bicycle with her earphones in.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 해안가 도로에서 한쪽 페달을 밟아 내린 자세로 상체를 기울인 수리영의 측면.\n\nLOCATION (lock): On the seaward lane of a coastal road 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: Track parallel at torso height from a direct side position, keeping 수리영 and the bicycle in a wide profile as she rides toward screen right. Place her leaned upper body in the middle-left and the lowered pedal and bent leg in the lower center, preserving enough road ahead on the right for her continued movement.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, moves toward open road at screen right.\n- KEY BACKGROUND ELEMENTS: 해안가 도로 (수리영 is riding along it); used as The road forms the horizontal travel axis and leaves visible space ahead of the bicycle; 바다 (Beside the road); used as The sea establishes the coastal setting behind the parallel side view; 자전거 (Moving forward under pedal power) — Its full side profile faces the tracking camera; used as The bicycle carries the downward pedal action central to the shot.\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 coastal road is rendered in a restrained, desaturated palette while preserving 수리영's bright, energetic presence.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young is riding her bicycle with her earphones in.\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 초반, 긴 검은 머리, 타원형의 젊은 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S6sh1__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S6sh1.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/groupbg_coastal_road_16062d.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "측면 트래킹 샷의 구도와 한쪽 페달을 밟은 자세를 잘 구현했으며, 레퍼런스와 일치하는 의상(청바지) 및 이어폰 착용 상태까지 정확히 묘사했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "지정된 구도와 동작은 무난하게 표현되었으나, 인물의 하의가 검은색 바지로 묘사되어 레퍼런스(청바지)와 불일치합니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "측면 트래킹 샷의 구도와 한쪽 페달을 밟은 자세를 잘 구현했으며, 레퍼런스와 일치하는 의상(청바지) 및 이어폰 착용 상태까지 정확히 묘사했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "지정된 구도와 동작은 무난하게 표현되었으나, 인물의 하의가 검은색 바지로 묘사되어 레퍼런스(청바지)와 불일치합니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "지시된 평행 트래킹 카메라의 완벽한 측면 구도를 구현했으며, 인물 레퍼런스의 복장과 이어폰 착용 상태까지 정확하게 반영했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "로케이션 레퍼런스의 원근감을 그대로 복사하여 '평행 측면 구도' 지시를 위반했으며, 인물의 복장이 레퍼런스와 일치하지 않습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지시된 평행 트래킹 카메라의 완벽한 측면 구도를 구현했으며, 인물 레퍼런스의 복장과 이어폰 착용 상태까지 정확하게 반영했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "로케이션 레퍼런스의 원근감을 그대로 복사하여 '평행 측면 구도' 지시를 위반했으며, 인물의 복장이 레퍼런스와 일치하지 않습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 10,
     "B": 18
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 10,
   "B": 18
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 9,
    "verdict_ko": "측면 트래킹 샷의 구도와 한쪽 페달을 밟은 자세를 잘 구현했으며, 레퍼런스와 일치하는 의상(청바지) 및 이어폰 착용 상태까지 정확히 묘사했습니다."
   },
   {
    "label": "A",
    "score": 6,
    "verdict_ko": "지정된 구도와 동작은 무난하게 표현되었으나, 인물의 하의가 검은색 바지로 묘사되어 레퍼런스(청바지)와 불일치합니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/groupbg_coastal_road_16062d.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "자전거 프레임의 안장 밑 기둥(시트 튜브)이 그려지지 않아 안장이 공중에 떠 있고, 상단 가로 막대(탑 튜브)가 인물의 다리 안으로 부자연스럽게 합쳐집니다.",
     "fix_en": "Render a proper bicycle seat tube extending vertically from under the saddle down to the bottom bracket, connecting the top tube to it so the frame passes believably behind the rider's leg."
    }
   ]
  },
  "composition_critique": {
   "issues": []
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Render a proper bicycle seat tube extending vertically from under the saddle down to the bottom bracket, connecting the top tube to it so the frame passes believably behind the rider's leg.\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": 3,
      "verdict_ko": "지시된 구도와 장소, 한쪽 페달을 내린 자세는 잘 구현되었으나, 자전거의 양쪽 크랭크(페달 축)가 180도를 이루지 않는 물리적으로 불가능한 구조로 묘사되어 크게 감점되었습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "A 후보와 시각적으로 동일한 이미지이며, 자전거의 페달 구조가 물리적으로 불가능하게 생성된 동일한 치명적 오류가 존재합니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "지시된 구도와 장소, 한쪽 페달을 내린 자세는 잘 구현되었으나, 자전거의 양쪽 크랭크(페달 축)가 180도를 이루지 않는 물리적으로 불가능한 구조로 묘사되어 크게 감점되었습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "A 후보와 시각적으로 동일한 이미지이며, 자전거의 페달 구조가 물리적으로 불가능하게 생성된 동일한 치명적 오류가 존재합니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "지정된 해안가 배경과 인물의 측면 구도, 앞으로 기울인 자세, 이어폰 착용 등 프롬프트의 요구사항을 충실히 반영함."
     },
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지정된 해안가 배경과 인물의 측면 구도, 앞으로 기울인 자세, 이어폰 착용 등 프롬프트의 요구사항을 충실히 반영함 (A와 동일한 이미지)."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지정된 해안가 배경과 인물의 측면 구도, 앞으로 기울인 자세, 이어폰 착용 등 프롬프트의 요구사항을 충실히 반영함."
     },
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "지정된 해안가 배경과 인물의 측면 구도, 앞으로 기울인 자세, 이어폰 착용 등 프롬프트의 요구사항을 충실히 반영함 (A와 동일한 이미지)."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 12,
     "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S6sh1__bgfirst_bg.png",
   "bg_asset_id": "f77c449e-22dd-46c9-ba52-b5d9566b2c99",
   "bg_record_key": "S6sh1::bgfirst_bg",
   "chain_winner": false,
   "authority": "groupbg",
   "group_key": "coastal_road",
   "groupbg_asset_id": "20148f1f-f264-4442-8f6d-57b8ec8f2c2e"
  },
  "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S7sh2::bgfirst_bg": {
  "input_fingerprint": "5dcfce5cc9d9c49c",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 로트와일러의 눈을 정면으로 강렬하게 응시하는 수리영의 굳은 표정.\n\nLOCATION (lock): Inside the shelter's dog-handling area, directly in front of the confined Rottweiler.\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: Place the camera statically at 수리영's eye height and close to her face, directly on the confrontation axis without becoming the dog's literal POV. Her rigid face occupies the center and most of the frame, while the 로트와일러 remains outside the frame just beside the lens so that her unwavering stare reads as a direct challenge.\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: Daytime ambient light is rendered with restrained saturation, low-to-mid-key contrast, and natural facial texture.\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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 로트와일러의 눈을 정면으로 강렬하게 응시하는 수리영의 굳은 표정.\n\nLOCATION (lock): Inside the shelter's dog-handling area, directly in front of the confined Rottweiler.\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: Place the camera statically at 수리영's eye height and close to her face, directly on the confrontation axis without becoming the dog's literal POV. Her rigid face occupies the center and most of the frame, while the 로트와일러 remains outside the frame just beside the lens so that her unwavering stare reads as a direct challenge.\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: Daytime ambient light is rendered with restrained saturation, low-to-mid-key contrast, and natural facial texture.\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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S7sh2__bgfirst_bg.png",
  "asset_id": "b05a0043-c43a-4abb-8715-701424be2f15",
  "input_asset_ids": [
   "185d4430-a971-4945-b5ef-12a6496077af",
   "252612ca-55b2-46b2-9a9f-0429d2c7d508"
  ]
 },
 "S7sh2": {
  "input_fingerprint": "7fb38d63646a1a56",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 로트와일러의 눈을 정면으로 강렬하게 응시하는 수리영의 굳은 표정.\n\nLOCATION (lock): Inside the shelter's dog-handling area, directly in front of the confined Rottweiler. 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: Place the camera statically at 수리영's eye height and close to her face, directly on the confrontation axis without becoming the dog's literal POV. Her rigid face occupies the center and most of the frame, while the 로트와일러 remains outside the frame just beside the lens so that her unwavering stare reads as a direct challenge.\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: Daytime ambient light is rendered with restrained saturation, low-to-mid-key contrast, and natural facial texture.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 로트와일러의 눈을 정면으로 강렬하게 응시하는 수리영의 굳은 표정.\n\nLOCATION (lock): Inside the shelter's dog-handling area, directly in front of the confined Rottweiler. 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: Place the camera statically at 수리영's eye height and close to her face, directly on the confrontation axis without becoming the dog's literal POV. Her rigid face occupies the center and most of the frame, while the 로트와일러 remains outside the frame just beside the lens so that her unwavering stare reads as a direct challenge.\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: Daytime ambient light is rendered with restrained saturation, low-to-mid-key contrast, and natural facial texture.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 로트와일러의 눈을 정면으로 강렬하게 응시하는 수리영의 굳은 표정.\n\nLOCATION (lock): Inside the shelter's dog-handling area, directly in front of the confined Rottweiler. 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: Place the camera statically at 수리영's eye height and close to her face, directly on the confrontation axis without becoming the dog's literal POV. Her rigid face occupies the center and most of the frame, while the 로트와일러 remains outside the frame just beside the lens so that her unwavering stare reads as a direct challenge.\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: Daytime ambient light is rendered with restrained saturation, low-to-mid-key contrast, and natural facial texture.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 초반, 긴 검은 머리, 타원형의 젊은 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S7sh2__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S7sh2.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L06B01.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "클로즈업 샷 크기와 인물의 화면 비중 지시를 B보다 잘 구현했으나, 개가 프레임 밖 렌즈 옆에 있어야 한다는 지시를 완벽히 따르지 못했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "로트와일러가 갇혀 있는 설정은 맞으나, 프레이밍이 미디엄 샷으로 넓어지고 개가 화면 안에 크게 등장하여 우선순위가 높은 카메라 지시를 어겼습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "클로즈업 샷 크기와 인물의 화면 비중 지시를 B보다 잘 구현했으나, 개가 프레임 밖 렌즈 옆에 있어야 한다는 지시를 완벽히 따르지 못했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "로트와일러가 갇혀 있는 설정은 맞으나, 프레이밍이 미디엄 샷으로 넓어지고 개가 화면 안에 크게 등장하여 우선순위가 높은 카메라 지시를 어겼습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "얼굴을 화면 중심에 채우는 클로즈업 프레이밍과 렌즈 옆으로 개를 배치하는 구도를 더 잘 구현했으나, 개가 프레임 밖으로 완전히 배제되지 않은 점이 아쉽습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "클로즈업 지시를 어기고 미디엄 샷으로 렌더링되었으며, 렌즈 바로 옆이 아닌 철창 안쪽에 개가 배치되어 텍스트가 요구한 강렬한 시선 교환의 구도가 약화되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "얼굴을 화면 중심에 채우는 클로즈업 프레이밍과 렌즈 옆으로 개를 배치하는 구도를 더 잘 구현했으나, 개가 프레임 밖으로 완전히 배제되지 않은 점이 아쉽습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "클로즈업 지시를 어기고 미디엄 샷으로 렌더링되었으며, 렌즈 바로 옆이 아닌 철창 안쪽에 개가 배치되어 텍스트가 요구한 강렬한 시선 교환의 구도가 약화되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 12,
     "B": 8
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 12,
   "B": 8
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 6,
    "verdict_ko": "클로즈업 샷 크기와 인물의 화면 비중 지시를 B보다 잘 구현했으나, 개가 프레임 밖 렌즈 옆에 있어야 한다는 지시를 완벽히 따르지 못했습니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "로트와일러가 갇혀 있는 설정은 맞으나, 프레이밍이 미디엄 샷으로 넓어지고 개가 화면 안에 크게 등장하여 우선순위가 높은 카메라 지시를 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L06B01.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "첨부된 배경 레퍼런스의 배경을 그대로 유지하지 않고 천장 조명, 철창의 크기, 테이블 위의 휴지걸이 등 구조와 사물을 임의로 변경했습니다.",
     "fix_en": "Restore the exact background from the SHOT BACKGROUND reference image, keeping its original lighting, perspective, and fixtures untouched."
    },
    {
     "issue_ko": "지시문에 로트와일러가 프레임 바깥에 있어야 한다고 명시되어 있으나, 화면 왼쪽에 강아지의 머리가 나타나 있습니다.",
     "fix_en": "Remove the dog from the frame entirely, as it must remain outside the frame."
    },
    {
     "issue_ko": "레이아웃 스케치의 시선 화살표가 왼쪽을 향하고 있는데, 인물이 이를 따르지 않고 카메라 렌즈를 정면으로 응시하고 있습니다.",
     "fix_en": "Redirect the woman's eyes to look to the left, following the direction of the arrow in the layout sketch."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Move the Rottweiler completely outside the frame to a position immediately beside the lens, preserving all other framing.",
     "issue_ko": "로트와일러가 화면 왼쪽 전경을 크게 차지하지만, 계약상 개는 렌즈 바로 옆의 프레임 밖에 있어야 한다."
    },
    {
     "fix_en": "Redirect Suri-young’s gaze slightly toward the Rottweiler’s off-frame position beside the lens without changing her head position or expression.",
     "issue_ko": "수리영의 시선이 정면 렌즈를 향해 있어, 렌즈 옆 프레임 밖의 로트와일러를 응시해야 하는 대치 축과 맞지 않는다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Restore the exact background from the SHOT BACKGROUND reference image, keeping its original lighting, perspective, and fixtures untouched.\n- Remove the dog from the frame entirely, as it must remain outside the frame.\n- Redirect the woman's eyes to look to the left, following the direction of the arrow in the layout sketch.\n- Move the Rottweiler completely outside the frame to a position immediately beside the lens, preserving all other framing.\n- Redirect Suri-young’s gaze slightly toward the Rottweiler’s off-frame position beside the lens without changing her head position or expression.\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": "스케치의 시선 방향(왼쪽)을 정확히 따랐으며, 지문대로 개를 프레임 외부에 배치하여 지시를 충실히 이행했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "프레임 외부에 있어야 할 개가 등장했고, 스케치의 시선 화살표를 무시한 채 렌즈를 응시하여 연출 의도와 어긋납니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "스케치의 시선 방향(왼쪽)을 정확히 따랐으며, 지문대로 개를 프레임 외부에 배치하여 지시를 충실히 이행했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "프레임 외부에 있어야 할 개가 등장했고, 스케치의 시선 화살표를 무시한 채 렌즈를 응시하여 연출 의도와 어긋납니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "로트와일러를 화면 밖으로 배제하라는 카메라 프레이밍 지시를 정확히 준수했으며, 배경과 인물 조건도 충실하게 반영했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "로트와일러가 화면 밖 렌즈 옆에 있어야 한다는 명시적인 프레이밍 지시를 어기고 개의 모습을 화면 안에 포함시켰습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "로트와일러를 화면 밖으로 배제하라는 카메라 프레이밍 지시를 정확히 준수했으며, 배경과 인물 조건도 충실하게 반영했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "로트와일러가 화면 밖 렌즈 옆에 있어야 한다는 명시적인 프레이밍 지시를 어기고 개의 모습을 화면 안에 포함시켰습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 8,
     "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S7sh2__bgfirst_bg.png",
   "bg_asset_id": "b05a0043-c43a-4abb-8715-701424be2f15",
   "bg_record_key": "S7sh2::bgfirst_bg",
   "chain_winner": true,
   "authority": "plate"
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S7sh3::variants": {
  "author_fp": "1a67079e93e32048",
  "author": {
   "variants": [
    {
     "approach_ko": "낮은 정면의 절제된 대칭 구도와 깊은 초점으로 복종한 자세와 보호소의 냉정한 공간감을 함께 드러낸다.",
     "prompt_en": "Use a restrained, low frontal wide view with a natural, minimally distorted lens feel. Hold the Rottweiler precisely centered in the lower-middle frame, its entire body and tucked tail clearly separated against the shelter floor, while deep focus preserves the enclosure’s physical depth and worn surfaces. Keep the daytime ambient light broad and unsentimental, with muted color and firm but natural tonal separation."
    },
    {
     "approach_ko": "바닥 가까이의 넓은 원근감과 선택적 초점으로 개의 낮아진 몸과 주변의 빈 공간을 정서적으로 강조한다.",
     "prompt_en": "Photograph from the same low frontal position with a more expansive wide-angle sense of floor depth, letting the ground plane lead toward the Rottweiler while maintaining ample clear space around its complete body and tucked tail. Keep the dog crisply resolved as the enclosure recedes with gentle focus falloff, emphasizing how low and still its posture has become. Shape the available daytime ambient light softly across its coat and the floor, retaining desaturated color and grounded contrast."
    }
   ]
  },
  "reused": false
 },
 "S7sh3": {
  "input_fingerprint": "d454852cbec03c80",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 꼬리를 내리고 바닥에 얌전히 엎드린 로트와일러의 전신.\n\nLOCATION (lock): Inside a dog enclosure within the animal shelter. 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 dolly backward from near the dog's head at body height, settling into a low frontal wide composition that contains its entire body. Keep the 로트와일러 centered in the lower-middle frame with visible space around its tucked tail and lowered posture, preparing the camera to rise and leave it behind.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: shelter floor (Visible beneath the dog); used as Provides the ground plane that makes the dog's complete submissive posture legible.\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: Daytime ambient light preserves a restrained, desaturated palette and grounded contrast without glamorizing the animal's submission.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The Rottweiler's aggression has subsided under Suri-young's gaze, leaving it docile for the injection.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 꼬리를 내리고 바닥에 얌전히 엎드린 로트와일러의 전신.\n\nLOCATION (lock): Inside a dog enclosure within the animal shelter. 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 dolly backward from near the dog's head at body height, settling into a low frontal wide composition that contains its entire body. Keep the 로트와일러 centered in the lower-middle frame with visible space around its tucked tail and lowered posture, preparing the camera to rise and leave it behind.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: shelter floor (Visible beneath the dog); used as Provides the ground plane that makes the dog's complete submissive posture legible.\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: Daytime ambient light preserves a restrained, desaturated palette and grounded contrast without glamorizing the animal's submission.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The Rottweiler's aggression has subsided under Suri-young's gaze, leaving it docile for the injection.\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 frontal wide view with a natural, minimally distorted lens feel. Hold the Rottweiler precisely centered in the lower-middle frame, its entire body and tucked tail clearly separated against the shelter floor, while deep focus preserves the enclosure’s physical depth and worn surfaces. Keep the daytime ambient light broad and unsentimental, with muted color and firm but natural tonal separation.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 꼬리를 내리고 바닥에 얌전히 엎드린 로트와일러의 전신.\n\nLOCATION (lock): Inside a dog enclosure within the animal shelter. 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 dolly backward from near the dog's head at body height, settling into a low frontal wide composition that contains its entire body. Keep the 로트와일러 centered in the lower-middle frame with visible space around its tucked tail and lowered posture, preparing the camera to rise and leave it behind.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: shelter floor (Visible beneath the dog); used as Provides the ground plane that makes the dog's complete submissive posture legible.\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: Daytime ambient light preserves a restrained, desaturated palette and grounded contrast without glamorizing the animal's submission.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The Rottweiler's aggression has subsided under Suri-young's gaze, leaving it docile for the injection.\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 same low frontal position with a more expansive wide-angle sense of floor depth, letting the ground plane lead toward the Rottweiler while maintaining ample clear space around its complete body and tucked tail. Keep the dog crisply resolved as the enclosure recedes with gentle focus falloff, emphasizing how low and still its posture has become. Shape the available daytime ambient light softly across its coat and the floor, retaining desaturated color and grounded 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."
  },
  "totals": {
   "A": 7,
   "B": 5
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "지정된 로우 정면 와이드 샷으로 엎드린 로트와일러를 잘 담아냈으며, 레퍼런스의 배경 디테일(싱크대, 선반 등)을 매우 정확하게 유지했습니다."
   },
   {
    "label": "B",
    "score": 5,
    "verdict_ko": "강아지의 엎드린 자세는 적절히 묘사되었으나, 우측 싱크대의 형태가 레퍼런스와 어긋나며 프레이밍이 다소 좁게 잡혔습니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S7sh2_sel.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "정면 앵글로 인해 개의 몸 뒷부분과 꼬리가 가려져 '전신'과 '말린 꼬리'가 보이지 않습니다.",
     "fix_en": "Angle the dog's body slightly so its entire length, hindquarters, and tucked tail are clearly visible against the floor."
    },
    {
     "issue_ko": "텍스트 금지 지시가 있음에도 배경 창문 아래 벽면에 정체불명의 문자가 포함되어 있습니다.",
     "fix_en": "Remove the illegible text signs from the wall below the background windows."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Shift the low camera slightly off the dog's head-on axis while keeping a frontal wide view, so the entire body and tucked tail are visibly separated against the floor, with the dog still centered in the lower-middle frame.",
     "issue_ko": "정면 축의 과도한 단축 때문에 뒷몸과 접힌 꼬리가 몸통 뒤에 가려져 전신과 꼬리의 여백이 명확하게 보이지 않는다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Angle the dog's body slightly so its entire length, hindquarters, and tucked tail are clearly visible against the floor.\n- Remove the illegible text signs from the wall below the background windows.\n- Shift the low camera slightly off the dog's head-on axis while keeping a frontal wide view, so the entire body and tucked tail are visibly separated against the floor, with the dog still centered in the lower-middle 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": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지정된 로우 앵글의 와이드 샷을 잘 구현하였으며, 꼬리를 내리고 엎드린 개의 전신을 텍스트 생성 오류 없이 완벽하게 표현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "배경 창문 아래에 지시되지 않은 텍스트/표지판이 생성되는 하드 위반이 발생했으며, 지시된 꼬리의 형태가 보이지 않습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지정된 로우 앵글의 와이드 샷을 잘 구현하였으며, 꼬리를 내리고 엎드린 개의 전신을 텍스트 생성 오류 없이 완벽하게 표현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "배경 창문 아래에 지시되지 않은 텍스트/표지판이 생성되는 하드 위반이 발생했으며, 지시된 꼬리의 형태가 보이지 않습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 프레이밍과 로트와일러의 자세를 잘 구현했으며, 꼬리가 보이도록 주변 공간을 확보하라는 지침을 충실히 따랐습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "강아지의 꼬리가 보이지 않아 지시사항을 누락했으며, 배경 벽에 레퍼런스에 없는 텍스트 표지판이 생성되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 프레이밍과 로트와일러의 자세를 잘 구현했으며, 꼬리가 보이도록 주변 공간을 확보하라는 지침을 충실히 따랐습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "강아지의 꼬리가 보이지 않아 지시사항을 누락했으며, 배경 벽에 레퍼런스에 없는 텍스트 표지판이 생성되었습니다."
     }
    ]
   },
   "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": "prev만 (배경 전용·공유 계획)",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S7sh2"
  },
  "lane_policy": "share_plan_prev_bgonly"
 },
 "S7sh6::variants": {
  "author_fp": "6d92eb69b2682b4f",
  "author": {
   "variants": [
    {
     "approach_ko": "얕은 초점과 절제된 정면광으로 굳게 다문 입술과 전화기 쪽 얼굴의 긴장을 밀착 포착한다.",
     "prompt_en": "Use an intimate, shallow-focus close-up with a natural perspective, settling at her eye height in the required three-quarter frontal view. Let soft daytime ambient light skim across her face, preserving restrained color and gentle shadow while giving strongest visual emphasis to the compressed line of her lips; keep the phone crisp along the near frame edge and let the receding occupied enclosures dissolve into muted peripheral softness."
    },
    {
     "approach_ko": "우리의 사선 후퇴선을 활용한 층 깊은 구도로 고립감과 통화의 불안을 강조한다.",
     "prompt_en": "Build the close-up around the diagonal recession of the enclosure fronts, using their soft repeating structure to pull depth away behind her off-center face. Hold the eye-height three-quarter frontal angle with a slightly compressed spatial feel: her pressed lips and listening expression remain dominant, the phone anchors the near edge, and subdued directional daylight creates low-to-mid-key modeling while the occupied enclosures linger as recognizable but unfocused context."
    }
   ]
  },
  "reused": false
 },
 "S7sh6": {
  "input_fingerprint": "6a852479612621fd",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 수리영이 휴대전화를 귀에 댄 채 입술을 굳게 다문 표정.\n\nLOCATION (lock): In the animal shelter's work area beside the dog enclosures. 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 at 수리영's eye height from the inherited medium position and settle into a close three-quarter frontal view as the phone reaches her ear. Place her face slightly off-center with the phone along the near edge of the frame, holding on her pressed lips while the occupied dog enclosures remain soft and peripheral.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: confined-dog enclosures (Occupied by confined dogs) — Their front sides recede behind 수리영; used as Soft background context behind her listening 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: Daytime ambient light is held in restrained saturation and low-to-mid-key contrast, emphasizing her tightening 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\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young has her mobile phone with her and is using it to call Minsuk, who remains unreachable.\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 초반, 긴 검은 머리, 타원형의 젊은 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 수리영이 휴대전화를 귀에 댄 채 입술을 굳게 다문 표정.\n\nLOCATION (lock): In the animal shelter's work area beside the dog enclosures. 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 at 수리영's eye height from the inherited medium position and settle into a close three-quarter frontal view as the phone reaches her ear. Place her face slightly off-center with the phone along the near edge of the frame, holding on her pressed lips while the occupied dog enclosures remain soft and peripheral.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: confined-dog enclosures (Occupied by confined dogs) — Their front sides recede behind 수리영; used as Soft background context behind her listening 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: Daytime ambient light is held in restrained saturation and low-to-mid-key contrast, emphasizing her tightening 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\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young has her mobile phone with her and is using it to call Minsuk, who remains unreachable.\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 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an intimate, shallow-focus close-up with a natural perspective, settling at her eye height in the required three-quarter frontal view. Let soft daytime ambient light skim across her face, preserving restrained color and gentle shadow while giving strongest visual emphasis to the compressed line of her lips; keep the phone crisp along the near frame edge and let the receding occupied enclosures dissolve into muted peripheral 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.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 수리영이 휴대전화를 귀에 댄 채 입술을 굳게 다문 표정.\n\nLOCATION (lock): In the animal shelter's work area beside the dog enclosures. 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 at 수리영's eye height from the inherited medium position and settle into a close three-quarter frontal view as the phone reaches her ear. Place her face slightly off-center with the phone along the near edge of the frame, holding on her pressed lips while the occupied dog enclosures remain soft and peripheral.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: confined-dog enclosures (Occupied by confined dogs) — Their front sides recede behind 수리영; used as Soft background context behind her listening 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: Daytime ambient light is held in restrained saturation and low-to-mid-key contrast, emphasizing her tightening 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\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young has her mobile phone with her and is using it to call Minsuk, who remains unreachable.\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 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nBuild the close-up around the diagonal recession of the enclosure fronts, using their soft repeating structure to pull depth away behind her off-center face. Hold the eye-height three-quarter frontal angle with a slightly compressed spatial feel: her pressed lips and listening expression remain dominant, the phone anchors the near edge, and subdued directional daylight creates low-to-mid-key modeling while the occupied enclosures linger as recognizable but unfocused context.\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 클로즈업 구도와 3/4 측면 앵글을 정확히 따랐으며, 굳게 다문 입술의 표정과 휴대전화의 위치가 완벽하게 일치합니다."
   },
   {
    "label": "B",
    "score": 5,
    "verdict_ko": "요구된 앵글보다 너무 정면을 향하고 중앙에 치우쳐 있으며, 굳게 다문 입술의 긴장감이 상대적으로 덜 표현되었습니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S7sh3_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "휴대전화의 화면이 귀 쪽이 아닌 바깥쪽(카메라 쪽)을 향하도록 거꾸로 들려 있습니다.",
     "fix_en": "Flip the mobile phone so the screen faces her head and the back panel faces outward."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Reframe so the phone sits along the near left edge of the frame while preserving the eye-level close-up, three-quarter frontal angle, and her slightly off-center face.",
     "issue_ko": "휴대전화가 프레임의 가까운 가장자리가 아니라 화면 안쪽 중앙부에 놓여 있다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Flip the mobile phone so the screen faces her head and the back panel faces outward.\n- Reframe so the phone sits along the near left edge of the frame while preserving the eye-level close-up, three-quarter frontal angle, and her slightly off-center 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": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "지정된 인물의 외모, 휴대전화를 귀에 댄 자세, 굳게 다문 입술 표정과 배경의 개집까지 프롬프트의 모든 세부 사항을 완벽하게 충족했습니다."
     },
     {
      "label": "B",
      "score": 2,
      "verdict_ko": "화면 좌측 하단에 프롬프트에 없는 두 번째 휴대전화를 든 정체불명의 손이 나타나 심각한 지시 위반(발명된 인물/사물)에 해당합니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "지정된 인물의 외모, 휴대전화를 귀에 댄 자세, 굳게 다문 입술 표정과 배경의 개집까지 프롬프트의 모든 세부 사항을 완벽하게 충족했습니다."
     },
     {
      "label": "B",
      "score": 2,
      "verdict_ko": "화면 좌측 하단에 프롬프트에 없는 두 번째 휴대전화를 든 정체불명의 손이 나타나 심각한 지시 위반(발명된 인물/사물)에 해당합니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 0,
      "verdict_ko": "화면 좌측에 설명에 없는 정체불명의 손과 두 번째 휴대전화가 등장하는 치명적인 오류(Hard Violation)가 있습니다."
     },
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "인물의 감정선, 요구된 클로즈업 구도, 배경의 동물 보호소 환경 및 지정된 의상을 모두 완벽하고 사실적으로 구현했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 0,
      "verdict_ko": "화면 좌측에 설명에 없는 정체불명의 손과 두 번째 휴대전화가 등장하는 치명적인 오류(Hard Violation)가 있습니다."
     },
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "인물의 감정선, 요구된 클로즈업 구도, 배경의 동물 보호소 환경 및 지정된 의상을 모두 완벽하고 사실적으로 구현했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 20,
     "B": 2
    },
    "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": "S7sh3"
  }
 },
 "S8sh1::variants": {
  "author_fp": "b33fb0fbf4f66568",
  "author": {
   "variants": [
    {
     "approach_ko": "눈높이 측면 트래킹 구도로 수리영의 전신과 페달링 자세를 또렷하게 읽히며 거리의 깊이를 길게 펼친다.",
     "prompt_en": "Photograph the moment as an eye-level lateral tracking view, keeping Suri-young’s entire figure and bicycle clearly legible in a clean full-body composition. Let the commercial street recede in layered depth behind her, with restrained dusk ambience and muted neon reflections separating her from the street-food stalls. Preserve natural directional movement with a crisp subject and only subtle softness in the receding surroundings."
    },
    {
     "approach_ko": "낮은 전방 사선 시점과 압축된 배경으로 자전거 위 전신 자세를 강조하고 네온과 포장마차를 조밀하게 겹친다.",
     "prompt_en": "Use a low frontal three-quarter viewpoint that gives the bicycle and Suri-young’s full pedaling posture immediate visual weight while retaining her complete body in frame. Compress the neon signs and street-food stalls into a dense backdrop, using shallow but believable depth so her raised and lowered knees read distinctly against separated planes. Shape the dusk light softly across her and allow restrained neon accents to edge the real metal, fabric, and street surfaces."
    }
   ]
  },
  "reused": false
 },
 "S8sh1": {
  "input_fingerprint": "1b2f3eb28166efae",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 네온사인과 포장마차가 늘어선 번화가 거리에서 자전거 안장에 앉아 한쪽 무릎은 내려가고 반대쪽 무릎은 올라온 페달링 중 자세의 수리영 전신.\n\nLOCATION (lock): On the busy commercial street outside a large supermarket, between neon signs and street-food stalls at dusk. The shot takes place here — the attached STORYBOARD SKETCH fixes the staging, camera and figure placement of this exact place. No location photograph is attached — build the location itself strictly from the location text above and the shot text, inventing nothing beyond them.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk ambient light and the explicitly present neon signage are balanced in a restrained, desaturated grade with selective muted color 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young continues traveling on her bicycle through the shopping district.\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 초반, 긴 검은 머리, 타원형의 젊은 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 네온사인과 포장마차가 늘어선 번화가 거리에서 자전거 안장에 앉아 한쪽 무릎은 내려가고 반대쪽 무릎은 올라온 페달링 중 자세의 수리영 전신.\n\nLOCATION (lock): On the busy commercial street outside a large supermarket, between neon signs and street-food stalls at dusk. The shot takes place here — the attached STORYBOARD SKETCH fixes the staging, camera and figure placement of this exact place. No location photograph is attached — build the location itself strictly from the location text above and the shot text, inventing nothing beyond them.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk ambient light and the explicitly present neon signage are balanced in a restrained, desaturated grade with selective muted color 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young continues traveling on her bicycle through the shopping district.\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 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment as an eye-level lateral tracking view, keeping Suri-young’s entire figure and bicycle clearly legible in a clean full-body composition. Let the commercial street recede in layered depth behind her, with restrained dusk ambience and muted neon reflections separating her from the street-food stalls. Preserve natural directional movement with a crisp subject and only subtle softness in the receding 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.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 네온사인과 포장마차가 늘어선 번화가 거리에서 자전거 안장에 앉아 한쪽 무릎은 내려가고 반대쪽 무릎은 올라온 페달링 중 자세의 수리영 전신.\n\nLOCATION (lock): On the busy commercial street outside a large supermarket, between neon signs and street-food stalls at dusk. The shot takes place here — the attached STORYBOARD SKETCH fixes the staging, camera and figure placement of this exact place. No location photograph is attached — build the location itself strictly from the location text above and the shot text, inventing nothing beyond them.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk ambient light and the explicitly present neon signage are balanced in a restrained, desaturated grade with selective muted color 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young continues traveling on her bicycle through the shopping district.\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 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a low frontal three-quarter viewpoint that gives the bicycle and Suri-young’s full pedaling posture immediate visual weight while retaining her complete body in frame. Compress the neon signs and street-food stalls into a dense backdrop, using shallow but believable depth so her raised and lowered knees read distinctly against separated planes. Shape the dusk light softly across her and allow restrained neon accents to edge the real metal, fabric, and street 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": "스토리보드의 카메라 구도와 배경 요소를 정확히 구현했으며, 지시된 페달링 자세와 인물의 인상착의가 물리적 오류 없이 자연스럽게 표현되었습니다."
   },
   {
    "label": "B",
    "score": 3,
    "verdict_ko": "인물의 오른쪽 발이 얹힌 페달이 자전거 크랭크와 연결되어 있지 않고 허공에 떠 있는 물리적 구조 오류가 발생하여 치명적인 감점 요인입니다."
   }
  ],
  "refs": [
   {
    "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/lane_sketch_S8sh1.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "샷 텍스트에 명시되지 않은 여러 배경 인물들이 거리에 포함되어 있습니다.",
     "fix_en": "Remove all other people from the background, leaving only Suri-young in the scene."
    },
    {
     "issue_ko": "프롬프트의 '텍스트 없음' 규칙에도 불구하고 간판과 상점에 글자들이 선명하게 나타납니다.",
     "fix_en": "Remove all legible text and characters from the neon signs and storefronts, replacing them with abstract light shapes."
    }
   ]
  },
  "composition_critique": {
   "issues": []
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all other people from the background, leaving only Suri-young in the scene.\n- Remove all legible text and characters from the neon signs and storefronts, replacing them with abstract light shapes.\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": "프롬프트의 엄격한 제약 조건에 따라 샷 텍스트에 없는 배경 인물과 간판의 텍스트를 완벽히 배제하면서 요구된 구도와 포즈를 정확히 구현했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "주인공 외에 인물을 추가하지 말라는 지시를 어기고 다수의 배경 인물을 포함했으며, 텍스트 금지 규칙을 위반하고 간판에 글자를 생성했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "프롬프트의 엄격한 제약 조건에 따라 샷 텍스트에 없는 배경 인물과 간판의 텍스트를 완벽히 배제하면서 요구된 구도와 포즈를 정확히 구현했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "주인공 외에 인물을 추가하지 말라는 지시를 어기고 다수의 배경 인물을 포함했으며, 텍스트 금지 규칙을 위반하고 간판에 글자를 생성했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "텍스트 금지 및 단일 인물 제약 조건을 잘 준수하여 샷 텍스트의 프레이밍과 자세를 정확하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "명시적인 텍스트 금지 규칙을 어기고 간판에 글자를 포함했으며, 지시되지 않은 다수의 인물을 배경에 추가하여 감점되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "텍스트 금지 및 단일 인물 제약 조건을 잘 준수하여 샷 텍스트의 프레이밍과 자세를 정확하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "명시적인 텍스트 금지 규칙을 어기고 간판에 글자를 포함했으며, 지시되지 않은 다수의 인물을 배경에 추가하여 감점되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 6,
     "B": 14
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "lane(map_marker): 스케치+엔티티",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S8sh6::bgfirst_bg": {
  "input_fingerprint": "18c4529eeb43eaf8",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 눈을 동그랗게 뜨고 멈칫한 수리영의 얼굴.\n\nLOCATION (lock): Inside the supermarket's retail floor, beside the employees stocking merchandise.\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: Finish the eye-line dolly-in inside the store at face-close distance and stop on 수리영's widened eyes. Center her face tightly with only narrow, softened traces of the store aisles at the edges, holding the startled pause before the exterior cut.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: store display aisles (In use for merchandise display) — Their visible sides recede away behind her; used as Soft edge context that identifies the store while isolating her 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: Setting-appropriate interior ambient light is rendered with restrained saturation and controlled low-to-mid-key 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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 눈을 동그랗게 뜨고 멈칫한 수리영의 얼굴.\n\nLOCATION (lock): Inside the supermarket's retail floor, beside the employees stocking merchandise.\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: Finish the eye-line dolly-in inside the store at face-close distance and stop on 수리영's widened eyes. Center her face tightly with only narrow, softened traces of the store aisles at the edges, holding the startled pause before the exterior cut.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: store display aisles (In use for merchandise display) — Their visible sides recede away behind her; used as Soft edge context that identifies the store while isolating her 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: Setting-appropriate interior ambient light is rendered with restrained saturation and controlled low-to-mid-key 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S8sh6__bgfirst_bg.png",
  "asset_id": "57b0a7e9-1676-422c-9838-c88bad1ef077",
  "input_asset_ids": [
   "539bb4cc-81a3-4224-ba5d-812722b5c562",
   "5022255a-e2c0-453a-a183-bcbbd7e3e146"
  ]
 },
 "S8sh6": {
  "input_fingerprint": "bef392498f984e81",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 눈을 동그랗게 뜨고 멈칫한 수리영의 얼굴.\n\nLOCATION (lock): Inside the supermarket's retail floor, beside the employees stocking merchandise. 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: Finish the eye-line dolly-in inside the store at face-close distance and stop on 수리영's widened eyes. Center her face tightly with only narrow, softened traces of the store aisles at the edges, holding the startled pause before the exterior cut.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: store display aisles (In use for merchandise display) — Their visible sides recede away behind her; used as Soft edge context that identifies the store while isolating her 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: Setting-appropriate interior ambient light is rendered with restrained saturation and controlled low-to-mid-key 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young has already parked her bicycle outside the mart, and Minsuk remains absent from her workplace.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 눈을 동그랗게 뜨고 멈칫한 수리영의 얼굴.\n\nLOCATION (lock): Inside the supermarket's retail floor, beside the employees stocking merchandise. 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: Finish the eye-line dolly-in inside the store at face-close distance and stop on 수리영's widened eyes. Center her face tightly with only narrow, softened traces of the store aisles at the edges, holding the startled pause before the exterior cut.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: store display aisles (In use for merchandise display) — Their visible sides recede away behind her; used as Soft edge context that identifies the store while isolating her 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: Setting-appropriate interior ambient light is rendered with restrained saturation and controlled low-to-mid-key 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young has already parked her bicycle outside the mart, and Minsuk remains absent from her workplace.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 눈을 동그랗게 뜨고 멈칫한 수리영의 얼굴.\n\nLOCATION (lock): Inside the supermarket's retail floor, beside the employees stocking merchandise. 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: Finish the eye-line dolly-in inside the store at face-close distance and stop on 수리영's widened eyes. Center her face tightly with only narrow, softened traces of the store aisles at the edges, holding the startled pause before the exterior cut.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: store display aisles (In use for merchandise display) — Their visible sides recede away behind her; used as Soft edge context that identifies the store while isolating her 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: Setting-appropriate interior ambient light is rendered with restrained saturation and controlled low-to-mid-key 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young has already parked her bicycle outside the mart, and Minsuk remains absent from her workplace.\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 초반, 긴 검은 머리, 타원형의 젊은 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S8sh6__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S8sh6.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L08B03.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 타이트한 클로즈업 프레이밍과 부드럽게 날아간 배경 심도를 정확히 구현했으며, 놀란 눈빛의 인물에 집중하여 프롬프트를 충실히 따름."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "프레이밍이 지시된 것보다 넓으며, 샷 텍스트에 명시되지 않은 배경 인물들을 추가하여 등장인물 제약 조건을 위반함."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 타이트한 클로즈업 프레이밍과 부드럽게 날아간 배경 심도를 정확히 구현했으며, 놀란 눈빛의 인물에 집중하여 프롬프트를 충실히 따름."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "프레이밍이 지시된 것보다 넓으며, 샷 텍스트에 명시되지 않은 배경 인물들을 추가하여 등장인물 제약 조건을 위반함."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 타이트한 클로즈업 샷 크기와 흐릿하고 좁은 배경 프레이밍 지침을 완벽하게 준수했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "프레이밍이 너무 넓어 얼굴을 타이트하게 중앙에 배치하라는 카메라 및 프레임 지침을 어겼습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지정된 타이트한 클로즈업 샷 크기와 흐릿하고 좁은 배경 프레이밍 지침을 완벽하게 준수했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "프레이밍이 너무 넓어 얼굴을 타이트하게 중앙에 배치하라는 카메라 및 프레임 지침을 어겼습니다."
     }
    ]
   },
   "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": "지시된 타이트한 클로즈업 프레이밍과 부드럽게 날아간 배경 심도를 정확히 구현했으며, 놀란 눈빛의 인물에 집중하여 프롬프트를 충실히 따름."
   },
   {
    "label": "A",
    "score": 3,
    "verdict_ko": "프레이밍이 지시된 것보다 넓으며, 샷 텍스트에 명시되지 않은 배경 인물들을 추가하여 등장인물 제약 조건을 위반함."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L08B03.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": []
  },
  "composition_critique": {
   "issues": []
  },
  "fix_skipped": true,
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S8sh6__bgfirst_bg.png",
   "bg_asset_id": "57b0a7e9-1676-422c-9838-c88bad1ef077",
   "bg_record_key": "S8sh6::bgfirst_bg",
   "chain_winner": false,
   "authority": "plate"
  },
  "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L08B01",
    "B": "L08B02",
    "C": "L08B03"
   },
   "assigned": "L08B03",
   "choice": "Candidate C",
   "confident": false,
   "reason_ko": "수리영의 표정(눈을 동그랗게 뜨고 멈칫함)만 묘사되어 있어, 특정 서브공간을 지시하는 단서가 부족하므로 현재 할당된 후보를 유지함.",
   "kept": "L08B03"
  }
 },
 "S8sh7::bgfirst_bg": {
  "input_fingerprint": "b7c9051af06e9a39",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 마트 건물 밖으로 나온 수리영이 휴대전화를 귀에 댄 채 서 있는 정지 구도.\n\nLOCATION (lock): Immediately outside the supermarket entrance at dusk, facing the adjoining commercial street.\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 the fixed eye-level position near the entrance, observe 수리영 at a medium-wide distance after she has come outside. Keep her full figure slightly to one side of center with the phone at her ear, using the store entrance and commercial street as stable spatial context rather than moving with her.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: store entrance (Visible behind her) — The entrance front faces the camera at a slight angle; used as Frames the boundary between the store and exterior street behind 수리영; street signage (Present outside the store) — Display faces are directed toward the street; specific content is not established; used as Provides distant commercial context around the static call.\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 and the commercial signage remain subdued within a desaturated, low-to-mid-key treatment.\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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 마트 건물 밖으로 나온 수리영이 휴대전화를 귀에 댄 채 서 있는 정지 구도.\n\nLOCATION (lock): Immediately outside the supermarket entrance at dusk, facing the adjoining commercial street.\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 the fixed eye-level position near the entrance, observe 수리영 at a medium-wide distance after she has come outside. Keep her full figure slightly to one side of center with the phone at her ear, using the store entrance and commercial street as stable spatial context rather than moving with her.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: store entrance (Visible behind her) — The entrance front faces the camera at a slight angle; used as Frames the boundary between the store and exterior street behind 수리영; street signage (Present outside the store) — Display faces are directed toward the street; specific content is not established; used as Provides distant commercial context around the static call.\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 and the commercial signage remain subdued within a desaturated, low-to-mid-key treatment.\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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S8sh7__bgfirst_bg.png",
  "asset_id": "5c8142d1-6c5d-4ff9-b337-a46d6af07a0b",
  "input_asset_ids": [
   "1a8bfeac-4818-4658-9281-8b1c73f280d9",
   "24819b8e-2f46-4fec-9830-66720db14400",
   "719cdcdf-d466-4592-a7b5-dce92e27fa6e"
  ]
 },
 "S8sh7": {
  "input_fingerprint": "f81585a7e0bf5473",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 마트 건물 밖으로 나온 수리영이 휴대전화를 귀에 댄 채 서 있는 정지 구도.\n\nLOCATION (lock): Immediately outside the supermarket entrance at dusk, facing the adjoining commercial street. 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 fixed eye-level position near the entrance, observe 수리영 at a medium-wide distance after she has come outside. Keep her full figure slightly to one side of center with the phone at her ear, using the store entrance and commercial street as stable spatial context rather than moving with her.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: store entrance (Visible behind her) — The entrance front faces the camera at a slight angle; used as Frames the boundary between the store and exterior street behind 수리영; street signage (Present outside the store) — Display faces are directed toward the street; specific content is not established; used as Provides distant commercial context around the static call.\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 and the commercial signage remain subdued within a desaturated, low-to-mid-key treatment.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young's bicycle remains parked outside while she calls Minsuk on her mobile phone; Minsuk still does not answer.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 마트 건물 밖으로 나온 수리영이 휴대전화를 귀에 댄 채 서 있는 정지 구도.\n\nLOCATION (lock): Immediately outside the supermarket entrance at dusk, facing the adjoining commercial street. 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 fixed eye-level position near the entrance, observe 수리영 at a medium-wide distance after she has come outside. Keep her full figure slightly to one side of center with the phone at her ear, using the store entrance and commercial street as stable spatial context rather than moving with her.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: store entrance (Visible behind her) — The entrance front faces the camera at a slight angle; used as Frames the boundary between the store and exterior street behind 수리영; street signage (Present outside the store) — Display faces are directed toward the street; specific content is not established; used as Provides distant commercial context around the static call.\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 and the commercial signage remain subdued within a desaturated, low-to-mid-key treatment.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young's bicycle remains parked outside while she calls Minsuk on her mobile phone; Minsuk still does not answer.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 마트 건물 밖으로 나온 수리영이 휴대전화를 귀에 댄 채 서 있는 정지 구도.\n\nLOCATION (lock): Immediately outside the supermarket entrance at dusk, facing the adjoining commercial street. 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 fixed eye-level position near the entrance, observe 수리영 at a medium-wide distance after she has come outside. Keep her full figure slightly to one side of center with the phone at her ear, using the store entrance and commercial street as stable spatial context rather than moving with her.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: store entrance (Visible behind her) — The entrance front faces the camera at a slight angle; used as Frames the boundary between the store and exterior street behind 수리영; street signage (Present outside the store) — Display faces are directed toward the street; specific content is not established; used as Provides distant commercial context around the static call.\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 and the commercial signage remain subdued within a desaturated, low-to-mid-key treatment.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young's bicycle remains parked outside while she calls Minsuk on her mobile phone; Minsuk still does not answer.\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 초반, 긴 검은 머리, 타원형의 젊은 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S8sh7__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S8sh7.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L08B02.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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/background_chain/seed_bg_supermarket_complex_sel.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지정된 건물 외관(STRUCTURE LOOK)과 주변 환경(LOCATION)의 조합 조건을 완벽히 충족하며 프레이밍과 동작 지시를 잘 따랐습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "건물 외관 지정 이미지(STRUCTURE LOOK)를 따르지 않고 위치 참고 사진의 낡은 건물 형태와 어닝을 그대로 반영하여 우선순위 조건에 위배됩니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지정된 건물 외관(STRUCTURE LOOK)과 주변 환경(LOCATION)의 조합 조건을 완벽히 충족하며 프레이밍과 동작 지시를 잘 따랐습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "건물 외관 지정 이미지(STRUCTURE LOOK)를 따르지 않고 위치 참고 사진의 낡은 건물 형태와 어닝을 그대로 반영하여 우선순위 조건에 위배됩니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 건물 외관(Structure Look)을 완벽히 반영했으며 인물의 와이드 샷 구도와 동작을 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "건물 외관 참조 이미지(Structure Look)를 무시하고 장소 참조 이미지의 형태와 어닝을 그대로 유지해 최우선 지침을 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 건물 외관(Structure Look)을 완벽히 반영했으며 인물의 와이드 샷 구도와 동작을 정확히 구현했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "건물 외관 참조 이미지(Structure Look)를 무시하고 장소 참조 이미지의 형태와 어닝을 그대로 유지해 최우선 지침을 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": "B",
    "score": 8,
    "verdict_ko": "지정된 건물 외관(STRUCTURE LOOK)과 주변 환경(LOCATION)의 조합 조건을 완벽히 충족하며 프레이밍과 동작 지시를 잘 따랐습니다."
   },
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "건물 외관 지정 이미지(STRUCTURE LOOK)를 따르지 않고 위치 참고 사진의 낡은 건물 형태와 어닝을 그대로 반영하여 우선순위 조건에 위배됩니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L08B02.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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/background_chain/seed_bg_supermarket_complex_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "카메라 구도가 레퍼런스 사진의 원경을 그대로 복사하여 인물이 너무 작게 묘사되었으며, 지시사항과 달리 마트 입구가 인물 뒤 배경으로 배치되지 않았습니다.",
     "fix_en": "Move the camera closer to an eye-level medium-wide distance so the full-figure character is larger in the frame, and position the store entrance directly behind her."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Reposition the camera to a fixed eye-level viewpoint near the supermarket entrance, maintaining Suri-young’s full figure slightly off-center with the entrance behind her and the commercial street visible in depth.",
     "issue_ko": "카메라가 입구 근처의 눈높이 시점이 아니라 길 건너편의 높은 위치에서 내려다보는 원경 시점으로 배치되어 있다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Move the camera closer to an eye-level medium-wide distance so the full-figure character is larger in the frame, and position the store entrance directly behind her.\n- Reposition the camera to a fixed eye-level viewpoint near the supermarket entrance, maintaining Suri-young’s full figure slightly off-center with the entrance behind her and the commercial street visible in depth.\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 프레임(미디엄 와이드/와이드 샷)보다 다소 넓게 연출되었으나, 금지된 추가 인물 없이 건물의 외형(Structure Look)과 주변 환경, 인물의 동작을 충실히 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "프레임 크기는 지시사항에 더 가깝지만, 우측 배경 거리에 프롬프트에 없는 행인을 임의로 추가하여 '추가 인물 금지' 조항을 위반한 하드 바이올레이션입니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 프레임(미디엄 와이드/와이드 샷)보다 다소 넓게 연출되었으나, 금지된 추가 인물 없이 건물의 외형(Structure Look)과 주변 환경, 인물의 동작을 충실히 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "프레임 크기는 지시사항에 더 가깝지만, 우측 배경 거리에 프롬프트에 없는 행인을 임의로 추가하여 '추가 인물 금지' 조항을 위반한 하드 바이올레이션입니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "요청된 미디엄 와이드 거리와 와이드 샷 프레이밍을 정확히 구현하여 인물을 적절한 비율로 배치했으며, 참조된 구조물의 외관과 시간대 설정을 훌륭하게 반영했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "지정된 카메라 거리보다 훨씬 멀리서 촬영된 익스트림 와이드 샷으로 렌더링되어, 인물이 프레임 내에서 지나치게 작게 표현되는 프레이밍 오류가 있습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "요청된 미디엄 와이드 거리와 와이드 샷 프레이밍을 정확히 구현하여 인물을 적절한 비율로 배치했으며, 참조된 구조물의 외관과 시간대 설정을 훌륭하게 반영했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "지정된 카메라 거리보다 훨씬 멀리서 촬영된 익스트림 와이드 샷으로 렌더링되어, 인물이 프레임 내에서 지나치게 작게 표현되는 프레이밍 오류가 있습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 13,
     "B": 12
    },
    "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S8sh7__bgfirst_bg.png",
   "bg_asset_id": "5c8142d1-6c5d-4ff9-b337-a46d6af07a0b",
   "bg_record_key": "S8sh7::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": "L08B01",
    "B": "L08B02",
    "C": "L08B03"
   },
   "assigned": "L08B02",
   "choice": "Candidate B",
   "confident": true,
   "reason_ko": "'마트 건물 밖'이라는 조건만 주어졌으며, 제시된 후보들(자전거 주차 구역, 노점 구역 등) 중 특정 공간을 요구하는 단서가 없으므로 기존 할당을 유지함.",
   "kept": "L08B02"
  }
 },
 "S9sh1::bgfirst_bg": {
  "input_fingerprint": "837a09fc11378bd3",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어두운 밤, 해안도로 위에서 상체를 숙인 채 자전거 핸들을 꽉 쥐고 한쪽 무릎을 굽혀 페달을 밟아 내리는 중간 상태인 수리영, 바람에 옷깃이 뒤로 펄럭이는 모습.\n\nLOCATION (lock): On the dark coastal road at night, along the route back home.\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 beside 수리영 at upper-body height and medium distance, holding a strong side profile while retaining the handlebars and the active pedal downstroke in frame. Her forward-canted torso occupies the center-left, with the coastal road extending through the remaining frame to preserve speed and direction.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: coastal road (Visible at night); used as Provides the directional line along which the parallel tracking move travels; bicycle (Being ridden at speed) — Its side profile faces the camera during the parallel track; used as Connects her clenched grip, seated posture, and pedal action within one readable silhouette.\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 nighttime ambient light keeps the rider legible in a restrained, desaturated image without specifying an unsupported 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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어두운 밤, 해안도로 위에서 상체를 숙인 채 자전거 핸들을 꽉 쥐고 한쪽 무릎을 굽혀 페달을 밟아 내리는 중간 상태인 수리영, 바람에 옷깃이 뒤로 펄럭이는 모습.\n\nLOCATION (lock): On the dark coastal road at night, along the route back home.\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 beside 수리영 at upper-body height and medium distance, holding a strong side profile while retaining the handlebars and the active pedal downstroke in frame. Her forward-canted torso occupies the center-left, with the coastal road extending through the remaining frame to preserve speed and direction.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: coastal road (Visible at night); used as Provides the directional line along which the parallel tracking move travels; bicycle (Being ridden at speed) — Its side profile faces the camera during the parallel track; used as Connects her clenched grip, seated posture, and pedal action within one readable silhouette.\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 nighttime ambient light keeps the rider legible in a restrained, desaturated image without specifying an unsupported 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S9sh1__bgfirst_bg.png",
  "asset_id": "d0799ea3-ac1b-4830-b134-895794717425",
  "input_asset_ids": [
   "dc095876-90a5-4187-ae34-e96d037690b8",
   "20148f1f-f264-4442-8f6d-57b8ec8f2c2e"
  ]
 },
 "S9sh1": {
  "input_fingerprint": "96e4296f29c075cb",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤, 해안도로 위에서 상체를 숙인 채 자전거 핸들을 꽉 쥐고 한쪽 무릎을 굽혀 페달을 밟아 내리는 중간 상태인 수리영, 바람에 옷깃이 뒤로 펄럭이는 모습.\n\nLOCATION (lock): On the dark coastal road at night, along the route back home. 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 beside 수리영 at upper-body height and medium distance, holding a strong side profile while retaining the handlebars and the active pedal downstroke in frame. Her forward-canted torso occupies the center-left, with the coastal road extending through the remaining frame to preserve speed and direction.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: coastal road (Visible at night); used as Provides the directional line along which the parallel tracking move travels; bicycle (Being ridden at speed) — Its side profile faces the camera during the parallel track; used as Connects her clenched grip, seated posture, and pedal action within one readable silhouette.\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 nighttime ambient light keeps the rider legible in a restrained, desaturated image without specifying an unsupported 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young is still using the same bicycle as she hurries home, with her mobile phone remaining among her belongings.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤, 해안도로 위에서 상체를 숙인 채 자전거 핸들을 꽉 쥐고 한쪽 무릎을 굽혀 페달을 밟아 내리는 중간 상태인 수리영, 바람에 옷깃이 뒤로 펄럭이는 모습.\n\nLOCATION (lock): On the dark coastal road at night, along the route back home. 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 beside 수리영 at upper-body height and medium distance, holding a strong side profile while retaining the handlebars and the active pedal downstroke in frame. Her forward-canted torso occupies the center-left, with the coastal road extending through the remaining frame to preserve speed and direction.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: coastal road (Visible at night); used as Provides the directional line along which the parallel tracking move travels; bicycle (Being ridden at speed) — Its side profile faces the camera during the parallel track; used as Connects her clenched grip, seated posture, and pedal action within one readable silhouette.\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 nighttime ambient light keeps the rider legible in a restrained, desaturated image without specifying an unsupported 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young is still using the same bicycle as she hurries home, with her mobile phone remaining among her belongings.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤, 해안도로 위에서 상체를 숙인 채 자전거 핸들을 꽉 쥐고 한쪽 무릎을 굽혀 페달을 밟아 내리는 중간 상태인 수리영, 바람에 옷깃이 뒤로 펄럭이는 모습.\n\nLOCATION (lock): On the dark coastal road at night, along the route back home. 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 beside 수리영 at upper-body height and medium distance, holding a strong side profile while retaining the handlebars and the active pedal downstroke in frame. Her forward-canted torso occupies the center-left, with the coastal road extending through the remaining frame to preserve speed and direction.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: coastal road (Visible at night); used as Provides the directional line along which the parallel tracking move travels; bicycle (Being ridden at speed) — Its side profile faces the camera during the parallel track; used as Connects her clenched grip, seated posture, and pedal action within one readable silhouette.\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 nighttime ambient light keeps the rider legible in a restrained, desaturated image without specifying an unsupported 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young is still using the same bicycle as she hurries home, with her mobile phone remaining among her belongings.\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 초반, 긴 검은 머리, 타원형의 젊은 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S9sh1__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S9sh1.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/groupbg_coastal_road_16062d.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 완벽한 측면 프로파일과 페달링 동작, 배경의 뻗은 도로를 잘 구현했으나 미디엄 샷보다 다소 넓게 촬영되었습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "카메라 앵글이 요구된 평행 측면이 아닌 사선(3/4) 구도이며, 페달을 밟는 하단부 동작이 프레임 밖으로 잘렸습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 완벽한 측면 프로파일과 페달링 동작, 배경의 뻗은 도로를 잘 구현했으나 미디엄 샷보다 다소 넓게 촬영되었습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "카메라 앵글이 요구된 평행 측면이 아닌 사선(3/4) 구도이며, 페달을 밟는 하단부 동작이 프레임 밖으로 잘렸습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "요구된 완벽한 측면 프로필(side profile)이 아닌 45도 앞쪽 측면에서 촬영되어 카메라 앵글 지침을 어겼습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지정된 완벽한 측면 앵글, 자전거 탑승 자세, 배경 위치 및 화면 구도를 모두 훌륭하게 구현했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "요구된 완벽한 측면 프로필(side profile)이 아닌 45도 앞쪽 측면에서 촬영되어 카메라 앵글 지침을 어겼습니다."
     },
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 완벽한 측면 앵글, 자전거 탑승 자세, 배경 위치 및 화면 구도를 모두 훌륭하게 구현했습니다."
     }
    ]
   },
   "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": "지시된 완벽한 측면 프로파일과 페달링 동작, 배경의 뻗은 도로를 잘 구현했으나 미디엄 샷보다 다소 넓게 촬영되었습니다."
   },
   {
    "label": "B",
    "score": 3,
    "verdict_ko": "카메라 앵글이 요구된 평행 측면이 아닌 사선(3/4) 구도이며, 페달을 밟는 하단부 동작이 프레임 밖으로 잘렸습니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/groupbg_coastal_road_16062d.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "자전거 체인링 아래 허공에 불필요한 세 번째 신발(발)이 떠 있습니다.",
     "fix_en": "Remove the extra floating shoe near the ground below the bicycle's chainring."
    },
    {
     "issue_ko": "레이아웃 스케치와 다르게 오른 다리가 위로 구부러지고 왼 다리가 아래로 뻗어 있습니다.",
     "fix_en": "Swap the leg positions so the right leg is extended downward and the left leg is bent upward, exactly matching the layout sketch."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Restore the shot background to its exact reference camera alignment and perspective, placing the horizon and road vanishing point at their original positions while preserving the rider unchanged.",
     "issue_ko": "배경의 카메라 정렬과 원근이 기준 이미지와 다릅니다. 수평선과 도로 소실점이 기준보다 위쪽·왼쪽으로 이동해 해안도로의 화면 구도가 바뀌었습니다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the extra floating shoe near the ground below the bicycle's chainring.\n- Swap the leg positions so the right leg is extended downward and the left leg is bent upward, exactly matching the layout sketch.\n- Restore the shot background to its exact reference camera alignment and perspective, placing the horizon and road vanishing point at their original positions while preserving the rider unchanged.\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": "지정된 자세와 구도를 잘 따랐으나, 배경 하늘 영역에 부자연스러운 사각형 형태의 노이즈 아티팩트가 발생하여 감점되었습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 스케치의 구도, 인물의 역동적인 자세, 그리고 참조 배경을 시각적 오류나 아티팩트 없이 가장 충실하게 구현했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "지정된 자세와 구도를 잘 따랐으나, 배경 하늘 영역에 부자연스러운 사각형 형태의 노이즈 아티팩트가 발생하여 감점되었습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 스케치의 구도, 인물의 역동적인 자세, 그리고 참조 배경을 시각적 오류나 아티팩트 없이 가장 충실하게 구현했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 원본 배경을 온전히 유지하면서 인물의 포즈, 의상, 자전거의 디테일을 스케치와 지시문에 맞게 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "인물과 구도는 지시문을 잘 따랐으나, 손대지 않아야 할 배경 하늘 부분에 사각형 형태의 노이즈(아티팩트)가 발생하여 감점되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지정된 원본 배경을 온전히 유지하면서 인물의 포즈, 의상, 자전거의 디테일을 스케치와 지시문에 맞게 정확히 구현했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "인물과 구도는 지시문을 잘 따랐으나, 손대지 않아야 할 배경 하늘 부분에 사각형 형태의 노이즈(아티팩트)가 발생하여 감점되었습니다."
     }
    ]
   },
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S9sh1__bgfirst_bg.png",
   "bg_asset_id": "d0799ea3-ac1b-4830-b134-895794717425",
   "bg_record_key": "S9sh1::bgfirst_bg",
   "chain_winner": true,
   "authority": "groupbg",
   "group_key": "coastal_road",
   "groupbg_asset_id": "20148f1f-f264-4442-8f6d-57b8ec8f2c2e"
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S10sh5::bgfirst_bg": {
  "input_fingerprint": "06db5e40c6bc0da9",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 정체불명의 남자의 손가락이 종이 전단지를 문틈 사이로 밀어 넣은 채 멈춘 클로즈업.\n\nLOCATION (lock): On the rooftop landing directly outside the rooftop apartment's metal entrance door 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: End the upward tilt at hand height beside the iron door frame, then hold completely still on the finger pinning the leaflet into the narrow gap. The finger and paper occupy less than the central third, with the door edge remaining visible as a scale reference and the man's face and body withheld.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: narrow iron-door gap in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: iron entrance door and frame (Closed until the lock releases) — The door edge and frame face the camera obliquely, exposing the gap between them; used as Forms the narrow vertical frame around the finger and leaflet; advertising leaflet (Pressed into the door gap) — A partial printed face of the paper is visible to camera; its specific content is not established; used as Isolated focal detail held inside the door gap.\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 nighttime ambient light isolates the tactile hand action with restrained color and solemn 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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 정체불명의 남자의 손가락이 종이 전단지를 문틈 사이로 밀어 넣은 채 멈춘 클로즈업.\n\nLOCATION (lock): On the rooftop landing directly outside the rooftop apartment's metal entrance door 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: End the upward tilt at hand height beside the iron door frame, then hold completely still on the finger pinning the leaflet into the narrow gap. The finger and paper occupy less than the central third, with the door edge remaining visible as a scale reference and the man's face and body withheld.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: narrow iron-door gap in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: iron entrance door and frame (Closed until the lock releases) — The door edge and frame face the camera obliquely, exposing the gap between them; used as Forms the narrow vertical frame around the finger and leaflet; advertising leaflet (Pressed into the door gap) — A partial printed face of the paper is visible to camera; its specific content is not established; used as Isolated focal detail held inside the door gap.\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 nighttime ambient light isolates the tactile hand action with restrained color and solemn 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S10sh5__bgfirst_bg.png",
  "asset_id": "fbdde791-1012-4866-bbed-56bd45bb8532",
  "input_asset_ids": [
   "cbdeae18-7140-42c3-8e8d-6577914f02ec",
   "0eaa28dc-0f9a-484d-926f-f719496746f7",
   "43e496a8-bffd-428d-bf77-b87551da16f1"
  ]
 },
 "S10sh5": {
  "input_fingerprint": "71407c1cb080cda9",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 정체불명의 남자의 손가락이 종이 전단지를 문틈 사이로 밀어 넣은 채 멈춘 클로즈업.\n\nLOCATION (lock): On the rooftop landing directly outside the rooftop apartment's metal entrance door at night. 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: End the upward tilt at hand height beside the iron door frame, then hold completely still on the finger pinning the leaflet into the narrow gap. The finger and paper occupy less than the central third, with the door edge remaining visible as a scale reference and the man's face and body withheld.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: narrow iron-door gap in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: iron entrance door and frame (Closed until the lock releases) — The door edge and frame face the camera obliquely, exposing the gap between them; used as Forms the narrow vertical frame around the finger and leaflet; advertising leaflet (Pressed into the door gap) — A partial printed face of the paper is visible to camera; its specific content is not established; used as Isolated focal detail held inside the door gap.\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 nighttime ambient light isolates the tactile hand action with restrained color and solemn 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The advertising flyer has been removed from the door gap and is being used to force the lock.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정체불명의 남자 (한국인 남성, 성인, 검은 머리, 평범한 남성 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 정체불명의 남자의 손가락이 종이 전단지를 문틈 사이로 밀어 넣은 채 멈춘 클로즈업.\n\nLOCATION (lock): On the rooftop landing directly outside the rooftop apartment's metal entrance door at night. 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: End the upward tilt at hand height beside the iron door frame, then hold completely still on the finger pinning the leaflet into the narrow gap. The finger and paper occupy less than the central third, with the door edge remaining visible as a scale reference and the man's face and body withheld.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: narrow iron-door gap in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: iron entrance door and frame (Closed until the lock releases) — The door edge and frame face the camera obliquely, exposing the gap between them; used as Forms the narrow vertical frame around the finger and leaflet; advertising leaflet (Pressed into the door gap) — A partial printed face of the paper is visible to camera; its specific content is not established; used as Isolated focal detail held inside the door gap.\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 nighttime ambient light isolates the tactile hand action with restrained color and solemn 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The advertising flyer has been removed from the door gap and is being used to force the lock.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정체불명의 남자 (한국인 남성, 성인, 검은 머리, 평범한 남성 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 정체불명의 남자의 손가락이 종이 전단지를 문틈 사이로 밀어 넣은 채 멈춘 클로즈업.\n\nLOCATION (lock): On the rooftop landing directly outside the rooftop apartment's metal entrance door at night. 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: End the upward tilt at hand height beside the iron door frame, then hold completely still on the finger pinning the leaflet into the narrow gap. The finger and paper occupy less than the central third, with the door edge remaining visible as a scale reference and the man's face and body withheld.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: narrow iron-door gap in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: iron entrance door and frame (Closed until the lock releases) — The door edge and frame face the camera obliquely, exposing the gap between them; used as Forms the narrow vertical frame around the finger and leaflet; advertising leaflet (Pressed into the door gap) — A partial printed face of the paper is visible to camera; its specific content is not established; used as Isolated focal detail held inside the door gap.\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 nighttime ambient light isolates the tactile hand action with restrained color and solemn 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The advertising flyer has been removed from the door gap and is being used to force the lock.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정체불명의 남자 (한국인 남성, 성인, 검은 머리, 평범한 남성 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S10sh5__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S10sh5.png"
    },
    {
     "label": "CHARACTER REFERENCE — 정체불명의 남자: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:418704>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L03B03.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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/background_chain/seed_bg_rooftop_villa_sel.png"
    },
    {
     "label": "CHARACTER REFERENCE — 정체불명의 남자: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:418704>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지정된 초록색 철문의 질감과 색상을 정확히 반영했으며, 문틈으로 전단지를 밀어 넣는 손가락의 클로즈업 앵글과 야간 조명 등 프레이밍 요구사항을 훌륭하게 충족합니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "클로즈업 프레이밍과 동작은 대체로 맞으나, 레퍼런스에 명시된 초록색 철문 대신 회색 문을 생성하여 구조물 외형 조건(Structure Look Authority)을 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지정된 초록색 철문의 질감과 색상을 정확히 반영했으며, 문틈으로 전단지를 밀어 넣는 손가락의 클로즈업 앵글과 야간 조명 등 프레이밍 요구사항을 훌륭하게 충족합니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "클로즈업 프레이밍과 동작은 대체로 맞으나, 레퍼런스에 명시된 초록색 철문 대신 회색 문을 생성하여 구조물 외형 조건(Structure Look Authority)을 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시문에 명시된 좁은 문틈과 손가락 클로즈업 구도를 정확히 따랐으며, 레퍼런스의 녹색 철문 질감과 야간 옥상 배경을 충실히 반영했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "문틈에 전단지를 밀어넣는 동작의 디테일이 부족하며, 구조물 레퍼런스의 녹색 철문 대신 지정되지 않은 회색 문이 등장해 감점되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시문에 명시된 좁은 문틈과 손가락 클로즈업 구도를 정확히 따랐으며, 레퍼런스의 녹색 철문 질감과 야간 옥상 배경을 충실히 반영했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "문틈에 전단지를 밀어넣는 동작의 디테일이 부족하며, 구조물 레퍼런스의 녹색 철문 대신 지정되지 않은 회색 문이 등장해 감점되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 8,
     "B": 15
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 8,
   "B": 15
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 8,
    "verdict_ko": "지정된 초록색 철문의 질감과 색상을 정확히 반영했으며, 문틈으로 전단지를 밀어 넣는 손가락의 클로즈업 앵글과 야간 조명 등 프레이밍 요구사항을 훌륭하게 충족합니다."
   },
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "클로즈업 프레이밍과 동작은 대체로 맞으나, 레퍼런스에 명시된 초록색 철문 대신 회색 문을 생성하여 구조물 외형 조건(Structure Look Authority)을 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L03B03.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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/background_chain/seed_bg_rooftop_villa_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 정체불명의 남자: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:418704>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "전단지 표면에 글자가 인쇄되어 있어 '텍스트 금지' 지침을 위반했습니다.",
     "fix_en": "Remove all legible text and characters from the surface of the flyer."
    },
    {
     "issue_ko": "문틈 오른쪽의 문틀이 문과 동일한 녹색 철제로 표현되어, 회색/흰색 프레임인 구조물 레퍼런스와 일치하지 않습니다.",
     "fix_en": "Change the material and color of the door frame on the right side of the gap to grey concrete or painted wall to match the structure reference."
    }
   ]
  },
  "composition_critique": {
   "issues": []
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all legible text and characters from the surface of the flyer.\n- Change the material and color of the door frame on the right side of the gap to grey concrete or painted wall to match the structure 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": "지정된 클로즈업 프레이밍을 정확히 구현했으며, 요구사항대로 인쇄된 문구가 노출된 전단지와 철제 문틀의 재질을 충실히 반영했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "프레이밍은 적절하나, 인쇄된 면이 보여야 하는 전단지가 백지로 렌더링되었고 문틀의 재질이 철제가 아닌 콘크리트로 잘못 묘사되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 클로즈업 프레이밍을 정확히 구현했으며, 요구사항대로 인쇄된 문구가 노출된 전단지와 철제 문틀의 재질을 충실히 반영했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "프레이밍은 적절하나, 인쇄된 면이 보여야 하는 전단지가 백지로 렌더링되었고 문틀의 재질이 철제가 아닌 콘크리트로 잘못 묘사되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "전단지의 인쇄된 면이 보이도록 한 세부 지시사항을 정확히 구현했으며, 프레이밍과 조명도 완벽하게 일치합니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "구도와 배경 등은 훌륭하게 연출되었으나, 전단지가 백지로 처리되어 인쇄물이 보인다는 조건을 충족하지 못했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "전단지의 인쇄된 면이 보이도록 한 세부 지시사항을 정확히 구현했으며, 프레이밍과 조명도 완벽하게 일치합니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "구도와 배경 등은 훌륭하게 연출되었으나, 전단지가 백지로 처리되어 인쇄물이 보인다는 조건을 충족하지 못했습니다."
     }
    ]
   },
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S10sh5__bgfirst_bg.png",
   "bg_asset_id": "fbdde791-1012-4866-bbed-56bd45bb8532",
   "bg_record_key": "S10sh5::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": "L03B01",
    "B": "L03B02",
    "C": "L03B03"
   },
   "assigned": "L03B03",
   "choice": "Candidate C",
   "confident": true,
   "reason_ko": "샷 텍스트에 '문틈'이 명시되어 있어 옥탑방의 철문 공간이 필요합니다. 세 후보 모두 해당 철문을 포함한 동일한 옥상 공간을 보여주고 있어 서브공간 상의 명확한 차이가 없으므로, 규칙에 따라 현재 배정된 후보를 유지합니다.",
   "kept": "L03B03"
  }
 },
 "S10sh7::variants": {
  "author_fp": "f4f91f8298946368",
  "author": {
   "variants": [
    {
     "approach_ko": "긴 렌즈의 압축감으로 유리 반사면을 밀도 높게 붙잡아, 위에서 짓누르는 남자의 체격과 아래에 갇힌 민숙의 정지된 긴장을 강조한다.",
     "prompt_en": "Use a compressed telephoto feel, holding the reflective glass nearly flat against the frame so the tableau feels trapped within its surface. Keep the larger upper figure dense and weighty above the compressed lower figure, with recognizable human posture and clothing contours but no resolved faces. Let soft nighttime ambience skim the glass at an oblique angle, preserving restrained reflections, deep desaturated blacks, and subtle material texture without revealing the room directly."
    },
    {
     "approach_ko": "완만한 광각의 사선 원근으로 창유리의 물리적 깊이와 불안정한 반사를 살려, 공격 장면을 차갑고 관찰적인 정지 이미지로 만든다.",
     "prompt_en": "Use a restrained wide-angle perspective from chest height, emphasizing the angled geometry and physical depth of the reflective pane while maintaining the prescribed medium scale and placements. Compose the upper silhouette as a looming diagonal pressure over the lower figure, with slight perspective tension across the glass rather than optical compression. Shape the low-key night light as a faint lateral gradient across the pane, allowing small shifts in reflectivity and surface wear to make the frozen human forms feel disturbingly real."
    }
   ]
  },
  "reused": false
 },
 "S10sh7": {
  "input_fingerprint": "6d416402a1d0aa37",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 옥탑방 유리창 안쪽으로 민숙(강민숙)의 위를 덮친 정체불명의 남자의 거대한 실루엣이 비친 정지 상태.\n\nLOCATION (lock): Outside the rooftop apartment window at night, looking at silhouettes visible through 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: Stop the lateral track close to the window at chest height and hold on its angled reflective plane as a static tableau. Within the reflection, 정체불명의 남자 forms the larger upper silhouette bent over 강민숙, while her compressed figure remains beneath him in the lower portion of the window; neither face is resolved.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 정체불명의 남자 reflected silhouette in the upper-center of the frame, midground; 강민숙 reflected figure in the lower-center of the frame, midground; window reflective plane in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: rooftop-room window (Intact and reflecting the attack silhouettes) — Its angled surface reflects the oversized silhouette of 정체불명의 남자 over 강민숙 inside the room; used as The sole visual plane carrying the reflected attack tableau.\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 nighttime ambient light reduces both reflected figures to oppressive, desaturated silhouettes with controlled 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop-room door has been forced open, and Minsuk still bears her small circular wrist mark as the intruder attacks and bites her 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: 정체불명의 남자 (한국인 남성, 성인, 검은 머리, 평범한 남성 얼굴); 강민숙 (한국인 여성, 40대 초반, 자연스러운 검은 머리, 부드러운 중년 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 옥탑방 유리창 안쪽으로 민숙(강민숙)의 위를 덮친 정체불명의 남자의 거대한 실루엣이 비친 정지 상태.\n\nLOCATION (lock): Outside the rooftop apartment window at night, looking at silhouettes visible through 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: Stop the lateral track close to the window at chest height and hold on its angled reflective plane as a static tableau. Within the reflection, 정체불명의 남자 forms the larger upper silhouette bent over 강민숙, while her compressed figure remains beneath him in the lower portion of the window; neither face is resolved.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 정체불명의 남자 reflected silhouette in the upper-center of the frame, midground; 강민숙 reflected figure in the lower-center of the frame, midground; window reflective plane in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: rooftop-room window (Intact and reflecting the attack silhouettes) — Its angled surface reflects the oversized silhouette of 정체불명의 남자 over 강민숙 inside the room; used as The sole visual plane carrying the reflected attack tableau.\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 nighttime ambient light reduces both reflected figures to oppressive, desaturated silhouettes with controlled 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop-room door has been forced open, and Minsuk still bears her small circular wrist mark as the intruder attacks and bites her 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: 정체불명의 남자 (한국인 남성, 성인, 검은 머리, 평범한 남성 얼굴); 강민숙 (한국인 여성, 40대 초반, 자연스러운 검은 머리, 부드러운 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a compressed telephoto feel, holding the reflective glass nearly flat against the frame so the tableau feels trapped within its surface. Keep the larger upper figure dense and weighty above the compressed lower figure, with recognizable human posture and clothing contours but no resolved faces. Let soft nighttime ambience skim the glass at an oblique angle, preserving restrained reflections, deep desaturated blacks, and subtle material texture without revealing the room directly.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 옥탑방 유리창 안쪽으로 민숙(강민숙)의 위를 덮친 정체불명의 남자의 거대한 실루엣이 비친 정지 상태.\n\nLOCATION (lock): Outside the rooftop apartment window at night, looking at silhouettes visible through 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: Stop the lateral track close to the window at chest height and hold on its angled reflective plane as a static tableau. Within the reflection, 정체불명의 남자 forms the larger upper silhouette bent over 강민숙, while her compressed figure remains beneath him in the lower portion of the window; neither face is resolved.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 정체불명의 남자 reflected silhouette in the upper-center of the frame, midground; 강민숙 reflected figure in the lower-center of the frame, midground; window reflective plane in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: rooftop-room window (Intact and reflecting the attack silhouettes) — Its angled surface reflects the oversized silhouette of 정체불명의 남자 over 강민숙 inside the room; used as The sole visual plane carrying the reflected attack tableau.\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 nighttime ambient light reduces both reflected figures to oppressive, desaturated silhouettes with controlled 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop-room door has been forced open, and Minsuk still bears her small circular wrist mark as the intruder attacks and bites her 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: 정체불명의 남자 (한국인 남성, 성인, 검은 머리, 평범한 남성 얼굴); 강민숙 (한국인 여성, 40대 초반, 자연스러운 검은 머리, 부드러운 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained wide-angle perspective from chest height, emphasizing the angled geometry and physical depth of the reflective pane while maintaining the prescribed medium scale and placements. Compose the upper silhouette as a looming diagonal pressure over the lower figure, with slight perspective tension across the glass rather than optical compression. Shape the low-key night light as a faint lateral gradient across the pane, allowing small shifts in reflectivity and surface wear to make the frozen human forms feel disturbingly real.\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": 8,
   "A": 4
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 8,
    "verdict_ko": "기울어진 창문 표면에 반사된 억압적인 실루엣의 대비를 정확한 앵글로 담아냈으며, 레퍼런스의 배경 디테일(초록색 문과 문틈의 종이)을 완벽하게 이어받았습니다."
   },
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "기울어진 반사면이라는 명확한 프레이밍 지시를 어기고 창문 너머 내부를 투시하듯 묘사했으며, 남자 몸에 출처를 알 수 없는 흰색 사각형 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S10sh5_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 정체불명의 남자: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:418704>"
   },
   {
    "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:1364281>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "프롬프트에서 창문 유리가 '온전한 상태(Intact)'여야 한다고 명시했지만, 생성된 이미지의 유리는 왼쪽이 크게 깨져 있습니다.",
     "fix_en": "Render the window glass completely intact, removing the large diagonal break and missing piece on the left side."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Render both figures as images reflected on the angled glass plane rather than as silhouettes seen directly through it, preserving their current scale and placement.",
     "issue_ko": "두 인물이 기울어진 유리 표면의 반사상이라기보다 창 뒤 실내에 직접 보이는 실루엣처럼 배치되어 반사면 중심의 깊이 관계가 어긋난다."
    },
    {
     "fix_en": "Lower and angle the man's head until his mouth contacts Minsuk's shoulder while keeping him looming above her in the upper-center.",
     "issue_ko": "남자의 머리가 민숙의 어깨에 닿지 않고 위쪽에 떨어져 있어, 어깨를 물며 덮친 동작의 접촉 관계가 구현되지 않았다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Render the window glass completely intact, removing the large diagonal break and missing piece on the left side.\n- Render both figures as images reflected on the angled glass plane rather than as silhouettes seen directly through it, preserving their current scale and placement.\n- Lower and angle the man's head until his mouth contacts Minsuk's shoulder while keeping him looming above her in the upper-center.\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": "창문에 비친 억압적인 실루엣이라는 핵심 지시를 정확히 구현했으며, 두 인물의 얼굴이 드러나지 않게 완벽히 처리했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "인물들의 얼굴 이목구비와 옷의 색상 등 디테일이 너무 선명하게 보여 '얼굴이 보이지 않는 실루엣'이라는 명시적 조건을 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "창문에 비친 억압적인 실루엣이라는 핵심 지시를 정확히 구현했으며, 두 인물의 얼굴이 드러나지 않게 완벽히 처리했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "인물들의 얼굴 이목구비와 옷의 색상 등 디테일이 너무 선명하게 보여 '얼굴이 보이지 않는 실루엣'이라는 명시적 조건을 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지시된 프레이밍을 정확히 따르며, 인물들의 얼굴이 보이지 않는 억압적인 실루엣(거대한 실루엣이 비친 정지 상태)을 성공적으로 구현했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "프레이밍과 배경 요소는 일치하나, 인물들이 실루엣이 아닌 구체적인 디테일과 얼굴이 보이는 상태로 렌더링되어 지시사항을 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "지시된 프레이밍을 정확히 따르며, 인물들의 얼굴이 보이지 않는 억압적인 실루엣(거대한 실루엣이 비친 정지 상태)을 성공적으로 구현했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "프레이밍과 배경 요소는 일치하나, 인물들이 실루엣이 아닌 구체적인 디테일과 얼굴이 보이는 상태로 렌더링되어 지시사항을 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 18,
     "B": 10
    },
    "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": "S10sh5"
  },
  "lane_policy": "ab_select_bypass:prev"
 },
 "S11sh4::bgfirst_bg": {
  "input_fingerprint": "1e663f63f35dd609",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 계단 중간에 선 채 코끝을 찡그린 수리영의 얼굴 클로즈업.\n\nLOCATION (lock): Midway up the exterior staircase connecting the residential building's yard to 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: Hold statically one step above 수리영 at face height, looking back toward her in a subtly high close angle. Her halted face occupies most of the frame with her wrinkled nose as the focal point, while a small portion of the stair flight remains behind her to preserve the interrupted ascent.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: stair flight (Visible beneath and behind her) — The lower steps recede behind 수리영 while the camera remains one step above her; used as Provides the receding spatial line behind her face and confirms that she has stopped midway upward.\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 nighttime ambient light holds her expression in restrained, desaturated contrast without introducing an unsupported 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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 계단 중간에 선 채 코끝을 찡그린 수리영의 얼굴 클로즈업.\n\nLOCATION (lock): Midway up the exterior staircase connecting the residential building's yard to 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: Hold statically one step above 수리영 at face height, looking back toward her in a subtly high close angle. Her halted face occupies most of the frame with her wrinkled nose as the focal point, while a small portion of the stair flight remains behind her to preserve the interrupted ascent.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: stair flight (Visible beneath and behind her) — The lower steps recede behind 수리영 while the camera remains one step above her; used as Provides the receding spatial line behind her face and confirms that she has stopped midway upward.\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 nighttime ambient light holds her expression in restrained, desaturated contrast without introducing an unsupported 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S11sh4__bgfirst_bg.png",
  "asset_id": "0c8c555f-2daf-431d-bb29-f36d60baee32",
  "input_asset_ids": [
   "7480506e-f0cc-41bb-a0fc-1629ec66105d",
   "cd29adc3-073c-4b80-8d6b-b2ef81f67e09"
  ]
 },
 "S11sh4": {
  "input_fingerprint": "be2b6fb79f933739",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 계단 중간에 선 채 코끝을 찡그린 수리영의 얼굴 클로즈업.\n\nLOCATION (lock): Midway up the exterior staircase connecting the residential building's yard to 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: Hold statically one step above 수리영 at face height, looking back toward her in a subtly high close angle. Her halted face occupies most of the frame with her wrinkled nose as the focal point, while a small portion of the stair flight remains behind her to preserve the interrupted ascent.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: stair flight (Visible beneath and behind her) — The lower steps recede behind 수리영 while the camera remains one step above her; used as Provides the receding spatial line behind her face and confirms that she has stopped midway upward.\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 nighttime ambient light holds her expression in restrained, desaturated contrast without introducing an unsupported 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young's bicycle has been left roughly parked in the rear corner of the courtyard while she climbs toward the rooftop room.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 계단 중간에 선 채 코끝을 찡그린 수리영의 얼굴 클로즈업.\n\nLOCATION (lock): Midway up the exterior staircase connecting the residential building's yard to 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: Hold statically one step above 수리영 at face height, looking back toward her in a subtly high close angle. Her halted face occupies most of the frame with her wrinkled nose as the focal point, while a small portion of the stair flight remains behind her to preserve the interrupted ascent.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: stair flight (Visible beneath and behind her) — The lower steps recede behind 수리영 while the camera remains one step above her; used as Provides the receding spatial line behind her face and confirms that she has stopped midway upward.\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 nighttime ambient light holds her expression in restrained, desaturated contrast without introducing an unsupported 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young's bicycle has been left roughly parked in the rear corner of the courtyard while she climbs toward the rooftop room.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 계단 중간에 선 채 코끝을 찡그린 수리영의 얼굴 클로즈업.\n\nLOCATION (lock): Midway up the exterior staircase connecting the residential building's yard to 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: Hold statically one step above 수리영 at face height, looking back toward her in a subtly high close angle. Her halted face occupies most of the frame with her wrinkled nose as the focal point, while a small portion of the stair flight remains behind her to preserve the interrupted ascent.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: stair flight (Visible beneath and behind her) — The lower steps recede behind 수리영 while the camera remains one step above her; used as Provides the receding spatial line behind her face and confirms that she has stopped midway upward.\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 nighttime ambient light holds her expression in restrained, desaturated contrast without introducing an unsupported 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young's bicycle has been left roughly parked in the rear corner of the courtyard while she climbs toward the rooftop room.\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 초반, 긴 검은 머리, 타원형의 젊은 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S11sh4__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S11sh4.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L11B01.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 클로즈업 프레이밍과 코를 찡그린 표정을 매우 잘 구현했으나, 마당에 있어야 할 자전거가 옥상에 잘못 배치되었습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "표정은 반영되었으나 프레임이 너무 넓어 '얼굴이 프레임의 대부분을 차지해야 한다'는 지시를 충족하지 못했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 클로즈업 프레이밍과 코를 찡그린 표정을 매우 잘 구현했으나, 마당에 있어야 할 자전거가 옥상에 잘못 배치되었습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "표정은 반영되었으나 프레임이 너무 넓어 '얼굴이 프레임의 대부분을 차지해야 한다'는 지시를 충족하지 못했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 약간 높은 앵글의 클로즈업 프레이밍과 코를 찡그린 표정을 정확히 구현했으며, 불필요한 사물을 추가하지 않아 공간적 논리를 잘 유지했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "표정과 프레이밍은 훌륭하지만, 설정상 마당에 있어야 할 자전거를 아직 도착하지 않은 옥상 공간에 잘못 배치하여 서사적 오류를 범했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 약간 높은 앵글의 클로즈업 프레이밍과 코를 찡그린 표정을 정확히 구현했으며, 불필요한 사물을 추가하지 않아 공간적 논리를 잘 유지했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "표정과 프레이밍은 훌륭하지만, 설정상 마당에 있어야 할 자전거를 아직 도착하지 않은 옥상 공간에 잘못 배치하여 서사적 오류를 범했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 12,
     "B": 12
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": false,
    "policy": 1
   }
  },
  "totals": {
   "A": 12,
   "B": 12
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "지정된 클로즈업 프레이밍과 코를 찡그린 표정을 매우 잘 구현했으나, 마당에 있어야 할 자전거가 옥상에 잘못 배치되었습니다."
   },
   {
    "label": "B",
    "score": 5,
    "verdict_ko": "표정은 반영되었으나 프레임이 너무 넓어 '얼굴이 프레임의 대부분을 차지해야 한다'는 지시를 충족하지 못했습니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L11B01.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "원본 배경 이미지에 있던 옥탑방의 열린 문(오른쪽으로 열려 있는 상태)이 사라졌습니다.",
     "fix_en": "Restore the open door on the rooftop structure exactly as it appears in the background reference image."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Push in to a true close-up so Suri-young’s face occupies most of the frame, cropping out most of her torso while retaining only a small visible portion of the receding stair flight behind her.",
     "issue_ko": "현재는 상반신과 넓은 배경이 많이 보여 미디엄 클로즈업에 가깝고, 얼굴이 프레임 대부분을 차지해야 하는 지정 클로즈업보다 너무 작다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Restore the open door on the rooftop structure exactly as it appears in the background reference image.\n- Push in to a true close-up so Suri-young’s face occupies most of the frame, cropping out most of her torso while retaining only a small visible portion of the receding stair flight behind her.\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 원본 배경을 그대로 유지하면서 인물의 스케치 위치, 찡그린 표정, 그리고 배경의 자전거까지 매우 정확하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "클로즈업 앵글은 좋으나 '절대 변경하지 말 것'을 지시한 배경 구조(건물, 문, 조명 등)를 완전히 변형시키는 치명적인 오류를 범했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 원본 배경을 그대로 유지하면서 인물의 스케치 위치, 찡그린 표정, 그리고 배경의 자전거까지 매우 정확하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "클로즈업 앵글은 좋으나 '절대 변경하지 말 것'을 지시한 배경 구조(건물, 문, 조명 등)를 완전히 변형시키는 치명적인 오류를 범했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지정된 원본 배경을 훼손하지 않고 정확한 스케일과 위치에 인물을 합성했으며, 요구된 표정과 자전거 디테일도 잘 살렸습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "인물의 클로즈업 비율은 스케치와 유사하나, 고정되어야 할 배경 이미지를 임의로 확대 및 크롭하여 카메라 원근감을 심각하게 훼손했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 원본 배경을 훼손하지 않고 정확한 스케일과 위치에 인물을 합성했으며, 요구된 표정과 자전거 디테일도 잘 살렸습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "인물의 클로즈업 비율은 스케치와 유사하나, 고정되어야 할 배경 이미지를 임의로 확대 및 크롭하여 카메라 원근감을 심각하게 훼손했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 16,
     "B": 7
    },
    "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S11sh4__bgfirst_bg.png",
   "bg_asset_id": "0c8c555f-2daf-431d-bb29-f36d60baee32",
   "bg_record_key": "S11sh4::bgfirst_bg",
   "chain_winner": true,
   "authority": "plate"
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S11sh6::variants": {
  "author_fp": "4362656dd70e4465",
  "author": {
   "variants": [
    {
     "approach_ko": "정면의 건축적 대칭과 깊은 초점으로 반쯤 열린 철문 속 암흑을 냉정한 중심 공백으로 강조한다.",
     "prompt_en": "Hold a disciplined eye-level wide view with near-symmetrical architectural geometry, keeping the half-open iron door small and centered within the surrounding entrance. Use deep focus and restrained nighttime exposure so the worn metal, threshold, and adjacent surfaces remain physically legible while the interior gap falls into a dense, clearly bounded black void. Let subtle tonal separation rather than dramatic highlights pull the eye inward."
    },
    {
     "approach_ko": "넓은 전경의 바닥 면과 사선형 구조를 활용해 시선을 중앙의 열린 틈으로 끌어가는 긴장감 있는 원근 구도다.",
     "prompt_en": "Build the wide frame around strong foreground-to-background perspective, allowing the rooftop landing and entrance architecture to form quiet converging lines toward the middle-center doorway. Keep the camera at the fixed eye line and the door under two-fifths of the image, with the oblique metal slab creating a sharp layered edge against the recessed darkness. Use soft, low-key night ambience with gentle sideward falloff across the metal and surrounding surfaces, preserving a deep tonal plunge beyond the threshold."
    }
   ]
  },
  "reused": false
 },
 "S11sh6": {
  "input_fingerprint": "d5a846dcc0152a72",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 수리영의 시점, 반쯤 열린 채 내부의 캄캄한 어둠을 드러낸 옥탑방 현관 철문.\n\nLOCATION (lock): At the top of the exterior rooftop stairs, facing the half-open metal entrance door and the darkness 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: From 수리영's exact eye line at rooftop height, the camera faces the rooftop-room entrance directly as the restrained dolly-in reaches its endpoint. The half-open iron door sits near the center and occupies less than two-fifths of the frame, with the visible interior gap held as the deepest point while the surrounding entrance provides scale.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: half-open rooftop-room entrance in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: rooftop-room iron entrance door (half-open) — Its exterior-facing side is visible at an oblique opening angle, exposing part of the entrance beyond; used as Central architectural frame around the dark interior gap; rooftop-room interior (dark); used as Deep focal void visible through the doorway.\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 nighttime ambience keeps the palette restrained while preserving the doorway and its dark interior as distinct tonal planes.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The forced rooftop-room entrance remains half open, exposing the dark interior; Suri-young's bicycle stays parked below in the courtyard.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 수리영의 시점, 반쯤 열린 채 내부의 캄캄한 어둠을 드러낸 옥탑방 현관 철문.\n\nLOCATION (lock): At the top of the exterior rooftop stairs, facing the half-open metal entrance door and the darkness 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: From 수리영's exact eye line at rooftop height, the camera faces the rooftop-room entrance directly as the restrained dolly-in reaches its endpoint. The half-open iron door sits near the center and occupies less than two-fifths of the frame, with the visible interior gap held as the deepest point while the surrounding entrance provides scale.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: half-open rooftop-room entrance in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: rooftop-room iron entrance door (half-open) — Its exterior-facing side is visible at an oblique opening angle, exposing part of the entrance beyond; used as Central architectural frame around the dark interior gap; rooftop-room interior (dark); used as Deep focal void visible through the doorway.\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 nighttime ambience keeps the palette restrained while preserving the doorway and its dark interior as distinct tonal planes.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The forced rooftop-room entrance remains half open, exposing the dark interior; Suri-young's bicycle stays parked below in the courtyard.\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):\nHold a disciplined eye-level wide view with near-symmetrical architectural geometry, keeping the half-open iron door small and centered within the surrounding entrance. Use deep focus and restrained nighttime exposure so the worn metal, threshold, and adjacent surfaces remain physically legible while the interior gap falls into a dense, clearly bounded black void. Let subtle tonal separation rather than dramatic highlights pull the eye inward.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 수리영의 시점, 반쯤 열린 채 내부의 캄캄한 어둠을 드러낸 옥탑방 현관 철문.\n\nLOCATION (lock): At the top of the exterior rooftop stairs, facing the half-open metal entrance door and the darkness 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: From 수리영's exact eye line at rooftop height, the camera faces the rooftop-room entrance directly as the restrained dolly-in reaches its endpoint. The half-open iron door sits near the center and occupies less than two-fifths of the frame, with the visible interior gap held as the deepest point while the surrounding entrance provides scale.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: half-open rooftop-room entrance in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: rooftop-room iron entrance door (half-open) — Its exterior-facing side is visible at an oblique opening angle, exposing part of the entrance beyond; used as Central architectural frame around the dark interior gap; rooftop-room interior (dark); used as Deep focal void visible through the doorway.\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 nighttime ambience keeps the palette restrained while preserving the doorway and its dark interior as distinct tonal planes.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The forced rooftop-room entrance remains half open, exposing the dark interior; Suri-young's bicycle stays parked below in the courtyard.\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):\nBuild the wide frame around strong foreground-to-background perspective, allowing the rooftop landing and entrance architecture to form quiet converging lines toward the middle-center doorway. Keep the camera at the fixed eye line and the door under two-fifths of the image, with the oblique metal slab creating a sharp layered edge against the recessed darkness. Use soft, low-key night ambience with gentle sideward falloff across the metal and surrounding surfaces, preserving a deep tonal plunge beyond the threshold.\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": 8,
   "A": 6
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 8,
    "verdict_ko": "요청된 1인칭 시점 구도를 정확히 구현했으며, 레퍼런스의 벽면 조명과 자전거 등 장소 디테일을 완벽하게 유지했습니다."
   },
   {
    "label": "A",
    "score": 6,
    "verdict_ko": "시점과 구도는 좋으나, 레퍼런스에 명확히 존재하던 옥탑방 벽면 조명이 누락되어 장소 일관성을 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S11sh4_sel.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "현관문이 철제 금속 판이 아닌 세로 홈이 패인 나무 판자 문처럼 묘사되었습니다.",
     "fix_en": "Change the door material to a solid, flat iron/metal slab."
    },
    {
     "issue_ko": "자전거가 아래 마당에 주차되어 있어야 한다는 지시와 달리 옥상 위에 그대로 있습니다.",
     "fix_en": "Remove the bicycle from the rooftop."
    }
   ]
  },
  "composition_critique": {
   "issues": []
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the door material to a solid, flat iron/metal slab.\n- Remove the bicycle from the rooftop.\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": "지문이 요구한 평평한 철문(metal slab)의 질감을 잘 살렸으며, 자전거가 아래 마당에 있다는 설정에 맞게 옥상에서 자전거를 배제하여 지시사항을 정확히 따랐습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "시점과 샷 크기는 적절하나, 철문이 나무 패널처럼 묘사되었고 아래 마당에 있어야 할 자전거가 옥상에 그대로 남아있어 설정과 충돌합니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지문이 요구한 평평한 철문(metal slab)의 질감을 잘 살렸으며, 자전거가 아래 마당에 있다는 설정에 맞게 옥상에서 자전거를 배제하여 지시사항을 정확히 따랐습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "시점과 샷 크기는 적절하나, 철문이 나무 패널처럼 묘사되었고 아래 마당에 있어야 할 자전거가 옥상에 그대로 남아있어 설정과 충돌합니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "철제 문의 재질과 프레이밍이 지시사항에 부합하며, 자전거가 마당에 있다는 텍스트 설정에 따라 옥상에서 자전거를 올바르게 배제했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "텍스트의 설정과 달리 레퍼런스처럼 자전거가 여전히 옥상에 남아 있으며, 현관문이 평평한 철제(metal slab)라기보다 패널 형태에 가깝습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "철제 문의 재질과 프레이밍이 지시사항에 부합하며, 자전거가 마당에 있다는 텍스트 설정에 따라 옥상에서 자전거를 올바르게 배제했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "텍스트의 설정과 달리 레퍼런스처럼 자전거가 여전히 옥상에 남아 있으며, 현관문이 평평한 철제(metal slab)라기보다 패널 형태에 가깝습니다."
     }
    ]
   },
   "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": "prev만 (배경 전용·공유 계획)",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S11sh4"
  },
  "lane_policy": "ab_select_bypass:bg_only:share_plan_prev_bgonly"
 },
 "S12sh7::variants": {
  "author_fp": "a78bda96b6f65625",
  "author": {
   "variants": [
    {
     "approach_ko": "깊은 초점과 넓은 원근감으로 피 묻은 바닥의 유도선과 시신의 고립을 냉정하게 드러낸다.",
     "prompt_en": "Use a restrained wide-lens feel with deep focus and clear spatial separation from the parted curtain in the left foreground, across the glossy blood-marked floor, to the entire seated body in the right-center midground and the bed beyond. Keep the slightly high three-quarter viewpoint clinically stable, letting floor lines, pooled blood, and footprints pull the eye toward the lowered hand and crumpled photograph. Shape the low-key illumination with soft directional falloff, preserving tactile fabric, skin, and liquid detail while the pale corpse stands in stark, desaturated isolation."
    },
    {
     "approach_ko": "압축된 원근과 얕은 초점 전이로 커튼 너머 시신을 은밀하고 숨 막히는 발견처럼 포착한다.",
     "prompt_en": "Give the wide composition a compressed, observational lens feel, with the soft curtain edge looming as an intimate foreground veil while the entire seated body remains sharply resolved behind it and the bed settles close in depth. Hold the static slightly elevated three-quarter angle so her collapsed posture, torn shoulder and collarbone, closed hand, wrist mark, and absolute stillness dominate without changing her right-center placement. Use dense shadow and narrowly controlled low-key light across her pallor, allowing the blood to catch sparse wet highlights as the only forceful color in the room."
    }
   ]
  },
  "reused": false
 },
 "S12sh7": {
  "input_fingerprint": "100ba586eed2bc73",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 젖혀진 커튼 뒤, 머리를 푹 숙인 채 방바닥에 앉아 미동도 없는 민숙(강민숙)의 창백한 시신.\n\nLOCATION (lock): Behind the window curtain in the daughter's bedroom of the rooftop apartment, above a blood-soaked floor. 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: Inside 수리영's room, the advancing camera settles into a static, slightly high three-quarter view at the end of the dolly, revealing 강민숙's entire seated body behind the displaced curtain. She occupies the right-center midground while the curtain edge frames the left foreground and the blood-marked floor leads toward her lowered hand, preparing the next downward move without changing the viewing axis.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: curtain (drawn aside); used as Parted foreground frame that completes the corpse reveal; room floor with blood and footprints (blood pooled and marked by bloody footprints); used as Ground plane leading the eye from the room toward the body; bed (present) — The side nearest 강민숙 and the wall above it are visible behind her; used as Spatial reference behind the seated body and for the later wall reveal.\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 interior illumination renders the corpse pale against a restrained, desaturated room, with the blood retained as the deepest selective red.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk's corpse remains seated motionless on the bedroom floor behind the curtain with her head dropped forward over her body, her shoulder and collarbone torn open amid pooled blood and bloody tracks. One lifeless hand remains closed around a crumpled photograph, and the circular mark remains on her wrist.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 젖혀진 커튼 뒤, 머리를 푹 숙인 채 방바닥에 앉아 미동도 없는 민숙(강민숙)의 창백한 시신.\n\nLOCATION (lock): Behind the window curtain in the daughter's bedroom of the rooftop apartment, above a blood-soaked floor. 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: Inside 수리영's room, the advancing camera settles into a static, slightly high three-quarter view at the end of the dolly, revealing 강민숙's entire seated body behind the displaced curtain. She occupies the right-center midground while the curtain edge frames the left foreground and the blood-marked floor leads toward her lowered hand, preparing the next downward move without changing the viewing axis.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: curtain (drawn aside); used as Parted foreground frame that completes the corpse reveal; room floor with blood and footprints (blood pooled and marked by bloody footprints); used as Ground plane leading the eye from the room toward the body; bed (present) — The side nearest 강민숙 and the wall above it are visible behind her; used as Spatial reference behind the seated body and for the later wall reveal.\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 interior illumination renders the corpse pale against a restrained, desaturated room, with the blood retained as the deepest selective red.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk's corpse remains seated motionless on the bedroom floor behind the curtain with her head dropped forward over her body, her shoulder and collarbone torn open amid pooled blood and bloody tracks. One lifeless hand remains closed around a crumpled photograph, and the circular mark remains on her wrist.\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 wide-lens feel with deep focus and clear spatial separation from the parted curtain in the left foreground, across the glossy blood-marked floor, to the entire seated body in the right-center midground and the bed beyond. Keep the slightly high three-quarter viewpoint clinically stable, letting floor lines, pooled blood, and footprints pull the eye toward the lowered hand and crumpled photograph. Shape the low-key illumination with soft directional falloff, preserving tactile fabric, skin, and liquid detail while the pale corpse stands in stark, desaturated isolation.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 젖혀진 커튼 뒤, 머리를 푹 숙인 채 방바닥에 앉아 미동도 없는 민숙(강민숙)의 창백한 시신.\n\nLOCATION (lock): Behind the window curtain in the daughter's bedroom of the rooftop apartment, above a blood-soaked floor. 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: Inside 수리영's room, the advancing camera settles into a static, slightly high three-quarter view at the end of the dolly, revealing 강민숙's entire seated body behind the displaced curtain. She occupies the right-center midground while the curtain edge frames the left foreground and the blood-marked floor leads toward her lowered hand, preparing the next downward move without changing the viewing axis.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: curtain (drawn aside); used as Parted foreground frame that completes the corpse reveal; room floor with blood and footprints (blood pooled and marked by bloody footprints); used as Ground plane leading the eye from the room toward the body; bed (present) — The side nearest 강민숙 and the wall above it are visible behind her; used as Spatial reference behind the seated body and for the later wall reveal.\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 interior illumination renders the corpse pale against a restrained, desaturated room, with the blood retained as the deepest selective red.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk's corpse remains seated motionless on the bedroom floor behind the curtain with her head dropped forward over her body, her shoulder and collarbone torn open amid pooled blood and bloody tracks. One lifeless hand remains closed around a crumpled photograph, and the circular mark remains on her wrist.\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 wide composition a compressed, observational lens feel, with the soft curtain edge looming as an intimate foreground veil while the entire seated body remains sharply resolved behind it and the bed settles close in depth. Hold the static slightly elevated three-quarter angle so her collapsed posture, torn shoulder and collarbone, closed hand, wrist mark, and absolute stillness dominate without changing her right-center placement. Use dense shadow and narrowly controlled low-key light across her pallor, allowing the blood to catch sparse wet highlights as the only forceful color in the room.\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": 8
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 3,
    "verdict_ko": "지정된 핏자국 대신 물자국이 있으며, 찢겨진 어깨 상처와 피가 묻은 바닥 등 핵심 상태가 누락되었습니다."
   },
   {
    "label": "B",
    "score": 8,
    "verdict_ko": "왼쪽 커튼 프레이밍, 핏자국, 구겨진 사진, 어깨의 상처 등 프롬프트의 요구사항을 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L04B07.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "오른쪽 손목에 있는 자국이 지시된 '원형 자국(circular mark)'이 아니라 단순한 붉은 선 또는 얼룩입니다.",
     "fix_en": "Change the red smudge on the right wrist into a clearly drawn circular mark."
    },
    {
     "issue_ko": "손이 구겨진 사진을 '꽉 쥐고(closed around)' 있지 않으며, 들고 있는 물건도 사진이 아니라 헝겊이나 휴지 조각처럼 보입니다.",
     "fix_en": "Make the hand tightly closed around a clearly recognizable crumpled photograph."
    },
    {
     "issue_ko": "바닥에 찍힌 핏빛 발자국들의 형태가 사람의 발 모양이 아니며, 발가락이 비정상적으로 크고 개수가 맞지 않습니다.",
     "fix_en": "Redraw the bloody footprints on the floor to match anatomically correct human foot proportions and toe counts."
    },
    {
     "issue_ko": "화면 왼쪽에 보이는 시신의 발가락 형태가 뭉개져 있어 정상적인 인체 해부학에 어긋납니다.",
     "fix_en": "Correct the toes on the visible foot to have normal human anatomy."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Shift the parted curtain farther left so Minsuk’s entire seated body, including both feet and all extremities, is unobstructed while preserving her right-center midground placement and the same camera axis.",
     "issue_ko": "왼쪽 전경의 커튼이 민숙의 다리와 발 일부를 가려, 계약에서 요구한 앉은 전신 공개가 완성되지 않았다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the red smudge on the right wrist into a clearly drawn circular mark.\n- Make the hand tightly closed around a clearly recognizable crumpled photograph.\n- Redraw the bloody footprints on the floor to match anatomically correct human foot proportions and toe counts.\n- Correct the toes on the visible foot to have normal human anatomy.\n- Shift the parted curtain farther left so Minsuk’s entire seated body, including both feet and all extremities, is unobstructed while preserving her right-center midground placement and the same camera axis.\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": "프롬프트에 명시된 손목의 뚜렷한 원형 자국을 완벽하게 구현하였으며, 찢어진 어깨와 구겨진 사진 등 세부 지시문도 충실히 반영되었습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "대부분의 지시사항과 구도를 잘 따랐으나, 손목에 있어야 할 원형 자국(circular mark)이 누락되어 감점되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "프롬프트에 명시된 손목의 뚜렷한 원형 자국을 완벽하게 구현하였으며, 찢어진 어깨와 구겨진 사진 등 세부 지시문도 충실히 반영되었습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "대부분의 지시사항과 구도를 잘 따랐으나, 손목에 있어야 할 원형 자국(circular mark)이 누락되어 감점되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "찢겨진 어깨에서 두 개의 오른팔이 뻗어나오는 치명적인 해부학적 오류가 있어 실격입니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 구도와 조명 분위기를 잘 살렸으며, 신체 구조의 오류 없이 시신과 소품의 위치를 정확히 묘사했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "찢겨진 어깨에서 두 개의 오른팔이 뻗어나오는 치명적인 해부학적 오류가 있어 실격입니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 구도와 조명 분위기를 잘 살렸으며, 신체 구조의 오류 없이 시신과 소품의 위치를 정확히 묘사했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 13,
     "B": 11
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": false,
    "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": "L04B01",
    "B": "L04B02",
    "C": "L04B03",
    "D": "L04B04",
    "E": "L04B05",
    "F": "L04B06",
    "G": "L04B07"
   },
   "assigned": "L04B06",
   "choice": "Candidate F",
   "confident": true,
   "reason_ko": "\"젖혀진 커튼 뒤\"와 딸의 방이라는 공간 설정에 부합하며, 기존 할당을 변경할 뚜렷한 이유가 없습니다.",
   "kept": "L04B06"
  }
 },
 "S12sh12::variants": {
  "author_fp": "3d1b7e7a0eff233a",
  "author": {
   "variants": [
    {
     "approach_ko": "얕은 초점으로 손가락 사이 사진의 구겨진 모서리에 시선을 고정하는 압축된 촬영.",
     "prompt_en": "Use a compressed, intimate lens feel with extremely selective focus placed on the creased photograph edge where it emerges between the fingers. Let the nearer hand contours and the forearm fall away softly, with the wrist already beyond focus; restrained low-key illumination catches only the paper ridges, pale knuckles, and slight skin texture against a nearly submerged floor plane."
    },
    {
     "approach_ko": "손과 바닥의 물리적 질감을 함께 읽히게 하는 낮고 자연스러운 원근의 냉정한 디테일 촬영.",
     "prompt_en": "Use a natural close-focus perspective from the locked hand-height position, rendering the closed hand, crushed paper, and thin surrounding floor plane with sober physical clarity while the wrist slips gently beyond the focal plane. Emphasize the diagonal weight of the lowered forearm and the rigid enclosure of the fingers around the photograph; soft directional falloff from the existing scene light reveals paper compression, skin folds, and floor texture without exposing the concealed image."
    }
   ]
  },
  "reused": false
 },
 "S12sh12": {
  "input_fingerprint": "5c83144973dd4759",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 축 늘어진 민숙(강민숙)의 죽은 손아귀 안에 구겨진 사진 한 장이 살짝 삐져나온 클로즈업.\n\nLOCATION (lock): On the floor behind the curtain in the daughter's bedroom, beside the seated body. 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 hand height beside 강민숙, the dolly-in ends on an isolated detail of her lowered hand and the crumpled photograph protruding between closed fingers. The hand crosses the lower center while the photograph remains below one-third of the frame, with a narrow strip of floor and forearm preserving realistic scale and the wrist waiting just beyond the focal plane.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: crumpled photograph (crumpled and partially concealed) — The camera sees part of its image-bearing face protruding from the closed hand; the depicted content is not yet fully exposed; used as Primary insert detail held between the corpse's fingers; room floor (visible around the lowered hand); used as Thin contextual plane around the hand that maintains physical 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: Restrained low-key illumination isolates the pale hand and photograph without introducing a new light source or color beyond the scene's muted palette.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk is seated motionless on the bedroom floor behind the curtain with her head dropped forward over her body, with the torn shoulder wound and surrounding blood; one lifeless hand remains closed around a crumpled photograph, and her wrist bears the circular mark.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 축 늘어진 민숙(강민숙)의 죽은 손아귀 안에 구겨진 사진 한 장이 살짝 삐져나온 클로즈업.\n\nLOCATION (lock): On the floor behind the curtain in the daughter's bedroom, beside the seated body. 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 hand height beside 강민숙, the dolly-in ends on an isolated detail of her lowered hand and the crumpled photograph protruding between closed fingers. The hand crosses the lower center while the photograph remains below one-third of the frame, with a narrow strip of floor and forearm preserving realistic scale and the wrist waiting just beyond the focal plane.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: crumpled photograph (crumpled and partially concealed) — The camera sees part of its image-bearing face protruding from the closed hand; the depicted content is not yet fully exposed; used as Primary insert detail held between the corpse's fingers; room floor (visible around the lowered hand); used as Thin contextual plane around the hand that maintains physical 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: Restrained low-key illumination isolates the pale hand and photograph without introducing a new light source or color beyond the scene's muted palette.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk is seated motionless on the bedroom floor behind the curtain with her head dropped forward over her body, with the torn shoulder wound and surrounding blood; one lifeless hand remains closed around a crumpled photograph, and her wrist bears the circular mark.\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, intimate lens feel with extremely selective focus placed on the creased photograph edge where it emerges between the fingers. Let the nearer hand contours and the forearm fall away softly, with the wrist already beyond focus; restrained low-key illumination catches only the paper ridges, pale knuckles, and slight skin texture against a nearly submerged floor plane.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 축 늘어진 민숙(강민숙)의 죽은 손아귀 안에 구겨진 사진 한 장이 살짝 삐져나온 클로즈업.\n\nLOCATION (lock): On the floor behind the curtain in the daughter's bedroom, beside the seated body. 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 hand height beside 강민숙, the dolly-in ends on an isolated detail of her lowered hand and the crumpled photograph protruding between closed fingers. The hand crosses the lower center while the photograph remains below one-third of the frame, with a narrow strip of floor and forearm preserving realistic scale and the wrist waiting just beyond the focal plane.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: crumpled photograph (crumpled and partially concealed) — The camera sees part of its image-bearing face protruding from the closed hand; the depicted content is not yet fully exposed; used as Primary insert detail held between the corpse's fingers; room floor (visible around the lowered hand); used as Thin contextual plane around the hand that maintains physical 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: Restrained low-key illumination isolates the pale hand and photograph without introducing a new light source or color beyond the scene's muted palette.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk is seated motionless on the bedroom floor behind the curtain with her head dropped forward over her body, with the torn shoulder wound and surrounding blood; one lifeless hand remains closed around a crumpled photograph, and her wrist bears the circular mark.\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 natural close-focus perspective from the locked hand-height position, rendering the closed hand, crushed paper, and thin surrounding floor plane with sober physical clarity while the wrist slips gently beyond the focal plane. Emphasize the diagonal weight of the lowered forearm and the rigid enclosure of the fingers around the photograph; soft directional falloff from the existing scene light reveals paper compression, skin folds, and floor texture without exposing the concealed image.\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": 8
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 5,
    "verdict_ko": "클로즈업 샷의 의도는 충족했으나, 바닥에 수평으로 놓인 팔의 각도가 레퍼런스에 명시된 '앉은 자세'와 모순됩니다."
   },
   {
    "label": "B",
    "score": 8,
    "verdict_ko": "레퍼런스의 앉은 자세에서 늘어뜨린 팔의 각도를 정확히 유지하며, 구겨진 사진과 손목의 흉터 등 디테일을 훌륭하게 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S12sh7_sel.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "레퍼런스 이미지에서는 인물의 왼손이 사진을 쥐고 있으나, 생성된 이미지에서는 오른손(엄지가 왼쪽에 위치한 손등)이 사진을 쥐고 있어 이전 샷의 자세가 유지되지 않았습니다.",
     "fix_en": "Replace the right hand with a left hand holding the photograph to match the pose in the reference image."
    },
    {
     "issue_ko": "손등이 카메라를 향하고 있음에도 손가락들이 반대 방향으로 꺾여 손톱이 정면을 향해 보이는 해부학적 오류가 있습니다.",
     "fix_en": "Redraw the fingers to curl naturally inward toward the hidden palm, removing the visible fingernails from this angle."
    }
   ]
  },
  "composition_critique": {
   "issues": []
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Replace the right hand with a left hand holding the photograph to match the pose in the reference image.\n- Redraw the fingers to curl naturally inward toward the hidden palm, removing the visible fingernails from this angle.\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 인서트 클로즈업 앵글, 구겨진 사진, 손목의 표식 등 모든 디테일을 정확하게 묘사했으며 해부학적으로도 자연스럽습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "검지에 불필요한 관절과 손톱이 중복으로 묘사되어 해부학적으로 불가능한 손의 형태를 보이는 하드 위반입니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 인서트 클로즈업 앵글, 구겨진 사진, 손목의 표식 등 모든 디테일을 정확하게 묘사했으며 해부학적으로도 자연스럽습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "검지에 불필요한 관절과 손톱이 중복으로 묘사되어 해부학적으로 불가능한 손의 형태를 보이는 하드 위반입니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "요구된 클로즈업 샷의 구도와 소품을 정확히 구현하였으며, 손가락 해부학이 정상적입니다."
     },
     {
      "label": "B",
      "score": 2,
      "verdict_ko": "치명적 오류: 손가락이 6개로 묘사되어 해부학적으로 불가능한 모습입니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "요구된 클로즈업 샷의 구도와 소품을 정확히 구현하였으며, 손가락 해부학이 정상적입니다."
     },
     {
      "label": "A",
      "score": 2,
      "verdict_ko": "치명적 오류: 손가락이 6개로 묘사되어 해부학적으로 불가능한 모습입니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 5,
     "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": "S12sh7"
  },
  "lane_policy": "share_plan_prev_bgonly",
  "locked_char_refs_excluded": [
   "강민숙(C02)"
  ],
  "plate_select": {
   "candidates": {
    "A": "L04B01",
    "B": "L04B02",
    "C": "L04B03",
    "D": "L04B04",
    "E": "L04B05",
    "F": "L04B06",
    "G": "L04B07"
   },
   "assigned": "L04B06",
   "choice": "Candidate F",
   "confident": true,
   "reason_ko": "로케이션 설명에 '딸의 침실 커튼 뒤'라고 명시되어 있으므로, 침실과 커튼이 보이는 Candidate F가 적합하며 현재 할당된 것을 유지합니다.",
   "kept": "L04B06"
  }
 },
 "S12sh15::variants": {
  "author_fp": "2cf819fb76f0cd50",
  "author": {
   "variants": [
    {
     "approach_ko": "정면적이고 절제된 심도 깊은 구도로 벽의 붉은 원과 침대의 공간 관계를 냉정하게 고정한다.",
     "prompt_en": "Use an austere, near-symmetrical medium composition from the locked low upward angle, with a natural perspective and restrained depth extending from the bed edge to the wall. Hold the complete brush-stroked red ring crisply in the upper center while the bed forms a quiet, clearly legible anchor across the lower quarter. Dim, soft interior light falls evenly enough to preserve the wall’s real texture and the ring’s uneven painted edges, with surrounding colors muted into cool shadow so the red alone commands attention."
    },
    {
     "approach_ko": "압축된 원근과 선택적 초점, 벽을 스치는 빛으로 붉은 붓자국의 물성을 불길하게 부각한다.",
     "prompt_en": "Photograph the locked medium framing with a compressed lens feel, isolating the wall mark as the plane of sharpest focus while allowing the bed edge below to fall gently soft without losing its spatial identity. Let low, directional light graze across the wall, revealing the physical drag, thickness, and irregular absorption of the red brush stroke against the surface. Keep the lower frame dense and subdued, creating a tense depth separation in which the complete red ring appears starkly present rather than graphic or luminous."
    }
   ]
  },
  "reused": false
 },
 "S12sh15": {
  "input_fingerprint": "0a9bff820d34186d",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 침대 위쪽 벽면에 붓으로 칠한 듯한 붉은색 원형 표식이 선명하게 드러난 정지 상태.\n\nLOCATION (lock): At the wall directly above the bed in the daughter's bedroom of the rooftop apartment. 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 the low position near 강민숙, the completed upward tilt holds absolutely still on the wall above the bed. The fully formed red circle sits in the upper-middle portion and remains under one-third of the frame, while the top edge of the bed anchors the lower quarter and confirms the mark's location.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: wall bearing the red circular mark in the upper-center of the frame, background; bed below the mark in the lower-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: wall above the bed (red circular mark fully visible) — The camera-facing wall surface clearly bears a complete, brush-like red circular mark; used as Locked focal surface carrying the supernatural reveal; bed (present) — Its upper edge and the side nearest the wall face the camera from below the mark; used as Lower-frame spatial anchor establishing that the mark is above it.\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 still, low-key interior rendering suppresses surrounding color so the fully visible red circle becomes the sole emphatic 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\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk's seated corpse and bloody wounds remain behind the curtain, but the photograph has now been removed from her hand by Suri-young. The wrist mark has turned red, and a new red circle remains visible on the wall above the bed.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 침대 위쪽 벽면에 붓으로 칠한 듯한 붉은색 원형 표식이 선명하게 드러난 정지 상태.\n\nLOCATION (lock): At the wall directly above the bed in the daughter's bedroom of the rooftop apartment. 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 the low position near 강민숙, the completed upward tilt holds absolutely still on the wall above the bed. The fully formed red circle sits in the upper-middle portion and remains under one-third of the frame, while the top edge of the bed anchors the lower quarter and confirms the mark's location.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: wall bearing the red circular mark in the upper-center of the frame, background; bed below the mark in the lower-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: wall above the bed (red circular mark fully visible) — The camera-facing wall surface clearly bears a complete, brush-like red circular mark; used as Locked focal surface carrying the supernatural reveal; bed (present) — Its upper edge and the side nearest the wall face the camera from below the mark; used as Lower-frame spatial anchor establishing that the mark is above it.\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 still, low-key interior rendering suppresses surrounding color so the fully visible red circle becomes the sole emphatic 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\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk's seated corpse and bloody wounds remain behind the curtain, but the photograph has now been removed from her hand by Suri-young. The wrist mark has turned red, and a new red circle remains visible on the wall above the bed.\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 austere, near-symmetrical medium composition from the locked low upward angle, with a natural perspective and restrained depth extending from the bed edge to the wall. Hold the complete brush-stroked red ring crisply in the upper center while the bed forms a quiet, clearly legible anchor across the lower quarter. Dim, soft interior light falls evenly enough to preserve the wall’s real texture and the ring’s uneven painted edges, with surrounding colors muted into cool shadow so the red alone commands attention.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 침대 위쪽 벽면에 붓으로 칠한 듯한 붉은색 원형 표식이 선명하게 드러난 정지 상태.\n\nLOCATION (lock): At the wall directly above the bed in the daughter's bedroom of the rooftop apartment. 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 the low position near 강민숙, the completed upward tilt holds absolutely still on the wall above the bed. The fully formed red circle sits in the upper-middle portion and remains under one-third of the frame, while the top edge of the bed anchors the lower quarter and confirms the mark's location.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: wall bearing the red circular mark in the upper-center of the frame, background; bed below the mark in the lower-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: wall above the bed (red circular mark fully visible) — The camera-facing wall surface clearly bears a complete, brush-like red circular mark; used as Locked focal surface carrying the supernatural reveal; bed (present) — Its upper edge and the side nearest the wall face the camera from below the mark; used as Lower-frame spatial anchor establishing that the mark is above it.\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 still, low-key interior rendering suppresses surrounding color so the fully visible red circle becomes the sole emphatic 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\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk's seated corpse and bloody wounds remain behind the curtain, but the photograph has now been removed from her hand by Suri-young. The wrist mark has turned red, and a new red circle remains visible on the wall above the bed.\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 locked medium framing with a compressed lens feel, isolating the wall mark as the plane of sharpest focus while allowing the bed edge below to fall gently soft without losing its spatial identity. Let low, directional light graze across the wall, revealing the physical drag, thickness, and irregular absorption of the red brush stroke against the surface. Keep the lower frame dense and subdued, creating a tense depth separation in which the complete red ring appears starkly present rather than graphic or luminous.\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": 6,
   "B": 8
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 6,
    "verdict_ko": "붉은 원이 화면의 1/3 이하를 차지해야 한다는 비율 조건을 벗어나 너무 크게 묘사되었으며, 카메라의 로우 앵글 틸트업 느낌이 부족합니다."
   },
   {
    "label": "B",
    "score": 8,
    "verdict_ko": "침대 상단이 화면 하단을 잡아주는 로우 앵글 구도를 잘 구현했으며, 붓으로 칠한 듯한 붉은 원의 질감과 크기 비율이 지시사항에 부합합니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S12sh12_sel.png"
   }
  ],
  "critique": {
   "issues": []
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Lower the bed and its headboard so their upper edge begins within the lower quarter of the frame, preserving the centered wall mark and all other framing.",
     "issue_ko": "침대와 헤드보드의 상단이 화면 중간 부근까지 올라와 하단 1/4에 고정되어야 한다는 구도보다 지나치게 큰 영역을 차지한다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Lower the bed and its headboard so their upper edge begins within the lower quarter of the frame, preserving the centered wall mark and all other framing.\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 붉은 원형 표식과 로우 앵글 구도를 정확히 구현했으며, 침대 프레임 등 배경 요소가 물리적으로 자연스럽습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "구도와 표식의 표현은 훌륭하나, 왼쪽의 침대 헤드보드가 중간에 허공에서 끊어지는 치명적인 물리적 오류가 있습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 붉은 원형 표식과 로우 앵글 구도를 정확히 구현했으며, 침대 프레임 등 배경 요소가 물리적으로 자연스럽습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "구도와 표식의 표현은 훌륭하나, 왼쪽의 침대 헤드보드가 중간에 허공에서 끊어지는 치명적인 물리적 오류가 있습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "프롬프트에서 요구한 침대 위 벽면의 붉은색 원형 표식과 매체의 질감, 정지된 구도를 매우 사실적으로 잘 구현했습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구사항을 훌륭하게 충족했으나, 침대 프레임이 가로지르는 형태보다 A의 자연스러운 침대 배치가 약간 더 시각적으로 설득력이 있습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "프롬프트에서 요구한 침대 위 벽면의 붉은색 원형 표식과 매체의 질감, 정지된 구도를 매우 사실적으로 잘 구현했습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요구사항을 훌륭하게 충족했으나, 침대 프레임이 가로지르는 형태보다 A의 자연스러운 침대 배치가 약간 더 시각적으로 설득력이 있습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 14,
     "B": 11
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": false,
    "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": "S12sh12"
  },
  "lane_policy": "share_plan_prev_bgonly",
  "plate_select": {
   "candidates": {
    "A": "L04B01",
    "B": "L04B02",
    "C": "L04B03",
    "D": "L04B04",
    "E": "L04B05",
    "F": "L04B06",
    "G": "L04B07"
   },
   "assigned": "L04B06",
   "choice": "Candidate F",
   "confident": true,
   "reason_ko": "원문의 '침대 위쪽 벽면에'라는 묘사에 따라 침대와 그 주변 벽면이 온전히 드러난 공간이어야 하므로, 현재 할당된 침실 후보가 적합함.",
   "kept": "L04B06"
  }
 },
 "S13sh3::bgfirst_bg": {
  "input_fingerprint": "91f3f75cb1e9c442",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 초점을 잃은 눈으로 가로등 불빛 아래 멍하니 서 있는 수리영의 넋이 나간 상체.\n\nLOCATION (lock): In the residential building's exterior yard at night, under the nearby streetlight.\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 directly before 수리영, the static camera holds her motionless upper body in a severe centered portrait, with only limited nighttime exterior context around her shoulders. Her body occupies roughly two-thirds of the frame, and the unbroken eye-level axis makes her unfocused stare feel exposed rather than confrontational.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: rooftop-room exterior area (visible at night); used as Visible exterior context behind the isolated figure; streetlight (on) — Its lit side faces toward 수리영's position; used as Motivated illumination on 수리영's arrested expression.\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: Streetlight illumination holds 수리영's drained face against a low-key, desaturated nighttime 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\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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 초점을 잃은 눈으로 가로등 불빛 아래 멍하니 서 있는 수리영의 넋이 나간 상체.\n\nLOCATION (lock): In the residential building's exterior yard at night, under the nearby streetlight.\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 directly before 수리영, the static camera holds her motionless upper body in a severe centered portrait, with only limited nighttime exterior context around her shoulders. Her body occupies roughly two-thirds of the frame, and the unbroken eye-level axis makes her unfocused stare feel exposed rather than confrontational.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: rooftop-room exterior area (visible at night); used as Visible exterior context behind the isolated figure; streetlight (on) — Its lit side faces toward 수리영's position; used as Motivated illumination on 수리영's arrested expression.\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: Streetlight illumination holds 수리영's drained face against a low-key, desaturated nighttime 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\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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S13sh3__bgfirst_bg.png",
  "asset_id": "282172ae-0870-45d5-9f46-07ccc1665b01",
  "input_asset_ids": [
   "0afb165a-0f59-4983-b435-5b16d5320f4e",
   "d43c6043-2a30-440f-81c8-0c1057323188"
  ]
 },
 "S13sh3": {
  "input_fingerprint": "6049813fb5f4582e",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 초점을 잃은 눈으로 가로등 불빛 아래 멍하니 서 있는 수리영의 넋이 나간 상체.\n\nLOCATION (lock): In the residential building's exterior yard at night, under the nearby streetlight. 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 directly before 수리영, the static camera holds her motionless upper body in a severe centered portrait, with only limited nighttime exterior context around her shoulders. Her body occupies roughly two-thirds of the frame, and the unbroken eye-level axis makes her unfocused stare feel exposed rather than confrontational.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: rooftop-room exterior area (visible at night); used as Visible exterior context behind the isolated figure; streetlight (on) — Its lit side faces toward 수리영's position; used as Motivated illumination on 수리영's arrested expression.\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: Streetlight illumination holds 수리영's drained face against a low-key, desaturated nighttime 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\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Until the later unexplained cleanup is revealed, Minsuk's immobile corpse remains seated behind the curtain inside, with the blood, bite wounds, red wrist mark, and red wall circle still present.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 초점을 잃은 눈으로 가로등 불빛 아래 멍하니 서 있는 수리영의 넋이 나간 상체.\n\nLOCATION (lock): In the residential building's exterior yard at night, under the nearby streetlight. 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 directly before 수리영, the static camera holds her motionless upper body in a severe centered portrait, with only limited nighttime exterior context around her shoulders. Her body occupies roughly two-thirds of the frame, and the unbroken eye-level axis makes her unfocused stare feel exposed rather than confrontational.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: rooftop-room exterior area (visible at night); used as Visible exterior context behind the isolated figure; streetlight (on) — Its lit side faces toward 수리영's position; used as Motivated illumination on 수리영's arrested expression.\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: Streetlight illumination holds 수리영's drained face against a low-key, desaturated nighttime 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\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Until the later unexplained cleanup is revealed, Minsuk's immobile corpse remains seated behind the curtain inside, with the blood, bite wounds, red wrist mark, and red wall circle still present.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 초점을 잃은 눈으로 가로등 불빛 아래 멍하니 서 있는 수리영의 넋이 나간 상체.\n\nLOCATION (lock): In the residential building's exterior yard at night, under the nearby streetlight. 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 directly before 수리영, the static camera holds her motionless upper body in a severe centered portrait, with only limited nighttime exterior context around her shoulders. Her body occupies roughly two-thirds of the frame, and the unbroken eye-level axis makes her unfocused stare feel exposed rather than confrontational.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: rooftop-room exterior area (visible at night); used as Visible exterior context behind the isolated figure; streetlight (on) — Its lit side faces toward 수리영's position; used as Motivated illumination on 수리영's arrested expression.\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: Streetlight illumination holds 수리영's drained face against a low-key, desaturated nighttime 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\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Until the later unexplained cleanup is revealed, Minsuk's immobile corpse remains seated behind the curtain inside, with the blood, bite wounds, red wrist mark, and red wall circle still present.\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 초반, 긴 검은 머리, 타원형의 젊은 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S13sh3__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S13sh3.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L10B01.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "지정된 정면 구도와 렌즈를 향한 시선 지침을 어기고 측면을 응시하고 있으며, 배경과의 원근감이 부자연스럽습니다."
     },
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "요구된 완벽한 정면 구도와 초점을 잃은 멍한 시선을 정확히 연출했으며, 레퍼런스 장소의 디테일과 스케일을 사실적으로 구현했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "지정된 정면 구도와 렌즈를 향한 시선 지침을 어기고 측면을 응시하고 있으며, 배경과의 원근감이 부자연스럽습니다."
     },
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "요구된 완벽한 정면 구도와 초점을 잃은 멍한 시선을 정확히 연출했으며, 레퍼런스 장소의 디테일과 스케일을 사실적으로 구현했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "카메라 정면을 향한 시선 축과 미디엄 샷 구도를 정확히 지켰으며, 배경과 인물의 물리적 비율이 자연스럽습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "인물이 배경에 비해 거대하게 표현되어 물리적으로 불가능한 비율을 보이며, 지정된 카메라 정면 응시 축을 벗어났습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "카메라 정면을 향한 시선 축과 미디엄 샷 구도를 정확히 지켰으며, 배경과 인물의 물리적 비율이 자연스럽습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "인물이 배경에 비해 거대하게 표현되어 물리적으로 불가능한 비율을 보이며, 지정된 카메라 정면 응시 축을 벗어났습니다."
     }
    ]
   },
   "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": "A",
    "score": 5,
    "verdict_ko": "지정된 정면 구도와 렌즈를 향한 시선 지침을 어기고 측면을 응시하고 있으며, 배경과의 원근감이 부자연스럽습니다."
   },
   {
    "label": "B",
    "score": 9,
    "verdict_ko": "요구된 완벽한 정면 구도와 초점을 잃은 멍한 시선을 정확히 연출했으며, 레퍼런스 장소의 디테일과 스케일을 사실적으로 구현했습니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L10B01.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "화면 왼쪽(인물의 오른쪽 귀)에 참조 이미지에 없는 회색 원형 장치(인공물)가 임의로 생성되었습니다.",
     "fix_en": "Remove the grey circular artifact from the subject's right ear."
    },
    {
     "issue_ko": "인물의 상체가 프레임의 약 3분의 2를 차지해야 한다는 카메라 지시사항과 달리, 인물이 너무 작게 렌더링되어 프레임 내 비중이 작습니다.",
     "fix_en": "Scale up the subject so her upper body occupies roughly two-thirds of the frame."
    }
   ]
  },
  "composition_critique": {
   "issues": []
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the grey circular artifact from the subject's right ear.\n- Scale up the subject so her upper body occupies roughly two-thirds 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": 7,
      "verdict_ko": "지시된 미디엄 샷 구도, 야외 배경, 조명, 인물의 공허한 표정과 외양을 모두 정확하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 미디엄 샷 구도, 야외 배경, 조명, 인물의 공허한 표정과 외양을 모두 정확하게 구현했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 미디엄 샷 구도, 야외 배경, 조명, 인물의 공허한 표정과 외양을 모두 정확하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 미디엄 샷 구도, 야외 배경, 조명, 인물의 공허한 표정과 외양을 모두 정확하게 구현했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 앵글과 스케일을 정확히 구현했으며, 넋이 나간 표정과 배경 디테일을 시각적 오류 없이 완벽하게 표현했습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "프레이밍과 배경 등 대부분의 요구사항을 잘 충족했으나, 인물의 오른쪽 귀/머리카락 부근에 부자연스러운 시각적 아티팩트가 존재합니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지정된 앵글과 스케일을 정확히 구현했으며, 넋이 나간 표정과 배경 디테일을 시각적 오류 없이 완벽하게 표현했습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "프레이밍과 배경 등 대부분의 요구사항을 잘 충족했으나, 인물의 오른쪽 귀/머리카락 부근에 부자연스러운 시각적 아티팩트가 존재합니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 14,
     "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S13sh3__bgfirst_bg.png",
   "bg_asset_id": "282172ae-0870-45d5-9f46-07ccc1665b01",
   "bg_record_key": "S13sh3::bgfirst_bg",
   "chain_winner": false,
   "authority": "plate"
  },
  "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S13sh4::variants": {
  "author_fp": "2690fdfbbf5ec1f6",
  "author": {
   "variants": [
    {
     "approach_ko": "절제된 관찰자 시점과 비교적 깊은 초점으로 두 사람의 대치와 야간 마당의 현실감을 함께 유지한다.",
     "prompt_en": "Use a natural perspective with restrained depth compression, holding both profiles and the locked shoulder grip in crisp focus while the rooftop exterior remains quietly legible behind them. Keep the composition laterally balanced and static, with available streetlamp illumination grazing faces, hands, hair, and fabric in firm but realistic contrast. Favor muted night color, tactile surface detail, and an unsensational observational tension."
    },
    {
     "approach_ko": "얕은 심도와 압축된 공간감으로 배경을 가라앉히고 어깨를 붙든 손과 두 사람의 충격적인 대치에 집중한다.",
     "prompt_en": "Use a subtly compressed lens feel and shallow depth of field to isolate the two profiles and the pressure of the hands on the shoulders, allowing the nighttime exterior context to recede into soft, unobtrusive shapes. Preserve the fixed eye-level medium two-shot while concentrating visual weight in the narrow space between their faces and along the rigid shoulder contact. Let the available exterior light create delicate edge separation and subdued facial modeling, with restrained highlights and dense natural shadows."
    }
   ]
  },
  "reused": false
 },
 "S13sh4": {
  "input_fingerprint": "b61b9f6c7016a4e7",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 혜수가 두 손으로 수리영의 양어깨를 단단히 붙잡고 마주 선 정면 구도.\n\nLOCATION (lock): In the exterior yard beside the residential building, near the newly arrived vehicle at night. 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 the pair, the camera holds a static two-shot across their face-to-face axis rather than presenting either woman frontally. 혜수 occupies the left half in profile with both hands locked around 수리영's shoulders, while 수리영 remains on the right, slightly lower and rigid within the same midground plane.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 혜수 in the middle-left of the frame, midground, reaches for 수리영's shoulders; 수리영 in the middle-right of the frame, midground, looks toward 혜수.\n- KEY BACKGROUND ELEMENTS: rooftop-room exterior area (visible at night); used as Unobtrusive nighttime exterior context retained behind the confrontation; staircase (present) — Its approach lies behind the pair and away from the current frontal camera axis; used as Background route toward which the camera will later reposition.\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: Available nighttime exterior illumination keeps both profiles readable in restrained, low-to-mid-key contrast without softening the shock.\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 villa yard, alley surfaces, streetlamp lighting, and nighttime atmosphere. Preserve the young woman's clothing and fixed standing position in that spot. Exclude the earlier solitary composition and introduce only the approaching officer who grips her shoulders.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk's corpse and the bloody, marked bedroom remain inside the rooftop room at this point, with no removal yet shown.\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 초반, 긴 검은 머리, 타원형의 젊은 얼굴); 혜수 (한국인 여성, 40대 중반, 단정한 검은 머리, 성숙한 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 혜수가 두 손으로 수리영의 양어깨를 단단히 붙잡고 마주 선 정면 구도.\n\nLOCATION (lock): In the exterior yard beside the residential building, near the newly arrived vehicle at night. 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 the pair, the camera holds a static two-shot across their face-to-face axis rather than presenting either woman frontally. 혜수 occupies the left half in profile with both hands locked around 수리영's shoulders, while 수리영 remains on the right, slightly lower and rigid within the same midground plane.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 혜수 in the middle-left of the frame, midground, reaches for 수리영's shoulders; 수리영 in the middle-right of the frame, midground, looks toward 혜수.\n- KEY BACKGROUND ELEMENTS: rooftop-room exterior area (visible at night); used as Unobtrusive nighttime exterior context retained behind the confrontation; staircase (present) — Its approach lies behind the pair and away from the current frontal camera axis; used as Background route toward which the camera will later reposition.\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: Available nighttime exterior illumination keeps both profiles readable in restrained, low-to-mid-key contrast without softening the shock.\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 villa yard, alley surfaces, streetlamp lighting, and nighttime atmosphere. Preserve the young woman's clothing and fixed standing position in that spot. Exclude the earlier solitary composition and introduce only the approaching officer who grips her shoulders.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk's corpse and the bloody, marked bedroom remain inside the rooftop room at this point, with no removal yet shown.\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 초반, 긴 검은 머리, 타원형의 젊은 얼굴); 혜수 (한국인 여성, 40대 중반, 단정한 검은 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a natural perspective with restrained depth compression, holding both profiles and the locked shoulder grip in crisp focus while the rooftop exterior remains quietly legible behind them. Keep the composition laterally balanced and static, with available streetlamp illumination grazing faces, hands, hair, and fabric in firm but realistic contrast. Favor muted night color, tactile surface detail, and an unsensational observational 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.",
   "B": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 혜수가 두 손으로 수리영의 양어깨를 단단히 붙잡고 마주 선 정면 구도.\n\nLOCATION (lock): In the exterior yard beside the residential building, near the newly arrived vehicle at night. 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 the pair, the camera holds a static two-shot across their face-to-face axis rather than presenting either woman frontally. 혜수 occupies the left half in profile with both hands locked around 수리영's shoulders, while 수리영 remains on the right, slightly lower and rigid within the same midground plane.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 혜수 in the middle-left of the frame, midground, reaches for 수리영's shoulders; 수리영 in the middle-right of the frame, midground, looks toward 혜수.\n- KEY BACKGROUND ELEMENTS: rooftop-room exterior area (visible at night); used as Unobtrusive nighttime exterior context retained behind the confrontation; staircase (present) — Its approach lies behind the pair and away from the current frontal camera axis; used as Background route toward which the camera will later reposition.\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: Available nighttime exterior illumination keeps both profiles readable in restrained, low-to-mid-key contrast without softening the shock.\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 villa yard, alley surfaces, streetlamp lighting, and nighttime atmosphere. Preserve the young woman's clothing and fixed standing position in that spot. Exclude the earlier solitary composition and introduce only the approaching officer who grips her shoulders.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsuk's corpse and the bloody, marked bedroom remain inside the rooftop room at this point, with no removal yet shown.\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 초반, 긴 검은 머리, 타원형의 젊은 얼굴); 혜수 (한국인 여성, 40대 중반, 단정한 검은 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a subtly compressed lens feel and shallow depth of field to isolate the two profiles and the pressure of the hands on the shoulders, allowing the nighttime exterior context to recede into soft, unobtrusive shapes. Preserve the fixed eye-level medium two-shot while concentrating visual weight in the narrow space between their faces and along the rigid shoulder contact. Let the available exterior light create delicate edge separation and subdued facial modeling, with restrained highlights and dense natural 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": 8,
   "A": 6
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 8,
    "verdict_ko": "요청된 미디엄 샷 프레이밍을 정확히 맞추었으며, 두 손으로 양어깨를 단단히 잡는 동작이 명확하게 묘사되었습니다."
   },
   {
    "label": "A",
    "score": 6,
    "verdict_ko": "지정된 미디엄 샷보다 프레임이 너무 넓게 잡혀(허벅지까지 노출됨) 샷 스케일 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S13sh3_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   },
   {
    "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:735029>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "혜수의 오른쪽 몸통 아래에서 불필요한 여분의 팔(재킷 소매)이 뻗어 나와 주된 팔 아래에 겹쳐 있습니다.",
     "fix_en": "Remove the extra jacket sleeve and arm segment protruding from Hyesu's lower torso so she has only one normal right arm."
    }
   ]
  },
  "composition_critique": {
   "issues": []
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the extra jacket sleeve and arm segment protruding from Hyesu's lower torso so she has only one normal right arm.\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 측면 투샷 구도와 두 인물의 자세, 이전 컷의 배경 및 의상 레퍼런스를 매우 충실하게 재현했습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "A와 동일한 이미지로, 프롬프트가 요구한 구도와 배경, 인물의 위치 및 동작을 훌륭하게 반영했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 측면 투샷 구도와 두 인물의 자세, 이전 컷의 배경 및 의상 레퍼런스를 매우 충실하게 재현했습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "A와 동일한 이미지로, 프롬프트가 요구한 구도와 배경, 인물의 위치 및 동작을 훌륭하게 반영했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "두 인물이 마주 보고 혜수가 수리영의 양어깨를 잡는 측면 미디엄 샷 구도와 지정된 배경 조건을 매우 충실히 재현했습니다."
     },
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "두 인물이 마주 보고 혜수가 수리영의 양어깨를 잡는 측면 미디엄 샷 구도와 지정된 배경 조건을 매우 충실히 재현했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "두 인물이 마주 보고 혜수가 수리영의 양어깨를 잡는 측면 미디엄 샷 구도와 지정된 배경 조건을 매우 충실히 재현했습니다."
     },
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "두 인물이 마주 보고 혜수가 수리영의 양어깨를 잡는 측면 미디엄 샷 구도와 지정된 배경 조건을 매우 충실히 재현했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 17,
     "B": 17
    },
    "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": "S13sh3"
  }
 },
 "S14sh4::variants": {
  "author_fp": "0e39ada449ca5230",
  "author": {
   "variants": [
    {
     "approach_ko": "정면성과 깊은 초점으로 방의 비정상적인 질서를 냉정한 증거 사진처럼 포착한다.",
     "prompt_en": "Use a restrained, rectilinear wide-lens feel with near-frontal geometry and deep focus from foreground to back wall. Hold the composition rigid and balanced around the central depth anchor, letting straight architectural lines and evenly legible surfaces create a clinical, evidence-like stillness. Keep the neutral ambient illumination broad and controlled, with subdued contrast, desaturated color, and precise material detail."
    },
    {
     "approach_ko": "넓은 전경 바닥과 강한 원근선으로 지워진 흔적의 부재를 불안하게 강조한다.",
     "prompt_en": "Favor a subtly more expansive wide-angle perspective while retaining the locked standing-height threshold view, giving the open foreground plane greater visual weight and allowing converging room lines to draw the eye into the central depth. Compose with deliberate asymmetrical tension between the central anchor and the visible side area, creating unease through negative space rather than dramatic lighting. Maintain natural depth, gentle edge falloff, controlled mid-key exposure, and cool-neutral, desaturated surfaces."
    }
   ]
  },
  "reused": false
 },
 "S14sh4": {
  "input_fingerprint": "0f9073b3c660c767",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 핏자국이나 벽의 붉은 표식, 시신 하나 없이 지나치게 말끔히 정돈된 수리영의 빈 방 안 전경.\n\nLOCATION (lock): Inside the daughter's bedroom of the rooftop apartment, now unnaturally clean and orderly with the curtain and bed area empty. 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 standing eye level on the threshold, the static wide frame presents the full room as an unnervingly orderly arrangement. The bed and bare wall occupy the central depth, the curtain area remains visible to one side, and open floor space in the foreground confirms that no body, blood, footprints, or red mark remains.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: bed (carefully arranged) — Its long side and the wall-facing end are visible from the threshold; used as Central spatial anchor proving the room has been reset; wall above the bed (blank and clean) — Its camera-facing surface is blank, with no red circular mark visible; used as Exposed content surface where the red mark should have been; curtain area (no body visible); used as Side-frame area where the corpse had been found; room floor (clean, with no blood or bloody footprints); used as Broad foreground evidence plane.\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 interior ambient light presents the desaturated, carefully ordered room with controlled mid-key clarity and no 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The bedroom has been unnaturally cleaned and reordered: Minsuk's corpse, all blood, and the red wall mark are gone, and the furniture and belongings are slightly displaced from their earlier positions.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 핏자국이나 벽의 붉은 표식, 시신 하나 없이 지나치게 말끔히 정돈된 수리영의 빈 방 안 전경.\n\nLOCATION (lock): Inside the daughter's bedroom of the rooftop apartment, now unnaturally clean and orderly with the curtain and bed area empty. 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 standing eye level on the threshold, the static wide frame presents the full room as an unnervingly orderly arrangement. The bed and bare wall occupy the central depth, the curtain area remains visible to one side, and open floor space in the foreground confirms that no body, blood, footprints, or red mark remains.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: bed (carefully arranged) — Its long side and the wall-facing end are visible from the threshold; used as Central spatial anchor proving the room has been reset; wall above the bed (blank and clean) — Its camera-facing surface is blank, with no red circular mark visible; used as Exposed content surface where the red mark should have been; curtain area (no body visible); used as Side-frame area where the corpse had been found; room floor (clean, with no blood or bloody footprints); used as Broad foreground evidence plane.\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 interior ambient light presents the desaturated, carefully ordered room with controlled mid-key clarity and no 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The bedroom has been unnaturally cleaned and reordered: Minsuk's corpse, all blood, and the red wall mark are gone, and the furniture and belongings are slightly displaced from their earlier positions.\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 near-frontal geometry and deep focus from foreground to back wall. Hold the composition rigid and balanced around the central depth anchor, letting straight architectural lines and evenly legible surfaces create a clinical, evidence-like stillness. Keep the neutral ambient illumination broad and controlled, with subdued contrast, desaturated color, and precise material 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 핏자국이나 벽의 붉은 표식, 시신 하나 없이 지나치게 말끔히 정돈된 수리영의 빈 방 안 전경.\n\nLOCATION (lock): Inside the daughter's bedroom of the rooftop apartment, now unnaturally clean and orderly with the curtain and bed area empty. 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 standing eye level on the threshold, the static wide frame presents the full room as an unnervingly orderly arrangement. The bed and bare wall occupy the central depth, the curtain area remains visible to one side, and open floor space in the foreground confirms that no body, blood, footprints, or red mark remains.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: bed (carefully arranged) — Its long side and the wall-facing end are visible from the threshold; used as Central spatial anchor proving the room has been reset; wall above the bed (blank and clean) — Its camera-facing surface is blank, with no red circular mark visible; used as Exposed content surface where the red mark should have been; curtain area (no body visible); used as Side-frame area where the corpse had been found; room floor (clean, with no blood or bloody footprints); used as Broad foreground evidence plane.\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 interior ambient light presents the desaturated, carefully ordered room with controlled mid-key clarity and no 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The bedroom has been unnaturally cleaned and reordered: Minsuk's corpse, all blood, and the red wall mark are gone, and the furniture and belongings are slightly displaced from their earlier positions.\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 subtly more expansive wide-angle perspective while retaining the locked standing-height threshold view, giving the open foreground plane greater visual weight and allowing converging room lines to draw the eye into the central depth. Compose with deliberate asymmetrical tension between the central anchor and the visible side area, creating unease through negative space rather than dramatic lighting. Maintain natural depth, gentle edge falloff, controlled mid-key exposure, and cool-neutral, desaturated 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": 3,
   "B": 7
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 3,
    "verdict_ko": "창문과 침대의 위치가 레퍼런스와 다르게 배치되어 장소의 일관성을 위반했습니다."
   },
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "레퍼런스의 방 구조를 정확히 유지하면서, 요구된 넓은 화각과 비정상적으로 깔끔한 빈 방의 상태를 훌륭하게 구현했습니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L04B07.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "기준 사진에 있던 침대의 어두운 나무 헤드보드가 누락되었습니다.",
     "fix_en": "Restore the dark wooden headboard to the far end of the bed against the wall."
    },
    {
     "issue_ko": "기준 사진에서는 창문 오른쪽 가장자리가 오른쪽 벽과 바로 맞닿아 있지만, 생성된 이미지에서는 창문 오른쪽에 넓은 벽면이 추가되어 방의 건축 구조가 달라졌습니다.",
     "fix_en": "Narrow the back wall so the right edge of the window and curtain are immediately adjacent to the room's right corner, matching the reference architecture."
    },
    {
     "issue_ko": "기준 사진에 존재하지 않는 작은 나무 수납장이 창문 아래쪽 벽에 임의로 생성되었습니다.",
     "fix_en": "Remove the small wooden cabinet located against the back wall beneath the window."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Reframe slightly to the right so the bed occupies the central depth as the primary spatial anchor, with its long side and wall-facing end fully visible, while retaining the curtain area at one side and the broad open foreground floor.",
     "issue_ko": "침대가 화면 오른쪽 가장자리로 밀려 일부 잘려 있어 중앙 깊이의 공간적 앵커가 되지 못하고, 커튼 영역이 중앙을 차지해 계약의 비대칭 배치와 어긋난다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Restore the dark wooden headboard to the far end of the bed against the wall.\n- Narrow the back wall so the right edge of the window and curtain are immediately adjacent to the room's right corner, matching the reference architecture.\n- Remove the small wooden cabinet located against the back wall beneath the window.\n- Reframe slightly to the right so the bed occupies the central depth as the primary spatial anchor, with its long side and wall-facing end fully visible, while retaining the curtain area at one side and the broad open foreground floor.\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": "레퍼런스의 공간 구조(우측 문 등)를 정확히 유지하면서 말끔히 치워진 방의 상태를 프롬프트에 맞게 잘 구현함."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "침대 위치가 지나치게 바뀌었고, 우측 전경의 문이 커튼으로 잘못 대체되는 공간 왜곡이 발생함."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "레퍼런스의 공간 구조(우측 문 등)를 정확히 유지하면서 말끔히 치워진 방의 상태를 프롬프트에 맞게 잘 구현함."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "침대 위치가 지나치게 바뀌었고, 우측 전경의 문이 커튼으로 잘못 대체되는 공간 왜곡이 발생함."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 앵글과 공간 구조(우측 전경의 문 등)를 정확히 유지하면서 가구들이 원래 위치에서 살짝 이동된 상태를 잘 표현했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "방이 깔끔하게 정돈된 모습은 구현되었으나, 우측 전경의 문 대신 커튼이 배치되어 공간 구조(Location Lock)에 어긋납니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 앵글과 공간 구조(우측 전경의 문 등)를 정확히 유지하면서 가구들이 원래 위치에서 살짝 이동된 상태를 잘 표현했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "방이 깔끔하게 정돈된 모습은 구현되었으나, 우측 전경의 문 대신 커튼이 배치되어 공간 구조(Location Lock)에 어긋납니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 14,
     "B": 9
    },
    "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": "L04B01",
    "B": "L04B02",
    "C": "L04B03",
    "D": "L04B04",
    "E": "L04B05",
    "F": "L04B06",
    "G": "L04B07"
   },
   "assigned": "L04B07",
   "choice": "Candidate G",
   "confident": true,
   "reason_ko": "\"말끔히 정돈된 수리영의 빈 방 안 전경\"이라는 지문에 맞춰, 침구가 정돈되고 시신이나 핏자국 없이 깨끗한 침실 전경을 보여주므로 현재 할당된 샷이 적합함.",
   "kept": "L04B07"
  }
 },
 "S14sh5::variants": {
  "author_fp": "eca2ba7434054cdd",
  "author": {
   "variants": [
    {
     "approach_ko": "얕은 심도로 눈과 굳어버린 표정에 밀착하고, 정돈된 방은 어깨 너머의 부드러운 증거로 남긴 압축된 초상 접근.",
     "prompt_en": "Use an intimate, gently compressed close-up with extremely selective focus: her nearest eye and the minute tension around her mouth are critically sharp, while her hair and the immaculate room fall away gradually into softness. Keep the softly recognizable bed and orderly space as restrained evidence over her far shoulder. Neutral ambient light wraps across her face with delicate shadow modeling, preserving natural skin texture while making her pallor and frozen disbelief the emotional center."
    },
    {
     "approach_ko": "근접한 광각 원근과 깊은 공간감을 이용해 창백한 얼굴과 어깨 너머의 비어 있고 깨끗한 방 사이의 불길한 대비를 강조한 접근.",
     "prompt_en": "Photograph the close-up with a subtly wider, spatially immediate lens feel, letting her foreground face feel present while the room recedes visibly beyond her far shoulder. Use the narrow background opening as deliberate negative space, with the arranged bed softly defocused yet clearly readable as part of the spotless room. Shape the neutral interior ambience with gentle side direction, holding restrained contrast across her rigid three-quarter expression and emphasizing the disconnect between what she sees and the clean physical evidence behind her."
    }
   ]
  },
  "reused": false
 },
 "S14sh5": {
  "input_fingerprint": "b7a3af1bd9864669",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 도저히 믿을 수 없다는 듯 당혹감에 사로잡혀 창백하게 질린 수리영의 얼굴 클로즈업.\n\nLOCATION (lock): Inside the daughter's newly cleaned bedroom, facing the empty area where the blood and body had appeared. 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: Just inside the room at 수리영's eye level, the dolly-in finishes at close facial distance as she freezes after entering. Her pale face occupies the left-center of the frame in three-quarter view, with a narrow, softly legible portion of the immaculate room retained over her far shoulder as the object of her disbelief.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: carefully arranged room (clean, with no blood, body, or red mark visible); used as Soft background evidence contrasting with 수리영's shocked expression; bed (carefully arranged) — Only part of the side facing the threshold remains visible behind 수리영; used as Defocused spatial cue behind her shoulder.\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 interior ambience keeps her pallor legible in restrained color and low-to-mid-key contrast, with no supernatural red remaining in frame.\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 spotless bedroom, its nighttime lighting, and the clean wall and curtain state. Preserve the young woman's clothing from the ongoing inspection. Exclude all blood, the corpse, and the red mark, and do not reproduce the preceding wide empty-room framing.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The spotless, altered room remains empty of Minsuk's corpse, bloodstains, and the red 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: 수리영 (한국인 여성, 20대 초반, 긴 검은 머리, 타원형의 젊은 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 도저히 믿을 수 없다는 듯 당혹감에 사로잡혀 창백하게 질린 수리영의 얼굴 클로즈업.\n\nLOCATION (lock): Inside the daughter's newly cleaned bedroom, facing the empty area where the blood and body had appeared. 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: Just inside the room at 수리영's eye level, the dolly-in finishes at close facial distance as she freezes after entering. Her pale face occupies the left-center of the frame in three-quarter view, with a narrow, softly legible portion of the immaculate room retained over her far shoulder as the object of her disbelief.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: carefully arranged room (clean, with no blood, body, or red mark visible); used as Soft background evidence contrasting with 수리영's shocked expression; bed (carefully arranged) — Only part of the side facing the threshold remains visible behind 수리영; used as Defocused spatial cue behind her shoulder.\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 interior ambience keeps her pallor legible in restrained color and low-to-mid-key contrast, with no supernatural red remaining in frame.\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 spotless bedroom, its nighttime lighting, and the clean wall and curtain state. Preserve the young woman's clothing from the ongoing inspection. Exclude all blood, the corpse, and the red mark, and do not reproduce the preceding wide empty-room framing.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The spotless, altered room remains empty of Minsuk's corpse, bloodstains, and the red 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: 수리영 (한국인 여성, 20대 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an intimate, gently compressed close-up with extremely selective focus: her nearest eye and the minute tension around her mouth are critically sharp, while her hair and the immaculate room fall away gradually into softness. Keep the softly recognizable bed and orderly space as restrained evidence over her far shoulder. Neutral ambient light wraps across her face with delicate shadow modeling, preserving natural skin texture while making her pallor and frozen disbelief the emotional center.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 도저히 믿을 수 없다는 듯 당혹감에 사로잡혀 창백하게 질린 수리영의 얼굴 클로즈업.\n\nLOCATION (lock): Inside the daughter's newly cleaned bedroom, facing the empty area where the blood and body had appeared. 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: Just inside the room at 수리영's eye level, the dolly-in finishes at close facial distance as she freezes after entering. Her pale face occupies the left-center of the frame in three-quarter view, with a narrow, softly legible portion of the immaculate room retained over her far shoulder as the object of her disbelief.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: carefully arranged room (clean, with no blood, body, or red mark visible); used as Soft background evidence contrasting with 수리영's shocked expression; bed (carefully arranged) — Only part of the side facing the threshold remains visible behind 수리영; used as Defocused spatial cue behind her shoulder.\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 interior ambience keeps her pallor legible in restrained color and low-to-mid-key contrast, with no supernatural red remaining in frame.\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 spotless bedroom, its nighttime lighting, and the clean wall and curtain state. Preserve the young woman's clothing from the ongoing inspection. Exclude all blood, the corpse, and the red mark, and do not reproduce the preceding wide empty-room framing.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The spotless, altered room remains empty of Minsuk's corpse, bloodstains, and the red 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: 수리영 (한국인 여성, 20대 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the close-up with a subtly wider, spatially immediate lens feel, letting her foreground face feel present while the room recedes visibly beyond her far shoulder. Use the narrow background opening as deliberate negative space, with the arranged bed softly defocused yet clearly readable as part of the spotless room. Shape the neutral interior ambience with gentle side direction, holding restrained contrast across her rigid three-quarter expression and emphasizing the disconnect between what she sees and the clean physical evidence behind her.\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": "얼굴 클로즈업 구도를 더 정확히 구현했으며, 창백하고 당혹스러운 표정을 잘 살린 동시에 레퍼런스의 공간 디테일(추가 조명 없음)을 완벽히 유지했습니다."
   },
   {
    "label": "A",
    "score": 5,
    "verdict_ko": "클로즈업 지시보다 프레임이 넓게 잡혔으며, 레퍼런스에 없는 켜진 조명이 침대 옆 탁자에 임의로 추가되어 공간 일관성(Location lock)을 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S14sh4_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "프레임에서 제외하도록 지시된 '이전 빈 방의 와이드 샷 구도'가 배경에 그대로 복제되어, 수리영 어깨 너머로 좁게 보여야 할 방 내부와 침대 전체가 어색한 비율로 모두 노출되었습니다.",
     "fix_en": "Replace the replicated wide-shot background plate with a correctly scaled, narrow, softly defocused portion of the room over her far shoulder, obscuring the left hallway and showing only a partial view of the bed."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Move the camera closer and tighten the close-up so only a narrow, softly defocused slice of the immaculate room and part of the bed’s threshold-facing side remain visible over her far shoulder, while keeping her face in the left-center.",
     "issue_ko": "배경 침실이 프레임 전반에 너무 넓고 선명하게 펼쳐지고 침대도 거의 전체가 보여, 얼굴 뒤 먼쪽 어깨 너머에 좁고 부드럽게 흐린 공간만 남겨야 하는 클로즈업 구도보다 지나치게 넓다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Replace the replicated wide-shot background plate with a correctly scaled, narrow, softly defocused portion of the room over her far shoulder, obscuring the left hallway and showing only a partial view of the bed.\n- Move the camera closer and tighten the close-up so only a narrow, softly defocused slice of the immaculate room and part of the bed’s threshold-facing side remain visible over her far shoulder, while keeping her face in the left-center.\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 클로즈업 구도와 피사계 심도(부드럽게 아웃포커싱된 배경)를 정확히 구현하며 주인공의 창백한 표정을 사실적으로 묘사했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "인물의 프레이밍과 복장, 장소는 정확히 일치하지만, 배경이 지나치게 선명하여 지시된 아웃포커싱(Defocused) 조건을 충족하지 못했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 클로즈업 구도와 피사계 심도(부드럽게 아웃포커싱된 배경)를 정확히 구현하며 주인공의 창백한 표정을 사실적으로 묘사했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "인물의 프레이밍과 복장, 장소는 정확히 일치하지만, 배경이 지나치게 선명하여 지시된 아웃포커싱(Defocused) 조건을 충족하지 못했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "클로즈업 샷과 아웃포커싱된 부드러운 배경 지침을 정확히 따랐으며, 인물의 충격받은 표정과 창백한 질감이 잘 표현되었습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "인물과 배경의 배치는 좋으나, 배경이 너무 선명하게 초점이 맞아 '부드럽게 읽히는 아웃포커싱된 공간'이라는 지침에 어긋납니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "클로즈업 샷과 아웃포커싱된 부드러운 배경 지침을 정확히 따랐으며, 인물의 충격받은 표정과 창백한 질감이 잘 표현되었습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "인물과 배경의 배치는 좋으나, 배경이 너무 선명하게 초점이 맞아 '부드럽게 읽히는 아웃포커싱된 공간'이라는 지침에 어긋납니다."
     }
    ]
   },
   "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": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S14sh4"
  }
 },
 "groupbg::coastal_bus_stop": {
  "input_fingerprint": "dad8e0dcfd4ebaea",
  "meta": {
   "model": "gpt-image-2",
   "size": "1536x864",
   "pack": "11.202607220237",
   "contract": "bgfirst_full_v3",
   "group_sig": {
    "key": "coastal_bus_stop",
    "tags": [
     "S15sh3",
     "S15sh6"
    ]
   },
   "context_sig": "6b674b4205ff2667"
  },
  "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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — contemporary South Korea, 2026; all people are Korean unless stated: At the bench of a coastal bus stop at night, facing the sea.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n해안가 버스정류장: 바다가 보이는 도로변에 벤치와 함께 설치된 버스 정류장 구조물. (특징: 정류장 벤치; 표지판 구조물; 배경으로 보이는 바다)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#15. 해안가. 버스정류장 부근 / N\n- 바다를 바라보며 멍한 채로 앉아있는 수리영\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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — contemporary South Korea, 2026; all people are Korean unless stated: At the bench of a coastal bus stop at night, facing the sea.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n해안가 버스정류장: 바다가 보이는 도로변에 벤치와 함께 설치된 버스 정류장 구조물. (특징: 정류장 벤치; 표지판 구조물; 배경으로 보이는 바다)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#15. 해안가. 버스정류장 부근 / N\n- 바다를 바라보며 멍한 채로 앉아있는 수리영\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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/groupbg_coastal_bus_stop_61fc1b.png",
  "asset_id": "58fe147b-e7f8-4e5c-93e5-1464e22ff60b",
  "input_asset_ids": [
   "77b39823-4766-4956-8303-e66359075bae"
  ],
  "origin_tag": "S15sh3",
  "place_text": "At the bench of a coastal bus stop at night, facing the sea.",
  "origin_inputs": {
   "place_text": "At the bench of a coastal bus stop at night, facing the sea.",
   "time_of_day_en": "night",
   "conti_asset_id": "77b39823-4766-4956-8303-e66359075bae"
  }
 },
 "S15sh3::bgfirst_bg": {
  "input_fingerprint": "4b88baf23913d955",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 벤치에서 허리를 반쯤 일으켜 세운 채 결연하고 단호한 눈빛으로 혜수를 올려다보는 수리영의 자세.\n\nLOCATION (lock): At the bench of a coastal bus stop at night, facing the 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: At seated eye height beside the bench, the final increment of the dolly-in settles below 수리영's eye line as she lifts her torso halfway from the seat. She occupies the left foreground while 혜수 remains on the right in the midground, leaving a clear upward diagonal between 수리영's resolved gaze and 혜수's face.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-left of the frame, foreground, looks toward 혜수; 혜수 in the middle-right of the frame, midground, looks toward 수리영.\n- KEY BACKGROUND ELEMENTS: bench (occupied by 수리영) — Its seat edge runs beneath 수리영 and toward the camera-side end used for the next pivot; used as Physical pivot and lower-frame anchor for 수리영's half-risen action; sea (visible from the bus-stop area); used as Distant setting context behind the pair.\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 nighttime ambience and restrained color hold the exchange in solemn contrast without specifying an unestablished 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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 벤치에서 허리를 반쯤 일으켜 세운 채 결연하고 단호한 눈빛으로 혜수를 올려다보는 수리영의 자세.\n\nLOCATION (lock): At the bench of a coastal bus stop at night, facing the 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: At seated eye height beside the bench, the final increment of the dolly-in settles below 수리영's eye line as she lifts her torso halfway from the seat. She occupies the left foreground while 혜수 remains on the right in the midground, leaving a clear upward diagonal between 수리영's resolved gaze and 혜수's face.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-left of the frame, foreground, looks toward 혜수; 혜수 in the middle-right of the frame, midground, looks toward 수리영.\n- KEY BACKGROUND ELEMENTS: bench (occupied by 수리영) — Its seat edge runs beneath 수리영 and toward the camera-side end used for the next pivot; used as Physical pivot and lower-frame anchor for 수리영's half-risen action; sea (visible from the bus-stop area); used as Distant setting context behind the pair.\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 nighttime ambience and restrained color hold the exchange in solemn contrast without specifying an unestablished 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S15sh3__bgfirst_bg.png",
  "asset_id": "1bce749b-1a1e-48a2-98ca-df4cb157a794",
  "input_asset_ids": [
   "77b39823-4766-4956-8303-e66359075bae",
   "58fe147b-e7f8-4e5c-93e5-1464e22ff60b"
  ]
 },
 "S15sh3": {
  "input_fingerprint": "3c21f5c957a1cafc",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 벤치에서 허리를 반쯤 일으켜 세운 채 결연하고 단호한 눈빛으로 혜수를 올려다보는 수리영의 자세.\n\nLOCATION (lock): At the bench of a coastal bus stop at night, facing the 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: At seated eye height beside the bench, the final increment of the dolly-in settles below 수리영's eye line as she lifts her torso halfway from the seat. She occupies the left foreground while 혜수 remains on the right in the midground, leaving a clear upward diagonal between 수리영's resolved gaze and 혜수's face.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-left of the frame, foreground, looks toward 혜수; 혜수 in the middle-right of the frame, midground, looks toward 수리영.\n- KEY BACKGROUND ELEMENTS: bench (occupied by 수리영) — Its seat edge runs beneath 수리영 and toward the camera-side end used for the next pivot; used as Physical pivot and lower-frame anchor for 수리영's half-risen action; sea (visible from the bus-stop area); used as Distant setting context behind the pair.\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 nighttime ambience and restrained color hold the exchange in solemn contrast without specifying an unestablished 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 벤치에서 허리를 반쯤 일으켜 세운 채 결연하고 단호한 눈빛으로 혜수를 올려다보는 수리영의 자세.\n\nLOCATION (lock): At the bench of a coastal bus stop at night, facing the 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: At seated eye height beside the bench, the final increment of the dolly-in settles below 수리영's eye line as she lifts her torso halfway from the seat. She occupies the left foreground while 혜수 remains on the right in the midground, leaving a clear upward diagonal between 수리영's resolved gaze and 혜수's face.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-left of the frame, foreground, looks toward 혜수; 혜수 in the middle-right of the frame, midground, looks toward 수리영.\n- KEY BACKGROUND ELEMENTS: bench (occupied by 수리영) — Its seat edge runs beneath 수리영 and toward the camera-side end used for the next pivot; used as Physical pivot and lower-frame anchor for 수리영's half-risen action; sea (visible from the bus-stop area); used as Distant setting context behind the pair.\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 nighttime ambience and restrained color hold the exchange in solemn contrast without specifying an unestablished 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 벤치에서 허리를 반쯤 일으켜 세운 채 결연하고 단호한 눈빛으로 혜수를 올려다보는 수리영의 자세.\n\nLOCATION (lock): At the bench of a coastal bus stop at night, facing the 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: At seated eye height beside the bench, the final increment of the dolly-in settles below 수리영's eye line as she lifts her torso halfway from the seat. She occupies the left foreground while 혜수 remains on the right in the midground, leaving a clear upward diagonal between 수리영's resolved gaze and 혜수's face.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-left of the frame, foreground, looks toward 혜수; 혜수 in the middle-right of the frame, midground, looks toward 수리영.\n- KEY BACKGROUND ELEMENTS: bench (occupied by 수리영) — Its seat edge runs beneath 수리영 and toward the camera-side end used for the next pivot; used as Physical pivot and lower-frame anchor for 수리영's half-risen action; sea (visible from the bus-stop area); used as Distant setting context behind the pair.\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 nighttime ambience and restrained color hold the exchange in solemn contrast without specifying an unestablished 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 초반, 긴 검은 머리, 타원형의 젊은 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S15sh3__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S15sh3.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/groupbg_coastal_bus_stop_61fc1b.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "벤치에서 엉덩이를 떼고 반쯤 일어선 수리영의 동작을 정확히 구현했으나, 요청된 미디엄 샷보다 프레임이 다소 넓습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "수리영이 벤치에 완전히 앉아 있어 '반쯤 일어선' 핵심 동작을 놓쳤으며, 풀 샷에 가깝게 너무 넓게 촬영되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "벤치에서 엉덩이를 떼고 반쯤 일어선 수리영의 동작을 정확히 구현했으나, 요청된 미디엄 샷보다 프레임이 다소 넓습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "수리영이 벤치에 완전히 앉아 있어 '반쯤 일어선' 핵심 동작을 놓쳤으며, 풀 샷에 가깝게 너무 넓게 촬영되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 미디엄 샷 프레이밍과 로우 앵글을 잘 구현했으며, 벤치에서 반쯤 일어난 수리영의 역동적인 자세와 두 인물 간의 물리적 스케일이 자연스럽습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "수리영의 자세와 샷의 기본 구성은 충족하나, 혜수가 전신으로 잡혀 미디엄 샷 기준보다 프레임이 넓어지고 주변 환경 대비 인물의 스케일이 다소 어색합니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 미디엄 샷 프레이밍과 로우 앵글을 잘 구현했으며, 벤치에서 반쯤 일어난 수리영의 역동적인 자세와 두 인물 간의 물리적 스케일이 자연스럽습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "수리영의 자세와 샷의 기본 구성은 충족하나, 혜수가 전신으로 잡혀 미디엄 샷 기준보다 프레임이 넓어지고 주변 환경 대비 인물의 스케일이 다소 어색합니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 9,
     "B": 14
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 9,
   "B": 14
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "벤치에서 엉덩이를 떼고 반쯤 일어선 수리영의 동작을 정확히 구현했으나, 요청된 미디엄 샷보다 프레임이 다소 넓습니다."
   },
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "수리영이 벤치에 완전히 앉아 있어 '반쯤 일어선' 핵심 동작을 놓쳤으며, 풀 샷에 가깝게 너무 넓게 촬영되었습니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/groupbg_coastal_bus_stop_61fc1b.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "벤치 좌석의 나무 판자 방향이 등받이와 수직으로 배열되어 있어, 등받이와 평행하게 가로로 배열된 로케이션 레퍼런스와 불일치합니다.",
     "fix_en": "Change the wooden planks of the bench seat to run horizontally, parallel to the backrest, to match the location reference."
    },
    {
     "issue_ko": "벤치 앞 가장자리를 짚고 있는 수리영의 왼손 손가락이 뭉개져서 나무 표면과 융합된 것처럼 보입니다.",
     "fix_en": "Redraw Suri-young's left hand to clearly show individual fingers anatomically resting on the bench edge."
    }
   ]
  },
  "composition_critique": {
   "issues": []
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the wooden planks of the bench seat to run horizontally, parallel to the backrest, to match the location reference.\n- Redraw Suri-young's left hand to clearly show individual fingers anatomically resting on the bench 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": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "벤치에서 반쯤 일어난 수리영과 혜수의 구도 및 눈빛 교환을 명시된 카메라 앵글로 정확히 구현했으며, 인체 해부학적 오류가 없습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "구도와 배경은 우수하나, 벤치를 짚고 있는 수리영의 왼손에 손가락이 추가로 묘사되는 치명적인 해부학적 오류(Hard Violation)가 있습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "벤치에서 반쯤 일어난 수리영과 혜수의 구도 및 눈빛 교환을 명시된 카메라 앵글로 정확히 구현했으며, 인체 해부학적 오류가 없습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "구도와 배경은 우수하나, 벤치를 짚고 있는 수리영의 왼손에 손가락이 추가로 묘사되는 치명적인 해부학적 오류(Hard Violation)가 있습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 구도와 벤치에서 반쯤 일어난 동작을 정확히 구현하였으며, 지정된 배경과 인물의 디테일이 충실히 반영되었습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 구도와 벤치에서 반쯤 일어난 동작을 정확히 구현하였으며, 지정된 배경과 인물의 디테일이 충실히 반영되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 구도와 벤치에서 반쯤 일어난 동작을 정확히 구현하였으며, 지정된 배경과 인물의 디테일이 충실히 반영되었습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지시된 구도와 벤치에서 반쯤 일어난 동작을 정확히 구현하였으며, 지정된 배경과 인물의 디테일이 충실히 반영되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 14,
     "B": 10
    },
    "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S15sh3__bgfirst_bg.png",
   "bg_asset_id": "1bce749b-1a1e-48a2-98ca-df4cb157a794",
   "bg_record_key": "S15sh3::bgfirst_bg",
   "chain_winner": false,
   "authority": "groupbg",
   "group_key": "coastal_bus_stop",
   "groupbg_asset_id": "58fe147b-e7f8-4e5c-93e5-1464e22ff60b"
  },
  "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S15sh6::variants": {
  "author_fp": "1bd36733c8d3283c",
  "author": {
   "variants": [
    {
     "approach_ko": "지면 가까운 은폐 시점과 압축된 망원 구도로 두 여자의 분리와 수리영의 공중 동작을 또렷하게 포착한다.",
     "prompt_en": "Photograph from a very low concealed eye line with a compressed long-lens feel, peering through a narrow, irregular gap in the shadowed grass. Keep the foreground vegetation soft and heavily occluding, while the distant Hye-su and Su-ri-young remain clearly readable as separate human figures; hold Su-ri-young’s airborne leg at the visual peak of the frame and counterbalance her motion with Hye-su’s stillness. Use restrained low-key night illumination and compressed depth to create the ominous sense of a hidden watcher."
    },
    {
     "approach_ko": "풀숲을 크게 전경화한 넓고 비대칭적인 주관숏으로, 어둠 속 멀리 떨어진 두 인물을 엿보는 불안감을 강조한다.",
     "prompt_en": "Use a wider subjective view from within the vegetation, with dark blades and clumps of grass occupying much of the near frame and leaving an off-center opening toward the distant women. Build strong layered depth: tactile foreground grass in partial focus, a broad zone of nocturnal darkness, then Hye-su standing and Su-ri-young caught running with one leg lifted in the far plane. Frame them small and asymmetrically with ample surrounding darkness, preserving enough muted edge light and tonal separation for their real bodies, clothing, and contrasting postures to remain legible."
    }
   ]
  },
  "reused": false
 },
 "S15sh6": {
  "input_fingerprint": "41c4ecb98a6e95df",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 풀숲 그림자에 반쯤 가려진 불길한 주관적 시야 틈새로, 멀찍이 서 있는 혜수와 한쪽 다리를 공중에 띄운 채 달려가는 자세인 수리영의 실루엣이 담긴 구도.\n\nLOCATION (lock): From concealed vegetation near the coastal bus stop, looking toward the bench area and the road where the two women separate. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key nighttime contrast leaves the distant figures as muted silhouettes while the foreground grass remains in concealing shadow.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 초반, 긴 검은 머리, 타원형의 젊은 얼굴); 혜수 (한국인 여성, 40대 중반, 단정한 검은 머리, 성숙한 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 풀숲 그림자에 반쯤 가려진 불길한 주관적 시야 틈새로, 멀찍이 서 있는 혜수와 한쪽 다리를 공중에 띄운 채 달려가는 자세인 수리영의 실루엣이 담긴 구도.\n\nLOCATION (lock): From concealed vegetation near the coastal bus stop, looking toward the bench area and the road where the two women separate. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key nighttime contrast leaves the distant figures as muted silhouettes while the foreground grass remains in concealing shadow.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 초반, 긴 검은 머리, 타원형의 젊은 얼굴); 혜수 (한국인 여성, 40대 중반, 단정한 검은 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from a very low concealed eye line with a compressed long-lens feel, peering through a narrow, irregular gap in the shadowed grass. Keep the foreground vegetation soft and heavily occluding, while the distant Hye-su and Su-ri-young remain clearly readable as separate human figures; hold Su-ri-young’s airborne leg at the visual peak of the frame and counterbalance her motion with Hye-su’s stillness. Use restrained low-key night illumination and compressed depth to create the ominous sense of a hidden watcher.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 풀숲 그림자에 반쯤 가려진 불길한 주관적 시야 틈새로, 멀찍이 서 있는 혜수와 한쪽 다리를 공중에 띄운 채 달려가는 자세인 수리영의 실루엣이 담긴 구도.\n\nLOCATION (lock): From concealed vegetation near the coastal bus stop, looking toward the bench area and the road where the two women separate. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key nighttime contrast leaves the distant figures as muted silhouettes while the foreground grass remains in concealing shadow.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 초반, 긴 검은 머리, 타원형의 젊은 얼굴); 혜수 (한국인 여성, 40대 중반, 단정한 검은 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a wider subjective view from within the vegetation, with dark blades and clumps of grass occupying much of the near frame and leaving an off-center opening toward the distant women. Build strong layered depth: tactile foreground grass in partial focus, a broad zone of nocturnal darkness, then Hye-su standing and Su-ri-young caught running with one leg lifted in the far plane. Frame them small and asymmetrically with ample surrounding darkness, preserving enough muted edge light and tonal separation for their real bodies, clothing, and contrasting postures to remain legible.\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": "전경의 풀숲을 어두운 그림자로 처리하여 지시된 주관적 시야와 조명 분위기를 잘 살렸으며 인물의 자세도 적절합니다."
   },
   {
    "label": "A",
    "score": 3,
    "verdict_ko": "전경의 풀이 밝게 노출되어 조명 지시를 어겼으며, 이전 샷에 없던 비닐 커버가 벤치 위에 생성된 오류가 있습니다."
   }
  ],
  "refs": [
   {
    "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.",
    "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/lane_sketch_S15sh6.png"
   },
   {
    "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S15sh3_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   },
   {
    "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:735029>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "두 인물 사이의 바닥에 프롬프트에 없는 길고 굵은 검은 선이 그려져 있습니다.",
     "fix_en": "Remove the long, thick dark line stretching across the ground between the two characters."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Widen the view and place both women farther into the background so they appear substantially smaller within a broader zone of darkness.",
     "issue_ko": "두 여성이 원경 인물치고 화면에서 너무 크고 가까워 보여, 요구된 넓은 주관적 시야와 원경의 작은 실루엣 구도가 약하다."
    },
    {
     "fix_en": "Shift the vegetation gap and the distant figures off-center to create a more asymmetrical concealed viewpoint while preserving all poses and motion directions.",
     "issue_ko": "풀숲의 시야 틈이 화면 중앙에 형성되고 양옆 식생이 대칭적으로 둘러싸여, 요구된 비대칭의 화면 밖 중심 개구부 구도와 다르다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the long, thick dark line stretching across the ground between the two characters.\n- Widen the view and place both women farther into the background so they appear substantially smaller within a broader zone of darkness.\n- Shift the vegetation gap and the distant figures off-center to create a more asymmetrical concealed viewpoint while preserving all poses and motion directions.\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": "풀숲 너머의 주관적 시점과 서 있는 혜수, 한쪽 다리를 들고 달리는 수리영의 실루엣 및 자세를 사실적으로 잘 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "요구사항을 전반적으로 충족했으나, 인물들 발아래 바닥을 가로지르는 불상(글리치/선)이 잘못 생성되어 감점되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "풀숲 너머의 주관적 시점과 서 있는 혜수, 한쪽 다리를 들고 달리는 수리영의 실루엣 및 자세를 사실적으로 잘 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "요구사항을 전반적으로 충족했으나, 인물들 발아래 바닥을 가로지르는 불상(글리치/선)이 잘못 생성되어 감점되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "풀숲을 통한 주관적 시점과 두 인물의 위치, 자세(달려가는 수리영과 서 있는 혜수)를 지시사항에 맞게 사실적으로 잘 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "구도와 인물의 배치는 유사하나, 바닥에 두 인물의 발을 잇는 부자연스럽고 설명할 수 없는 검은 직선이 렌더링 오류처럼 나타나 감점 요소가 되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "풀숲을 통한 주관적 시점과 두 인물의 위치, 자세(달려가는 수리영과 서 있는 혜수)를 지시사항에 맞게 사실적으로 잘 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "구도와 인물의 배치는 유사하나, 바닥에 두 인물의 발을 잇는 부자연스럽고 설명할 수 없는 검은 직선이 렌더링 오류처럼 나타나 감점 요소가 되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 8,
     "B": 14
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   },
   "winner": "B",
   "fix_won": true,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "lane(map_marker): 스케치+prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S15sh3"
  }
 },
 "S16sh1::bgfirst_bg": {
  "input_fingerprint": "3a5fbde8d10595c2",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 휴대전화를 귀에 바짝 댄 채 미간을 좁히고 경찰서 사무실 문고리를 꽉 틀어쥔 혜수의 정지된 손과 심각한 상체.\n\nLOCATION (lock): At the doorway into the police office at night, with the interior duty area beyond.\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 chest height just inside the office entrance, hold a medium, shallow three-quarter view of 혜수, placing her tense upper body on the right and her hand locked around the open door handle in the lower-left foreground. The static observer framing keeps the phone tight to her ear and reserves deeper office space behind her for the impending retreat along her entry path.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 경찰서 사무실 문 (open as 혜수 enters) — Its open edge angles away from the camera into the office; used as The entrance edge forms a rigid lateral frame around 혜수 and visually arrests her movement; 문고리 (being tightly twisted and held) — Its graspable side faces 혜수 while remaining visible to the camera; used as Held in the lower-left foreground as the physical focus of her halted entry; 휴대전화 (in use during an unanswered call) — Its side and back are visible while the screen faces 혜수; used as Pressed beside 혜수's face as the unanswered call continues.\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 low-to-mid-key office illumination at night keeps the palette desaturated and her expression solemn without specifying an unestablished 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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 휴대전화를 귀에 바짝 댄 채 미간을 좁히고 경찰서 사무실 문고리를 꽉 틀어쥔 혜수의 정지된 손과 심각한 상체.\n\nLOCATION (lock): At the doorway into the police office at night, with the interior duty area beyond.\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 chest height just inside the office entrance, hold a medium, shallow three-quarter view of 혜수, placing her tense upper body on the right and her hand locked around the open door handle in the lower-left foreground. The static observer framing keeps the phone tight to her ear and reserves deeper office space behind her for the impending retreat along her entry path.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 경찰서 사무실 문 (open as 혜수 enters) — Its open edge angles away from the camera into the office; used as The entrance edge forms a rigid lateral frame around 혜수 and visually arrests her movement; 문고리 (being tightly twisted and held) — Its graspable side faces 혜수 while remaining visible to the camera; used as Held in the lower-left foreground as the physical focus of her halted entry; 휴대전화 (in use during an unanswered call) — Its side and back are visible while the screen faces 혜수; used as Pressed beside 혜수's face as the unanswered call continues.\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 low-to-mid-key office illumination at night keeps the palette desaturated and her expression solemn without specifying an unestablished 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S16sh1__bgfirst_bg.png",
  "asset_id": "31d5001a-2060-4122-8ffd-6a92e7e18a92",
  "input_asset_ids": [
   "fb019fde-caa6-44be-b9d0-a5cd24aafe2d",
   "c70d5404-e782-42f5-9171-23bd635197c5"
  ]
 },
 "S16sh1": {
  "input_fingerprint": "0a9dfe0ae131f15f",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 휴대전화를 귀에 바짝 댄 채 미간을 좁히고 경찰서 사무실 문고리를 꽉 틀어쥔 혜수의 정지된 손과 심각한 상체.\n\nLOCATION (lock): At the doorway into the police office at night, with the interior duty area beyond. 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 inside the office entrance, hold a medium, shallow three-quarter view of 혜수, placing her tense upper body on the right and her hand locked around the open door handle in the lower-left foreground. The static observer framing keeps the phone tight to her ear and reserves deeper office space behind her for the impending retreat along her entry path.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 경찰서 사무실 문 (open as 혜수 enters) — Its open edge angles away from the camera into the office; used as The entrance edge forms a rigid lateral frame around 혜수 and visually arrests her movement; 문고리 (being tightly twisted and held) — Its graspable side faces 혜수 while remaining visible to the camera; used as Held in the lower-left foreground as the physical focus of her halted entry; 휴대전화 (in use during an unanswered call) — Its side and back are visible while the screen faces 혜수; used as Pressed beside 혜수's face as the unanswered call continues.\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 low-to-mid-key office illumination at night keeps the palette desaturated and her expression solemn without specifying an unestablished 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hye-soo has her mobile phone with her and is calling Minsuk, whose phone remains unanswered.\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 중반, 단정한 검은 머리, 성숙한 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 휴대전화를 귀에 바짝 댄 채 미간을 좁히고 경찰서 사무실 문고리를 꽉 틀어쥔 혜수의 정지된 손과 심각한 상체.\n\nLOCATION (lock): At the doorway into the police office at night, with the interior duty area beyond. 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 inside the office entrance, hold a medium, shallow three-quarter view of 혜수, placing her tense upper body on the right and her hand locked around the open door handle in the lower-left foreground. The static observer framing keeps the phone tight to her ear and reserves deeper office space behind her for the impending retreat along her entry path.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 경찰서 사무실 문 (open as 혜수 enters) — Its open edge angles away from the camera into the office; used as The entrance edge forms a rigid lateral frame around 혜수 and visually arrests her movement; 문고리 (being tightly twisted and held) — Its graspable side faces 혜수 while remaining visible to the camera; used as Held in the lower-left foreground as the physical focus of her halted entry; 휴대전화 (in use during an unanswered call) — Its side and back are visible while the screen faces 혜수; used as Pressed beside 혜수's face as the unanswered call continues.\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 low-to-mid-key office illumination at night keeps the palette desaturated and her expression solemn without specifying an unestablished 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hye-soo has her mobile phone with her and is calling Minsuk, whose phone remains unanswered.\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 중반, 단정한 검은 머리, 성숙한 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 휴대전화를 귀에 바짝 댄 채 미간을 좁히고 경찰서 사무실 문고리를 꽉 틀어쥔 혜수의 정지된 손과 심각한 상체.\n\nLOCATION (lock): At the doorway into the police office at night, with the interior duty area beyond. 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 inside the office entrance, hold a medium, shallow three-quarter view of 혜수, placing her tense upper body on the right and her hand locked around the open door handle in the lower-left foreground. The static observer framing keeps the phone tight to her ear and reserves deeper office space behind her for the impending retreat along her entry path.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 경찰서 사무실 문 (open as 혜수 enters) — Its open edge angles away from the camera into the office; used as The entrance edge forms a rigid lateral frame around 혜수 and visually arrests her movement; 문고리 (being tightly twisted and held) — Its graspable side faces 혜수 while remaining visible to the camera; used as Held in the lower-left foreground as the physical focus of her halted entry; 휴대전화 (in use during an unanswered call) — Its side and back are visible while the screen faces 혜수; used as Pressed beside 혜수's face as the unanswered call continues.\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 low-to-mid-key office illumination at night keeps the palette desaturated and her expression solemn without specifying an unestablished 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hye-soo has her mobile phone with her and is calling Minsuk, whose phone remains unanswered.\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 중반, 단정한 검은 머리, 성숙한 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S16sh1__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S16sh1.png"
    },
    {
     "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:735029>"
    },
    {
     "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.",
     "path": "<bytes:524464>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L13B04.png"
    },
    {
     "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:735029>"
    },
    {
     "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.",
     "path": "<bytes:524464>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "좁혀진 미간, 귀에 댄 휴대전화, 문고리를 쥔 정지된 동작과 지정된 화면 구도를 사실적으로 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "문고리가 부착된 문의 형태가 물리적으로 어색하며, 프레임 내 인물과 문의 공간적 관계가 지시문과 덜 일치합니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "좁혀진 미간, 귀에 댄 휴대전화, 문고리를 쥔 정지된 동작과 지정된 화면 구도를 사실적으로 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "문고리가 부착된 문의 형태가 물리적으로 어색하며, 프레임 내 인물과 문의 공간적 관계가 지시문과 덜 일치합니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "유리문으로 인해 측면 프레임 효과가 약하며, 배경에 지정되지 않은 대형 정수기가 등장하여 감점됨."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "단단한 문을 통한 프레임 효과가 지시문과 일치하며, 인물의 표정과 지정된 배경 및 소품을 가장 정확하게 구현함."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "유리문으로 인해 측면 프레임 효과가 약하며, 배경에 지정되지 않은 대형 정수기가 등장하여 감점됨."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "단단한 문을 통한 프레임 효과가 지시문과 일치하며, 인물의 표정과 지정된 배경 및 소품을 가장 정확하게 구현함."
     }
    ]
   },
   "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": "좁혀진 미간, 귀에 댄 휴대전화, 문고리를 쥔 정지된 동작과 지정된 화면 구도를 사실적으로 정확히 구현했습니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "문고리가 부착된 문의 형태가 물리적으로 어색하며, 프레임 내 인물과 문의 공간적 관계가 지시문과 덜 일치합니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L13B04.png"
   },
   {
    "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:735029>"
   },
   {
    "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.",
    "path": "<bytes:524464>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "문고리 레버가 아래로 틀어진(twisted) 상태가 아니라 수평으로 평행하게 멈춰 있습니다.",
     "fix_en": "Angle the door handle lever downward to show it being actively twisted."
    },
    {
     "issue_ko": "오른손이 문고리 레버를 꽉 틀어쥐지 않고 손잡이 끝부분에 느슨하게 얹혀 있습니다.",
     "fix_en": "Wrap the right fingers firmly and completely around the cylindrical lever of the door handle."
    },
    {
     "issue_ko": "휴대전화를 들고 있는 왼손 손가락(특히 검지와 중지)의 관절 형태가 뭉개져 해부학적으로 기형적입니다.",
     "fix_en": "Correct the anatomy of the left fingers holding the phone to show distinct, natural knuckles and structure."
    }
   ]
  },
  "composition_critique": {
   "issues": []
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Angle the door handle lever downward to show it being actively twisted.\n- Wrap the right fingers firmly and completely around the cylindrical lever of the door handle.\n- Correct the anatomy of the left fingers holding the phone to show distinct, natural knuckles and structure.\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": "지시된 구도와 인물, 배경을 정확히 반영했으며 문고리를 잡은 손의 형태가 자연스럽습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "문고리를 쥔 손 아래로 금속 파편 혹은 손가락이 기형적으로 생성되는 치명적인 오류가 발생했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지시된 구도와 인물, 배경을 정확히 반영했으며 문고리를 잡은 손의 형태가 자연스럽습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "문고리를 쥔 손 아래로 금속 파편 혹은 손가락이 기형적으로 생성되는 치명적인 오류가 발생했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 구도와 샷 크기를 완벽히 준수하며 인물의 자세, 소품, 찡그린 표정을 물리적 왜곡 없이 훌륭하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "문고리를 쥔 오른손 아래로 정체불명의 금속 형태가 튀어나와 구조가 왜곡되는 치명적인 오류가 발생했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요구된 구도와 샷 크기를 완벽히 준수하며 인물의 자세, 소품, 찡그린 표정을 물리적 왜곡 없이 훌륭하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "문고리를 쥔 오른손 아래로 정체불명의 금속 형태가 튀어나와 구조가 왜곡되는 치명적인 오류가 발생했습니다."
     }
    ]
   },
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S16sh1__bgfirst_bg.png",
   "bg_asset_id": "31d5001a-2060-4122-8ffd-6a92e7e18a92",
   "bg_record_key": "S16sh1::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": "L13B01",
    "B": "L13B02",
    "C": "L13B03",
    "D": "L13B04",
    "E": "L13B05",
    "F": "L13B06"
   },
   "assigned": "L13B05",
   "choice": "Candidate C",
   "confident": true,
   "reason_ko": "원문의 '경찰서 사무실 문고리를 꽉 틀어쥔'이라는 묘사에 따라 출입문이 포함된 배경이 필요하므로 문이 명확히 보이는 공간을 선택함.",
   "kept": "L13B05"
  }
 },
 "S16sh3::variants": {
  "author_fp": "adc8dc30acc994f0",
  "author": {
   "variants": [
    {
     "approach_ko": "정면에 가까운 깊은 초점과 엄격한 기하학으로 휴대전화와 김형사의 눈을 동시에 붙잡는 압박감 있는 구도.",
     "prompt_en": "Use a near-frontal, moderately wide perspective with disciplined deep focus, making the sharply rendered phone screen a blunt lower-central block while Detective Kim’s eyes remain equally legible directly beyond it. Let the desk edge form a firm receding line into the workstation, and hold Hye-soo’s left-edge lean in tense profile. Shape the desaturated office light into controlled facial contrast, with restrained screen glow and crisp, realistic surface detail."
    },
    {
     "approach_ko": "얕고 압축된 공간감과 비대칭 레이어링으로 혜수의 돌진과 김형사의 반응을 더 즉각적으로 느끼게 하는 구도.",
     "prompt_en": "Favor a compressed lens feel and tight asymmetrical layering: the phone looms in the lower center, Hye-soo presses in from the left edge, and Detective Kim’s sharply visible eyes appear narrowly trapped above and behind the screen. Keep the workstation context subdued but readable through the receding side of the desk and softly separated background planes. Use low-to-mid-key office illumination with a gentle falloff across the faces, allowing the active screen to provide the cleanest local emphasis without looking artificially luminous."
    }
   ]
  },
  "reused": false
 },
 "S16sh3": {
  "input_fingerprint": "bd0e71d65cf903c1",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 혜수가 특정 전화번호가 띄워진 자신의 휴대전화 화면을 의자에 앉은 김형사의 눈앞으로 불쑥 내민 찰나.\n\nLOCATION (lock): Beside the duty desk inside the police office at night. 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 김형사's seated eye level beside the desk, complete the dolly-in within arm's reach as 혜수 thrusts the phone into the lower-central foreground, where its screen occupies roughly one third of the frame. Keep 김형사's eyes sharply visible just beyond the phone and 혜수 leaning in from the left edge, creating a compressed two-person confrontation without losing the workstation context.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 혜수의 휴대전화 화면 (on and displaying 민숙's saved number) — The screen face is turned toward 김형사 and the camera, displaying the saved telephone number; used as The active screen is the sharp foreground focal plane, with 김형사's eyes retained beyond it; 김형사의 책상 (in use during duty work) — Its side edge recedes behind the raised phone toward 김형사; used as Provides the shared spatial boundary beside both characters and guides the camera's curved endpoint.\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: Desaturated, low-to-mid-key office illumination holds both faces in grounded contrast while the active phone screen receives selective 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hye-soo's phone continues to display Minsuk's saved number as she presents it to Detective Kim for location tracing.\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 중반, 단정한 검은 머리, 성숙한 얼굴); 김형사 (한국인 남성, 30대, 짧은 검은 머리, 평범한 성인 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 혜수가 특정 전화번호가 띄워진 자신의 휴대전화 화면을 의자에 앉은 김형사의 눈앞으로 불쑥 내민 찰나.\n\nLOCATION (lock): Beside the duty desk inside the police office at night. 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 김형사's seated eye level beside the desk, complete the dolly-in within arm's reach as 혜수 thrusts the phone into the lower-central foreground, where its screen occupies roughly one third of the frame. Keep 김형사's eyes sharply visible just beyond the phone and 혜수 leaning in from the left edge, creating a compressed two-person confrontation without losing the workstation context.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 혜수의 휴대전화 화면 (on and displaying 민숙's saved number) — The screen face is turned toward 김형사 and the camera, displaying the saved telephone number; used as The active screen is the sharp foreground focal plane, with 김형사's eyes retained beyond it; 김형사의 책상 (in use during duty work) — Its side edge recedes behind the raised phone toward 김형사; used as Provides the shared spatial boundary beside both characters and guides the camera's curved endpoint.\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: Desaturated, low-to-mid-key office illumination holds both faces in grounded contrast while the active phone screen receives selective 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hye-soo's phone continues to display Minsuk's saved number as she presents it to Detective Kim for location tracing.\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 중반, 단정한 검은 머리, 성숙한 얼굴); 김형사 (한국인 남성, 30대, 짧은 검은 머리, 평범한 성인 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a near-frontal, moderately wide perspective with disciplined deep focus, making the sharply rendered phone screen a blunt lower-central block while Detective Kim’s eyes remain equally legible directly beyond it. Let the desk edge form a firm receding line into the workstation, and hold Hye-soo’s left-edge lean in tense profile. Shape the desaturated office light into controlled facial contrast, with restrained screen glow and crisp, realistic 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 혜수가 특정 전화번호가 띄워진 자신의 휴대전화 화면을 의자에 앉은 김형사의 눈앞으로 불쑥 내민 찰나.\n\nLOCATION (lock): Beside the duty desk inside the police office at night. 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 김형사's seated eye level beside the desk, complete the dolly-in within arm's reach as 혜수 thrusts the phone into the lower-central foreground, where its screen occupies roughly one third of the frame. Keep 김형사's eyes sharply visible just beyond the phone and 혜수 leaning in from the left edge, creating a compressed two-person confrontation without losing the workstation context.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 혜수의 휴대전화 화면 (on and displaying 민숙's saved number) — The screen face is turned toward 김형사 and the camera, displaying the saved telephone number; used as The active screen is the sharp foreground focal plane, with 김형사's eyes retained beyond it; 김형사의 책상 (in use during duty work) — Its side edge recedes behind the raised phone toward 김형사; used as Provides the shared spatial boundary beside both characters and guides the camera's curved endpoint.\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: Desaturated, low-to-mid-key office illumination holds both faces in grounded contrast while the active phone screen receives selective 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hye-soo's phone continues to display Minsuk's saved number as she presents it to Detective Kim for location tracing.\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 중반, 단정한 검은 머리, 성숙한 얼굴); 김형사 (한국인 남성, 30대, 짧은 검은 머리, 평범한 성인 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFavor a compressed lens feel and tight asymmetrical layering: the phone looms in the lower center, Hye-soo presses in from the left edge, and Detective Kim’s sharply visible eyes appear narrowly trapped above and behind the screen. Keep the workstation context subdued but readable through the receding side of the desk and softly separated background planes. Use low-to-mid-key office illumination with a gentle falloff across the faces, allowing the active screen to provide the cleanest local emphasis without looking artificially luminous.\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": 8,
   "A": 5
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 8,
    "verdict_ko": "요구된 화면 구도(전경 하단 중앙, 프레임 1/3 비율)를 정확히 따랐으며, 스마트폰 화면의 내용이 레퍼런스와 완벽하게 일치합니다."
   },
   {
    "label": "A",
    "score": 5,
    "verdict_ko": "휴대전화의 위치와 크기가 지시사항에 미치지 못하며, 화면 인터페이스가 제공된 소품 레퍼런스를 반영하지 않고 임의로 생성되었습니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S16sh1_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:735029>"
   },
   {
    "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.",
    "path": "<bytes:524464>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "휴대전화를 쥐고 있는 혜수의 손가락 개수가 정상보다 많고 관절이 비정상적으로 길게 일그러져 해부학적으로 크게 왜곡되어 있습니다.",
     "fix_en": "Redraw Hye-soo's hand holding the phone to have correct human anatomy with a natural structure and exactly five properly proportioned fingers."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Dolly in tighter while maintaining Detective Kim’s seated eye-level axis, enlarging the phone screen to roughly one third of the frame in the lower center and retaining his eyes immediately above it.",
     "issue_ko": "현재 프레이밍이 지시보다 넓고 휴대전화 화면이 프레임의 약 3분의 1보다 훨씬 작아, 전경에서 압도적으로 돌출되는 구도가 아니다."
    },
    {
     "fix_en": "Redirect Detective Kim’s gaze downward onto the presented phone screen without changing any other pose or framing element.",
     "issue_ko": "김형사의 시선이 눈앞의 휴대전화 화면이 아니라 왼쪽의 혜수 얼굴을 향하고 있다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Redraw Hye-soo's hand holding the phone to have correct human anatomy with a natural structure and exactly five properly proportioned fingers.\n- Dolly in tighter while maintaining Detective Kim’s seated eye-level axis, enlarging the phone screen to roughly one third of the frame in the lower center and retaining his eyes immediately above it.\n- Redirect Detective Kim’s gaze downward onto the presented phone screen without changing any other pose or framing element.\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 미디엄 샷 구도(김형사 눈높이, 하단 중앙의 휴대전화, 왼쪽에서 다가오는 혜수)를 정확히 구현했으며, 휴대전화 화면의 텍스트와 UI를 레퍼런스와 매우 흡사하게 재현했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "구도는 압축적이나 휴대전화 화면의 연락처 텍스트가 비정상적으로 중복 생성되었고, 혜수의 얼굴이 프레임에서 과도하게 잘려나가 아쉽습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "요구된 미디엄 샷 구도(김형사 눈높이, 하단 중앙의 휴대전화, 왼쪽에서 다가오는 혜수)를 정확히 구현했으며, 휴대전화 화면의 텍스트와 UI를 레퍼런스와 매우 흡사하게 재현했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "구도는 압축적이나 휴대전화 화면의 연락처 텍스트가 비정상적으로 중복 생성되었고, 혜수의 얼굴이 프레임에서 과도하게 잘려나가 아쉽습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "제시된 미디엄 샷 프레이밍을 정확히 구현하여 혜수의 얼굴과 정체성을 명확히 보여주며, 두 인물의 압축된 대치 상황과 휴대폰 화면의 디테일을 레퍼런스에 맞게 훌륭히 표현했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "혜수의 얼굴이 프레임 밖으로 잘려나가 캐릭터의 정체성을 확인할 수 없으며, 휴대폰 화면 내 연락처 텍스트가 비정상적으로 중복 출력되는 오류가 있습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "제시된 미디엄 샷 프레이밍을 정확히 구현하여 혜수의 얼굴과 정체성을 명확히 보여주며, 두 인물의 압축된 대치 상황과 휴대폰 화면의 디테일을 레퍼런스에 맞게 훌륭히 표현했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "혜수의 얼굴이 프레임 밖으로 잘려나가 캐릭터의 정체성을 확인할 수 없으며, 휴대폰 화면 내 연락처 텍스트가 비정상적으로 중복 출력되는 오류가 있습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 18,
     "B": 11
    },
    "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": "S16sh1"
  }
 },
 "S17sh1::variants": {
  "author_fp": "4805b2262d4013de",
  "author": {
   "variants": [
    {
     "approach_ko": "깊은 초점과 엄격한 대각선 구도로 고양이와 높은 창문 사이의 거리와 고요를 강조한다.",
     "prompt_en": "Use deep focus and a restrained, rectilinear wide composition, keeping the crouched cat, the rising boundary, and the elevated window clearly legible across the full depth. Emphasize the severe lower-right-to-upper-left diagonal of the cat’s gaze, with broad areas of cool dawn-toned architecture creating isolation around the two anchors. Let the yellow window light remain contained and softly luminous against crisp bluish ambient detail, with natural shadow density and minimal bloom."
    },
    {
     "approach_ko": "전경 고양이에 선택적으로 초점을 두고 부드럽게 빛나는 창문을 먼 시선의 대상으로 만들어 긴장감을 높인다.",
     "prompt_en": "Photograph from the locked near-ground viewpoint with selective focus favoring the cat’s rear three-quarter form and tactile foreground detail, while the high window remains unmistakable but gently softened in the background. Use the boundary as a strong layered plane crossing the depth between cat and room, preserving the lateral route while making the upward eyeline feel steep and expectant. Allow the yellow light to bloom subtly into the cool dawn atmosphere, with restrained contrast and a quiet, watchful mood."
    }
   ]
  },
  "reused": false
 },
 "S17sh1": {
  "input_fingerprint": "7fd153eb8d2bbf14",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dawn.\n\nSHOT TEXT (authoritative, Korean): 푸르스름한 새벽 여명이 밝아오는 하늘 아래, 노란 불빛이 새어 나오는 옥탑방 창문을 올려다보는 검은 고양이의 모습.\n\nLOCATION (lock): On the residential building's rooftop at dawn, outside a rooftop room whose window emits yellow interior light. 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 black cat's rear three-quarter side near ground level, hold a wide static composition with the cat crouched low in the lower-right and the lit rooftop-room window high in the upper-left. The steep upward relationship between animal and window establishes the cat's eyeline while preserving lateral space for its coming turn toward the wall.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 검은 고양이 in the lower-right of the frame, foreground, looks toward lit rooftop-room window; lit rooftop-room window in the upper-left of the frame, background.\n- KEY BACKGROUND ELEMENTS: 옥탑방 창문 (lit from inside) — The camera sees the illuminated room side through the window from below; used as The elevated visual anchor receiving the cat's upward look; 담 (fixed boundary beside the rooftop room) — Its side rises across the cat's intended lateral path; used as Defines the route the cat will cross after turning away from the window.\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: Bluish dawn light is held in a restrained palette against the muted yellow illumination emerging from the rooftop-room window.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains lit through dawn after the night's events.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dawn.\n\nSHOT TEXT (authoritative, Korean): 푸르스름한 새벽 여명이 밝아오는 하늘 아래, 노란 불빛이 새어 나오는 옥탑방 창문을 올려다보는 검은 고양이의 모습.\n\nLOCATION (lock): On the residential building's rooftop at dawn, outside a rooftop room whose window emits yellow interior light. 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 black cat's rear three-quarter side near ground level, hold a wide static composition with the cat crouched low in the lower-right and the lit rooftop-room window high in the upper-left. The steep upward relationship between animal and window establishes the cat's eyeline while preserving lateral space for its coming turn toward the wall.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 검은 고양이 in the lower-right of the frame, foreground, looks toward lit rooftop-room window; lit rooftop-room window in the upper-left of the frame, background.\n- KEY BACKGROUND ELEMENTS: 옥탑방 창문 (lit from inside) — The camera sees the illuminated room side through the window from below; used as The elevated visual anchor receiving the cat's upward look; 담 (fixed boundary beside the rooftop room) — Its side rises across the cat's intended lateral path; used as Defines the route the cat will cross after turning away from the window.\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: Bluish dawn light is held in a restrained palette against the muted yellow illumination emerging from the rooftop-room window.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains lit through dawn after the night's events.\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 deep focus and a restrained, rectilinear wide composition, keeping the crouched cat, the rising boundary, and the elevated window clearly legible across the full depth. Emphasize the severe lower-right-to-upper-left diagonal of the cat’s gaze, with broad areas of cool dawn-toned architecture creating isolation around the two anchors. Let the yellow window light remain contained and softly luminous against crisp bluish ambient detail, with natural shadow density and minimal bloom.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dawn.\n\nSHOT TEXT (authoritative, Korean): 푸르스름한 새벽 여명이 밝아오는 하늘 아래, 노란 불빛이 새어 나오는 옥탑방 창문을 올려다보는 검은 고양이의 모습.\n\nLOCATION (lock): On the residential building's rooftop at dawn, outside a rooftop room whose window emits yellow interior light. 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 black cat's rear three-quarter side near ground level, hold a wide static composition with the cat crouched low in the lower-right and the lit rooftop-room window high in the upper-left. The steep upward relationship between animal and window establishes the cat's eyeline while preserving lateral space for its coming turn toward the wall.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 검은 고양이 in the lower-right of the frame, foreground, looks toward lit rooftop-room window; lit rooftop-room window in the upper-left of the frame, background.\n- KEY BACKGROUND ELEMENTS: 옥탑방 창문 (lit from inside) — The camera sees the illuminated room side through the window from below; used as The elevated visual anchor receiving the cat's upward look; 담 (fixed boundary beside the rooftop room) — Its side rises across the cat's intended lateral path; used as Defines the route the cat will cross after turning away from the window.\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: Bluish dawn light is held in a restrained palette against the muted yellow illumination emerging from the rooftop-room window.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains lit through dawn after the night's events.\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 locked near-ground viewpoint with selective focus favoring the cat’s rear three-quarter form and tactile foreground detail, while the high window remains unmistakable but gently softened in the background. Use the boundary as a strong layered plane crossing the depth between cat and room, preserving the lateral route while making the upward eyeline feel steep and expectant. Allow the yellow light to bloom subtly into the cool dawn atmosphere, with restrained contrast and a quiet, watchful mood.\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": "'우측 하단 고양이, 좌측 상단 창문'의 레이아웃과 지면 근처의 로우 앵글 등 명시된 카메라 프레이밍 지시를 가장 충실히 구현했습니다."
   },
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "지면 근처의 로우 앵글 지시를 무시하고 금지된 레퍼런스 사진의 하이 앵글 구도를 거의 그대로 답습하여 프레이밍 우선순위를 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L03B01.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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/background_chain/seed_bg_rooftop_villa_sel.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "옥탑방의 문이 녹슨 철문으로 생성되었으나, 지침에 따라 구조물 참조 사진(STRUCTURE LOOK)에 있는 녹색 문으로 렌더링되어야 합니다.",
     "fix_en": "Change the rusty metal door on the rooftop structure to match the green door shown in the STRUCTURE LOOK reference photograph."
    },
    {
     "issue_ko": "참조 사진의 공간 구조에는 없는 벽돌과 콘크리트 가벽이 옥상 한가운데에 임의로 생성되어 원래의 탁 트인 공간 레이아웃을 위반했습니다.",
     "fix_en": "Remove the freestanding brick and concrete wall in the middle of the rooftop to restore the open spatial layout of the location reference photograph."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Lower the cat into a compact crouching posture in the lower-right while preserving its upward gaze, screen position, and all other framing.",
     "issue_ko": "고양이가 낮게 웅크린 자세가 아니라 상체를 세운 채 앉아 있어 계약의 낮은 전경 실루엣과 다릅니다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the rusty metal door on the rooftop structure to match the green door shown in the STRUCTURE LOOK reference photograph.\n- Remove the freestanding brick and concrete wall in the middle of the rooftop to restore the open spatial layout of the location reference photograph.\n- Lower the cat into a compact crouching posture in the lower-right while preserving its upward gaze, screen position, and all other framing.\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 배치를 잘 구현했으며 구조물 레퍼런스의 초록색 문을 정확히 반영했으나, 주변 배경 일부(물탱크 등)가 누락되었습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "구조물 레퍼런스의 문 색상을 따르지 않았으며, 옥상 한가운데에 존재하지 않는 거대한 벽돌 구조물을 임의로 생성하는 치명적인 오류가 있습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 앵글과 배치를 잘 구현했으며 구조물 레퍼런스의 초록색 문을 정확히 반영했으나, 주변 배경 일부(물탱크 등)가 누락되었습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "구조물 레퍼런스의 문 색상을 따르지 않았으며, 옥상 한가운데에 존재하지 않는 거대한 벽돌 구조물을 임의로 생성하는 치명적인 오류가 있습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "고양이가 낮게 웅크린 자세(crouched low)를 프레임 우측 하단에 잘 구현했으며, 구조물 우선순위에 따라 초록색 문을 올바르게 반영했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "고양이가 웅크리지 않고 앉아 있으며, 레퍼런스에 없는 정체불명의 벽돌 구조물을 옥상 중앙에 임의로 생성하여 장소의 형태를 훼손했습니다(Hard Violation)."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "고양이가 낮게 웅크린 자세(crouched low)를 프레임 우측 하단에 잘 구현했으며, 구조물 우선순위에 따라 초록색 문을 올바르게 반영했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "고양이가 웅크리지 않고 앉아 있으며, 레퍼런스에 없는 정체불명의 벽돌 구조물을 옥상 중앙에 임의로 생성하여 장소의 형태를 훼손했습니다(Hard Violation)."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 8,
     "B": 14
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "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": "L03B01",
    "B": "L03B02",
    "C": "L03B03"
   },
   "assigned": "L03B01",
   "choice": "Candidate A",
   "confident": true,
   "reason_ko": "지문에서 명시한 '검은 고양이의 모습'과 '푸르스름한 새벽 여명' 조건에 정확히 부합하며, 동일한 옥탑방 공간을 보여주는 현재 배정된 후보를 변경할 이유가 없습니다.",
   "kept": "L03B01"
  }
 },
 "S18sh1::bgfirst_bg": {
  "input_fingerprint": "81ff774b63cb1c40",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 창문으로 햇살이 스며드는 방 안, 침대 모서리에 이마를 기댄 채 바닥에 웅크려 잠들어 있는 수리영의 모습.\n\nLOCATION (lock): On the floor beside the bed in the rooftop apartment's main bedroom, with early daylight entering through a window.\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: Begin with a held low wide frame from a shallow angle across the bedroom, placing 수리영 curled beside the bed in the lower-middle area while the bed corner provides scale above her. After the stillness registers, start the slow dolly toward her sleeping form, keeping the window and surrounding room secondary to the contact between her forehead and the bed edge.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 침대 (used as support while 수리영 sleeps) — Its corner angles toward the camera, with 수리영 resting against its edge; used as The bed corner forms the dominant spatial anchor above 수리영 and supports her sleeping posture; 창문 (allowing sunlight into the room) — Its plane faces the room at an oblique angle; the exterior view is not emphasized; used as Kept in the background as the established path for incoming sunlight.\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 filtering through the window is rendered in a restrained, desaturated low-to-mid-key balance that preserves the exhausted stillness.\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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 창문으로 햇살이 스며드는 방 안, 침대 모서리에 이마를 기댄 채 바닥에 웅크려 잠들어 있는 수리영의 모습.\n\nLOCATION (lock): On the floor beside the bed in the rooftop apartment's main bedroom, with early daylight entering through a window.\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: Begin with a held low wide frame from a shallow angle across the bedroom, placing 수리영 curled beside the bed in the lower-middle area while the bed corner provides scale above her. After the stillness registers, start the slow dolly toward her sleeping form, keeping the window and surrounding room secondary to the contact between her forehead and the bed edge.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 침대 (used as support while 수리영 sleeps) — Its corner angles toward the camera, with 수리영 resting against its edge; used as The bed corner forms the dominant spatial anchor above 수리영 and supports her sleeping posture; 창문 (allowing sunlight into the room) — Its plane faces the room at an oblique angle; the exterior view is not emphasized; used as Kept in the background as the established path for incoming sunlight.\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 filtering through the window is rendered in a restrained, desaturated low-to-mid-key balance that preserves the exhausted stillness.\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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S18sh1__bgfirst_bg.png",
  "asset_id": "38ff24f0-4b20-4eb5-8e78-f32a0de538a6",
  "input_asset_ids": [
   "f378e729-1f2b-4d84-b247-9b2ce8975da5",
   "b6092cac-71ec-4de8-a87e-7f7aba5d573f"
  ]
 },
 "S18sh1": {
  "input_fingerprint": "9d6a337a363d3721",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 창문으로 햇살이 스며드는 방 안, 침대 모서리에 이마를 기댄 채 바닥에 웅크려 잠들어 있는 수리영의 모습.\n\nLOCATION (lock): On the floor beside the bed in the rooftop apartment's main bedroom, with early daylight entering through a window. 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: Begin with a held low wide frame from a shallow angle across the bedroom, placing 수리영 curled beside the bed in the lower-middle area while the bed corner provides scale above her. After the stillness registers, start the slow dolly toward her sleeping form, keeping the window and surrounding room secondary to the contact between her forehead and the bed edge.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 침대 (used as support while 수리영 sleeps) — Its corner angles toward the camera, with 수리영 resting against its edge; used as The bed corner forms the dominant spatial anchor above 수리영 and supports her sleeping posture; 창문 (allowing sunlight into the room) — Its plane faces the room at an oblique angle; the exterior view is not emphasized; used as Kept in the background as the established path for incoming sunlight.\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 filtering through the window is rendered in a restrained, desaturated low-to-mid-key balance that preserves the exhausted stillness.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The house remains in its unnaturally clean, reordered state, with Minsuk's corpse, blood, and bedroom wall mark still absent.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 창문으로 햇살이 스며드는 방 안, 침대 모서리에 이마를 기댄 채 바닥에 웅크려 잠들어 있는 수리영의 모습.\n\nLOCATION (lock): On the floor beside the bed in the rooftop apartment's main bedroom, with early daylight entering through a window. 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: Begin with a held low wide frame from a shallow angle across the bedroom, placing 수리영 curled beside the bed in the lower-middle area while the bed corner provides scale above her. After the stillness registers, start the slow dolly toward her sleeping form, keeping the window and surrounding room secondary to the contact between her forehead and the bed edge.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 침대 (used as support while 수리영 sleeps) — Its corner angles toward the camera, with 수리영 resting against its edge; used as The bed corner forms the dominant spatial anchor above 수리영 and supports her sleeping posture; 창문 (allowing sunlight into the room) — Its plane faces the room at an oblique angle; the exterior view is not emphasized; used as Kept in the background as the established path for incoming sunlight.\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 filtering through the window is rendered in a restrained, desaturated low-to-mid-key balance that preserves the exhausted stillness.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The house remains in its unnaturally clean, reordered state, with Minsuk's corpse, blood, and bedroom wall mark still absent.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 창문으로 햇살이 스며드는 방 안, 침대 모서리에 이마를 기댄 채 바닥에 웅크려 잠들어 있는 수리영의 모습.\n\nLOCATION (lock): On the floor beside the bed in the rooftop apartment's main bedroom, with early daylight entering through a window. 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: Begin with a held low wide frame from a shallow angle across the bedroom, placing 수리영 curled beside the bed in the lower-middle area while the bed corner provides scale above her. After the stillness registers, start the slow dolly toward her sleeping form, keeping the window and surrounding room secondary to the contact between her forehead and the bed edge.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 침대 (used as support while 수리영 sleeps) — Its corner angles toward the camera, with 수리영 resting against its edge; used as The bed corner forms the dominant spatial anchor above 수리영 and supports her sleeping posture; 창문 (allowing sunlight into the room) — Its plane faces the room at an oblique angle; the exterior view is not emphasized; used as Kept in the background as the established path for incoming sunlight.\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 filtering through the window is rendered in a restrained, desaturated low-to-mid-key balance that preserves the exhausted stillness.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The house remains in its unnaturally clean, reordered state, with Minsuk's corpse, blood, and bedroom wall mark still absent.\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 초반, 긴 검은 머리, 타원형의 젊은 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S18sh1__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S18sh1.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L04B05.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "제공된 로케이션 레퍼런스의 공간(창문, 침대, 문 너머의 구조)을 완벽하게 재현했으며, 지정된 로우 앵글 와이드 샷과 인물의 수면 자세를 훌륭하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "인물의 포즈는 구체적이나, 레퍼런스와 달리 창문이 두 개로 나뉘고 침대 프레임과 배경 공간이 완전히 변형되어 로케이션 고정 조건을 심각하게 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "제공된 로케이션 레퍼런스의 공간(창문, 침대, 문 너머의 구조)을 완벽하게 재현했으며, 지정된 로우 앵글 와이드 샷과 인물의 수면 자세를 훌륭하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "인물의 포즈는 구체적이나, 레퍼런스와 달리 창문이 두 개로 나뉘고 침대 프레임과 배경 공간이 완전히 변형되어 로케이션 고정 조건을 심각하게 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "요청된 로우 앵글 와이드 샷을 정확히 구현했으며, 레퍼런스 이미지의 공간(창문, 침대, 조명 등)과 인물의 디테일 및 자세를 완벽하게 재현했습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "인물의 복장과 자세는 대체로 일치하나, 레퍼런스에 없는 높은 나무 침대 프레임이 추가되어 공간의 정확도가 떨어집니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "요청된 로우 앵글 와이드 샷을 정확히 구현했으며, 레퍼런스 이미지의 공간(창문, 침대, 조명 등)과 인물의 디테일 및 자세를 완벽하게 재현했습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "인물의 복장과 자세는 대체로 일치하나, 레퍼런스에 없는 높은 나무 침대 프레임이 추가되어 공간의 정확도가 떨어집니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 19,
     "B": 11
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 19,
   "B": 11
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 9,
    "verdict_ko": "제공된 로케이션 레퍼런스의 공간(창문, 침대, 문 너머의 구조)을 완벽하게 재현했으며, 지정된 로우 앵글 와이드 샷과 인물의 수면 자세를 훌륭하게 구현했습니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "인물의 포즈는 구체적이나, 레퍼런스와 달리 창문이 두 개로 나뉘고 침대 프레임과 배경 공간이 완전히 변형되어 로케이션 고정 조건을 심각하게 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L04B05.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "수리영이 침대 모서리에 이마를 기대지 않고 뺨을 댄 채 고개를 옆으로 돌리고 있습니다.",
     "fix_en": "Rotate Suri-young's head downward so that her forehead rests directly against her hands or the edge of the bed, matching the bowed pose in the sketch."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Restore the exact camera position, height, axis, and wide-frame perspective of the shot-background reference while preserving everything else.",
     "issue_ko": "카메라 위치와 원근이 배경 레퍼런스와 달라 창문과 출입문은 왼쪽으로 치우치고 침대는 화면에서 더 크게 나타난다."
    },
    {
     "fix_en": "Move Suriyoung downward to match the layout sketch’s lower-middle screen position and keep her scale unchanged.",
     "issue_ko": "수리영이 레이아웃 스케치의 지정 위치보다 화면 위쪽에 배치되어 있다."
    },
    {
     "fix_en": "Rotate and lower her head so her forehead directly contacts the bed edge, retaining the curled sleeping pose and all other composition.",
     "issue_ko": "이마를 침대 모서리에 기대야 하지만 현재는 손 위에 뺨과 관자놀이를 대고 있다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Rotate Suri-young's head downward so that her forehead rests directly against her hands or the edge of the bed, matching the bowed pose in the sketch.\n- Restore the exact camera position, height, axis, and wide-frame perspective of the shot-background reference while preserving everything else.\n- Move Suriyoung downward to match the layout sketch’s lower-middle screen position and keep her scale unchanged.\n- Rotate and lower her head so her forehead directly contacts the bed edge, retaining the curled sleeping pose and all other composition.\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 배경과 스케치 구도를 정확히 유지하며 캐릭터의 지시된 포즈를 자연스럽게 구현했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "구도와 배경 보존은 우수하나, 손과 얼굴 등 일부 세부 묘사의 자연스러움이 A에 비해 다소 부족합니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 배경과 스케치 구도를 정확히 유지하며 캐릭터의 지시된 포즈를 자연스럽게 구현했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "구도와 배경 보존은 우수하나, 손과 얼굴 등 일부 세부 묘사의 자연스러움이 A에 비해 다소 부족합니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 원본 배경과 스케치의 구도를 정확히 유지하며 캐릭터의 자세를 자연스럽게 구현했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "전반적인 구도와 배경 유지는 훌륭하나, 뒤로 뻗은 발목과 신발의 형태가 다소 어색합니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 원본 배경과 스케치의 구도를 정확히 유지하며 캐릭터의 자세를 자연스럽게 구현했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "전반적인 구도와 배경 유지는 훌륭하나, 뒤로 뻗은 발목과 신발의 형태가 다소 어색합니다."
     }
    ]
   },
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S18sh1__bgfirst_bg.png",
   "bg_asset_id": "38ff24f0-4b20-4eb5-8e78-f32a0de538a6",
   "bg_record_key": "S18sh1::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",
    "E": "L04B05",
    "F": "L04B06",
    "G": "L04B07"
   },
   "assigned": "L04B05",
   "choice": "Candidate E",
   "confident": true,
   "reason_ko": "햇살이 스며드는 방 안과 침대가 있다는 묘사(창문으로 햇살이 스며드는 방 안, 침대 모서리)를 충족하며, 현재 할당된 후보가 이를 정확히 보여주므로 변경할 이유가 없습니다.",
   "kept": "L04B05"
  }
 },
 "S18sh11::variants": {
  "author_fp": "b7bc01a3eaf2169c",
  "author": {
   "variants": [
    {
     "approach_ko": "거울을 냉정한 증거 화면처럼 평면적으로 구성해 수리영과 붉은 원형 표식을 동시에 또렷하게 포착한다.",
     "prompt_en": "Treat the mirror as a cold, forensic image plane, with restrained depth and near-even clarity across the reflected upper body and tiled wall. Keep Suri-young visually subordinate and off-center while the red outlined circle holds the strongest compositional weight over her shoulder; use flat, cool early-morning ambience and precise surface detail for an unnervingly objective stillness."
    },
    {
     "approach_ko": "거울 프레임과 세면대의 수렴선을 활용한 깊고 폐쇄적인 구도로 수리영의 반응과 등 뒤 표식 사이의 긴장을 강조한다.",
     "prompt_en": "Build a claustrophobic layered composition through the visible mirror frame and the receding sink edge, drawing the eye inward toward Suri-young’s reflected upper body before it lands on the unmistakable red ring behind her. Use gentle focus separation while preserving the mark’s clear shape, with subtly directional early-morning ambience leaving the reflected tile field in controlled falloff and the deep red stroke as the only forceful color."
    }
   ]
  },
  "reused": false
 },
 "S18sh11": {
  "input_fingerprint": "d4a58407cdce54fa",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 거울 표면에 비친 수리영의 등 뒤 욕실 타일 벽면에 선명하게 드러난 붉은 원형 표식.\n\nLOCATION (lock): Inside the rooftop apartment's bathroom, facing the mirror and the tiled wall reflected behind it. 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 just behind and slightly beside 수리영 at shoulder height, push directly toward the mirror until her reflected upper body sits off-center and the reflected red circle becomes fully legible over her shoulder. Settle into a brief, unnervingly static hold, keeping the mirror frame visible enough to establish that both 수리영 and the mark are seen only within its reflected space.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 욕실 거울 (showing the sudden reflected anomaly) — The mirror reflects 수리영's upper body and the tiled wall behind her, where the red circular mark is visible; the mirror frame remains partially in view; used as The primary image plane containing 수리영 and the otherwise impossible red mark; 욕실 타일 벽면 (a red circular mark is visible in the reflection) — Reflected behind 수리영, with the marked area held clearly in the composition; used as Appears inside the mirror behind 수리영 as the location of the circular mark; 세면대 (being used for support) — Its front edge sits below 수리영 and recedes toward the mirror; used as Provides a physical support point beneath the reflected figure.\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 bathroom ambience remains desaturated and controlled while the reflected deep red circle becomes the sole forceful color 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young now has her doll-adorned backpack with her. The prior corpse and blood remain absent from the cleaned house.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 거울 표면에 비친 수리영의 등 뒤 욕실 타일 벽면에 선명하게 드러난 붉은 원형 표식.\n\nLOCATION (lock): Inside the rooftop apartment's bathroom, facing the mirror and the tiled wall reflected behind it. 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 just behind and slightly beside 수리영 at shoulder height, push directly toward the mirror until her reflected upper body sits off-center and the reflected red circle becomes fully legible over her shoulder. Settle into a brief, unnervingly static hold, keeping the mirror frame visible enough to establish that both 수리영 and the mark are seen only within its reflected space.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 욕실 거울 (showing the sudden reflected anomaly) — The mirror reflects 수리영's upper body and the tiled wall behind her, where the red circular mark is visible; the mirror frame remains partially in view; used as The primary image plane containing 수리영 and the otherwise impossible red mark; 욕실 타일 벽면 (a red circular mark is visible in the reflection) — Reflected behind 수리영, with the marked area held clearly in the composition; used as Appears inside the mirror behind 수리영 as the location of the circular mark; 세면대 (being used for support) — Its front edge sits below 수리영 and recedes toward the mirror; used as Provides a physical support point beneath the reflected figure.\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 bathroom ambience remains desaturated and controlled while the reflected deep red circle becomes the sole forceful color 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young now has her doll-adorned backpack with her. The prior corpse and blood remain absent from the cleaned house.\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):\nTreat the mirror as a cold, forensic image plane, with restrained depth and near-even clarity across the reflected upper body and tiled wall. Keep Suri-young visually subordinate and off-center while the red outlined circle holds the strongest compositional weight over her shoulder; use flat, cool early-morning ambience and precise surface detail for an unnervingly objective 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 거울 표면에 비친 수리영의 등 뒤 욕실 타일 벽면에 선명하게 드러난 붉은 원형 표식.\n\nLOCATION (lock): Inside the rooftop apartment's bathroom, facing the mirror and the tiled wall reflected behind it. 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 just behind and slightly beside 수리영 at shoulder height, push directly toward the mirror until her reflected upper body sits off-center and the reflected red circle becomes fully legible over her shoulder. Settle into a brief, unnervingly static hold, keeping the mirror frame visible enough to establish that both 수리영 and the mark are seen only within its reflected space.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 욕실 거울 (showing the sudden reflected anomaly) — The mirror reflects 수리영's upper body and the tiled wall behind her, where the red circular mark is visible; the mirror frame remains partially in view; used as The primary image plane containing 수리영 and the otherwise impossible red mark; 욕실 타일 벽면 (a red circular mark is visible in the reflection) — Reflected behind 수리영, with the marked area held clearly in the composition; used as Appears inside the mirror behind 수리영 as the location of the circular mark; 세면대 (being used for support) — Its front edge sits below 수리영 and recedes toward the mirror; used as Provides a physical support point beneath the reflected figure.\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 bathroom ambience remains desaturated and controlled while the reflected deep red circle becomes the sole forceful color 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young now has her doll-adorned backpack with her. The prior corpse and blood remain absent from the cleaned house.\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):\nBuild a claustrophobic layered composition through the visible mirror frame and the receding sink edge, drawing the eye inward toward Suri-young’s reflected upper body before it lands on the unmistakable red ring behind her. Use gentle focus separation while preserving the mark’s clear shape, with subtly directional early-morning ambience leaving the reflected tile field in controlled falloff and the deep red stroke as the only forceful color.\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": "거울 안팎의 세면대 구조와 피사체의 팔 위치가 물리적으로 전혀 일치하지 않는 치명적인 반사 오류가 있습니다."
   },
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "요구된 숄더 뷰 앵글, 오프센터로 배치된 반사 이미지, 열린 형태의 붉은 원형 표식 및 인형 배낭을 사실적으로 잘 구현했습니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L04B03.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "카메라 프레이밍 지시사항에 수리영과 붉은 표식이 '오직 거울에 반사된 공간 안에서만(seen only within its reflected space)' 보여야 한다고 명시되어 있으나, 우측 전경에 수리영의 실제 뒷모습이 화면에 포함되어 있습니다.",
     "fix_en": "Remove the physical foreground figure of Suri-young and her backpack from the right side of the frame so she appears solely within the mirror's reflection."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Lower or widen the framing slightly to include the sink’s front edge beneath Suri-young, visibly receding toward the mirror, while preserving the current camera axis and reflected placements.",
     "issue_ko": "세면대의 앞쪽 가장자리가 화면 아래에서 거의 잘려 있어, 수리영 아래에서 거울 쪽으로 후퇴하는 구도 요소가 드러나지 않는다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the physical foreground figure of Suri-young and her backpack from the right side of the frame so she appears solely within the mirror's reflection.\n- Lower or widen the framing slightly to include the sink’s front edge beneath Suri-young, visibly receding toward the mirror, while preserving the current camera axis and reflected placements.\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": "피사체가 거울에 비친 공간에만 존재해야 한다는 프레이밍 지시를 정확히 따랐으며, 카메라 구도와 붉은 표식의 렌더링이 훌륭합니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "인물과 표식이 오직 반사된 공간에서만 보여야 한다는 카메라 지시를 어기고 전경에 인물의 실제 뒷모습을 포함시켰습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "피사체가 거울에 비친 공간에만 존재해야 한다는 프레이밍 지시를 정확히 따랐으며, 카메라 구도와 붉은 표식의 렌더링이 훌륭합니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "인물과 표식이 오직 반사된 공간에서만 보여야 한다는 카메라 지시를 어기고 전경에 인물의 실제 뒷모습을 포함시켰습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 수리영의 등 뒤 어깨 너머 구도를 정확하게 구현하였고, 인형이 달린 배낭 소품까지 충실히 반영했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "거울 앞에 인물의 실체가 보이지 않아 물리적으로 불가능한 구도를 연출했으며, 요구된 인형 소품이 누락되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요구된 수리영의 등 뒤 어깨 너머 구도를 정확하게 구현하였고, 인형이 달린 배낭 소품까지 충실히 반영했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "거울 앞에 인물의 실체가 보이지 않아 물리적으로 불가능한 구도를 연출했으며, 요구된 인형 소품이 누락되었습니다."
     }
    ]
   },
   "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": "플레이트만 (배경 전용)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L04B01",
    "B": "L04B02",
    "C": "L04B03",
    "D": "L04B04",
    "E": "L04B05",
    "F": "L04B06",
    "G": "L04B07"
   },
   "assigned": "L04B03",
   "choice": "Candidate C",
   "confident": true,
   "reason_ko": "문틈 너머 욕실 타일 벽면에 '붉은 원형 표식'이 선명하게 그려져 있는 공간이므로 이 숏에 가장 부합합니다.",
   "kept": "L04B03"
  }
 },
 "S18sh15::variants": {
  "author_fp": "535186079b232964",
  "author": {
   "variants": [
    {
     "approach_ko": "정면의 엄격한 대칭과 깊은 초점으로 환각의 변형을 냉정하고 도피할 수 없게 포착한다.",
     "prompt_en": "Use a rigid, centered close-up with a restrained normal-lens perspective, keeping the printed plane, its visible edges, and the narrow strip of table crisply resolved. Let the subdued early-morning ambience fall evenly across the image so the grotesque downward displacement of the bright smile reads with clinical clarity against the photograph’s otherwise stable physical geometry."
    },
    {
     "approach_ko": "비대칭에 가까운 밀착 구도와 얕은 초점, 사선의 부드러운 아침빛으로 얼굴의 붕괴를 압박감 있게 강조한다.",
     "prompt_en": "Favor a claustrophobic close-up with the distorted smiling face weighted strongly within the crop, using shallow optical falloff so the facial surface commands attention while the photograph edges and table strip recede softly without disappearing. Shape the existing desaturated morning daylight into a gentle oblique wash, allowing slight tonal falloff across the print to make the downward-sliding hallucination feel immediate and oppressive."
    }
   ]
  },
  "reused": false
 },
 "S18sh15": {
  "input_fingerprint": "5c7d706b071c7686",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 수리영의 주관적 시점, 사진 속 민숙(강민숙)의 환하게 웃는 얼굴 표면이 기괴하게 일그러진 채 반쯤 흘러내린 끔찍한 환각의 정지 순간.\n\nLOCATION (lock): At the dining table in the rooftop apartment's living area, where a family photograph is displayed. 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 수리영's exact subjective eye point, hold squarely on the photograph while cropping to the mother's side so the distorted face becomes the human primary subject and the POV character's pictured likeness remains outside the frame. Keep the photograph within a natural portion of the composition, with its edges and a narrow band of the table still visible as the smiling facial surface appears grotesquely displaced downward.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 식탁 위 사진 (subjectively distorted in 수리영's hallucination) — The printed front face is square to the camera, cropped to 강민숙's smiling portrait as it appears distorted and partially melted; used as The fixed content plane on which the facial hallucination occurs; 식탁 (holding the photograph); used as A narrow surrounding strip preserves the photograph's real-world scale and placement.\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 daylight remains desaturated and naturalistic while the facial deformation carries the oppressive visual emphasis 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young still has the doll-adorned backpack she took out before entering the bathroom; the apparent distortion of Minsuk's face is only a transient hallucination and leaves the photograph unchanged.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 수리영의 주관적 시점, 사진 속 민숙(강민숙)의 환하게 웃는 얼굴 표면이 기괴하게 일그러진 채 반쯤 흘러내린 끔찍한 환각의 정지 순간.\n\nLOCATION (lock): At the dining table in the rooftop apartment's living area, where a family photograph is displayed. 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 수리영's exact subjective eye point, hold squarely on the photograph while cropping to the mother's side so the distorted face becomes the human primary subject and the POV character's pictured likeness remains outside the frame. Keep the photograph within a natural portion of the composition, with its edges and a narrow band of the table still visible as the smiling facial surface appears grotesquely displaced downward.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 식탁 위 사진 (subjectively distorted in 수리영's hallucination) — The printed front face is square to the camera, cropped to 강민숙's smiling portrait as it appears distorted and partially melted; used as The fixed content plane on which the facial hallucination occurs; 식탁 (holding the photograph); used as A narrow surrounding strip preserves the photograph's real-world scale and placement.\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 daylight remains desaturated and naturalistic while the facial deformation carries the oppressive visual emphasis 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young still has the doll-adorned backpack she took out before entering the bathroom; the apparent distortion of Minsuk's face is only a transient hallucination and leaves the photograph unchanged.\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 rigid, centered close-up with a restrained normal-lens perspective, keeping the printed plane, its visible edges, and the narrow strip of table crisply resolved. Let the subdued early-morning ambience fall evenly across the image so the grotesque downward displacement of the bright smile reads with clinical clarity against the photograph’s otherwise stable physical geometry.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 수리영의 주관적 시점, 사진 속 민숙(강민숙)의 환하게 웃는 얼굴 표면이 기괴하게 일그러진 채 반쯤 흘러내린 끔찍한 환각의 정지 순간.\n\nLOCATION (lock): At the dining table in the rooftop apartment's living area, where a family photograph is displayed. 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 수리영's exact subjective eye point, hold squarely on the photograph while cropping to the mother's side so the distorted face becomes the human primary subject and the POV character's pictured likeness remains outside the frame. Keep the photograph within a natural portion of the composition, with its edges and a narrow band of the table still visible as the smiling facial surface appears grotesquely displaced downward.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 식탁 위 사진 (subjectively distorted in 수리영's hallucination) — The printed front face is square to the camera, cropped to 강민숙's smiling portrait as it appears distorted and partially melted; used as The fixed content plane on which the facial hallucination occurs; 식탁 (holding the photograph); used as A narrow surrounding strip preserves the photograph's real-world scale and placement.\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 daylight remains desaturated and naturalistic while the facial deformation carries the oppressive visual emphasis 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young still has the doll-adorned backpack she took out before entering the bathroom; the apparent distortion of Minsuk's face is only a transient hallucination and leaves the photograph unchanged.\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 claustrophobic close-up with the distorted smiling face weighted strongly within the crop, using shallow optical falloff so the facial surface commands attention while the photograph edges and table strip recede softly without disappearing. Shape the existing desaturated morning daylight into a gentle oblique wash, allowing slight tonal falloff across the print to make the downward-sliding hallucination feel immediate and oppressive.\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": 8,
   "A": 3
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 8,
    "verdict_ko": "지시된 클로즈업 프레이밍과 식탁의 좁은 여백을 완벽하게 구현했으며, 환각에 의한 왜곡이 사진 평면 내에서 일어나는 점을 잘 표현했습니다."
   },
   {
    "label": "A",
    "score": 3,
    "verdict_ko": "프레이밍이 지정된 클로즈업보다 훨씬 넓으며, 2D 사진 속 인물이 3D 공간인 식탁 위로 흘러내리는 물리적 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L04B03.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "환각으로 일그러진 얼굴이 사진의 2D 인화면을 벗어나 식탁 위로 입체적으로 흘러내리며 사진의 테두리 바깥으로 튀어나왔습니다.",
     "fix_en": "Confine the distorted, melted facial features completely to the flat 2D printed surface within the borders of the photograph."
    },
    {
     "issue_ko": "가족 사진을 카메라 프레임으로 잘라내어 수리영의 모습이 화면 밖으로 나가게 해야 하나, 사진의 네 모서리가 모두 화면 안에 들어온 단독 인물 사진으로 묘사되었습니다.",
     "fix_en": "Extend the physical photograph off one side of the screen so the camera framing crops out the rest of the family photo, rather than showing a complete four-sided portrait."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Reposition the camera directly perpendicular to the photograph so its printed face is square to the lens, preserving everything else.",
     "issue_ko": "카메라가 사진을 비스듬한 상부 각도에서 내려다봐 사진 면이 사다리꼴 원근으로 보이며, 사진 정면을 정사각으로 바라보라는 주관적 시점 구도와 다릅니다."
    },
    {
     "fix_en": "Move in tighter so the distorted smiling face dominates the close-up while leaving only the photograph edges and a narrow surrounding strip of table visible.",
     "issue_ko": "사진 주변의 식탁 면적이 지나치게 넓어 왜곡된 얼굴의 화면 비중이 계약에서 요구한 폐쇄적인 클로즈업보다 작습니다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Confine the distorted, melted facial features completely to the flat 2D printed surface within the borders of the photograph.\n- Extend the physical photograph off one side of the screen so the camera framing crops out the rest of the family photo, rather than showing a complete four-sided portrait.\n- Reposition the camera directly perpendicular to the photograph so its printed face is square to the lens, preserving everything else.\n- Move in tighter so the distorted smiling face dominates the close-up while leaving only the photograph edges and a narrow surrounding strip of table visible.\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 대로 사진이 카메라 정면(squarely)을 향하고 있으며, 왜곡된 인물과 식탁의 비율이 요청된 프레이밍(클로즈업)에 잘 부합합니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "사진의 일그러진 환각 효과는 잘 표현되었으나, 카메라 앵글이 정면(squarely)이 아닌 사선에서 촬영되어 지시사항을 다소 벗어났습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지시된 대로 사진이 카메라 정면(squarely)을 향하고 있으며, 왜곡된 인물과 식탁의 비율이 요청된 프레이밍(클로즈업)에 잘 부합합니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "사진의 일그러진 환각 효과는 잘 표현되었으나, 카메라 앵글이 정면(squarely)이 아닌 사선에서 촬영되어 지시사항을 다소 벗어났습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요청된 클로즈업 샷과 '피사체 정면 응시(squarely on the photograph)' 구도를 더 정확히 구현하여 환각의 질감이 잘 강조되었습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "사진의 질감과 왜곡된 얼굴은 잘 표현되었으나, 카메라 앵글이 비스듬하여 정면 구도(squarely) 지침에서 약간 벗어납니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요청된 클로즈업 샷과 '피사체 정면 응시(squarely on the photograph)' 구도를 더 정확히 구현하여 환각의 질감이 잘 강조되었습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "사진의 질감과 왜곡된 얼굴은 잘 표현되었으나, 카메라 앵글이 비스듬하여 정면 구도(squarely) 지침에서 약간 벗어납니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 10,
     "B": 14
    },
    "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": "L04B01",
    "B": "L04B02",
    "C": "L04B03",
    "D": "L04B04",
    "E": "L04B05",
    "F": "L04B06",
    "G": "L04B07"
   },
   "assigned": "L04B03",
   "choice": "Candidate C",
   "confident": true,
   "reason_ko": "지문에서 '사진 속'을 바라보는 시점을 묘사하며, 로케이션 설명에 따라 식탁(dining table)이 있는 공간이어야 하므로 현재 할당된 컷을 유지함.",
   "kept": "L04B03"
  }
 },
 "S19sh4::bgfirst_bg": {
  "input_fingerprint": "c2313c40b6e4f74f",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 마트 밖, 고개를 든 채 건물 천장 모서리에 매달린 CCTV 카메라를 올려다보는 수리영의 뒷모습.\n\nLOCATION (lock): Outside the supermarket, directly beneath a CCTV camera mounted high on an exterior corner of the building.\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 behind 수리영 at upper-torso height, complete the upward tilt with her back and craned head occupying the lower center while the mounted CCTV camera remains small but distinct in the upper-right corner. Hold both in the same medium-wide frame so her eyeline forms a clean diagonal toward the surveillance unit before the camera begins its later descent and arc.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the lower-center of the frame, midground, looks toward overhead CCTV camera; overhead CCTV camera in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: CCTV 카메라 (mounted at the building corner) — Its lens-bearing front is angled down from the building's upper corner toward the exterior area; used as The overhead endpoint of 수리영's gaze and the evidence she is evaluating; 마트 건물 외부 모서리 (supporting the mounted CCTV camera) — The adjoining faces converge above and to the right of 수리영; used as Provides the high corner that spatially fixes the CCTV above 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: Natural early-morning exterior light is kept restrained and desaturated, with grounded contrast between 수리영 and the building corner.\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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 마트 밖, 고개를 든 채 건물 천장 모서리에 매달린 CCTV 카메라를 올려다보는 수리영의 뒷모습.\n\nLOCATION (lock): Outside the supermarket, directly beneath a CCTV camera mounted high on an exterior corner of the building.\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 behind 수리영 at upper-torso height, complete the upward tilt with her back and craned head occupying the lower center while the mounted CCTV camera remains small but distinct in the upper-right corner. Hold both in the same medium-wide frame so her eyeline forms a clean diagonal toward the surveillance unit before the camera begins its later descent and arc.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the lower-center of the frame, midground, looks toward overhead CCTV camera; overhead CCTV camera in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: CCTV 카메라 (mounted at the building corner) — Its lens-bearing front is angled down from the building's upper corner toward the exterior area; used as The overhead endpoint of 수리영's gaze and the evidence she is evaluating; 마트 건물 외부 모서리 (supporting the mounted CCTV camera) — The adjoining faces converge above and to the right of 수리영; used as Provides the high corner that spatially fixes the CCTV above 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: Natural early-morning exterior light is kept restrained and desaturated, with grounded contrast between 수리영 and the building corner.\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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S19sh4__bgfirst_bg.png",
  "asset_id": "5be0fcaf-aabf-431f-88b1-408f5beb6739",
  "input_asset_ids": [
   "c87a928a-f60f-45f2-930b-791a8ae1e548",
   "951c4005-c33a-470a-86e2-8b52360a937b"
  ]
 },
 "S19sh4": {
  "input_fingerprint": "6018a5ac16a0e6c5",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 밖, 고개를 든 채 건물 천장 모서리에 매달린 CCTV 카메라를 올려다보는 수리영의 뒷모습.\n\nLOCATION (lock): Outside the supermarket, directly beneath a CCTV camera mounted high on an exterior corner of the building. 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 behind 수리영 at upper-torso height, complete the upward tilt with her back and craned head occupying the lower center while the mounted CCTV camera remains small but distinct in the upper-right corner. Hold both in the same medium-wide frame so her eyeline forms a clean diagonal toward the surveillance unit before the camera begins its later descent and arc.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the lower-center of the frame, midground, looks toward overhead CCTV camera; overhead CCTV camera in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: CCTV 카메라 (mounted at the building corner) — Its lens-bearing front is angled down from the building's upper corner toward the exterior area; used as The overhead endpoint of 수리영's gaze and the evidence she is evaluating; 마트 건물 외부 모서리 (supporting the mounted CCTV camera) — The adjoining faces converge above and to the right of 수리영; used as Provides the high corner that spatially fixes the CCTV above 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: Natural early-morning exterior light is kept restrained and desaturated, with grounded contrast between 수리영 and the building corner.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young continues to carry her doll-adorned backpack while investigating Minsuk's disappearance.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 밖, 고개를 든 채 건물 천장 모서리에 매달린 CCTV 카메라를 올려다보는 수리영의 뒷모습.\n\nLOCATION (lock): Outside the supermarket, directly beneath a CCTV camera mounted high on an exterior corner of the building. 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 behind 수리영 at upper-torso height, complete the upward tilt with her back and craned head occupying the lower center while the mounted CCTV camera remains small but distinct in the upper-right corner. Hold both in the same medium-wide frame so her eyeline forms a clean diagonal toward the surveillance unit before the camera begins its later descent and arc.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the lower-center of the frame, midground, looks toward overhead CCTV camera; overhead CCTV camera in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: CCTV 카메라 (mounted at the building corner) — Its lens-bearing front is angled down from the building's upper corner toward the exterior area; used as The overhead endpoint of 수리영's gaze and the evidence she is evaluating; 마트 건물 외부 모서리 (supporting the mounted CCTV camera) — The adjoining faces converge above and to the right of 수리영; used as Provides the high corner that spatially fixes the CCTV above 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: Natural early-morning exterior light is kept restrained and desaturated, with grounded contrast between 수리영 and the building corner.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young continues to carry her doll-adorned backpack while investigating Minsuk's disappearance.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 밖, 고개를 든 채 건물 천장 모서리에 매달린 CCTV 카메라를 올려다보는 수리영의 뒷모습.\n\nLOCATION (lock): Outside the supermarket, directly beneath a CCTV camera mounted high on an exterior corner of the building. 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 behind 수리영 at upper-torso height, complete the upward tilt with her back and craned head occupying the lower center while the mounted CCTV camera remains small but distinct in the upper-right corner. Hold both in the same medium-wide frame so her eyeline forms a clean diagonal toward the surveillance unit before the camera begins its later descent and arc.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the lower-center of the frame, midground, looks toward overhead CCTV camera; overhead CCTV camera in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: CCTV 카메라 (mounted at the building corner) — Its lens-bearing front is angled down from the building's upper corner toward the exterior area; used as The overhead endpoint of 수리영's gaze and the evidence she is evaluating; 마트 건물 외부 모서리 (supporting the mounted CCTV camera) — The adjoining faces converge above and to the right of 수리영; used as Provides the high corner that spatially fixes the CCTV above 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: Natural early-morning exterior light is kept restrained and desaturated, with grounded contrast between 수리영 and the building corner.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young continues to carry her doll-adorned backpack while investigating Minsuk's disappearance.\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 초반, 긴 검은 머리, 타원형의 젊은 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S19sh4__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S19sh4.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L09B01.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "요구된 상체 높이의 프레이밍과 대각선 시선, 인형이 달린 가방을 정확히 묘사했으며 로케이션의 건축 구조를 완벽하게 유지했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "카메라 높이가 지시된 '상체'보다 낮으며, 우측 벽면에 레퍼런스에 없는 커다란 창문을 임의로 생성하여 로케이션 고정 규칙을 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "요구된 상체 높이의 프레이밍과 대각선 시선, 인형이 달린 가방을 정확히 묘사했으며 로케이션의 건축 구조를 완벽하게 유지했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "카메라 높이가 지시된 '상체'보다 낮으며, 우측 벽면에 레퍼런스에 없는 커다란 창문을 임의로 생성하여 로케이션 고정 규칙을 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 상향 틸트(로우 앵글)를 적용하여 인물과 CCTV의 대각선 구도를 훌륭하게 연출했으며, 가방의 인형 디테일도 잘 살렸습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "프롬프트의 상향 틸트 지시를 무시하고 장소 레퍼런스 사진의 카메라 구도를 거의 그대로 모방하는 금지 사항을 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 상향 틸트(로우 앵글)를 적용하여 인물과 CCTV의 대각선 구도를 훌륭하게 연출했으며, 가방의 인형 디테일도 잘 살렸습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "프롬프트의 상향 틸트 지시를 무시하고 장소 레퍼런스 사진의 카메라 구도를 거의 그대로 모방하는 금지 사항을 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 11,
     "B": 11
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": false,
    "policy": 1
   }
  },
  "totals": {
   "A": 11,
   "B": 11
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 8,
    "verdict_ko": "요구된 상체 높이의 프레이밍과 대각선 시선, 인형이 달린 가방을 정확히 묘사했으며 로케이션의 건축 구조를 완벽하게 유지했습니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "카메라 높이가 지시된 '상체'보다 낮으며, 우측 벽면에 레퍼런스에 없는 커다란 창문을 임의로 생성하여 로케이션 고정 규칙을 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L09B01.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": []
  },
  "composition_critique": {
   "issues": []
  },
  "fix_skipped": true,
  "bgfirst": {
   "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S19sh4__bgfirst_bg.png",
   "bg_asset_id": "5be0fcaf-aabf-431f-88b1-408f5beb6739",
   "bg_record_key": "S19sh4::bgfirst_bg",
   "chain_winner": true,
   "authority": "plate"
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S19sh6::variants": {
  "author_fp": "d4f436d2cdf31c9a",
  "author": {
   "variants": [
    {
     "approach_ko": "얕은 심도와 정면 대칭에 가까운 구도로 미세한 표정과 통화의 긴장을 고립시키는 초근접 관찰.",
     "prompt_en": "Use a tightly compressed, intimate close-up with an almost symmetrical frontal composition, holding Suri-young’s face at the middle-center as the dominant visual mass. Keep the focus critically precise on her furrowed brow and eyes fixed beyond the lens, with the phone’s back and side edge sharply present against her ear; let her hair, the far cheek plane, and the supermarket exterior dissolve progressively into soft early-morning depth. Favor restrained contrast, cool desaturated skin tones, and delicate natural texture so her frozen concentration feels isolated and solemn."
    },
    {
     "approach_ko": "조금 더 공간감 있는 관찰형 클로즈업으로 얼굴과 휴대전화의 물리적 압박, 흐릿한 마트 외관의 방향성을 함께 살린다.",
     "prompt_en": "Photograph the close-up with a more immediate, spatially present lens feel, still directly frontal and at eye level, allowing the side of the phone, the hand holding it firmly to her face, and the full contour of her head to shape the frame without weakening the facial emphasis. Use the softened lines and tonal blocks of the supermarket exterior as subdued depth behind her, while maintaining clear separation from the background rather than reducing it to abstraction. Let the natural early-morning light rake gently across her features, revealing the tension between her knitted brows and motionless distant gaze with sober, realistic contrast."
    }
   ]
  },
  "reused": false
 },
 "S19sh6": {
  "input_fingerprint": "66451c3404378fa3",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 휴대전화를 귀에 바짝 댄 수리영이 미간을 찌푸린 채 먼 곳을 응시하는 정지된 얼굴.\n\nLOCATION (lock): Outside the supermarket near the exterior CCTV coverage area in daylight. 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: Settle directly before 수리영 at eye level in a close facial frame, allowing her face and the phone pressed to her ear to occupy most of the image while the store exterior falls soft behind her. Her eyes look past the lens rather than into it, turning the frontal camera position into a tense observation of her distant concentration.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 휴대전화 (in use during a call to 혜수) — Its back and side edge face the camera while its speaker is pressed to 수리영's ear; used as Held against the side of her face while she explains the CCTV discovery; 마트 외부 (open for early-morning business) — The building face lies behind 수리영 and outside the primary focus plane; used as Retained only as softened exterior context behind the close portrait.\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 natural early-morning light keeps the close portrait desaturated and solemn without adding an unestablished artificial 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\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 daytime market exterior, facade materials, overhead CCTV placement, and natural light. Preserve the young woman's clothing from the continuing inspection. Exclude the earlier rear-view pose and the male employee, and do not force the CCTV camera into the tight facial crop.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young has her mobile phone and doll-adorned backpack with her as she calls Hye-soo from 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 초반, 긴 검은 머리, 타원형의 젊은 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 휴대전화를 귀에 바짝 댄 수리영이 미간을 찌푸린 채 먼 곳을 응시하는 정지된 얼굴.\n\nLOCATION (lock): Outside the supermarket near the exterior CCTV coverage area in daylight. 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: Settle directly before 수리영 at eye level in a close facial frame, allowing her face and the phone pressed to her ear to occupy most of the image while the store exterior falls soft behind her. Her eyes look past the lens rather than into it, turning the frontal camera position into a tense observation of her distant concentration.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 휴대전화 (in use during a call to 혜수) — Its back and side edge face the camera while its speaker is pressed to 수리영's ear; used as Held against the side of her face while she explains the CCTV discovery; 마트 외부 (open for early-morning business) — The building face lies behind 수리영 and outside the primary focus plane; used as Retained only as softened exterior context behind the close portrait.\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 natural early-morning light keeps the close portrait desaturated and solemn without adding an unestablished artificial 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\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 daytime market exterior, facade materials, overhead CCTV placement, and natural light. Preserve the young woman's clothing from the continuing inspection. Exclude the earlier rear-view pose and the male employee, and do not force the CCTV camera into the tight facial crop.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young has her mobile phone and doll-adorned backpack with her as she calls Hye-soo from 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 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a tightly compressed, intimate close-up with an almost symmetrical frontal composition, holding Suri-young’s face at the middle-center as the dominant visual mass. Keep the focus critically precise on her furrowed brow and eyes fixed beyond the lens, with the phone’s back and side edge sharply present against her ear; let her hair, the far cheek plane, and the supermarket exterior dissolve progressively into soft early-morning depth. Favor restrained contrast, cool desaturated skin tones, and delicate natural texture so her frozen concentration feels isolated and solemn.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 휴대전화를 귀에 바짝 댄 수리영이 미간을 찌푸린 채 먼 곳을 응시하는 정지된 얼굴.\n\nLOCATION (lock): Outside the supermarket near the exterior CCTV coverage area in daylight. 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: Settle directly before 수리영 at eye level in a close facial frame, allowing her face and the phone pressed to her ear to occupy most of the image while the store exterior falls soft behind her. Her eyes look past the lens rather than into it, turning the frontal camera position into a tense observation of her distant concentration.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 휴대전화 (in use during a call to 혜수) — Its back and side edge face the camera while its speaker is pressed to 수리영's ear; used as Held against the side of her face while she explains the CCTV discovery; 마트 외부 (open for early-morning business) — The building face lies behind 수리영 and outside the primary focus plane; used as Retained only as softened exterior context behind the close portrait.\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 natural early-morning light keeps the close portrait desaturated and solemn without adding an unestablished artificial 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\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 daytime market exterior, facade materials, overhead CCTV placement, and natural light. Preserve the young woman's clothing from the continuing inspection. Exclude the earlier rear-view pose and the male employee, and do not force the CCTV camera into the tight facial crop.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young has her mobile phone and doll-adorned backpack with her as she calls Hye-soo from 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 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the close-up with a more immediate, spatially present lens feel, still directly frontal and at eye level, allowing the side of the phone, the hand holding it firmly to her face, and the full contour of her head to shape the frame without weakening the facial emphasis. Use the softened lines and tonal blocks of the supermarket exterior as subdued depth behind her, while maintaining clear separation from the background rather than reducing it to abstraction. Let the natural early-morning light rake gently across her features, revealing the tension between her knitted brows and motionless distant gaze with sober, realistic 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."
  },
  "totals": {
   "B": 7,
   "A": 5
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "지정된 샷의 클로즈업 구도와 피사체의 표정을 정확히 구현했으며, 배경의 마트 외부 구조물도 레퍼런스와 일치합니다."
   },
   {
    "label": "A",
    "score": 5,
    "verdict_ko": "표정과 구도는 좋으나, 가방 끈 없이 인형이 후드티에 직접 붙어 있는 형태적 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S19sh4_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "이전 장면에 이어 백팩을 메고 있어야 하나, 프레임에 보이는 어깨에 백팩 끈이 없습니다.",
     "fix_en": "Add the backpack's shoulder straps over her shoulders on top of the hoodie."
    }
   ]
  },
  "composition_critique": {
   "issues": []
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add the backpack's shoulder straps over her shoulders on top of the hoodie.\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 클로즈업 프레이밍, 시선 처리, 배경 묘사를 정확히 따랐으며, 설정된 백팩의 어깨끈까지 프레임 내에 충실히 반영했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "대부분의 지침을 잘 따랐으나, 함께 명시된 백팩의 어깨끈이 생략되어 유지 설정에서 감점되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 클로즈업 프레이밍, 시선 처리, 배경 묘사를 정확히 따랐으며, 설정된 백팩의 어깨끈까지 프레임 내에 충실히 반영했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "대부분의 지침을 잘 따랐으나, 함께 명시된 백팩의 어깨끈이 생략되어 유지 설정에서 감점되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "요청된 클로즈업 프레임, 인물의 표정과 시선, 이전 샷과 일치하는 배경 및 백팩 착용 상태(어깨 끈)를 모두 충실하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "대부분의 요구사항을 잘 따랐으나, 프레임 내에 보여야 할 백팩 어깨 끈이 누락되어 이전 샷과의 연결성이 다소 떨어집니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "요청된 클로즈업 프레임, 인물의 표정과 시선, 이전 샷과 일치하는 배경 및 백팩 착용 상태(어깨 끈)를 모두 충실하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "대부분의 요구사항을 잘 따랐으나, 프레임 내에 보여야 할 백팩 어깨 끈이 누락되어 이전 샷과의 연결성이 다소 떨어집니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 13,
     "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": "S19sh4"
  }
 },
 "S20sh2::bgfirst_bg": {
  "input_fingerprint": "5b360c4e027ffb9a",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 모니터 화면의 푸른빛이 혜수의 굳은 뺨을 비추는 측면 구도.\n\nLOCATION (lock): Inside the supermarket's administrative security office, beside the CCTV monitor whose blue screen light falls across the viewers.\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: Track laterally beside the workstation at seated eye level, then settle into a near-static close profile of 혜수 on the right, her cheek held between the camera and the monitor's blue cast. Keep a narrow oblique slice of the active monitor at the left edge so its unusually dark CCTV segment remains the source of her fixed attention.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: CCTV 모니터 (on and replaying the dark CCTV segment) — The active screen face is seen obliquely from the side, showing the exceptionally dark and unidentifiable portion of the footage; used as An oblique edge remains visible beside 혜수 and supplies the screen light across her cheek; 작업대 (in use for CCTV review) — Its near edge runs laterally beneath 혜수 and the monitor; used as Forms the lower spatial base beneath the close profile and marks the camera's tracking route.\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 monitor's blue light grazes 혜수's rigid cheek within an otherwise desaturated, low-to-mid-key interior treatment.\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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 모니터 화면의 푸른빛이 혜수의 굳은 뺨을 비추는 측면 구도.\n\nLOCATION (lock): Inside the supermarket's administrative security office, beside the CCTV monitor whose blue screen light falls across the viewers.\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: Track laterally beside the workstation at seated eye level, then settle into a near-static close profile of 혜수 on the right, her cheek held between the camera and the monitor's blue cast. Keep a narrow oblique slice of the active monitor at the left edge so its unusually dark CCTV segment remains the source of her fixed attention.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: CCTV 모니터 (on and replaying the dark CCTV segment) — The active screen face is seen obliquely from the side, showing the exceptionally dark and unidentifiable portion of the footage; used as An oblique edge remains visible beside 혜수 and supplies the screen light across her cheek; 작업대 (in use for CCTV review) — Its near edge runs laterally beneath 혜수 and the monitor; used as Forms the lower spatial base beneath the close profile and marks the camera's tracking route.\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 monitor's blue light grazes 혜수's rigid cheek within an otherwise desaturated, low-to-mid-key interior treatment.\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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S20sh2__bgfirst_bg.png",
  "asset_id": "a0f5da44-2b89-4a24-aa15-2a8edd25dcb9",
  "input_asset_ids": [
   "25e4c6fd-9d88-4458-86ee-a6a9ec761d75",
   "a53ce1a0-e024-4afd-97ce-cd02a0efd85f"
  ]
 },
 "S20sh2": {
  "input_fingerprint": "457d56386caf3aec",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 모니터 화면의 푸른빛이 혜수의 굳은 뺨을 비추는 측면 구도.\n\nLOCATION (lock): Inside the supermarket's administrative security office, beside the CCTV monitor whose blue screen light falls across the viewers. 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 laterally beside the workstation at seated eye level, then settle into a near-static close profile of 혜수 on the right, her cheek held between the camera and the monitor's blue cast. Keep a narrow oblique slice of the active monitor at the left edge so its unusually dark CCTV segment remains the source of her fixed attention.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: CCTV 모니터 (on and replaying the dark CCTV segment) — The active screen face is seen obliquely from the side, showing the exceptionally dark and unidentifiable portion of the footage; used as An oblique edge remains visible beside 혜수 and supplies the screen light across her cheek; 작업대 (in use for CCTV review) — Its near edge runs laterally beneath 혜수 and the monitor; used as Forms the lower spatial base beneath the close profile and marks the camera's tracking route.\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 monitor's blue light grazes 혜수's rigid cheek within an otherwise desaturated, low-to-mid-key interior treatment.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young remains at the CCTV review with her doll-adorned 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: 혜수 (한국인 여성, 40대 중반, 단정한 검은 머리, 성숙한 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 모니터 화면의 푸른빛이 혜수의 굳은 뺨을 비추는 측면 구도.\n\nLOCATION (lock): Inside the supermarket's administrative security office, beside the CCTV monitor whose blue screen light falls across the viewers. 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 laterally beside the workstation at seated eye level, then settle into a near-static close profile of 혜수 on the right, her cheek held between the camera and the monitor's blue cast. Keep a narrow oblique slice of the active monitor at the left edge so its unusually dark CCTV segment remains the source of her fixed attention.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: CCTV 모니터 (on and replaying the dark CCTV segment) — The active screen face is seen obliquely from the side, showing the exceptionally dark and unidentifiable portion of the footage; used as An oblique edge remains visible beside 혜수 and supplies the screen light across her cheek; 작업대 (in use for CCTV review) — Its near edge runs laterally beneath 혜수 and the monitor; used as Forms the lower spatial base beneath the close profile and marks the camera's tracking route.\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 monitor's blue light grazes 혜수's rigid cheek within an otherwise desaturated, low-to-mid-key interior treatment.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young remains at the CCTV review with her doll-adorned 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: 혜수 (한국인 여성, 40대 중반, 단정한 검은 머리, 성숙한 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 모니터 화면의 푸른빛이 혜수의 굳은 뺨을 비추는 측면 구도.\n\nLOCATION (lock): Inside the supermarket's administrative security office, beside the CCTV monitor whose blue screen light falls across the viewers. 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 laterally beside the workstation at seated eye level, then settle into a near-static close profile of 혜수 on the right, her cheek held between the camera and the monitor's blue cast. Keep a narrow oblique slice of the active monitor at the left edge so its unusually dark CCTV segment remains the source of her fixed attention.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: CCTV 모니터 (on and replaying the dark CCTV segment) — The active screen face is seen obliquely from the side, showing the exceptionally dark and unidentifiable portion of the footage; used as An oblique edge remains visible beside 혜수 and supplies the screen light across her cheek; 작업대 (in use for CCTV review) — Its near edge runs laterally beneath 혜수 and the monitor; used as Forms the lower spatial base beneath the close profile and marks the camera's tracking route.\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 monitor's blue light grazes 혜수's rigid cheek within an otherwise desaturated, low-to-mid-key interior treatment.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young remains at the CCTV review with her doll-adorned 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: 혜수 (한국인 여성, 40대 중반, 단정한 검은 머리, 성숙한 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S20sh2__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S20sh2.png"
    },
    {
     "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:735029>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L16B01.png"
    },
    {
     "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:735029>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지시된 클로즈업 샷 크기와 측면 구도를 정확히 구현했으며, 화면 왼쪽 가장자리에 모니터를 좁게 배치하여 뺨에 푸른빛이 떨어지는 조명 묘사가 훌륭합니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "요구된 클로즈업 앵글을 무시하고 상반신과 책상 전체가 보이는 미디엄 샷으로 렌더링하여 프레이밍 지침을 크게 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지시된 클로즈업 샷 크기와 측면 구도를 정확히 구현했으며, 화면 왼쪽 가장자리에 모니터를 좁게 배치하여 뺨에 푸른빛이 떨어지는 조명 묘사가 훌륭합니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "요구된 클로즈업 앵글을 무시하고 상반신과 책상 전체가 보이는 미디엄 샷으로 렌더링하여 프레이밍 지침을 크게 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "지시된 클로즈업 샷 크기를 완벽히 준수하였으며, 화면 좌측에 모니터 일부만 걸치고 인물의 측면에 푸른빛이 닿게 하라는 구도 지시를 정확하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "지시된 클로즈업을 무시하고 프레임을 넓혀 인물의 상반신과 가방까지 보여줌으로써 가장 중요한 카메라 및 구도 지시를 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지시된 클로즈업 샷 크기를 완벽히 준수하였으며, 화면 좌측에 모니터 일부만 걸치고 인물의 측면에 푸른빛이 닿게 하라는 구도 지시를 정확하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "지시된 클로즈업을 무시하고 프레임을 넓혀 인물의 상반신과 가방까지 보여줌으로써 가장 중요한 카메라 및 구도 지시를 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 9,
     "B": 18
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 9,
   "B": 18
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 9,
    "verdict_ko": "지시된 클로즈업 샷 크기와 측면 구도를 정확히 구현했으며, 화면 왼쪽 가장자리에 모니터를 좁게 배치하여 뺨에 푸른빛이 떨어지는 조명 묘사가 훌륭합니다."
   },
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "요구된 클로즈업 앵글을 무시하고 상반신과 책상 전체가 보이는 미디엄 샷으로 렌더링하여 프레이밍 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L16B01.png"
   },
   {
    "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:735029>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "모니터 화면 좌측 상단과 하단에 텍스트와 UI가 포함되어 있어 '텍스트 없음' 지시를 위반했습니다.",
     "fix_en": "Remove all text and UI elements from the monitor screen."
    },
    {
     "issue_ko": "왼쪽 가장자리의 모니터가 '좁고 비스듬한 조각(narrow oblique slice)'으로 보이지 않고 화면의 넓은 면이 정면에 가깝게 노출되어 있습니다.",
     "fix_en": "Adjust the monitor on the left so only a narrow, oblique slice of its screen is visible at the edge."
    },
    {
     "issue_ko": "모니터에 띄워진 CCTV 영상이 매우 어둡고 식별 불가능해야 한다는 지시와 달리, 밝고 뚜렷한 방 내부를 보여줍니다.",
     "fix_en": "Darken the CCTV footage on the monitor so the scene is exceptionally dark and unidentifiable."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Reduce the monitor to a narrow oblique slice confined to the left edge while preserving Hyesu’s close profile on the right and her gaze toward the screen.",
     "issue_ko": "왼쪽 CCTV 모니터가 화면의 약 4분의 1을 차지해, 계약에서 요구한 ‘왼쪽 가장자리의 좁고 비스듬한 일부’보다 지나치게 크게 보인다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all text and UI elements from the monitor screen.\n- Adjust the monitor on the left so only a narrow, oblique slice of its screen is visible at the edge.\n- Darken the CCTV footage on the monitor so the scene is exceptionally dark and unidentifiable.\n- Reduce the monitor to a narrow oblique slice confined to the left edge while preserving Hyesu’s close profile on the right and her gaze toward the 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": 10,
      "verdict_ko": "프롬프트에 명시된 '비정상적으로 어둡고 식별할 수 없는' CCTV 화면의 세부 사항을 완벽하게 묘사하여 지시사항을 정확히 이행했습니다."
     },
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "전반적인 구도와 조명은 훌륭하지만, 모니터 화면이 프롬프트가 요구한 '매우 어두운 CCTV 구간'에 비해 너무 밝고 형태가 뚜렷하게 보입니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "프롬프트에 명시된 '비정상적으로 어둡고 식별할 수 없는' CCTV 화면의 세부 사항을 완벽하게 묘사하여 지시사항을 정확히 이행했습니다."
     },
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "전반적인 구도와 조명은 훌륭하지만, 모니터 화면이 프롬프트가 요구한 '매우 어두운 CCTV 구간'에 비해 너무 밝고 형태가 뚜렷하게 보입니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "지시된 클로즈업 측면 구도, 모니터의 푸른 조명, 혜수의 외모와 배경 요소를 정확하고 사실적으로 구현했습니다."
     },
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "A와 시각적으로 동일하며, 프롬프트가 요구한 카메라 앵글, 인물의 표정, 조명 디테일을 모두 훌륭하게 충족합니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지시된 클로즈업 측면 구도, 모니터의 푸른 조명, 혜수의 외모와 배경 요소를 정확하고 사실적으로 구현했습니다."
     },
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "A와 시각적으로 동일하며, 프롬프트가 요구한 카메라 앵글, 인물의 표정, 조명 디테일을 모두 훌륭하게 충족합니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 17,
     "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S20sh2__bgfirst_bg.png",
   "bg_asset_id": "a0f5da44-2b89-4a24-aa15-2a8edd25dcb9",
   "bg_record_key": "S20sh2::bgfirst_bg",
   "chain_winner": false,
   "authority": "plate"
  },
  "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S20sh6::variants": {
  "author_fp": "2edd7ea784bb1034",
  "author": {
   "variants": [
    {
     "approach_ko": "얕은 심도로 보안직원의 멈춘 옆얼굴을 선명하게 고립시키고 혜수와 작업대를 부드러운 후경으로 연결한다.",
     "prompt_en": "Use a restrained shallow depth of field, holding the security employee’s arrested side profile in precise focus while Hye-soo remains recognizable but gently softened across the workstation. Let the workstation edge form a quiet diagonal bridge through the softer plane, with subtle monitor spill shaping his cheek and jaw against subdued office falloff. Preserve natural skin texture and an unembellished, solemn stillness."
    },
    {
     "approach_ko": "두 인물과 작업대의 관계를 함께 읽히는 깊은 초점의 압축된 정적 구도로 담는다.",
     "prompt_en": "Photograph the exchange as a compressed two-plane tableau with enough depth of field for both the security employee’s profile and Hye-soo’s answering gaze to remain clearly legible. Emphasize the workstation as a firm horizontal structure separating and linking them, while the angled active monitor stays visually subordinate. Keep the illumination even but low-key, allowing restrained contrast and desaturated physical textures to carry the gravity of the agreement."
    }
   ]
  },
  "reused": false
 },
 "S20sh6": {
  "input_fingerprint": "4f1dc1f7d1dc1b34",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 한국인 남성 보안직원이 혜수를 향해 고개를 끄덕인 채 멈춘 측면.\n\nLOCATION (lock): At the CCTV workstation inside the supermarket's administrative security office, lit in part by the monitor screens. 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 seated eye level beside 한국인 남성 보안직원, hold his close side profile in the left foreground at the arrested endpoint of a nod, with 혜수 retained across the workstation in the right midground. The static composition uses the shared work surface to connect his quiet agreement to her request without changing the established room axis.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 한국인 남성 보안직원 in the middle-left of the frame, foreground, looks toward Hyesu across the workstation; 혜수 in the middle-right of the frame, midground, looks toward security employee; shared workstation in the lower-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: CCTV 작업대 (in use during the CCTV review) — Its edge extends from the foreground employee toward 혜수 across the frame; used as The shared anchor separating and visually linking the employee's nod with 혜수's request; CCTV 모니터 (on) — The active screen face is angled away from the camera and toward the seated figures, with its footage not emphasized in this shot; used as Remains secondary beside the exchange as evidence of the footage they have reviewed.\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, desaturated interior illumination holds the nod in solemn low-to-mid-key contrast without assigning an unestablished light color or 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\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young remains behind Hye-soo and the security employee with her doll-adorned backpack as the same dark CCTV segment is reviewed.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 한국인 남성 보안직원 (한국인 남성, 성인, 단정한 짧은 머리, 평범한 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 한국인 남성 보안직원이 혜수를 향해 고개를 끄덕인 채 멈춘 측면.\n\nLOCATION (lock): At the CCTV workstation inside the supermarket's administrative security office, lit in part by the monitor screens. 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 seated eye level beside 한국인 남성 보안직원, hold his close side profile in the left foreground at the arrested endpoint of a nod, with 혜수 retained across the workstation in the right midground. The static composition uses the shared work surface to connect his quiet agreement to her request without changing the established room axis.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 한국인 남성 보안직원 in the middle-left of the frame, foreground, looks toward Hyesu across the workstation; 혜수 in the middle-right of the frame, midground, looks toward security employee; shared workstation in the lower-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: CCTV 작업대 (in use during the CCTV review) — Its edge extends from the foreground employee toward 혜수 across the frame; used as The shared anchor separating and visually linking the employee's nod with 혜수's request; CCTV 모니터 (on) — The active screen face is angled away from the camera and toward the seated figures, with its footage not emphasized in this shot; used as Remains secondary beside the exchange as evidence of the footage they have reviewed.\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, desaturated interior illumination holds the nod in solemn low-to-mid-key contrast without assigning an unestablished light color or 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\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young remains behind Hye-soo and the security employee with her doll-adorned backpack as the same dark CCTV segment is reviewed.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 한국인 남성 보안직원 (한국인 남성, 성인, 단정한 짧은 머리, 평범한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained shallow depth of field, holding the security employee’s arrested side profile in precise focus while Hye-soo remains recognizable but gently softened across the workstation. Let the workstation edge form a quiet diagonal bridge through the softer plane, with subtle monitor spill shaping his cheek and jaw against subdued office falloff. Preserve natural skin texture and an unembellished, solemn 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 한국인 남성 보안직원이 혜수를 향해 고개를 끄덕인 채 멈춘 측면.\n\nLOCATION (lock): At the CCTV workstation inside the supermarket's administrative security office, lit in part by the monitor screens. 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 seated eye level beside 한국인 남성 보안직원, hold his close side profile in the left foreground at the arrested endpoint of a nod, with 혜수 retained across the workstation in the right midground. The static composition uses the shared work surface to connect his quiet agreement to her request without changing the established room axis.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 한국인 남성 보안직원 in the middle-left of the frame, foreground, looks toward Hyesu across the workstation; 혜수 in the middle-right of the frame, midground, looks toward security employee; shared workstation in the lower-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: CCTV 작업대 (in use during the CCTV review) — Its edge extends from the foreground employee toward 혜수 across the frame; used as The shared anchor separating and visually linking the employee's nod with 혜수's request; CCTV 모니터 (on) — The active screen face is angled away from the camera and toward the seated figures, with its footage not emphasized in this shot; used as Remains secondary beside the exchange as evidence of the footage they have reviewed.\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, desaturated interior illumination holds the nod in solemn low-to-mid-key contrast without assigning an unestablished light color or 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\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young remains behind Hye-soo and the security employee with her doll-adorned backpack as the same dark CCTV segment is reviewed.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 한국인 남성 보안직원 (한국인 남성, 성인, 단정한 짧은 머리, 평범한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the exchange as a compressed two-plane tableau with enough depth of field for both the security employee’s profile and Hye-soo’s answering gaze to remain clearly legible. Emphasize the workstation as a firm horizontal structure separating and linking them, while the angled active monitor stays visually subordinate. Keep the illumination even but low-key, allowing restrained contrast and desaturated physical textures to carry the gravity of the agreement.\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 미디엄 샷과 좌측 전경의 보안직원, 고개를 끄덕이는 순간의 디테일을 정확히 구현했으며 이전 샷과의 연속성도 훌륭합니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "카메라가 너무 뒤로 물러나 프레이밍이 넓어졌고, 우측 전경에 불필요한 모니터가 추가되어 화면 구성 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S20sh2_sel.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "보안직원의 시선이 혜수를 향하지 않고 아래를 내려다보고 있습니다.",
     "fix_en": "Direct the male security guard's gaze forward so he is looking at Hyesu."
    },
    {
     "issue_ko": "화면 왼쪽 상단 배경의 벽에 붙은 종이에 표와 텍스트가 있습니다.",
     "fix_en": "Remove all text and grid lines from the papers pinned to the wall in the top left."
    },
    {
     "issue_ko": "샷 텍스트에 없는 어린이가 배경에 임의로 추가되었으며, 이는 레퍼런스(가방과 인형만 있음)와도 일치하지 않습니다.",
     "fix_en": "Remove the child in the background, leaving only the doll and backpack resting on the chair as seen in the reference image."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Raise and turn the security employee’s gaze toward Hye-soo while preserving his arrested nod, side profile, and all other framing.",
     "issue_ko": "보안직원의 시선이 혜수를 향하지 않고 아래쪽 작업대로 떨어져 있어, 두 사람이 서로 바라보도록 설정된 시선 축과 어긋난다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Direct the male security guard's gaze forward so he is looking at Hyesu.\n- Remove all text and grid lines from the papers pinned to the wall in the top left.\n- Remove the child in the background, leaving only the doll and backpack resting on the chair as seen in the reference image.\n- Raise and turn the security employee’s gaze toward Hye-soo while preserving his arrested nod, side profile, and all other framing.\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": "샷 텍스트에 명시된 구도와 인물 배치를 정확히 따랐으며, 이전 샷 레퍼런스에 맞춰 의자 위에 아이 없이 인형이 달린 가방만 일관되게 유지하여 지시사항을 완벽히 이행했습니다."
     },
     {
      "label": "A",
      "score": 2,
      "verdict_ko": "이전 샷 레퍼런스와 달리 의자에 앉아있는 어린아이(인물)를 임의로 추가하여, 샷 텍스트에 없는 인물을 등장시키지 말라는 제약을 어긴 하드 위반(Hard Violation)입니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "샷 텍스트에 명시된 구도와 인물 배치를 정확히 따랐으며, 이전 샷 레퍼런스에 맞춰 의자 위에 아이 없이 인형이 달린 가방만 일관되게 유지하여 지시사항을 완벽히 이행했습니다."
     },
     {
      "label": "A",
      "score": 2,
      "verdict_ko": "이전 샷 레퍼런스와 달리 의자에 앉아있는 어린아이(인물)를 임의로 추가하여, 샷 텍스트에 없는 인물을 등장시키지 말라는 제약을 어긴 하드 위반(Hard Violation)입니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "제시된 샷 텍스트의 구도와 인물 배치를 정확히 구현했으며, 레퍼런스와 동일하게 의자 위에 인형 가방만 남겨두어 샷 텍스트에 없는 인물의 등장을 금지한 규칙을 완벽히 준수했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "샷 텍스트에 명시되지 않은 어린아이를 배경에 임의로 추가하여 '샷 텍스트에 없는 인물 추가 금지'라는 강력한 지침과 이전 샷(레퍼런스)과의 일관성을 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "제시된 샷 텍스트의 구도와 인물 배치를 정확히 구현했으며, 레퍼런스와 동일하게 의자 위에 인형 가방만 남겨두어 샷 텍스트에 없는 인물의 등장을 금지한 규칙을 완벽히 준수했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "샷 텍스트에 명시되지 않은 어린아이를 배경에 임의로 추가하여 '샷 텍스트에 없는 인물 추가 금지'라는 강력한 지침과 이전 샷(레퍼런스)과의 일관성을 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 8,
     "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": "S20sh2"
  }
 },
 "S21sh3::bgfirst_bg": {
  "input_fingerprint": "17453574d753b68d",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 텅 빈 사무실 한가운데 책상 앞에 덩그러니 앉아있는 수리영의 뒷모습.\n\nLOCATION (lock): At a detective's desk in the middle of the police office, after the other officers have left for a meeting.\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: From the office entrance axis, hold a static wide frame at 수리영's seated shoulder height, directly behind her. She remains small and centered in the midground with the desk ahead, while vacant workstations spread across both sides to make the empty office outweigh her figure.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 혜수의 책상 (살인사건 확대 사진들이 어지럽게 놓여 있음) — The working side and desktop face 수리영 and remain visible beyond her shoulders; used as Forms the central destination of the rear-axis composition; 사무실 내 작업 공간 (형사들이 모두 빠져나가 비어 있음) — Vacated desks and seating are seen from the entrance side; used as Extends the negative space on both sides of 수리영; 살인사건 확대 사진들 (여러 장이 어지럽게 펼쳐져 있음) — The printed image sides face upward on the desktop, seen obliquely from behind 수리영; used as Creates scattered content across the desk without pulling focus from 수리영's isolation.\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 ambient illumination appropriate to the daytime office is restrained and desaturated, with low-to-mid-key contrast preserving a solemn sense of vacancy.\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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 텅 빈 사무실 한가운데 책상 앞에 덩그러니 앉아있는 수리영의 뒷모습.\n\nLOCATION (lock): At a detective's desk in the middle of the police office, after the other officers have left for a meeting.\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: From the office entrance axis, hold a static wide frame at 수리영's seated shoulder height, directly behind her. She remains small and centered in the midground with the desk ahead, while vacant workstations spread across both sides to make the empty office outweigh her figure.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 혜수의 책상 (살인사건 확대 사진들이 어지럽게 놓여 있음) — The working side and desktop face 수리영 and remain visible beyond her shoulders; used as Forms the central destination of the rear-axis composition; 사무실 내 작업 공간 (형사들이 모두 빠져나가 비어 있음) — Vacated desks and seating are seen from the entrance side; used as Extends the negative space on both sides of 수리영; 살인사건 확대 사진들 (여러 장이 어지럽게 펼쳐져 있음) — The printed image sides face upward on the desktop, seen obliquely from behind 수리영; used as Creates scattered content across the desk without pulling focus from 수리영's isolation.\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 ambient illumination appropriate to the daytime office is restrained and desaturated, with low-to-mid-key contrast preserving a solemn sense of vacancy.\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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S21sh3__bgfirst_bg.png",
  "asset_id": "7d05c644-ddeb-4fb3-9458-2c36bdb52301",
  "input_asset_ids": [
   "29285c8f-ec57-45cf-a7f6-834b21c842c4",
   "c70d5404-e782-42f5-9171-23bd635197c5"
  ]
 },
 "S21sh3": {
  "input_fingerprint": "8a0efb6241f9ec01",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 사무실 한가운데 책상 앞에 덩그러니 앉아있는 수리영의 뒷모습.\n\nLOCATION (lock): At a detective's desk in the middle of the police office, after the other officers have left for a meeting. 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 office entrance axis, hold a static wide frame at 수리영's seated shoulder height, directly behind her. She remains small and centered in the midground with the desk ahead, while vacant workstations spread across both sides to make the empty office outweigh her figure.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 혜수의 책상 (살인사건 확대 사진들이 어지럽게 놓여 있음) — The working side and desktop face 수리영 and remain visible beyond her shoulders; used as Forms the central destination of the rear-axis composition; 사무실 내 작업 공간 (형사들이 모두 빠져나가 비어 있음) — Vacated desks and seating are seen from the entrance side; used as Extends the negative space on both sides of 수리영; 살인사건 확대 사진들 (여러 장이 어지럽게 펼쳐져 있음) — The printed image sides face upward on the desktop, seen obliquely from behind 수리영; used as Creates scattered content across the desk without pulling focus from 수리영's isolation.\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 ambient illumination appropriate to the daytime office is restrained and desaturated, with low-to-mid-key contrast preserving a solemn sense of vacancy.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young is left alone in the office with her doll-adorned backpack, while the enlarged murder-scene photographs remain scattered across Hye-soo's desk.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 사무실 한가운데 책상 앞에 덩그러니 앉아있는 수리영의 뒷모습.\n\nLOCATION (lock): At a detective's desk in the middle of the police office, after the other officers have left for a meeting. 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 office entrance axis, hold a static wide frame at 수리영's seated shoulder height, directly behind her. She remains small and centered in the midground with the desk ahead, while vacant workstations spread across both sides to make the empty office outweigh her figure.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 혜수의 책상 (살인사건 확대 사진들이 어지럽게 놓여 있음) — The working side and desktop face 수리영 and remain visible beyond her shoulders; used as Forms the central destination of the rear-axis composition; 사무실 내 작업 공간 (형사들이 모두 빠져나가 비어 있음) — Vacated desks and seating are seen from the entrance side; used as Extends the negative space on both sides of 수리영; 살인사건 확대 사진들 (여러 장이 어지럽게 펼쳐져 있음) — The printed image sides face upward on the desktop, seen obliquely from behind 수리영; used as Creates scattered content across the desk without pulling focus from 수리영's isolation.\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 ambient illumination appropriate to the daytime office is restrained and desaturated, with low-to-mid-key contrast preserving a solemn sense of vacancy.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young is left alone in the office with her doll-adorned backpack, while the enlarged murder-scene photographs remain scattered across Hye-soo's desk.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 사무실 한가운데 책상 앞에 덩그러니 앉아있는 수리영의 뒷모습.\n\nLOCATION (lock): At a detective's desk in the middle of the police office, after the other officers have left for a meeting. 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 office entrance axis, hold a static wide frame at 수리영's seated shoulder height, directly behind her. She remains small and centered in the midground with the desk ahead, while vacant workstations spread across both sides to make the empty office outweigh her figure.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 혜수의 책상 (살인사건 확대 사진들이 어지럽게 놓여 있음) — The working side and desktop face 수리영 and remain visible beyond her shoulders; used as Forms the central destination of the rear-axis composition; 사무실 내 작업 공간 (형사들이 모두 빠져나가 비어 있음) — Vacated desks and seating are seen from the entrance side; used as Extends the negative space on both sides of 수리영; 살인사건 확대 사진들 (여러 장이 어지럽게 펼쳐져 있음) — The printed image sides face upward on the desktop, seen obliquely from behind 수리영; used as Creates scattered content across the desk without pulling focus from 수리영's isolation.\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 ambient illumination appropriate to the daytime office is restrained and desaturated, with low-to-mid-key contrast preserving a solemn sense of vacancy.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young is left alone in the office with her doll-adorned backpack, while the enlarged murder-scene photographs remain scattered across Hye-soo's desk.\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 초반, 긴 검은 머리, 타원형의 젊은 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S21sh3__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S21sh3.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L13B04.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "레퍼런스 사진의 공간 구조와 집기들을 완벽하게 재현했으며, 지정된 앵글과 흩어진 사건 사진들의 배치를 정확히 구현함."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 앵글과 피사체의 배치는 양호하나, 원본 로케이션에 없는 책상을 왼쪽에 추가하고 모니터가 사건 사진의 비중을 가림."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "레퍼런스 사진의 공간 구조와 집기들을 완벽하게 재현했으며, 지정된 앵글과 흩어진 사건 사진들의 배치를 정확히 구현함."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 앵글과 피사체의 배치는 양호하나, 원본 로케이션에 없는 책상을 왼쪽에 추가하고 모니터가 사건 사진의 비중을 가림."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "카메라 구도, 텅 빈 사무실의 적막한 분위기, 의자에 걸린 인형 달린 배낭과 책상 위에 흩어진 사건 사진 등 모든 지시 사항이 완벽하게 구현되었습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "전반적인 구도와 배경의 형태는 나쁘지 않으나, 책상의 모니터 배치와 바닥에 놓인 배낭이 지시된 미장센의 완성도 면에서 B에 비해 다소 떨어집니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "카메라 구도, 텅 빈 사무실의 적막한 분위기, 의자에 걸린 인형 달린 배낭과 책상 위에 흩어진 사건 사진 등 모든 지시 사항이 완벽하게 구현되었습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "전반적인 구도와 배경의 형태는 나쁘지 않으나, 책상의 모니터 배치와 바닥에 놓인 배낭이 지시된 미장센의 완성도 면에서 B에 비해 다소 떨어집니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 19,
     "B": 14
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 19,
   "B": 14
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 10,
    "verdict_ko": "레퍼런스 사진의 공간 구조와 집기들을 완벽하게 재현했으며, 지정된 앵글과 흩어진 사건 사진들의 배치를 정확히 구현함."
   },
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "지정된 앵글과 피사체의 배치는 양호하나, 원본 로케이션에 없는 책상을 왼쪽에 추가하고 모니터가 사건 사진의 비중을 가림."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L13B04.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "백팩 상단에 텍스트가 포함되어 있습니다.",
     "fix_en": "Remove the text from the upper section of the backpack."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Scale down Suri-young and her seated setup slightly, anchoring the chair near its current bottom position so her head and shoulders sit lower and occupy less of the frame.",
     "issue_ko": "수리영과 앉아 있는 구성이 레이아웃보다 화면에서 너무 크게 보여 머리와 어깨가 지나치게 위까지 올라와 있으며, 중경의 작고 고립된 인물 비중이 약해졌다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the text from the upper section of the backpack.\n- Scale down Suri-young and her seated setup slightly, anchoring the chair near its current bottom position so her head and shoulders sit lower and occupy less 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": 10,
      "verdict_ko": "지정된 원본 배경, 레이아웃 스케치, 인물 설정 및 백팩의 디테일까지 완벽하게 반영하여 의도된 샷을 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "전체적인 구도와 지침은 잘 따랐으나, 의자 하단 베이스 구조에 물리적으로 불가능한 형태의 왜곡이 발생했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "지정된 원본 배경, 레이아웃 스케치, 인물 설정 및 백팩의 디테일까지 완벽하게 반영하여 의도된 샷을 정확히 구현했습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "전체적인 구도와 지침은 잘 따랐으나, 의자 하단 베이스 구조에 물리적으로 불가능한 형태의 왜곡이 발생했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 앵글과 텅 빈 사무실 배경을 완벽히 유지하며, 인물과 가방 디테일이 자연스럽게 표현되었습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "구도와 배경 유지는 우수하나, 가방에 지시되지 않은 의미 없는 텍스트가 생성되어 감점되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 앵글과 텅 빈 사무실 배경을 완벽히 유지하며, 인물과 가방 디테일이 자연스럽게 표현되었습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "구도와 배경 유지는 우수하나, 가방에 지시되지 않은 의미 없는 텍스트가 생성되어 감점되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 16,
     "B": 15
    },
    "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S21sh3__bgfirst_bg.png",
   "bg_asset_id": "7d05c644-ddeb-4fb3-9458-2c36bdb52301",
   "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": "L13B01",
    "B": "L13B02",
    "C": "L13B03",
    "D": "L13B04",
    "E": "L13B05",
    "F": "L13B06"
   },
   "assigned": "L13B04",
   "choice": "Candidate D",
   "confident": true,
   "reason_ko": "\"사무실 한가운데 책상 앞\"이라는 묘사에 사무실 내부 책상들이 배치된 현재 할당 이미지가 잘 부합하며, 다른 후보로 변경할 명확한 공간적/시간적 근거가 텍스트에 없습니다.",
   "kept": "L13B04"
  }
 },
 "S21sh5::variants": {
  "author_fp": "0895a95c92f764dc",
  "author": {
   "variants": [
    {
     "approach_ko": "얕은 심도로 굳어버린 얼굴과 아래로 향한 공포의 시선을 고립시키는 밀착형 클로즈업.",
     "prompt_en": "Use an intimate, gently compressed close-up with the camera held steady at eye level and just off frontal, isolating her rigid face against a softly dissolved office. Keep both widened, downward-looking eyes and the minute tension around her lids, jaw, and lips critically sharp, while the oblique desk edge and the narrow strip of upward-facing photographs fall quickly out of focus along the bottom. Shape the existing neutral late-morning interior ambience into restrained facial modeling, with subdued color and natural skin texture emphasizing her arrested terror."
    },
    {
     "approach_ko": "비스듬한 책상과 사진 띠를 전경 단서로 삼아 얼굴까지 긴장을 끌어올리는 근접 원근형 클로즈업.",
     "prompt_en": "Favor a close-focus perspective that lets the oblique desk boundary and thin lower-edge band of photographs act as a restrained foreground vector rising toward her face, while still allowing her face to fill the frame from the required slight three-quarter angle. Hold enough depth for the photographic cue to remain faintly legible as physical prints, but make her enlarged downward gaze the decisive point of focus; crop tightly so her frozen expression feels pinned inside the frame. Keep the neutral daytime office light broad and controlled, using modest contrast and desaturated surfaces for a stark, observational realism."
    }
   ]
  },
  "reused": false
 },
 "S21sh5": {
  "input_fingerprint": "5c55c574d8e54a65",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 사진을 내려다보는 수리영의 두 눈이 공포로 크게 확장된 채 딱딱하게 굳어버린 얼굴 클로즈업.\n\nLOCATION (lock): At the detective's desk inside the otherwise empty police office, over crime-scene photographs spread across its surface. 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 수리영's eye level, complete the tracking arc by settling into a face-filling close-up from a slight three-quarter angle. Her rigid face occupies most of the frame, with her enlarged downward gaze held sharp and only a narrow, softened strip of photographs remaining along the lower edge.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 살인사건 확대 사진들 (책상 위에 펼쳐져 있음) — Their printed faces lie upward toward 수리영 and are only partially visible to the camera at the bottom edge; used as Appears as a narrow lower-frame cue for the source of her terror; 책상 가장자리 (사진들이 놓여 있음) — Its front edge crosses the lower frame at an oblique angle; used as Provides a restrained lower boundary beneath the close-up.\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 daytime-interior ambience is kept desaturated and controlled, allowing modest facial contrast to harden 수리영's terrified expression 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\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 daylight police office, desk area, institutional materials, and unattended atmosphere. Preserve the young woman's clothing and the crime-scene photographs spread across the desk. Exclude the preceding wide rear view and any officers from this tight facial reaction shot.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The gruesome enlarged murder photographs remain spread across Hye-soo's desk as Suri-young examines them; her doll-adorned backpack stays with her.\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 초반, 긴 검은 머리, 타원형의 젊은 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 사진을 내려다보는 수리영의 두 눈이 공포로 크게 확장된 채 딱딱하게 굳어버린 얼굴 클로즈업.\n\nLOCATION (lock): At the detective's desk inside the otherwise empty police office, over crime-scene photographs spread across its surface. 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 수리영's eye level, complete the tracking arc by settling into a face-filling close-up from a slight three-quarter angle. Her rigid face occupies most of the frame, with her enlarged downward gaze held sharp and only a narrow, softened strip of photographs remaining along the lower edge.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 살인사건 확대 사진들 (책상 위에 펼쳐져 있음) — Their printed faces lie upward toward 수리영 and are only partially visible to the camera at the bottom edge; used as Appears as a narrow lower-frame cue for the source of her terror; 책상 가장자리 (사진들이 놓여 있음) — Its front edge crosses the lower frame at an oblique angle; used as Provides a restrained lower boundary beneath the close-up.\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 daytime-interior ambience is kept desaturated and controlled, allowing modest facial contrast to harden 수리영's terrified expression 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\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 daylight police office, desk area, institutional materials, and unattended atmosphere. Preserve the young woman's clothing and the crime-scene photographs spread across the desk. Exclude the preceding wide rear view and any officers from this tight facial reaction shot.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The gruesome enlarged murder photographs remain spread across Hye-soo's desk as Suri-young examines them; her doll-adorned backpack stays with her.\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 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an intimate, gently compressed close-up with the camera held steady at eye level and just off frontal, isolating her rigid face against a softly dissolved office. Keep both widened, downward-looking eyes and the minute tension around her lids, jaw, and lips critically sharp, while the oblique desk edge and the narrow strip of upward-facing photographs fall quickly out of focus along the bottom. Shape the existing neutral late-morning interior ambience into restrained facial modeling, with subdued color and natural skin texture emphasizing her arrested terror.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 사진을 내려다보는 수리영의 두 눈이 공포로 크게 확장된 채 딱딱하게 굳어버린 얼굴 클로즈업.\n\nLOCATION (lock): At the detective's desk inside the otherwise empty police office, over crime-scene photographs spread across its surface. 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 수리영's eye level, complete the tracking arc by settling into a face-filling close-up from a slight three-quarter angle. Her rigid face occupies most of the frame, with her enlarged downward gaze held sharp and only a narrow, softened strip of photographs remaining along the lower edge.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 살인사건 확대 사진들 (책상 위에 펼쳐져 있음) — Their printed faces lie upward toward 수리영 and are only partially visible to the camera at the bottom edge; used as Appears as a narrow lower-frame cue for the source of her terror; 책상 가장자리 (사진들이 놓여 있음) — Its front edge crosses the lower frame at an oblique angle; used as Provides a restrained lower boundary beneath the close-up.\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 daytime-interior ambience is kept desaturated and controlled, allowing modest facial contrast to harden 수리영's terrified expression 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\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 daylight police office, desk area, institutional materials, and unattended atmosphere. Preserve the young woman's clothing and the crime-scene photographs spread across the desk. Exclude the preceding wide rear view and any officers from this tight facial reaction shot.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The gruesome enlarged murder photographs remain spread across Hye-soo's desk as Suri-young examines them; her doll-adorned backpack stays with her.\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 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFavor a close-focus perspective that lets the oblique desk boundary and thin lower-edge band of photographs act as a restrained foreground vector rising toward her face, while still allowing her face to fill the frame from the required slight three-quarter angle. Hold enough depth for the photographic cue to remain faintly legible as physical prints, but make her enlarged downward gaze the decisive point of focus; crop tightly so her frozen expression feels pinned inside the frame. Keep the neutral daytime office light broad and controlled, using modest contrast and desaturated surfaces for a stark, observational realism.\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": 10,
   "B": 8
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 10,
    "verdict_ko": "지시된 클로즈업 프레이밍과 공포에 질린 표정을 완벽하게 구현했으며, 레퍼런스와 일치하는 책상 재질과 흐릿하게 처리된 사진 표현이 정확합니다."
   },
   {
    "label": "B",
    "score": 8,
    "verdict_ko": "표정과 프레이밍은 훌륭하나, 책상의 재질이 레퍼런스의 회색톤과 다른 나무 재질로 묘사되었고 전경의 사진이 지시된 것보다 선명합니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S21sh3_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": []
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Tighten and recenter the close-up so her rigid face occupies most of the frame, leaving only minimal softened office background around it.",
     "issue_ko": "얼굴이 화면 오른쪽에 치우쳐 있고 왼쪽 배경 여백이 넓어, 얼굴이 프레임 대부분을 채우는 클로즈업 구도보다 느슨하다."
    },
    {
     "fix_en": "Reduce the visible tabletop and photographs to a narrow strip along the bottom edge, with the desk’s front edge crossing that strip at a clear oblique angle.",
     "issue_ko": "책상과 사진이 하단에서 지나치게 넓은 영역을 차지하며, 책상 앞쪽 가장자리가 하단을 사선으로 가로지르는 구도가 명확하지 않다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Tighten and recenter the close-up so her rigid face occupies most of the frame, leaving only minimal softened office background around it.\n- Reduce the visible tabletop and photographs to a narrow strip along the bottom edge, with the desk’s front edge crossing that strip at a clear oblique angle.\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": "지시된 4분의 3 측면 각도, 아래를 향한 시선, 비스듬하게 가로지르는 책상 가장자리를 정확하게 구현하여 프레이밍 조건에 완벽히 부합합니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "얼굴이 정면을 향하고 책상 가장자리가 수평으로 배치되어, 요구된 3/4 각도 및 비스듬한 하단 경계 구도 조건을 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지시된 4분의 3 측면 각도, 아래를 향한 시선, 비스듬하게 가로지르는 책상 가장자리를 정확하게 구현하여 프레이밍 조건에 완벽히 부합합니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "얼굴이 정면을 향하고 책상 가장자리가 수평으로 배치되어, 요구된 3/4 각도 및 비스듬한 하단 경계 구도 조건을 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "정면 구도로 촬영되어 프롬프트가 요구한 '약간의 4분의 3 측면 각도(slight three-quarter angle)' 지시를 따르지 않았으며, 아래를 향한 시선이 명확하지 않습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "요구된 4분의 3 측면 각도와 아래(사진)를 향하는 공포에 질린 시선, 타이트한 클로즈업 프레이밍을 가장 정확하게 구현했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "정면 구도로 촬영되어 프롬프트가 요구한 '약간의 4분의 3 측면 각도(slight three-quarter angle)' 지시를 따르지 않았으며, 아래를 향한 시선이 명확하지 않습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요구된 4분의 3 측면 각도와 아래(사진)를 향하는 공포에 질린 시선, 타이트한 클로즈업 프레이밍을 가장 정확하게 구현했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 15,
     "B": 11
    },
    "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": "S21sh3"
  }
 },
 "S21sh9::variants": {
  "author_fp": "42c43e680ddf87c0",
  "author": {
   "variants": [
    {
     "approach_ko": "정면 대칭과 깊은 초점으로 문턱의 경계와 빈 공간을 차갑고 긴장감 있게 강조한다.",
     "prompt_en": "Use a restrained, architectural composition with firm central symmetry and deep focus, keeping the doorway edges, her half-exited posture, and the vacant office surfaces crisply legible. Let the straight lines of the doorway pin her at the center while the empty interior holds as austere negative space; render the late-morning ambient light evenly, with subtle tonal separation across the threshold and cool, desaturated surface detail."
    },
    {
     "approach_ko": "얕은 초점과 밀착된 관찰감으로 몰래 주변을 살피는 수리영의 긴장된 몸짓에 집중한다.",
     "prompt_en": "Give the medium view an intimate, watchful lens feel with selective focus concentrated on her back, hair, clothing folds, and doll-adorned backpack, while the adjoining spaces fall gently softer without losing their physical identity. Emphasize the tentative weight shift of her half-concealed body and the narrow doorway boundary, using restrained late-morning ambience, soft falloff, and modest contrast to create quiet, immediate suspense."
    }
   ]
  },
  "reused": false
 },
 "S21sh9": {
  "input_fingerprint": "e0ee8c3f01ebf2c9",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 사무실 문 밖으로 반쯤 몸을 내민 채 주위를 훔쳐보는 수리영의 뒷모습.\n\nLOCATION (lock): At the open doorway of the empty police office, looking out into the adjoining station corridor. 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: Hold just inside the office at shoulder height, centered behind 수리영 on the doorway axis after the tracking move stops. Frame her back in a medium view with one half of her body beyond the threshold and the other half still inside, using the doorway to divide concealment from exposure.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: office doorway enclosing 수리영's half-exited body in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 사무실 출입문 (열려 있음) — The doorway is viewed straight along its opening axis from inside the office; used as Bisects 수리영's body and fixes the exit route at frame center; 사무실 내부 (아무도 없이 비어 있음); used as Remains behind 수리영 as negative space and confirms that nobody is watching.\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 ambient illumination appropriate to the daytime interior remains restrained and desaturated, with modest contrast across the doorway boundary.\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 daylight police office layout, desks, doorway, and empty unattended state. Preserve the young woman's clothing and the existing desk clutter. Exclude the seated pose and prominently displayed crime-scene photographs from the doorway-focused frame, and include no officers.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young carries the handwritten coordinate note folded in her pocket and keeps her doll-adorned backpack with her as she slips out.\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 초반, 긴 검은 머리, 타원형의 젊은 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 사무실 문 밖으로 반쯤 몸을 내민 채 주위를 훔쳐보는 수리영의 뒷모습.\n\nLOCATION (lock): At the open doorway of the empty police office, looking out into the adjoining station corridor. 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: Hold just inside the office at shoulder height, centered behind 수리영 on the doorway axis after the tracking move stops. Frame her back in a medium view with one half of her body beyond the threshold and the other half still inside, using the doorway to divide concealment from exposure.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: office doorway enclosing 수리영's half-exited body in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 사무실 출입문 (열려 있음) — The doorway is viewed straight along its opening axis from inside the office; used as Bisects 수리영's body and fixes the exit route at frame center; 사무실 내부 (아무도 없이 비어 있음); used as Remains behind 수리영 as negative space and confirms that nobody is watching.\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 ambient illumination appropriate to the daytime interior remains restrained and desaturated, with modest contrast across the doorway boundary.\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 daylight police office layout, desks, doorway, and empty unattended state. Preserve the young woman's clothing and the existing desk clutter. Exclude the seated pose and prominently displayed crime-scene photographs from the doorway-focused frame, and include no officers.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young carries the handwritten coordinate note folded in her pocket and keeps her doll-adorned backpack with her as she slips out.\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 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained, architectural composition with firm central symmetry and deep focus, keeping the doorway edges, her half-exited posture, and the vacant office surfaces crisply legible. Let the straight lines of the doorway pin her at the center while the empty interior holds as austere negative space; render the late-morning ambient light evenly, with subtle tonal separation across the threshold and cool, desaturated 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 사무실 문 밖으로 반쯤 몸을 내민 채 주위를 훔쳐보는 수리영의 뒷모습.\n\nLOCATION (lock): At the open doorway of the empty police office, looking out into the adjoining station corridor. 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: Hold just inside the office at shoulder height, centered behind 수리영 on the doorway axis after the tracking move stops. Frame her back in a medium view with one half of her body beyond the threshold and the other half still inside, using the doorway to divide concealment from exposure.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: office doorway enclosing 수리영's half-exited body in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 사무실 출입문 (열려 있음) — The doorway is viewed straight along its opening axis from inside the office; used as Bisects 수리영's body and fixes the exit route at frame center; 사무실 내부 (아무도 없이 비어 있음); used as Remains behind 수리영 as negative space and confirms that nobody is watching.\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 ambient illumination appropriate to the daytime interior remains restrained and desaturated, with modest contrast across the doorway boundary.\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 daylight police office layout, desks, doorway, and empty unattended state. Preserve the young woman's clothing and the existing desk clutter. Exclude the seated pose and prominently displayed crime-scene photographs from the doorway-focused frame, and include no officers.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young carries the handwritten coordinate note folded in her pocket and keeps her doll-adorned backpack with her as she slips out.\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 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nGive the medium view an intimate, watchful lens feel with selective focus concentrated on her back, hair, clothing folds, and doll-adorned backpack, while the adjoining spaces fall gently softer without losing their physical identity. Emphasize the tentative weight shift of her half-concealed body and the narrow doorway boundary, using restrained late-morning ambience, soft falloff, and modest contrast to create quiet, immediate suspense.\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": "지정된 앵글(사무실 내부에서 등 뒤를 잡는 구도)을 무시하고 측면 전신 샷으로 연출했으며, 요구된 배낭과 주머니 속 쪽지 또한 누락되었습니다."
   },
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "카메라 위치(사무실 내부에서 등 뒤를 바라보는 축), 문으로 이등분된 구도, 쪽지와 배낭 등 모든 주요 지시사항을 매우 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S21sh5_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "복도 배경의 파란색 간판과 뒷주머니에 꽂힌 종이에 금지된 텍스트가 노출되어 있습니다.",
     "fix_en": "Remove all text from the blue sign in the corridor background and from the paper in the back pocket."
    },
    {
     "issue_ko": "수리영이 배낭을 소지해야 한다는 지시와 달리, 인형이 달린 배낭이 문 손잡이에 걸려 있습니다.",
     "fix_en": "Move the doll-adorned backpack from the door onto Suri-young's back so she is carrying it."
    },
    {
     "issue_ko": "뒷주머니에 있는 쪽지가 접혀 있지(folded) 않고 넓게 펼쳐진 상태로 튀어나와 있습니다.",
     "fix_en": "Change the paper in the back pocket to be folded and tucked deeper inside."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Shift Suri-young left across the threshold so the doorway boundary bisects her torso, leaving approximately half of her body inside the office and half exposed in the corridor while preserving the camera and framing.",
     "issue_ko": "문틀 경계가 수리영의 몸을 반으로 가르지 않고 몸 대부분이 복도 쪽에 나가 있어, 반은 사무실 안에 숨고 반은 문밖에 노출되는 구도가 아니다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all text from the blue sign in the corridor background and from the paper in the back pocket.\n- Move the doll-adorned backpack from the door onto Suri-young's back so she is carrying it.\n- Change the paper in the back pocket to be folded and tucked deeper inside.\n- Shift Suri-young left across the threshold so the doorway boundary bisects her torso, leaving approximately half of her body inside the office and half exposed in the corridor while preserving the camera and framing.\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": "문틀이 인물의 몸을 반으로 나누는 구도를 정확히 구현했으며, 배낭과 쪽지 등 소품의 착용 상태도 자연스럽습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "인물의 몸이 문지방에 의해 양분되지 않았으며, 배낭이 허공에 떠 있는 심각한 물리적 오류가 있습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "문틀이 인물의 몸을 반으로 나누는 구도를 정확히 구현했으며, 배낭과 쪽지 등 소품의 착용 상태도 자연스럽습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "인물의 몸이 문지방에 의해 양분되지 않았으며, 배낭이 허공에 떠 있는 심각한 물리적 오류가 있습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "문의 경계를 활용해 몸의 절반만 드러내는 구도를 정확히 연출했으며, 인형 달린 가방을 메고 있는 설정도 충실히 반영했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "몸이 문에 의해 양분되지 않고 프레임 안에 온전히 노출되었으며, 가방을 메지 않고 문에 걸어두어 소지 상태 지시를 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "문의 경계를 활용해 몸의 절반만 드러내는 구도를 정확히 연출했으며, 인형 달린 가방을 메고 있는 설정도 충실히 반영했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "몸이 문에 의해 양분되지 않고 프레임 안에 온전히 노출되었으며, 가방을 메지 않고 문에 걸어두어 소지 상태 지시를 위반했습니다."
     }
    ]
   },
   "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": "S21sh5"
  }
 },
 "S22sh1::bgfirst_bg": {
  "input_fingerprint": "80dad9945944b33e",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 대낮의 경찰서 앞, 자전거 페달을 밟고 선 수리영의 상체가 앞으로 크게 쏠린 측면 구도.\n\nLOCATION (lock): On the pavement directly outside the police station in broad daylight, beside the bicycle parking area.\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: Track parallel to the bicycle at upper-torso height from a tight lateral position, holding 수리영 in profile as the environment slips behind her. Her pitched-forward torso occupies the central and upper frame while enough of the handlebar and pedaling leg remains visible below to preserve the force of acceleration.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 자전거 (주행 중) — Its lateral side faces the camera, with the handlebar and part of the pedal mechanism visible; used as Provides the moving support structure around 수리영's compressed riding posture; 경찰서 앞 공간 (대낮) — Seen laterally behind 수리영 as the camera moves parallel with her; used as Establishes the departure location behind the tracked rider.\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 ambient light is treated with a restrained, desaturated palette and grounded mid-level 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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 대낮의 경찰서 앞, 자전거 페달을 밟고 선 수리영의 상체가 앞으로 크게 쏠린 측면 구도.\n\nLOCATION (lock): On the pavement directly outside the police station in broad daylight, beside the bicycle parking area.\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: Track parallel to the bicycle at upper-torso height from a tight lateral position, holding 수리영 in profile as the environment slips behind her. Her pitched-forward torso occupies the central and upper frame while enough of the handlebar and pedaling leg remains visible below to preserve the force of acceleration.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 자전거 (주행 중) — Its lateral side faces the camera, with the handlebar and part of the pedal mechanism visible; used as Provides the moving support structure around 수리영's compressed riding posture; 경찰서 앞 공간 (대낮) — Seen laterally behind 수리영 as the camera moves parallel with her; used as Establishes the departure location behind the tracked rider.\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 ambient light is treated with a restrained, desaturated palette and grounded mid-level 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S22sh1__bgfirst_bg.png",
  "asset_id": "bfaf545e-df03-4354-baf4-8b3e72cf031e",
  "input_asset_ids": [
   "a50a5128-4361-48b9-be72-ab34a4a56bc2",
   "8e19dc34-8d4e-4793-b0ee-a0224c56a32f"
  ]
 },
 "S22sh1": {
  "input_fingerprint": "bcc792efa2fcabde",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 대낮의 경찰서 앞, 자전거 페달을 밟고 선 수리영의 상체가 앞으로 크게 쏠린 측면 구도.\n\nLOCATION (lock): On the pavement directly outside the police station in broad daylight, beside the bicycle parking 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: Track parallel to the bicycle at upper-torso height from a tight lateral position, holding 수리영 in profile as the environment slips behind her. Her pitched-forward torso occupies the central and upper frame while enough of the handlebar and pedaling leg remains visible below to preserve the force of acceleration.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 자전거 (주행 중) — Its lateral side faces the camera, with the handlebar and part of the pedal mechanism visible; used as Provides the moving support structure around 수리영's compressed riding posture; 경찰서 앞 공간 (대낮) — Seen laterally behind 수리영 as the camera moves parallel with her; used as Establishes the departure location behind the tracked rider.\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 ambient light is treated with a restrained, desaturated palette and grounded mid-level 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young departs on her bicycle with the coordinate note in her pocket and the doll-adorned backpack in her possession.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 대낮의 경찰서 앞, 자전거 페달을 밟고 선 수리영의 상체가 앞으로 크게 쏠린 측면 구도.\n\nLOCATION (lock): On the pavement directly outside the police station in broad daylight, beside the bicycle parking 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: Track parallel to the bicycle at upper-torso height from a tight lateral position, holding 수리영 in profile as the environment slips behind her. Her pitched-forward torso occupies the central and upper frame while enough of the handlebar and pedaling leg remains visible below to preserve the force of acceleration.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 자전거 (주행 중) — Its lateral side faces the camera, with the handlebar and part of the pedal mechanism visible; used as Provides the moving support structure around 수리영's compressed riding posture; 경찰서 앞 공간 (대낮) — Seen laterally behind 수리영 as the camera moves parallel with her; used as Establishes the departure location behind the tracked rider.\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 ambient light is treated with a restrained, desaturated palette and grounded mid-level 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young departs on her bicycle with the coordinate note in her pocket and the doll-adorned backpack in her possession.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning.\n\nSHOT TEXT (authoritative, Korean): 대낮의 경찰서 앞, 자전거 페달을 밟고 선 수리영의 상체가 앞으로 크게 쏠린 측면 구도.\n\nLOCATION (lock): On the pavement directly outside the police station in broad daylight, beside the bicycle parking 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: Track parallel to the bicycle at upper-torso height from a tight lateral position, holding 수리영 in profile as the environment slips behind her. Her pitched-forward torso occupies the central and upper frame while enough of the handlebar and pedaling leg remains visible below to preserve the force of acceleration.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 자전거 (주행 중) — Its lateral side faces the camera, with the handlebar and part of the pedal mechanism visible; used as Provides the moving support structure around 수리영's compressed riding posture; 경찰서 앞 공간 (대낮) — Seen laterally behind 수리영 as the camera moves parallel with her; used as Establishes the departure location behind the tracked rider.\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 ambient light is treated with a restrained, desaturated palette and grounded mid-level 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young departs on her bicycle with the coordinate note in her pocket and the doll-adorned backpack in her possession.\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 초반, 긴 검은 머리, 타원형의 젊은 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S22sh1__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S22sh1.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L13B01.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 로케이션 레퍼런스의 건축물과 자전거 보관소를 정확히 구현했으며, 요구된 측면 미디엄 샷과 상체가 쏠린 자세를 잘 표현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "자전거 핸들 부분의 구조가 물리적으로 어색하며, 배경의 경찰서 건물이 제공된 로케이션 레퍼런스와 전혀 일치하지 않습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 로케이션 레퍼런스의 건축물과 자전거 보관소를 정확히 구현했으며, 요구된 측면 미디엄 샷과 상체가 쏠린 자세를 잘 표현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "자전거 핸들 부분의 구조가 물리적으로 어색하며, 배경의 경찰서 건물이 제공된 로케이션 레퍼런스와 전혀 일치하지 않습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 프레이밍보다 다소 넓으나, '페달을 밟고 선' 핵심 동작을 정확히 표현했고 레퍼런스의 배경(자전거 거치대 및 건물 구조)을 완벽하게 재현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "타이트한 미디엄 샷의 프레이밍은 잘 따랐으나, 안장에 앉아 있어 지시된 동작을 누락했으며 경찰서 배경의 세부 형태가 레퍼런스와 다릅니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 프레이밍보다 다소 넓으나, '페달을 밟고 선' 핵심 동작을 정확히 표현했고 레퍼런스의 배경(자전거 거치대 및 건물 구조)을 완벽하게 재현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "타이트한 미디엄 샷의 프레이밍은 잘 따랐으나, 안장에 앉아 있어 지시된 동작을 누락했으며 경찰서 배경의 세부 형태가 레퍼런스와 다릅니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 15,
     "B": 7
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 15,
   "B": 7
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 8,
    "verdict_ko": "지정된 로케이션 레퍼런스의 건축물과 자전거 보관소를 정확히 구현했으며, 요구된 측면 미디엄 샷과 상체가 쏠린 자세를 잘 표현했습니다."
   },
   {
    "label": "B",
    "score": 3,
    "verdict_ko": "자전거 핸들 부분의 구조가 물리적으로 어색하며, 배경의 경찰서 건물이 제공된 로케이션 레퍼런스와 전혀 일치하지 않습니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L13B01.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "수리영이 자전거 안장에 앉아 있습니다. 샷 텍스트('페달을 밟고 선')와 스케치의 자세에 맞춰 안장에서 엉덩이를 떼고 페달을 딛고 일어선 상태여야 합니다.",
     "fix_en": "Lift Suri-young's hips completely off the saddle so she is standing on the pedals while leaning forward."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Raise her hips clearly off the saddle into a standing-on-the-pedals acceleration posture while preserving her profile, forward lean, scale, and all other framing.",
     "issue_ko": "수리영이 페달 위에 일어서 가속하는 자세가 아니라 안장에 앉아 주행하는 자세로 보인다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Lift Suri-young's hips completely off the saddle so she is standing on the pedals while leaning forward.\n- Raise her hips clearly off the saddle into a standing-on-the-pedals acceleration posture while preserving her profile, forward lean, scale, and all other framing.\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": "주어진 구도와 캐릭터 레퍼런스를 매우 정확하게 구현했으며, 핸들을 쥔 손과 자전거 부품의 묘사가 자연스럽습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지침을 전반적으로 잘 따랐으나, 핸들을 잡은 손가락의 형태와 앞바구니의 구조 등 세부 묘사에서 약간의 왜곡이 보입니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "주어진 구도와 캐릭터 레퍼런스를 매우 정확하게 구현했으며, 핸들을 쥔 손과 자전거 부품의 묘사가 자연스럽습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지침을 전반적으로 잘 따랐으나, 핸들을 잡은 손가락의 형태와 앞바구니의 구조 등 세부 묘사에서 약간의 왜곡이 보입니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "주어진 스케치의 구도와 배경을 완벽하게 유지했으며, 인물의 이목구비와 디테일이 레퍼런스에 더 가깝게 구현되었습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "지시사항을 전반적으로 잘 따랐으나, 얼굴의 생김새가 레퍼런스와 약간 차이가 나며 핸들을 잡은 손의 묘사가 다소 뭉개졌습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "주어진 스케치의 구도와 배경을 완벽하게 유지했으며, 인물의 이목구비와 디테일이 레퍼런스에 더 가깝게 구현되었습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "지시사항을 전반적으로 잘 따랐으나, 얼굴의 생김새가 레퍼런스와 약간 차이가 나며 핸들을 잡은 손의 묘사가 다소 뭉개졌습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 14,
     "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S22sh1__bgfirst_bg.png",
   "bg_asset_id": "bfaf545e-df03-4354-baf4-8b3e72cf031e",
   "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"
  },
  "plate_select": {
   "candidates": {
    "A": "L13B01",
    "B": "L13B02",
    "C": "L13B03",
    "D": "L13B04",
    "E": "L13B05",
    "F": "L13B06"
   },
   "assigned": "L13B01",
   "choice": "Candidate A",
   "confident": true,
   "reason_ko": "\"대낮의 경찰서 앞\"이라는 묘사와 자전거가 등장하는 것으로 보아, 경찰서 건물 외부 및 거리가 배경이 되어야 합니다.",
   "kept": "L13B01"
  }
 },
 "S23sh1::variants": {
  "author_fp": "07fbc9df25a34fc0",
  "author": {
   "variants": [
    {
     "approach_ko": "정면 대칭과 균일한 선명도로 좌표 발견을 냉정하고 증거물처럼 포착한다.",
     "prompt_en": "Use a restrained, forensic visual treatment with exact frontal symmetry and a natural-perspective lens feel. Keep the display surface and narrow surrounding bezel evenly crisp, with controlled screen exposure, minimal glare, and neutral ambient illumination falling uniformly across the device. Render the deep-red accent with clean separation from the desaturated image, making the discovery feel precise and evidentiary."
    },
    {
     "approach_ko": "화면에만 시선을 밀착시키고 주변부를 어둡고 부드럽게 가라앉혀 목적지의 불길한 의미를 강조한다.",
     "prompt_en": "Use a more compressed, intimate lens feel while preserving the required frontal close-up and visible monitor border. Hold the illuminated display as the sole sharp plane, allowing the bezel to fall subtly softer toward the outer edges; let the surrounding ambient light recede gently so the screen seems to emerge from the subdued café interior. Give the red accent slight luminous density without bloom or graphic artificiality, creating a quiet sense of ominous destination."
    }
   ]
  },
  "reused": false
 },
 "S23sh1": {
  "input_fingerprint": "71e9bd7b0d73ce06",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): PC방 모니터 화면에 바다 위 특정 지점에 붉은 핀이 꽂힌 지도가 띄워져 있는 모습.\n\nLOCATION (lock): At a computer cubicle inside an internet café, with a mapped sea coordinate displayed on the monitor. 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: Place the camera directly in front of the monitor at seated eye level and hold a static close view with the device frame still visible around the display. Center the sea map and its red pin, keeping the screen within a natural proportion rather than cropping away all surrounding monitor edges.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: PC방 모니터 화면 (지도와 붉은 위치 핀이 표시되어 있음) — The active front face is square to the camera and shows a map centered on a specific point at sea with a red pin; used as Acts as the sole focal plane and presents the mapped destination through a digital display; 모니터 외곽 (화면이 켜져 있음) — The front bezel faces the camera directly around the displayed map; used as Provides a narrow physical border that identifies the image as device-mediated.\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 illumination remains neutral and desaturated, while the displayed red pin supplies the single controlled 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The location entered from Suri-young's coordinate note is displayed on the map, and her doll-adorned backpack remains with her for the printed maps.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): PC방 모니터 화면에 바다 위 특정 지점에 붉은 핀이 꽂힌 지도가 띄워져 있는 모습.\n\nLOCATION (lock): At a computer cubicle inside an internet café, with a mapped sea coordinate displayed on the monitor. 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: Place the camera directly in front of the monitor at seated eye level and hold a static close view with the device frame still visible around the display. Center the sea map and its red pin, keeping the screen within a natural proportion rather than cropping away all surrounding monitor edges.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: PC방 모니터 화면 (지도와 붉은 위치 핀이 표시되어 있음) — The active front face is square to the camera and shows a map centered on a specific point at sea with a red pin; used as Acts as the sole focal plane and presents the mapped destination through a digital display; 모니터 외곽 (화면이 켜져 있음) — The front bezel faces the camera directly around the displayed map; used as Provides a narrow physical border that identifies the image as device-mediated.\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 illumination remains neutral and desaturated, while the displayed red pin supplies the single controlled 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The location entered from Suri-young's coordinate note is displayed on the map, and her doll-adorned backpack remains with her for the printed maps.\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, forensic visual treatment with exact frontal symmetry and a natural-perspective lens feel. Keep the display surface and narrow surrounding bezel evenly crisp, with controlled screen exposure, minimal glare, and neutral ambient illumination falling uniformly across the device. Render the deep-red accent with clean separation from the desaturated image, making the discovery feel precise and evidentiary.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): PC방 모니터 화면에 바다 위 특정 지점에 붉은 핀이 꽂힌 지도가 띄워져 있는 모습.\n\nLOCATION (lock): At a computer cubicle inside an internet café, with a mapped sea coordinate displayed on the monitor. 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: Place the camera directly in front of the monitor at seated eye level and hold a static close view with the device frame still visible around the display. Center the sea map and its red pin, keeping the screen within a natural proportion rather than cropping away all surrounding monitor edges.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: PC방 모니터 화면 (지도와 붉은 위치 핀이 표시되어 있음) — The active front face is square to the camera and shows a map centered on a specific point at sea with a red pin; used as Acts as the sole focal plane and presents the mapped destination through a digital display; 모니터 외곽 (화면이 켜져 있음) — The front bezel faces the camera directly around the displayed map; used as Provides a narrow physical border that identifies the image as device-mediated.\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 illumination remains neutral and desaturated, while the displayed red pin supplies the single controlled 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The location entered from Suri-young's coordinate note is displayed on the map, and her doll-adorned backpack remains with her for the printed maps.\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 more compressed, intimate lens feel while preserving the required frontal close-up and visible monitor border. Hold the illuminated display as the sole sharp plane, allowing the bezel to fall subtly softer toward the outer edges; let the surrounding ambient light recede gently so the screen seems to emerge from the subdued café interior. Give the red accent slight luminous density without bloom or graphic artificiality, creating a quiet sense of ominous destination.\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": 0
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 9,
    "verdict_ko": "모니터 주변 베젤이 보이는 클로즈업 프레이밍과 바다 위 붉은 핀이 꽂힌 지도의 묘사가 지시사항과 정확히 일치하며, 금지된 텍스트 오버레이가 없습니다."
   },
   {
    "label": "B",
    "score": 0,
    "verdict_ko": "화면 하단에 프롬프트 텍스트가 캡션으로 그대로 생성되어 '텍스트 및 캡션 금지' 규정을 어긴 치명적인 위반(Hard Violation)이 발생했습니다."
   }
  ],
  "refs": [],
  "critique": {
   "issues": [
    {
     "issue_ko": "모니터 화면의 지도와 인터페이스에 텍스트가 포함되어 있어 텍스트 금지 규칙을 위반합니다.",
     "fix_en": "Remove all text, labels, and numbers from the map, search bar, and user interface on the monitor screen."
    }
   ]
  },
  "composition_critique": {
   "issues": []
  },
  "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 numbers from the map, search bar, and user interface on the monitor 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": "모니터 화면의 지도와 붉은 핀을 요구된 클로즈업 구도로 잘 담아냈으며, 화면 내 텍스트 생성을 최소화하여 '텍스트 금지' 지침에 더 잘 부합합니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "지도의 붉은 핀과 모니터 구도는 훌륭하게 구현되었으나, UI와 지도 상에 텍스트가 명확히 생성되어 '텍스트 금지' 지침을 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "모니터 화면의 지도와 붉은 핀을 요구된 클로즈업 구도로 잘 담아냈으며, 화면 내 텍스트 생성을 최소화하여 '텍스트 금지' 지침에 더 잘 부합합니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "지도의 붉은 핀과 모니터 구도는 훌륭하게 구현되었으나, UI와 지도 상에 텍스트가 명확히 생성되어 '텍스트 금지' 지침을 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지정된 정면 클로즈업 프레이밍과 모니터 안의 붉은 핀을 정확하게 구현했으며, 금지된 워터마크나 불필요한 텍스트 없이 깔끔하게 렌더링되었습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "구도와 붉은 핀의 묘사는 우수하나, 화면 우측 하단에 지침에서 금지한 워터마크/캡션 형태의 불필요한 텍스트가 생성되어 감점되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지정된 정면 클로즈업 프레이밍과 모니터 안의 붉은 핀을 정확하게 구현했으며, 금지된 워터마크나 불필요한 텍스트 없이 깔끔하게 렌더링되었습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "구도와 붉은 핀의 묘사는 우수하나, 화면 우측 하단에 지침에서 금지한 워터마크/캡션 형태의 불필요한 텍스트가 생성되어 감점되었습니다."
     }
    ]
   },
   "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": "플레이트만 (배경 전용)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S24sh1::bgfirst_bg": {
  "input_fingerprint": "56a0f50fc68c1ff6",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 번개 빛이 번쩍여 해안가 도로 위 자전거 안장에 앉아 한쪽 다리로 페달을 힘껏 밟고 있는 수리영의 역동적인 실루엣이 드러난 찰나.\n\nLOCATION (lock): On an exposed coastal road during a heavy thunderstorm, under lightning and dense rain clouds.\n\nTIME OF DAY (lock): day, stormy overcast.\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 bicycle-saddle height, track parallel from the side in a wide profile that holds 수리영 and the full bicycle against the coastal road. Time the frame to the lightning flash as one leg drives the pedal downward, exposing her dynamic silhouette without changing the lateral tracking axis.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 자전거 (빗속에서 주행 중) — The full lateral side faces the camera during the downward pedal stroke; used as Carries 수리영 across the frame as a complete side-profile silhouette; 해안가 도로 (세찬 비가 내리고 있음); used as Establishes the route beneath and behind the tracked bicycle; 번개 (번쩍이는 순간); used as Temporarily reveals the rider's silhouette; 먹구름과 세찬 비 (천둥 번개와 함께 지속됨); used as Reduces clarity across the wider environment.\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: A lightning flash briefly cuts through the rain-darkened, desaturated scene, producing a stark but grounded silhouette with otherwise low-key storm 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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 번개 빛이 번쩍여 해안가 도로 위 자전거 안장에 앉아 한쪽 다리로 페달을 힘껏 밟고 있는 수리영의 역동적인 실루엣이 드러난 찰나.\n\nLOCATION (lock): On an exposed coastal road during a heavy thunderstorm, under lightning and dense rain clouds.\n\nTIME OF DAY (lock): day, stormy overcast.\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 bicycle-saddle height, track parallel from the side in a wide profile that holds 수리영 and the full bicycle against the coastal road. Time the frame to the lightning flash as one leg drives the pedal downward, exposing her dynamic silhouette without changing the lateral tracking axis.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 자전거 (빗속에서 주행 중) — The full lateral side faces the camera during the downward pedal stroke; used as Carries 수리영 across the frame as a complete side-profile silhouette; 해안가 도로 (세찬 비가 내리고 있음); used as Establishes the route beneath and behind the tracked bicycle; 번개 (번쩍이는 순간); used as Temporarily reveals the rider's silhouette; 먹구름과 세찬 비 (천둥 번개와 함께 지속됨); used as Reduces clarity across the wider environment.\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: A lightning flash briefly cuts through the rain-darkened, desaturated scene, producing a stark but grounded silhouette with otherwise low-key storm 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S24sh1__bgfirst_bg.png",
  "asset_id": "5f9831d6-4e8d-473d-bbdc-cf9fed12f329",
  "input_asset_ids": [
   "518c9de9-0967-4f6a-948a-081a9f82f0c9",
   "20148f1f-f264-4442-8f6d-57b8ec8f2c2e"
  ]
 },
 "S24sh1": {
  "input_fingerprint": "99086e2f9c2b9d42",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, stormy overcast.\n\nSHOT TEXT (authoritative, Korean): 번개 빛이 번쩍여 해안가 도로 위 자전거 안장에 앉아 한쪽 다리로 페달을 힘껏 밟고 있는 수리영의 역동적인 실루엣이 드러난 찰나.\n\nLOCATION (lock): On an exposed coastal road during a heavy thunderstorm, under lightning and dense rain 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: At bicycle-saddle height, track parallel from the side in a wide profile that holds 수리영 and the full bicycle against the coastal road. Time the frame to the lightning flash as one leg drives the pedal downward, exposing her dynamic silhouette without changing the lateral tracking axis.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 자전거 (빗속에서 주행 중) — The full lateral side faces the camera during the downward pedal stroke; used as Carries 수리영 across the frame as a complete side-profile silhouette; 해안가 도로 (세찬 비가 내리고 있음); used as Establishes the route beneath and behind the tracked bicycle; 번개 (번쩍이는 순간); used as Temporarily reveals the rider's silhouette; 먹구름과 세찬 비 (천둥 번개와 함께 지속됨); used as Reduces clarity across the wider environment.\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: A lightning flash briefly cuts through the rain-darkened, desaturated scene, producing a stark but grounded silhouette with otherwise low-key storm 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young rides with the doll-adorned backpack containing the folded printed maps, while continuing to use her bicycle.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, stormy overcast.\n\nSHOT TEXT (authoritative, Korean): 번개 빛이 번쩍여 해안가 도로 위 자전거 안장에 앉아 한쪽 다리로 페달을 힘껏 밟고 있는 수리영의 역동적인 실루엣이 드러난 찰나.\n\nLOCATION (lock): On an exposed coastal road during a heavy thunderstorm, under lightning and dense rain 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: At bicycle-saddle height, track parallel from the side in a wide profile that holds 수리영 and the full bicycle against the coastal road. Time the frame to the lightning flash as one leg drives the pedal downward, exposing her dynamic silhouette without changing the lateral tracking axis.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 자전거 (빗속에서 주행 중) — The full lateral side faces the camera during the downward pedal stroke; used as Carries 수리영 across the frame as a complete side-profile silhouette; 해안가 도로 (세찬 비가 내리고 있음); used as Establishes the route beneath and behind the tracked bicycle; 번개 (번쩍이는 순간); used as Temporarily reveals the rider's silhouette; 먹구름과 세찬 비 (천둥 번개와 함께 지속됨); used as Reduces clarity across the wider environment.\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: A lightning flash briefly cuts through the rain-darkened, desaturated scene, producing a stark but grounded silhouette with otherwise low-key storm 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young rides with the doll-adorned backpack containing the folded printed maps, while continuing to use her bicycle.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, stormy overcast.\n\nSHOT TEXT (authoritative, Korean): 번개 빛이 번쩍여 해안가 도로 위 자전거 안장에 앉아 한쪽 다리로 페달을 힘껏 밟고 있는 수리영의 역동적인 실루엣이 드러난 찰나.\n\nLOCATION (lock): On an exposed coastal road during a heavy thunderstorm, under lightning and dense rain 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: At bicycle-saddle height, track parallel from the side in a wide profile that holds 수리영 and the full bicycle against the coastal road. Time the frame to the lightning flash as one leg drives the pedal downward, exposing her dynamic silhouette without changing the lateral tracking axis.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 자전거 (빗속에서 주행 중) — The full lateral side faces the camera during the downward pedal stroke; used as Carries 수리영 across the frame as a complete side-profile silhouette; 해안가 도로 (세찬 비가 내리고 있음); used as Establishes the route beneath and behind the tracked bicycle; 번개 (번쩍이는 순간); used as Temporarily reveals the rider's silhouette; 먹구름과 세찬 비 (천둥 번개와 함께 지속됨); used as Reduces clarity across the wider environment.\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: A lightning flash briefly cuts through the rain-darkened, desaturated scene, producing a stark but grounded silhouette with otherwise low-key storm 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young rides with the doll-adorned backpack containing the folded printed maps, while continuing to use her bicycle.\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 초반, 긴 검은 머리, 타원형의 젊은 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S24sh1__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S24sh1.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/groupbg_coastal_road_16062d.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 해안가 도로의 디테일(가드레일, 배경)을 정확히 구현했으며, 요구된 실루엣 조명과 역동적인 페달링 동작을 훌륭하게 표현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "레퍼런스 이미지와 일치하지 않는 가드레일 디자인으로 장소 제약을 위반했으며, 배낭에 달린 인형과 지도의 묘사가 부자연스럽습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 해안가 도로의 디테일(가드레일, 배경)을 정확히 구현했으며, 요구된 실루엣 조명과 역동적인 페달링 동작을 훌륭하게 표현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "레퍼런스 이미지와 일치하지 않는 가드레일 디자인으로 장소 제약을 위반했으며, 배낭에 달린 인형과 지도의 묘사가 부자연스럽습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "레퍼런스의 해안가 도로 배경(가드레일, 가로등, 지형)을 정확히 재현했으며, 번개로 인한 실루엣과 배낭의 디테일(인형, 지도)이 프롬프트와 일치합니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "지정된 로케이션 레퍼런스의 디테일(가드레일, 도로 형태, 배경 지형)을 반영하지 못했고, 자전거 앞바퀴 포크의 구조가 왜곡되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "레퍼런스의 해안가 도로 배경(가드레일, 가로등, 지형)을 정확히 재현했으며, 번개로 인한 실루엣과 배낭의 디테일(인형, 지도)이 프롬프트와 일치합니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "지정된 로케이션 레퍼런스의 디테일(가드레일, 도로 형태, 배경 지형)을 반영하지 못했고, 자전거 앞바퀴 포크의 구조가 왜곡되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 15,
     "B": 7
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 15,
   "B": 7
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "지정된 해안가 도로의 디테일(가드레일, 배경)을 정확히 구현했으며, 요구된 실루엣 조명과 역동적인 페달링 동작을 훌륭하게 표현했습니다."
   },
   {
    "label": "B",
    "score": 3,
    "verdict_ko": "레퍼런스 이미지와 일치하지 않는 가드레일 디자인으로 장소 제약을 위반했으며, 배낭에 달린 인형과 지도의 묘사가 부자연스럽습니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/groupbg_coastal_road_16062d.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "원본 배경 이미지에 있던 좌우의 번개들이 지워지고 캐릭터 바로 뒤에 새로운 번개가 추가되어, 배경을 전혀 수정하지 말라는 지시를 위반했습니다.",
     "fix_en": "Restore the background exactly to the provided reference image, keeping the original lightning bolts on the far left and right, and remove the newly added lightning bolt behind the character."
    },
    {
     "issue_ko": "캐릭터의 왼쪽 다리(구부린 다리)가 자전거 프레임 앞쪽(카메라 방향)에 그려져 있어, 두 다리가 자전거의 같은 쪽에 위치하는 오류가 있습니다.",
     "fix_en": "Move the character's left leg to the far side of the bicycle frame so she correctly straddles the bike."
    },
    {
     "issue_ko": "샷 텍스트에 '자전거 안장에 앉아'라고 명시되어 있으나, 캐릭터의 엉덩이가 안장에 닿지 않고 위로 떠 있습니다.",
     "fix_en": "Lower the character's waist and hips so she is sitting securely on the bicycle saddle."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Restore the exact background camera and perspective so the sea horizon and road vanishing level match the supplied empty background and the camera sits at bicycle-saddle height.",
     "issue_ko": "배경의 수평선과 도로 소실점이 원본 배경보다 위로 올라가 있어 카메라 높이와 축이 고정된 배경 및 안장 높이 시점과 일치하지 않는다."
    },
    {
     "fix_en": "Move the complete rider-and-bicycle unit left and slightly downward to match the layout sketch, preserving its side-profile scale.",
     "issue_ko": "수리영과 자전거 전체가 레이아웃보다 화면 중앙 쪽으로 지나치게 오른쪽에 있고 위에도 떠 있다."
    },
    {
     "fix_en": "Lean the rider’s torso farther forward over the handlebars while keeping the bicycle in full rightward side profile and the driving leg in its downward pedal stroke.",
     "issue_ko": "수리영의 상체가 레이아웃보다 지나치게 수직으로 서 있어, 핸들 쪽으로 숙인 역동적인 주행 자세와 다르다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Restore the background exactly to the provided reference image, keeping the original lightning bolts on the far left and right, and remove the newly added lightning bolt behind the character.\n- Move the character's left leg to the far side of the bicycle frame so she correctly straddles the bike.\n- Lower the character's waist and hips so she is sitting securely on the bicycle saddle.\n- Restore the exact background camera and perspective so the sea horizon and road vanishing level match the supplied empty background and the camera sits at bicycle-saddle height.\n- Move the complete rider-and-bicycle unit left and slightly downward to match the layout sketch, preserving its side-profile scale.\n- Lean the rider’s torso farther forward over the handlebars while keeping the bicycle in full rightward side profile and the driving leg in its downward pedal stroke.\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": "지정된 배경 원본의 우측 번개를 임의로 삭제하여 '배경을 정확히 유지할 것'이라는 핵심 지침을 위반했습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "제공된 배경 원본(우측 번개 포함)을 온전히 유지하면서 인물의 배치, 포즈, 배낭의 디테일과 실루엣을 성공적으로 구현했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "지정된 배경 원본의 우측 번개를 임의로 삭제하여 '배경을 정확히 유지할 것'이라는 핵심 지침을 위반했습니다."
     },
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "제공된 배경 원본(우측 번개 포함)을 온전히 유지하면서 인물의 배치, 포즈, 배낭의 디테일과 실루엣을 성공적으로 구현했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "원본 배경 이미지를 완벽하게 유지하면서, 지시된 역동적인 실루엣과 배낭의 소품(지도, 인형)을 스케치에 맞게 잘 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "캐릭터와 소품의 구성은 양호하나, 원본 배경을 그대로 유지하라는 지시를 어기고 번개의 위치와 구름 등 배경을 임의로 변경했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "원본 배경 이미지를 완벽하게 유지하면서, 지시된 역동적인 실루엣과 배낭의 소품(지도, 인형)을 스케치에 맞게 잘 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "캐릭터와 소품의 구성은 양호하나, 원본 배경을 그대로 유지하라는 지시를 어기고 번개의 위치와 구름 등 배경을 임의로 변경했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 13,
     "B": 12
    },
    "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S24sh1__bgfirst_bg.png",
   "bg_asset_id": "5f9831d6-4e8d-473d-bbdc-cf9fed12f329",
   "bg_record_key": "S24sh1::bgfirst_bg",
   "chain_winner": true,
   "authority": "groupbg",
   "group_key": "coastal_road",
   "groupbg_asset_id": "20148f1f-f264-4442-8f6d-57b8ec8f2c2e"
  },
  "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)",
  "share_plan": {
   "ref_plan": "background"
  }
 },
 "S25sh1::bgfirst_bg": {
  "input_fingerprint": "cfc6df1c8df6214b",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 해질 무렵의 어두컴컴한 옥탑방 거실, 열린 창문과 바닥에 나뒹구는 의자들 앞에 멈춰 선 수리영의 굳은 뒷모습.\n\nLOCATION (lock): Inside the rooftop apartment's dark living room at dusk, facing an open window and overturned dining chairs.\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 near the living-room entrance at eye level, hold a static wide frame several paces behind 수리영. Her rigid back stays centered in the midground while the open window and overturned dining chairs remain fixed ahead, allowing the disturbed room to occupy most of the image.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 거실 창문 (열려 있음) — The room-facing frame and open aperture are visible ahead of 수리영; used as Creates the principal opening ahead of 수리영 and anchors the disrupted room layout; 식탁 의자들 (바닥에 쓰러져 있음) — Their sides and undersides are exposed at irregular angles toward the camera; used as Breaks the normal geometry of the living space around 수리영's path; 거실 바닥의 발자국 (누군가 다녀간 흔적으로 남아 있음) — The upward-facing floor surface carries visible footprints leading through the room; used as Leads from the entrance toward the disturbed interior; 헤집어진 가재도구 (집 안에 어질러져 있음); used as Provides environmental evidence of a search.\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 dusk is restrained and desaturated, with low-to-mid-key contrast holding 수리영 and the disturbed room in the same oppressive visual field.\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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 해질 무렵의 어두컴컴한 옥탑방 거실, 열린 창문과 바닥에 나뒹구는 의자들 앞에 멈춰 선 수리영의 굳은 뒷모습.\n\nLOCATION (lock): Inside the rooftop apartment's dark living room at dusk, facing an open window and overturned dining chairs.\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 near the living-room entrance at eye level, hold a static wide frame several paces behind 수리영. Her rigid back stays centered in the midground while the open window and overturned dining chairs remain fixed ahead, allowing the disturbed room to occupy most of the image.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 거실 창문 (열려 있음) — The room-facing frame and open aperture are visible ahead of 수리영; used as Creates the principal opening ahead of 수리영 and anchors the disrupted room layout; 식탁 의자들 (바닥에 쓰러져 있음) — Their sides and undersides are exposed at irregular angles toward the camera; used as Breaks the normal geometry of the living space around 수리영's path; 거실 바닥의 발자국 (누군가 다녀간 흔적으로 남아 있음) — The upward-facing floor surface carries visible footprints leading through the room; used as Leads from the entrance toward the disturbed interior; 헤집어진 가재도구 (집 안에 어질러져 있음); used as Provides environmental evidence of a search.\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 dusk is restrained and desaturated, with low-to-mid-key contrast holding 수리영 and the disturbed room in the same oppressive visual field.\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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S25sh1__bgfirst_bg.png",
  "asset_id": "8279f279-33a3-4505-9501-0b4d3b105320",
  "input_asset_ids": [
   "514f7d69-5f45-44ee-abd3-3be20cd5ddeb",
   "56a91bbe-83f6-4863-8952-ce28537e5796"
  ]
 },
 "S25sh1": {
  "input_fingerprint": "4ee0defdcee4b37d",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 해질 무렵의 어두컴컴한 옥탑방 거실, 열린 창문과 바닥에 나뒹구는 의자들 앞에 멈춰 선 수리영의 굳은 뒷모습.\n\nLOCATION (lock): Inside the rooftop apartment's dark living room at dusk, facing an open window and overturned dining chairs. 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 near the living-room entrance at eye level, hold a static wide frame several paces behind 수리영. Her rigid back stays centered in the midground while the open window and overturned dining chairs remain fixed ahead, allowing the disturbed room to occupy most of the image.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 거실 창문 (열려 있음) — The room-facing frame and open aperture are visible ahead of 수리영; used as Creates the principal opening ahead of 수리영 and anchors the disrupted room layout; 식탁 의자들 (바닥에 쓰러져 있음) — Their sides and undersides are exposed at irregular angles toward the camera; used as Breaks the normal geometry of the living space around 수리영's path; 거실 바닥의 발자국 (누군가 다녀간 흔적으로 남아 있음) — The upward-facing floor surface carries visible footprints leading through the room; used as Leads from the entrance toward the disturbed interior; 헤집어진 가재도구 (집 안에 어질러져 있음); used as Provides environmental evidence of a search.\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 dusk is restrained and desaturated, with low-to-mid-key contrast holding 수리영 and the disturbed room in the same oppressive visual field.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains ransacked, with intruder footprints, an open living-room window, overturned dining chairs, and disturbed household objects. Suri-young still has her backpack containing the printed maps.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 해질 무렵의 어두컴컴한 옥탑방 거실, 열린 창문과 바닥에 나뒹구는 의자들 앞에 멈춰 선 수리영의 굳은 뒷모습.\n\nLOCATION (lock): Inside the rooftop apartment's dark living room at dusk, facing an open window and overturned dining chairs. 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 near the living-room entrance at eye level, hold a static wide frame several paces behind 수리영. Her rigid back stays centered in the midground while the open window and overturned dining chairs remain fixed ahead, allowing the disturbed room to occupy most of the image.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 거실 창문 (열려 있음) — The room-facing frame and open aperture are visible ahead of 수리영; used as Creates the principal opening ahead of 수리영 and anchors the disrupted room layout; 식탁 의자들 (바닥에 쓰러져 있음) — Their sides and undersides are exposed at irregular angles toward the camera; used as Breaks the normal geometry of the living space around 수리영's path; 거실 바닥의 발자국 (누군가 다녀간 흔적으로 남아 있음) — The upward-facing floor surface carries visible footprints leading through the room; used as Leads from the entrance toward the disturbed interior; 헤집어진 가재도구 (집 안에 어질러져 있음); used as Provides environmental evidence of a search.\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 dusk is restrained and desaturated, with low-to-mid-key contrast holding 수리영 and the disturbed room in the same oppressive visual field.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains ransacked, with intruder footprints, an open living-room window, overturned dining chairs, and disturbed household objects. Suri-young still has her backpack containing the printed maps.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 해질 무렵의 어두컴컴한 옥탑방 거실, 열린 창문과 바닥에 나뒹구는 의자들 앞에 멈춰 선 수리영의 굳은 뒷모습.\n\nLOCATION (lock): Inside the rooftop apartment's dark living room at dusk, facing an open window and overturned dining chairs. 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 near the living-room entrance at eye level, hold a static wide frame several paces behind 수리영. Her rigid back stays centered in the midground while the open window and overturned dining chairs remain fixed ahead, allowing the disturbed room to occupy most of the image.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 거실 창문 (열려 있음) — The room-facing frame and open aperture are visible ahead of 수리영; used as Creates the principal opening ahead of 수리영 and anchors the disrupted room layout; 식탁 의자들 (바닥에 쓰러져 있음) — Their sides and undersides are exposed at irregular angles toward the camera; used as Breaks the normal geometry of the living space around 수리영's path; 거실 바닥의 발자국 (누군가 다녀간 흔적으로 남아 있음) — The upward-facing floor surface carries visible footprints leading through the room; used as Leads from the entrance toward the disturbed interior; 헤집어진 가재도구 (집 안에 어질러져 있음); used as Provides environmental evidence of a search.\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 dusk is restrained and desaturated, with low-to-mid-key contrast holding 수리영 and the disturbed room in the same oppressive visual field.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains ransacked, with intruder footprints, an open living-room window, overturned dining chairs, and disturbed household objects. Suri-young still has her backpack containing the printed maps.\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 초반, 긴 검은 머리, 타원형의 젊은 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S25sh1__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S25sh1.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L04B04.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "레퍼런스 이미지의 가구(구부러진 나무 의자)와 커튼의 위치까지 정확히 재현했으며, 옥탑방의 어두운 해질녘 분위기와 발자국 디테일이 매우 사실적입니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "전반적인 프레이밍과 요소들은 훌륭하나, 레퍼런스의 특정 의자 디자인을 놓쳤고 바닥의 발자국이 스탬프처럼 다소 인위적으로 보입니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "레퍼런스 이미지의 가구(구부러진 나무 의자)와 커튼의 위치까지 정확히 재현했으며, 옥탑방의 어두운 해질녘 분위기와 발자국 디테일이 매우 사실적입니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "전반적인 프레이밍과 요소들은 훌륭하나, 레퍼런스의 특정 의자 디자인을 놓쳤고 바닥의 발자국이 스탬프처럼 다소 인위적으로 보입니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "레퍼런스 이미지의 미닫이 창문 구조와 특정 가구(구부러진 나무 등받이 의자, 식탁 다리)를 정확히 반영했으며, 제시된 구도와 요소들을 충실히 구현했습니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "시간대 표현은 훌륭하나, 로케이션 레퍼런스의 미닫이 창문을 여닫이로 변형하고 가구의 형태를 다르게 묘사하여 공간적 일관성이 떨어집니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "레퍼런스 이미지의 미닫이 창문 구조와 특정 가구(구부러진 나무 등받이 의자, 식탁 다리)를 정확히 반영했으며, 제시된 구도와 요소들을 충실히 구현했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "시간대 표현은 훌륭하나, 로케이션 레퍼런스의 미닫이 창문을 여닫이로 변형하고 가구의 형태를 다르게 묘사하여 공간적 일관성이 떨어집니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 13,
     "B": 18
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 13,
   "B": 18
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 9,
    "verdict_ko": "레퍼런스 이미지의 가구(구부러진 나무 의자)와 커튼의 위치까지 정확히 재현했으며, 옥탑방의 어두운 해질녘 분위기와 발자국 디테일이 매우 사실적입니다."
   },
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "전반적인 프레이밍과 요소들은 훌륭하나, 레퍼런스의 특정 의자 디자인을 놓쳤고 바닥의 발자국이 스탬프처럼 다소 인위적으로 보입니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L04B04.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "기준 사진에서 창문이 있는 벽과 주방 싱크대가 있는 벽은 모서리에서 직각으로 만나지만, 생성된 이미지에서는 창문과 주방이 하나의 평평한 벽에 나란히 배치되어 공간의 건축 구조가 변형되었습니다.",
     "fix_en": "Restore the 90-degree corner from the reference, placing the window on the left wall and the kitchen counter on the perpendicular wall to the right."
    }
   ]
  },
  "composition_critique": {
   "issues": []
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Restore the 90-degree corner from the reference, placing the window on the left wall and the kitchen counter on the perpendicular wall to the 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": 10,
      "verdict_ko": "지시된 장소, 구도, 해질 무렵의 조명, 인물의 뒷모습과 배낭, 어질러진 방의 상태 등 모든 프롬프트 요구사항을 완벽하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "지시된 장소, 구도, 해질 무렵의 조명, 인물의 뒷모습과 배낭, 어질러진 방의 상태 등 모든 프롬프트 요구사항을 완벽하게 구현했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "지시된 장소, 구도, 해질 무렵의 조명, 인물의 뒷모습과 배낭, 어질러진 방의 상태 등 모든 프롬프트 요구사항을 완벽하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "지시된 장소, 구도, 해질 무렵의 조명, 인물의 뒷모습과 배낭, 어질러진 방의 상태 등 모든 프롬프트 요구사항을 완벽하게 구현했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "프롬프트에 명시된 넓은 촬영 구도, 어두운 조명, 캐릭터의 뒷모습과 어질러진 방의 상태를 완벽히 구현했으며 레퍼런스의 공간 구조를 정확히 따랐습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "대부분의 지시사항을 잘 따랐으나, 레퍼런스 이미지의 왼쪽 벽면 구조와 달리 임의로 새로운 문(개구부)을 생성하여 장소 일치도가 떨어집니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "프롬프트에 명시된 넓은 촬영 구도, 어두운 조명, 캐릭터의 뒷모습과 어질러진 방의 상태를 완벽히 구현했으며 레퍼런스의 공간 구조를 정확히 따랐습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "대부분의 지시사항을 잘 따랐으나, 레퍼런스 이미지의 왼쪽 벽면 구조와 달리 임의로 새로운 문(개구부)을 생성하여 장소 일치도가 떨어집니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 17,
     "B": 20
    },
    "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S25sh1__bgfirst_bg.png",
   "bg_asset_id": "8279f279-33a3-4505-9501-0b4d3b105320",
   "bg_record_key": "S25sh1::bgfirst_bg",
   "chain_winner": false,
   "authority": "plate"
  },
  "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L04B01",
    "B": "L04B02",
    "C": "L04B03",
    "D": "L04B04",
    "E": "L04B05",
    "F": "L04B06",
    "G": "L04B07"
   },
   "assigned": "L04B02",
   "choice": "Candidate B",
   "confident": true,
   "reason_ko": "\"옥탑방 거실\"과 \"바닥에 나뒹구는 의자들\"이라는 묘사를 통해 거실/주방 공간이 배경임을 알 수 있으며, 다른 뚜렷한 차이가 없으므로 거실 서브공간을 보여주는 기존 할당을 유지합니다.",
   "kept": "L04B02"
  }
 },
 "S25sh4::variants": {
  "author_fp": "2c25637ecb3090ce",
  "author": {
   "variants": [
    {
     "approach_ko": "엄격한 정면 대칭과 깊은 초점으로 표식을 차갑고 증거물처럼 제시한다.",
     "prompt_en": "Use a disciplined rectilinear lens feel with strict bilateral symmetry and deep focus from the doorway edges through the wall surface. Keep verticals true and the geometry rigid, allowing the rough brush ridges, uneven paint coverage, pinning point, dried blood, and surrounding wear to register with forensic clarity. Soft, cool dusk ambience spreads evenly across the room, while restrained contrast and natural color separation make the deep red feel stark against the desaturated setting."
    },
    {
     "approach_ko": "어두운 문틀과 중심부 선택 초점으로 방 안의 표식을 위협적인 발견처럼 응시한다.",
     "prompt_en": "Let the doorway edges fall into soft, dark foreground defocus, creating a tunnel-like visual pull toward the precise center while the marked wall remains square and undistorted. Hold the photograph and nearby painted strokes at the sharpest plane, with clarity easing gently toward the outer wall and room edges. Shape the dusk exposure with subtle peripheral falloff and low-key tonal depth so the rough red pigment catches faint grazing texture and the centered image feels ominously isolated within the surrounding emptiness."
    }
   ]
  },
  "reused": false
 },
 "S25sh4": {
  "input_fingerprint": "915c88966b8abe7e",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 수리영의 방 안쪽 벽면에 붉은색 도료로 거칠게 칠해진 커다란 원형 표식이 드러난 정지 구도.\n\nLOCATION (lock): At the interior wall of the daughter's bedroom in the rooftop apartment, where a large red circle has been painted. 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 just inside the bedroom doorway at eye level, hold a still, frontal wide composition of the marked wall. Center the large red circle in the middle of the frame with the bloodied photograph fixed at its center, excluding 수리영 entirely and preserving empty wall around the symbol.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: large red circular wall mark with photograph fixed at its center in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 붉은 원형 표식이 그려진 벽 (커다란 붉은 원이 그려져 있음) — The marked face of the wall is square to the camera, displaying a large roughly painted red circle; used as Forms the symmetrical central image and prepares the straight dolly toward its center; 피묻은 사진 (피가 묻은 채 벽에 꽂혀 있음) — The photograph's image side faces the camera from the center of the circle, though its fine details remain small at this distance; used as Creates the precise focal point at the center of the circular mark; 침실 방문 (열려 있음) — The open doorway edges frame the marked wall from the room entrance; used as Provides a rigid foreground boundary for the frontal reveal.\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 dusk ambience keeps the room desaturated and solemn, with the painted circle serving as the deliberate 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The ransacked state persists, and a large red circle is now painted on the bedroom wall with the bloodstained old photograph pinned at its center. Suri-young's backpack still contains the folded maps.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 수리영의 방 안쪽 벽면에 붉은색 도료로 거칠게 칠해진 커다란 원형 표식이 드러난 정지 구도.\n\nLOCATION (lock): At the interior wall of the daughter's bedroom in the rooftop apartment, where a large red circle has been painted. 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 just inside the bedroom doorway at eye level, hold a still, frontal wide composition of the marked wall. Center the large red circle in the middle of the frame with the bloodied photograph fixed at its center, excluding 수리영 entirely and preserving empty wall around the symbol.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: large red circular wall mark with photograph fixed at its center in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 붉은 원형 표식이 그려진 벽 (커다란 붉은 원이 그려져 있음) — The marked face of the wall is square to the camera, displaying a large roughly painted red circle; used as Forms the symmetrical central image and prepares the straight dolly toward its center; 피묻은 사진 (피가 묻은 채 벽에 꽂혀 있음) — The photograph's image side faces the camera from the center of the circle, though its fine details remain small at this distance; used as Creates the precise focal point at the center of the circular mark; 침실 방문 (열려 있음) — The open doorway edges frame the marked wall from the room entrance; used as Provides a rigid foreground boundary for the frontal reveal.\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 dusk ambience keeps the room desaturated and solemn, with the painted circle serving as the deliberate 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The ransacked state persists, and a large red circle is now painted on the bedroom wall with the bloodstained old photograph pinned at its center. Suri-young's backpack still contains the folded maps.\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 disciplined rectilinear lens feel with strict bilateral symmetry and deep focus from the doorway edges through the wall surface. Keep verticals true and the geometry rigid, allowing the rough brush ridges, uneven paint coverage, pinning point, dried blood, and surrounding wear to register with forensic clarity. Soft, cool dusk ambience spreads evenly across the room, while restrained contrast and natural color separation make the deep red feel stark against the desaturated setting.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 수리영의 방 안쪽 벽면에 붉은색 도료로 거칠게 칠해진 커다란 원형 표식이 드러난 정지 구도.\n\nLOCATION (lock): At the interior wall of the daughter's bedroom in the rooftop apartment, where a large red circle has been painted. 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 just inside the bedroom doorway at eye level, hold a still, frontal wide composition of the marked wall. Center the large red circle in the middle of the frame with the bloodied photograph fixed at its center, excluding 수리영 entirely and preserving empty wall around the symbol.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: large red circular wall mark with photograph fixed at its center in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 붉은 원형 표식이 그려진 벽 (커다란 붉은 원이 그려져 있음) — The marked face of the wall is square to the camera, displaying a large roughly painted red circle; used as Forms the symmetrical central image and prepares the straight dolly toward its center; 피묻은 사진 (피가 묻은 채 벽에 꽂혀 있음) — The photograph's image side faces the camera from the center of the circle, though its fine details remain small at this distance; used as Creates the precise focal point at the center of the circular mark; 침실 방문 (열려 있음) — The open doorway edges frame the marked wall from the room entrance; used as Provides a rigid foreground boundary for the frontal reveal.\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 dusk ambience keeps the room desaturated and solemn, with the painted circle serving as the deliberate 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The ransacked state persists, and a large red circle is now painted on the bedroom wall with the bloodstained old photograph pinned at its center. Suri-young's backpack still contains the folded maps.\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):\nLet the doorway edges fall into soft, dark foreground defocus, creating a tunnel-like visual pull toward the precise center while the marked wall remains square and undistorted. Hold the photograph and nearby painted strokes at the sharpest plane, with clarity easing gently toward the outer wall and room edges. Shape the dusk exposure with subtle peripheral falloff and low-key tonal depth so the rough red pigment catches faint grazing texture and the centered image feels ominously isolated within the surrounding 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."
  },
  "totals": {
   "B": 8,
   "A": 4
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 8,
    "verdict_ko": "레퍼런스 이미지의 방 구조(창문, 침대, 바닥 등)를 정확히 반영하면서 지시된 와이드 샷과 프레이밍을 완벽하게 구현했습니다."
   },
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "프레이밍이 너무 좁아 와이드 샷의 느낌을 주지 못하며, 방의 공간적 맥락(창문, 침대 등)이 전혀 보이지 않습니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L04B07.png"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "방이 난장판(ransacked)이라는 지시와 다르게 침대가 단정하게 정돈되어 있고 바닥이 너무 깨끗합니다.",
     "fix_en": "Mess up the bedsheets and scatter belongings on the floor to clearly show a ransacked state."
    },
    {
     "issue_ko": "유지되어야 할 상태(Carried State)에 명시된 수리영의 배낭이 프레임 안 바닥에 보이지 않습니다.",
     "fix_en": "Add Suri-young's backpack on the floor near the bed."
    },
    {
     "issue_ko": "오른쪽 창문 유리에 참조 사진 촬영 환경에 있던 카메라와 삼각대 실루엣이 잘못 반영되어 있습니다.",
     "fix_en": "Remove the camera and tripod silhouette from the window reflection."
    }
   ]
  },
  "composition_critique": {
   "issues": []
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Mess up the bedsheets and scatter belongings on the floor to clearly show a ransacked state.\n- Add Suri-young's backpack on the floor near the bed.\n- Remove the camera and tripod silhouette from the window reflection.\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 구도와 붉은 원형 표식을 잘 구현했으며, 프롬프트에 명시된 '어질러진 상태(ransacked state)'를 정확하게 반영하여 더 우수합니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "구도와 벽면의 표식은 잘 표현되었으나, 방이 어질러져 있어야 한다는 지시를 누락하여 단정하게 정돈된 상태로 묘사되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "요구된 구도와 붉은 원형 표식을 잘 구현했으며, 프롬프트에 명시된 '어질러진 상태(ransacked state)'를 정확하게 반영하여 더 우수합니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "구도와 벽면의 표식은 잘 표현되었으나, 방이 어질러져 있어야 한다는 지시를 누락하여 단정하게 정돈된 상태로 묘사되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "문틈으로 보이는 정면 구도, 붉은 원과 중앙의 사진을 정확히 구현했으며, 프롬프트에 명시된 '어질러진 상태(ransacked state)'와 배낭의 디테일까지 훌륭하게 반영했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "핵심 구도와 벽면의 표식은 잘 표현했으나, 필수 지시사항인 어질러진 방의 상태와 배낭이 완전히 누락되어 연속성 유지에 실패했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "문틈으로 보이는 정면 구도, 붉은 원과 중앙의 사진을 정확히 구현했으며, 프롬프트에 명시된 '어질러진 상태(ransacked state)'와 배낭의 디테일까지 훌륭하게 반영했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "핵심 구도와 벽면의 표식은 잘 표현했으나, 필수 지시사항인 어질러진 방의 상태와 배낭이 완전히 누락되어 연속성 유지에 실패했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 10,
     "B": 15
    },
    "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": "L04B01",
    "B": "L04B02",
    "C": "L04B03",
    "D": "L04B04",
    "E": "L04B05",
    "F": "L04B06",
    "G": "L04B07"
   },
   "assigned": "L04B01",
   "choice": "Candidate C",
   "confident": true,
   "reason_ko": "샷 설명에 명시된 '붉은색 도료로 거칠게 칠해진 커다란 원형 표식'이 유일하게 벽면에 그려져 있는 후보입니다.",
   "kept": "L04B01"
  }
 },
 "S25sh8::variants": {
  "author_fp": "33b045f6a8e841b5",
  "author": {
   "variants": [
    {
     "approach_ko": "수평적이고 절제된 심도 깊은 와이드 구도로 붕괴와 방의 참상을 동등하게 응시한다.",
     "prompt_en": "Use a level, rectilinear wide-lens perspective with restrained depth and minimal distortion, keeping the full figure, bed, and surrounding disorder clearly legible from front to back. Build the composition as a severe observational tableau, with her off-center placement counterweighted by the violated room and the red wall circle. Let soft lateral dusk ambience reveal fabric weight, mattress compression, and scattered material textures without beautifying the collapse."
    },
    {
     "approach_ko": "완만한 압축감과 선택적 초점으로 방은 겹쳐진 혼란으로 남기고 수리영의 육체적 붕괴에 시선을 붙든다.",
     "prompt_en": "Use a gently compressed wide-shot perspective from the locked waist height, arranging the room in overlapping depth layers while preserving the complete view of her body and the bed. Hold selective focus on her slackened posture and unfixed gaze, allowing the surrounding disorder and red wall circle to recede slightly softer but remain unmistakably readable. Keep the dusk exposure low and desaturated, with subtle edge falloff and quiet directional modeling across her face, clothing, backpack, clenched photograph, and the yielding mattress."
    }
   ]
  },
  "reused": false
 },
 "S25sh8": {
  "input_fingerprint": "538878e3e6b6fe2e",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 침대 매트리스 위에 두 다리의 힘이 풀려 주저앉은 채 초점을 잃은 수리영의 전신.\n\nLOCATION (lock): On the bed inside the daughter's ransacked bedroom at dusk. 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: Cut across the bedroom to a waist-height wide frame that contains 수리영's full body and the bed. Hold the exact instant her legs lose strength and her weight lands on the mattress, placing her slightly off-center while the disordered room remains legible around her.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 침대와 매트리스 (수리영이 주저앉음) — The long side of the bed faces the camera, with the mattress receiving 수리영's falling weight; used as Catches and contains the physical collapse in a full-body composition; 엉망이 된 방 안 (가재도구가 헤집어진 상태); used as Keeps the cause of her overwhelmed state present in the surrounding 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 dusk ambience remains desaturated and low-to-mid-key, holding her physical collapse in a tragic, unembellished register.\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 dim bedroom, wall finish, mattress placement, dusk lighting, and the large red circle as an ongoing state. Preserve the young woman's clothing and the disordered condition established in the room. Exclude the earlier wall-only composition and any close emphasis on the photograph, widening instead to her collapsed full-body pose on the mattress.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The room remains ransacked and the red wall circle remains, but Suri-young has removed the bloodstained photograph from its center. She wears the backpack containing the maps and keeps the photograph tightly in hand.\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 초반, 긴 검은 머리, 타원형의 젊은 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 침대 매트리스 위에 두 다리의 힘이 풀려 주저앉은 채 초점을 잃은 수리영의 전신.\n\nLOCATION (lock): On the bed inside the daughter's ransacked bedroom at dusk. 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: Cut across the bedroom to a waist-height wide frame that contains 수리영's full body and the bed. Hold the exact instant her legs lose strength and her weight lands on the mattress, placing her slightly off-center while the disordered room remains legible around her.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 침대와 매트리스 (수리영이 주저앉음) — The long side of the bed faces the camera, with the mattress receiving 수리영's falling weight; used as Catches and contains the physical collapse in a full-body composition; 엉망이 된 방 안 (가재도구가 헤집어진 상태); used as Keeps the cause of her overwhelmed state present in the surrounding 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 dusk ambience remains desaturated and low-to-mid-key, holding her physical collapse in a tragic, unembellished register.\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 dim bedroom, wall finish, mattress placement, dusk lighting, and the large red circle as an ongoing state. Preserve the young woman's clothing and the disordered condition established in the room. Exclude the earlier wall-only composition and any close emphasis on the photograph, widening instead to her collapsed full-body pose on the mattress.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The room remains ransacked and the red wall circle remains, but Suri-young has removed the bloodstained photograph from its center. She wears the backpack containing the maps and keeps the photograph tightly in hand.\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 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a level, rectilinear wide-lens perspective with restrained depth and minimal distortion, keeping the full figure, bed, and surrounding disorder clearly legible from front to back. Build the composition as a severe observational tableau, with her off-center placement counterweighted by the violated room and the red wall circle. Let soft lateral dusk ambience reveal fabric weight, mattress compression, and scattered material textures without beautifying the collapse.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 침대 매트리스 위에 두 다리의 힘이 풀려 주저앉은 채 초점을 잃은 수리영의 전신.\n\nLOCATION (lock): On the bed inside the daughter's ransacked bedroom at dusk. 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: Cut across the bedroom to a waist-height wide frame that contains 수리영's full body and the bed. Hold the exact instant her legs lose strength and her weight lands on the mattress, placing her slightly off-center while the disordered room remains legible around her.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 침대와 매트리스 (수리영이 주저앉음) — The long side of the bed faces the camera, with the mattress receiving 수리영's falling weight; used as Catches and contains the physical collapse in a full-body composition; 엉망이 된 방 안 (가재도구가 헤집어진 상태); used as Keeps the cause of her overwhelmed state present in the surrounding 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 dusk ambience remains desaturated and low-to-mid-key, holding her physical collapse in a tragic, unembellished register.\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 dim bedroom, wall finish, mattress placement, dusk lighting, and the large red circle as an ongoing state. Preserve the young woman's clothing and the disordered condition established in the room. Exclude the earlier wall-only composition and any close emphasis on the photograph, widening instead to her collapsed full-body pose on the mattress.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The room remains ransacked and the red wall circle remains, but Suri-young has removed the bloodstained photograph from its center. She wears the backpack containing the maps and keeps the photograph tightly in hand.\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 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a gently compressed wide-shot perspective from the locked waist height, arranging the room in overlapping depth layers while preserving the complete view of her body and the bed. Hold selective focus on her slackened posture and unfixed gaze, allowing the surrounding disorder and red wall circle to recede slightly softer but remain unmistakably readable. Keep the dusk exposure low and desaturated, with subtle edge falloff and quiet directional modeling across her face, clothing, backpack, clenched photograph, and the yielding mattress.\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": 10,
   "A": 2
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 10,
    "verdict_ko": "침대의 긴 면이 보이도록 카메라 구도를 완벽하게 변경했으며, 매트리스 위에 힘이 풀려 주저앉은 자세와 초점 잃은 시선, 소품(가방, 사진)의 상태까지 프롬프트를 정확히 구현했습니다."
   },
   {
    "label": "A",
    "score": 2,
    "verdict_ko": "카메라 구도를 변경하라는 지시를 무시하고 레퍼런스와 동일한 구도를 사용했으며, 매트리스에 주저앉지 않고 바닥에 어정쩡하게 서 있고 가방이 두 개로 복제되는 등 오류가 많습니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S25sh4_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "수리영이 이미 배낭을 메고 있는데, 바닥(왼쪽 다리 아래)에 똑같은 배낭이 하나 더 있습니다.",
     "fix_en": "Remove the duplicate green backpack sitting on the floor under her left leg."
    },
    {
     "issue_ko": "화면 왼쪽 바닥의 이불 밑으로 발목이 들어 있는 신발이 튀어나와 있어 존재하지 않는 인물이 추가된 것처럼 보입니다.",
     "fix_en": "Remove the foot and shoe sticking out from under the white blanket on the far left floor."
    },
    {
     "issue_ko": "사진을 쥐고 있는 오른손의 손가락 개수가 너무 많아 해부학적으로 틀렸습니다.",
     "fix_en": "Correct the anatomy of the right hand holding the photograph so it has a normal number of fingers."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Reframe slightly wider and to the left so the entire red wall circle is visible and clearly readable while preserving Suri-young’s full-body scale, her off-center placement, and the full bed.",
     "issue_ko": "화면 왼쪽에서 붉은 벽 원이 대부분 잘려 있어 원형으로 식별되지 않으며, 배경에서 명확히 읽혀야 한다는 구도 조건을 충족하지 못한다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the duplicate green backpack sitting on the floor under her left leg.\n- Remove the foot and shoe sticking out from under the white blanket on the far left floor.\n- Correct the anatomy of the right hand holding the photograph so it has a normal number of fingers.\n- Reframe slightly wider and to the left so the entire red wall circle is visible and clearly readable while preserving Suri-young’s full-body scale, her off-center placement, and the full bed.\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 프레이밍 내에서 피 묻은 사진을 들고 백팩을 멘 상태를 정확히 묘사했으며, 이전 샷의 난장판이 된 방과 붉은 원 등 배경 디테일을 소품 복제 없이 자연스럽게 유지했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "인물이 백팩을 메고 있음에도 불구하고 바닥(왼발 아래)에 또 다른 백팩이 복제되어 생성되는 치명적인 오류(Hard Violation)가 발생했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "지시된 프레이밍 내에서 피 묻은 사진을 들고 백팩을 멘 상태를 정확히 묘사했으며, 이전 샷의 난장판이 된 방과 붉은 원 등 배경 디테일을 소품 복제 없이 자연스럽게 유지했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "인물이 백팩을 메고 있음에도 불구하고 바닥(왼발 아래)에 또 다른 백팩이 복제되어 생성되는 치명적인 오류(Hard Violation)가 발생했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "프롬프트의 지시대로 엉망이 된 방 안에서 매트리스에 주저앉은 전신 와이드 샷을 잘 구현했으며, 백팩 착용과 사진을 쥔 손 등의 소품 상태가 정확합니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "인물이 백팩을 메고 있음에도 불구하고 바닥에 동일한 백팩이 하나 더 생성되는 소품 중복 오류(Hard Violation)가 발생했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "프롬프트의 지시대로 엉망이 된 방 안에서 매트리스에 주저앉은 전신 와이드 샷을 잘 구현했으며, 백팩 착용과 사진을 쥔 손 등의 소품 상태가 정확합니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "인물이 백팩을 메고 있음에도 불구하고 바닥에 동일한 백팩이 하나 더 생성되는 소품 중복 오류(Hard Violation)가 발생했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 12,
     "B": 18
    },
    "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": "S25sh4"
  }
 },
 "S26sh3::bgfirst_bg": {
  "input_fingerprint": "b4957347829420e5",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 수리영의 손에 들린 휴대전화 화면 위로 혜수의 부재중 전화 알림창이 떠 있는 클로즈업.\n\nLOCATION (lock): On the rocks at the edge of an empty, fogbound harbor at dusk.\n\nTIME OF DAY (lock): dusk, foggy.\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 dolly-in beside 수리영 at hand height, isolating her raised hand and phone while her seated body falls mostly outside the frame. Keep the phone below forty percent of the image, centered with the missed-call notification legible, while portions of her hand and lap provide human scale around it.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 휴대전화 화면 (부재중 전화 알림창이 떠 있음) — The active front face angles toward the camera and displays multiple missed calls from 혜수; used as Presents the missed calls as device-mediated information at the center of the insert; 빈 포구 (안개가 자욱하고 비가 그친 상태); used as Provides limited setting context beyond the hand without competing with the notification.\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 ambience filtered by the stated fog keeps the image muted and desaturated, with restrained contrast preserving the phone notification's readability.\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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 수리영의 손에 들린 휴대전화 화면 위로 혜수의 부재중 전화 알림창이 떠 있는 클로즈업.\n\nLOCATION (lock): On the rocks at the edge of an empty, fogbound harbor at dusk.\n\nTIME OF DAY (lock): dusk, foggy.\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 dolly-in beside 수리영 at hand height, isolating her raised hand and phone while her seated body falls mostly outside the frame. Keep the phone below forty percent of the image, centered with the missed-call notification legible, while portions of her hand and lap provide human scale around it.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 휴대전화 화면 (부재중 전화 알림창이 떠 있음) — The active front face angles toward the camera and displays multiple missed calls from 혜수; used as Presents the missed calls as device-mediated information at the center of the insert; 빈 포구 (안개가 자욱하고 비가 그친 상태); used as Provides limited setting context beyond the hand without competing with the notification.\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 ambience filtered by the stated fog keeps the image muted and desaturated, with restrained contrast preserving the phone notification's readability.\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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S26sh3__bgfirst_bg.png",
  "asset_id": "1b6dbc03-8b32-4c9c-a677-5274c64cbb1a",
  "input_asset_ids": [
   "455a53f9-8654-4f61-aaad-c8a53507128e",
   "f3d2879f-0bff-4a56-8944-26aa74d738ba"
  ]
 },
 "S26sh3": {
  "input_fingerprint": "1138066a96195475",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손에 들린 휴대전화 화면 위로 혜수의 부재중 전화 알림창이 떠 있는 클로즈업.\n\nLOCATION (lock): On the rocks at the edge of an empty, fogbound harbor 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: Complete the dolly-in beside 수리영 at hand height, isolating her raised hand and phone while her seated body falls mostly outside the frame. Keep the phone below forty percent of the image, centered with the missed-call notification legible, while portions of her hand and lap provide human scale around it.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 휴대전화 화면 (부재중 전화 알림창이 떠 있음) — The active front face angles toward the camera and displays multiple missed calls from 혜수; used as Presents the missed calls as device-mediated information at the center of the insert; 빈 포구 (안개가 자욱하고 비가 그친 상태); used as Provides limited setting context beyond the hand without competing with the notification.\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 ambience filtered by the stated fog keeps the image muted and desaturated, with restrained contrast preserving the phone notification's readability.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young still has her phone, the backpack containing the printed maps, and the bloodstained old photograph; Hye-soo's repeated missed calls remain logged on the phone.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손에 들린 휴대전화 화면 위로 혜수의 부재중 전화 알림창이 떠 있는 클로즈업.\n\nLOCATION (lock): On the rocks at the edge of an empty, fogbound harbor 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: Complete the dolly-in beside 수리영 at hand height, isolating her raised hand and phone while her seated body falls mostly outside the frame. Keep the phone below forty percent of the image, centered with the missed-call notification legible, while portions of her hand and lap provide human scale around it.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 휴대전화 화면 (부재중 전화 알림창이 떠 있음) — The active front face angles toward the camera and displays multiple missed calls from 혜수; used as Presents the missed calls as device-mediated information at the center of the insert; 빈 포구 (안개가 자욱하고 비가 그친 상태); used as Provides limited setting context beyond the hand without competing with the notification.\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 ambience filtered by the stated fog keeps the image muted and desaturated, with restrained contrast preserving the phone notification's readability.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young still has her phone, the backpack containing the printed maps, and the bloodstained old photograph; Hye-soo's repeated missed calls remain logged on the phone.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손에 들린 휴대전화 화면 위로 혜수의 부재중 전화 알림창이 떠 있는 클로즈업.\n\nLOCATION (lock): On the rocks at the edge of an empty, fogbound harbor 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: Complete the dolly-in beside 수리영 at hand height, isolating her raised hand and phone while her seated body falls mostly outside the frame. Keep the phone below forty percent of the image, centered with the missed-call notification legible, while portions of her hand and lap provide human scale around it.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 휴대전화 화면 (부재중 전화 알림창이 떠 있음) — The active front face angles toward the camera and displays multiple missed calls from 혜수; used as Presents the missed calls as device-mediated information at the center of the insert; 빈 포구 (안개가 자욱하고 비가 그친 상태); used as Provides limited setting context beyond the hand without competing with the notification.\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 ambience filtered by the stated fog keeps the image muted and desaturated, with restrained contrast preserving the phone notification's readability.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young still has her phone, the backpack containing the printed maps, and the bloodstained old photograph; Hye-soo's repeated missed calls remain logged on the phone.\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 초반, 긴 검은 머리, 타원형의 젊은 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S26sh3__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S26sh3.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    },
    {
     "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.",
     "path": "<bytes:524464>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L17B02.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    },
    {
     "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.",
     "path": "<bytes:524464>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "프롬프트 레퍼런스의 설명용 영문 텍스트가 스마트폰 화면에 그대로 합성되어 출력되는 치명적인 오류(Hard Violation)가 있습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "인물의 측면에서 손과 휴대전화를 강조하는 앵글 지시를 잘 따랐으며, 화면 내 부재중 전화 알림과 배경의 분위기가 충실히 구현되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "프롬프트 레퍼런스의 설명용 영문 텍스트가 스마트폰 화면에 그대로 합성되어 출력되는 치명적인 오류(Hard Violation)가 있습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "인물의 측면에서 손과 휴대전화를 강조하는 앵글 지시를 잘 따랐으며, 화면 내 부재중 전화 알림과 배경의 분위기가 충실히 구현되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "지정된 인서트 클로즈업 프레이밍을 어기고 인물의 상반신을 넓게 노출하였으나, 치명적인 1순위 위반 사항이 없어 우위입니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "레퍼런스 이미지의 영문 지시어(missed-call notification...)가 화면에 그대로 출력되는 치명적인 마커 유출(1순위 위반)로 탈락입니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "지정된 인서트 클로즈업 프레이밍을 어기고 인물의 상반신을 넓게 노출하였으나, 치명적인 1순위 위반 사항이 없어 우위입니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "레퍼런스 이미지의 영문 지시어(missed-call notification...)가 화면에 그대로 출력되는 치명적인 마커 유출(1순위 위반)로 탈락입니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 6,
     "B": 12
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 6,
   "B": 12
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 3,
    "verdict_ko": "프롬프트 레퍼런스의 설명용 영문 텍스트가 스마트폰 화면에 그대로 합성되어 출력되는 치명적인 오류(Hard Violation)가 있습니다."
   },
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "인물의 측면에서 손과 휴대전화를 강조하는 앵글 지시를 잘 따랐으며, 화면 내 부재중 전화 알림과 배경의 분위기가 충실히 구현되었습니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L17B02.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   },
   {
    "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.",
    "path": "<bytes:524464>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "화면에 비가 내리는 모습이 보이나, 프롬프트는 '비가 그친 상태'를 명시하고 있습니다.",
     "fix_en": "Remove the falling rain streaks to show that the rain has stopped."
    },
    {
     "issue_ko": "인물이 후드를 쓰고 있어 머리카락이 가려져 있으나, 레퍼런스 이미지의 긴 검은 머리와 정확히 일치해야 합니다.",
     "fix_en": "Lower the hood to fully reveal the character's long black hair matching the reference."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Reframe laterally so the phone is centered in the image while preserving its current angle, scale, and depth.",
     "issue_ko": "휴대전화가 화면 중앙이 아니라 오른쪽으로 치우쳐 있다."
    },
    {
     "fix_en": "Crop the seated figure more tightly out of the right side, leaving only the hand and limited lap portions around the phone.",
     "issue_ko": "수리영의 얼굴과 상체가 오른쪽 프레임을 크게 차지해, 앉은 몸이 대부분 프레임 밖에 있어야 한다는 인서트 구도와 어긋난다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the falling rain streaks to show that the rain has stopped.\n- Lower the hood to fully reveal the character's long black hair matching the reference.\n- Reframe laterally so the phone is centered in the image while preserving its current angle, scale, and depth.\n- Crop the seated figure more tightly out of the right side, leaving only the hand and limited lap portions around the phone.\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": "캐릭터의 긴 검은 머리와 의상(회색 후드 집업)을 정확히 반영하였으며, 화면 비율과 배경 묘사도 지시사항에 부합합니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "화면 구성은 유사하나 캐릭터가 후드를 쓰고 있어 지정된 외양(긴 검은 머리의 여성)과 일치하지 않습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "캐릭터의 긴 검은 머리와 의상(회색 후드 집업)을 정확히 반영하였으며, 화면 비율과 배경 묘사도 지시사항에 부합합니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "화면 구성은 유사하나 캐릭터가 후드를 쓰고 있어 지정된 외양(긴 검은 머리의 여성)과 일치하지 않습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "지정된 카메라 앵글과 프레이밍을 잘 따랐으며, 캐릭터의 긴 검은 머리와 비가 그친 안개 낀 배경을 정확하게 묘사했습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "프레이밍은 적절하나 '비가 그친 상태'라는 지시와 달리 화면에 비가 내리고 있으며, 후드를 착용해 지정된 캐릭터의 헤어스타일과 어긋납니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "지정된 카메라 앵글과 프레이밍을 잘 따랐으며, 캐릭터의 긴 검은 머리와 비가 그친 안개 낀 배경을 정확하게 묘사했습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "프레이밍은 적절하나 '비가 그친 상태'라는 지시와 달리 화면에 비가 내리고 있으며, 후드를 착용해 지정된 캐릭터의 헤어스타일과 어긋납니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 10,
     "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S26sh3__bgfirst_bg.png",
   "bg_asset_id": "1b6dbc03-8b32-4c9c-a677-5274c64cbb1a",
   "bg_record_key": "S26sh3::bgfirst_bg",
   "chain_winner": false,
   "authority": "plate"
  },
  "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)",
  "share_plan": {
   "ref_plan": "background"
  },
  "plate_select": {
   "candidates": {
    "A": "L17B01",
    "B": "L17B02"
   },
   "assigned": "L17B02",
   "choice": "Candidate B",
   "confident": true,
   "reason_ko": "\"휴대전화 화면... 클로즈업\" 샷으로 배경의 특정 구역이 명확히 요구되지 않으며, 제공된 두 후보가 동일한 이미지이므로 기존 할당을 유지합니다.",
   "kept": "L17B02"
  }
 },
 "S26sh7::variants": {
  "author_fp": "78afdb64235f7255",
  "author": {
   "variants": [
    {
     "approach_ko": "얕은 심도로 수리영의 절박한 시선과 긴장된 어깨에 집중하는 압축된 인물 중심 접근.",
     "prompt_en": "Use a gently compressed lens feel and very selective depth of field, holding critical focus on Suri-young’s lifted eyes and strained facial muscles while the elevated figure and boat edge fall slightly softer. Let the fog-diffused dusk wrap her face with subdued, cool tonal transitions, isolating the urgency of her performance against the quiet open water without turning the background into abstraction."
    },
    {
     "approach_ko": "배의 수직 구조와 안개 낀 수면까지 또렷하게 연결해 두 인물의 높이 차를 강조하는 공간 중심 접근.",
     "prompt_en": "Use a more spatial, moderately deep-focus rendering that keeps Suri-young, the boat’s shore-facing side, and the elevated position at the upper-right legible within one continuous plane of tension. Emphasize the boat edge as a firm compositional rise through the frame and preserve the fog-covered water as muted negative space; broad dusk ambience should reveal restrained surface wear and fabric texture with low-to-mid-key contrast."
    }
   ]
  },
  "reused": false
 },
 "S26sh7": {
  "input_fingerprint": "56b80921c1a7eb80",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy.\n\nSHOT TEXT (authoritative, Korean): 배 위에 선 인우를 향해 간절한 눈빛을 띤 채 마주 선 수리영의 상체.\n\nLOCATION (lock): At the foggy harbor's mooring edge beside the newly arrived 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: From the shore-side dialogue axis at chest height, hold a static medium frame angled slightly upward, tightening only the camera distance so 수리영's upper body occupies the left-center while 인우 remains elevated at the upper-right edge on the boat. Keep her lifted eyeline and tense shoulders clear against the open harbor, preserving the unequal vertical relationship without changing the established axis.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 소형 어선 (At the harbor beside the shore) — Its shore-facing side is visible behind 수리영, with 인우 positioned above that edge; used as The boat's side establishes 인우's elevated position behind 수리영; 바다 (Fog-covered); used as The open water remains visible as restrained negative space behind the dialogue; 안개 (Dense across the harbor); used as It softens the distant harbor and limits background visibility.\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: Desaturated dusk ambience filtered through dense fog keeps the contrast low-to-mid-key and the exchange solemn.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young retains the backpack with the folded maps and the old bloodstained photograph while asking In-woo for passage.\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 초반, 긴 검은 머리, 타원형의 젊은 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy.\n\nSHOT TEXT (authoritative, Korean): 배 위에 선 인우를 향해 간절한 눈빛을 띤 채 마주 선 수리영의 상체.\n\nLOCATION (lock): At the foggy harbor's mooring edge beside the newly arrived 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: From the shore-side dialogue axis at chest height, hold a static medium frame angled slightly upward, tightening only the camera distance so 수리영's upper body occupies the left-center while 인우 remains elevated at the upper-right edge on the boat. Keep her lifted eyeline and tense shoulders clear against the open harbor, preserving the unequal vertical relationship without changing the established axis.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 소형 어선 (At the harbor beside the shore) — Its shore-facing side is visible behind 수리영, with 인우 positioned above that edge; used as The boat's side establishes 인우's elevated position behind 수리영; 바다 (Fog-covered); used as The open water remains visible as restrained negative space behind the dialogue; 안개 (Dense across the harbor); used as It softens the distant harbor and limits background visibility.\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: Desaturated dusk ambience filtered through dense fog keeps the contrast low-to-mid-key and the exchange solemn.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young retains the backpack with the folded maps and the old bloodstained photograph while asking In-woo for passage.\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 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a gently compressed lens feel and very selective depth of field, holding critical focus on Suri-young’s lifted eyes and strained facial muscles while the elevated figure and boat edge fall slightly softer. Let the fog-diffused dusk wrap her face with subdued, cool tonal transitions, isolating the urgency of her performance against the quiet open water without turning the background into abstraction.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy.\n\nSHOT TEXT (authoritative, Korean): 배 위에 선 인우를 향해 간절한 눈빛을 띤 채 마주 선 수리영의 상체.\n\nLOCATION (lock): At the foggy harbor's mooring edge beside the newly arrived 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: From the shore-side dialogue axis at chest height, hold a static medium frame angled slightly upward, tightening only the camera distance so 수리영's upper body occupies the left-center while 인우 remains elevated at the upper-right edge on the boat. Keep her lifted eyeline and tense shoulders clear against the open harbor, preserving the unequal vertical relationship without changing the established axis.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 소형 어선 (At the harbor beside the shore) — Its shore-facing side is visible behind 수리영, with 인우 positioned above that edge; used as The boat's side establishes 인우's elevated position behind 수리영; 바다 (Fog-covered); used as The open water remains visible as restrained negative space behind the dialogue; 안개 (Dense across the harbor); used as It softens the distant harbor and limits background visibility.\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: Desaturated dusk ambience filtered through dense fog keeps the contrast low-to-mid-key and the exchange solemn.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young retains the backpack with the folded maps and the old bloodstained photograph while asking In-woo for passage.\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 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a more spatial, moderately deep-focus rendering that keeps Suri-young, the boat’s shore-facing side, and the elevated position at the upper-right legible within one continuous plane of tension. Emphasize the boat edge as a firm compositional rise through the frame and preserve the fog-covered water as muted negative space; broad dusk ambience should reveal restrained surface wear and fabric texture with low-to-mid-key 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."
  },
  "totals": {
   "A": 8,
   "B": 4
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 8,
    "verdict_ko": "명시된 카메라 앵글과 시선을 잘 살렸으며, 배낭에 꽂힌 지도와 사진 소품을 정확히 구현했습니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "지정된 소품(지도, 사진)이 배낭에서 누락되었고 인우의 상체가 프레임 밖으로 잘려 연출이 다소 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S26sh3_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "지도와 사진이 배낭 겉면에 입체감 없이 평면 스티커처럼 부자연스럽게 붙어 있습니다.",
     "fix_en": "Tuck the folded maps and photograph naturally into a physical pocket or compartment of the backpack."
    }
   ]
  },
  "composition_critique": {
   "issues": []
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Tuck the folded maps and photograph naturally into a physical pocket or compartment of the backpack.\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": "배낭 측면의 메쉬 주머니 안에 지도와 사진이 현실적으로 수납되어 물리적 사실성을 준수했으며, 요구된 미디엄 샷 구도와 분위기를 완벽히 구현했습니다."
     },
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "구도와 인물의 감정 표현은 훌륭하나, 지도와 사진이 배낭 표면에 주머니 없이 비현실적으로 붙어 있어 물리적 사실성 지침을 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "배낭 측면의 메쉬 주머니 안에 지도와 사진이 현실적으로 수납되어 물리적 사실성을 준수했으며, 요구된 미디엄 샷 구도와 분위기를 완벽히 구현했습니다."
     },
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "구도와 인물의 감정 표현은 훌륭하나, 지도와 사진이 배낭 표면에 주머니 없이 비현실적으로 붙어 있어 물리적 사실성 지침을 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 10,
      "verdict_ko": "지퍼의 물리적 상태가 올바르며, 카메라 앵글, 인물 배치, 배낭과 소품(지도, 피 묻은 사진) 등 모든 프롬프트 요구사항을 완벽하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "요구사항을 잘 따랐으나, 완전히 열린 후드 재킷 중간에 지퍼 슬라이더가 떠 있는 물리적 구조 오류가 있어 감점되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 10,
      "verdict_ko": "지퍼의 물리적 상태가 올바르며, 카메라 앵글, 인물 배치, 배낭과 소품(지도, 피 묻은 사진) 등 모든 프롬프트 요구사항을 완벽하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "요구사항을 잘 따랐으나, 완전히 열린 후드 재킷 중간에 지퍼 슬라이더가 떠 있는 물리적 구조 오류가 있어 감점되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 18,
     "B": 19
    },
    "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": "S26sh3"
  },
  "lane_policy": "ab_select_bypass:prev"
 },
 "S26sh9::variants": {
  "author_fp": "f87a733e81af1773",
  "author": {
   "variants": [
    {
     "approach_ko": "입술과 휴대전화의 간격, 눌린 엄지의 촉감을 얕은 심도로 강조하는 친밀한 측면 디테일.",
     "prompt_en": "Photograph the tight profile detail with an intimate, shallow-focus lens feel, keeping the pressed thumb, visible microphone icon, lower edge of the phone, and nearby lips crisply legible while her cheek falls gently softer. Hold the camera steady at mouth-and-hand height, with the fog-covered sea dissolving into a faint, desaturated spatial trace beyond her. Let the soft dusk ambience reveal natural skin texture, fingertip pressure, and the phone’s physical surface without heightened contrast."
    },
    {
     "approach_ko": "휴대전화 화면과 입술을 또렷한 평면적 긴장으로 묶고 안개 낀 바다까지 절제된 층위로 남기는 구성.",
     "prompt_en": "Use a restrained, slightly flatter close-up perspective that binds her lips, thumb, and angled phone screen into a precise profile composition, with the microphone icon clearly readable beneath the pressing thumb. Preserve a modest sense of depth through the partial cheek and hand at natural scale, while allowing the fog-covered sea to remain a soft but discernible layer behind her. Favor balanced focus across the central gesture and quiet, low-contrast dusk modeling rather than isolating a single detail."
    }
   ]
  },
  "reused": false
 },
 "S26sh9": {
  "input_fingerprint": "47eb6a1b38025424",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy.\n\nSHOT TEXT (authoritative, Korean): 수리영의 엄지손가락이 휴대전화 화면의 마이크 아이콘을 누른 채 기기 아랫부분을 입술 가까이 댄 클로즈업.\n\nLOCATION (lock): At the harbor edge beside the moored boat, facing the fog-covered sea at dusk. 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: Begin at mouth-and-hand height just behind 수리영's phone side, holding a tight profile detail in which her thumb, the visible microphone icon, and her lips share the central and lower portions of the frame. Pause before the prescribed dolly-out so the gesture reads clearly, with the phone kept at a natural scale and part of her cheek and hand providing human reference.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 휴대전화 (Recording a voice message) — The active screen face is angled toward the camera, showing the microphone icon beneath her pressing thumb; used as Primary foreground focus beside 수리영's lips; 바다 (Obscured by dense fog); used as Reduced to a soft spatial trace beyond 수리영 as she addresses the sea.\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, desaturated dusk ambience maintains restrained contrast across her skin, hand, and phone without adding an unstated 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\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 foggy dusk harbor, muted wet surfaces, dockside light, and nearby boat placement. Preserve the young woman's clothing and phone from the continuing moment. Exclude the boatman and the earlier face-to-face arrangement from this tight hand-and-phone insert.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young uses her phone to leave Hye-soo a voice message while retaining the backpack of maps and the old 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 초반, 긴 검은 머리, 타원형의 젊은 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy.\n\nSHOT TEXT (authoritative, Korean): 수리영의 엄지손가락이 휴대전화 화면의 마이크 아이콘을 누른 채 기기 아랫부분을 입술 가까이 댄 클로즈업.\n\nLOCATION (lock): At the harbor edge beside the moored boat, facing the fog-covered sea at dusk. 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: Begin at mouth-and-hand height just behind 수리영's phone side, holding a tight profile detail in which her thumb, the visible microphone icon, and her lips share the central and lower portions of the frame. Pause before the prescribed dolly-out so the gesture reads clearly, with the phone kept at a natural scale and part of her cheek and hand providing human reference.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 휴대전화 (Recording a voice message) — The active screen face is angled toward the camera, showing the microphone icon beneath her pressing thumb; used as Primary foreground focus beside 수리영's lips; 바다 (Obscured by dense fog); used as Reduced to a soft spatial trace beyond 수리영 as she addresses the sea.\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, desaturated dusk ambience maintains restrained contrast across her skin, hand, and phone without adding an unstated 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\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 foggy dusk harbor, muted wet surfaces, dockside light, and nearby boat placement. Preserve the young woman's clothing and phone from the continuing moment. Exclude the boatman and the earlier face-to-face arrangement from this tight hand-and-phone insert.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young uses her phone to leave Hye-soo a voice message while retaining the backpack of maps and the old 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 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the tight profile detail with an intimate, shallow-focus lens feel, keeping the pressed thumb, visible microphone icon, lower edge of the phone, and nearby lips crisply legible while her cheek falls gently softer. Hold the camera steady at mouth-and-hand height, with the fog-covered sea dissolving into a faint, desaturated spatial trace beyond her. Let the soft dusk ambience reveal natural skin texture, fingertip pressure, and the phone’s physical surface without heightened 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy.\n\nSHOT TEXT (authoritative, Korean): 수리영의 엄지손가락이 휴대전화 화면의 마이크 아이콘을 누른 채 기기 아랫부분을 입술 가까이 댄 클로즈업.\n\nLOCATION (lock): At the harbor edge beside the moored boat, facing the fog-covered sea at dusk. 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: Begin at mouth-and-hand height just behind 수리영's phone side, holding a tight profile detail in which her thumb, the visible microphone icon, and her lips share the central and lower portions of the frame. Pause before the prescribed dolly-out so the gesture reads clearly, with the phone kept at a natural scale and part of her cheek and hand providing human reference.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 휴대전화 (Recording a voice message) — The active screen face is angled toward the camera, showing the microphone icon beneath her pressing thumb; used as Primary foreground focus beside 수리영's lips; 바다 (Obscured by dense fog); used as Reduced to a soft spatial trace beyond 수리영 as she addresses the sea.\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, desaturated dusk ambience maintains restrained contrast across her skin, hand, and phone without adding an unstated 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\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 foggy dusk harbor, muted wet surfaces, dockside light, and nearby boat placement. Preserve the young woman's clothing and phone from the continuing moment. Exclude the boatman and the earlier face-to-face arrangement from this tight hand-and-phone insert.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young uses her phone to leave Hye-soo a voice message while retaining the backpack of maps and the old 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 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained, slightly flatter close-up perspective that binds her lips, thumb, and angled phone screen into a precise profile composition, with the microphone icon clearly readable beneath the pressing thumb. Preserve a modest sense of depth through the partial cheek and hand at natural scale, while allowing the fog-covered sea to remain a soft but discernible layer behind her. Favor balanced focus across the central gesture and quiet, low-contrast dusk modeling rather than isolating a single 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."
  },
  "totals": {
   "A": 7,
   "B": 3
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "요구된 입술과 스마트폰의 클로즈업 프레이밍, 안개 낀 바다 배경의 조도를 잘 구현하여 지시사항을 무난히 준수했습니다."
   },
   {
    "label": "B",
    "score": 3,
    "verdict_ko": "레퍼런스 이미지에 포함된 설명용 주석 텍스트가 스마트폰 화면에 그대로 유출되어 렌더링된 치명적인 규정 위반(Hard Violation)이 확인됩니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S26sh7_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   },
   {
    "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.",
    "path": "<bytes:524464>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "엄지손가락이 평면 화면 안의 마이크 아이콘이 아닌, 유리 위로 돌출된 입체적인 금속 버튼을 누르고 있습니다.",
     "fix_en": "Replace the protruding 3D button on the screen with a flat, 2D graphic of a microphone icon displayed natively on the screen."
    },
    {
     "issue_ko": "화면을 누르는 엄지손가락에 손톱이 없고 관절 형태가 무너져 기형적으로 보입니다.",
     "fix_en": "Redraw the pressing thumb with correct human anatomy, ensuring a natural fingernail and joint structure are visible."
    },
    {
     "issue_ko": "이전 샷 설정에 따라 메고 있어야 할 배낭의 어깨끈이 보이는 어깨 부위에 없습니다.",
     "fix_en": "Add the black backpack strap crossing down over the dark grey hoodie on her visible shoulder."
    },
    {
     "issue_ko": "텍스트 생성 금지 규칙에도 불구하고 휴대전화 화면에 읽을 수 없는 가짜 글씨와 파란색 말풍선이 있습니다.",
     "fix_en": "Remove the blue chat bubbles and illegible text scribbles from the phone screen, leaving a clean recording interface."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Raise and move the phone closer to her mouth so its lower edge sits immediately beside her lips, and reframe so the lips, pressing thumb, and microphone icon share the central and lower frame.",
     "issue_ko": "휴대전화 아랫부분이 입술에서 너무 멀리 떨어져 있고 입술도 화면의 우측 상단에 치우쳐 있어, 입술·누르는 엄지·마이크 아이콘이 중앙과 하단을 함께 차지해야 하는 구도와 다릅니다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Replace the protruding 3D button on the screen with a flat, 2D graphic of a microphone icon displayed natively on the screen.\n- Redraw the pressing thumb with correct human anatomy, ensuring a natural fingernail and joint structure are visible.\n- Add the black backpack strap crossing down over the dark grey hoodie on her visible shoulder.\n- Remove the blue chat bubbles and illegible text scribbles from the phone screen, leaving a clean recording interface.\n- Raise and move the phone closer to her mouth so its lower edge sits immediately beside her lips, and reframe so the lips, pressing thumb, and microphone icon share the central and lower 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": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "마이크 아이콘이 명확하게 렌더링되었으며 어깨의 백팩 스트랩이 나타나 소품 및 의상 유지 조건을 더 충실히 이행했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "화면의 마이크 아이콘이 뭉개져 알아볼 수 없으며, 착용하고 있어야 할 백팩의 끈이 프레임 내에 보이지 않습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "마이크 아이콘이 명확하게 렌더링되었으며 어깨의 백팩 스트랩이 나타나 소품 및 의상 유지 조건을 더 충실히 이행했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "화면의 마이크 아이콘이 뭉개져 알아볼 수 없으며, 착용하고 있어야 할 백팩의 끈이 프레임 내에 보이지 않습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "화면 인터페이스가 다소 왜곡되었으나 물리적인 질감과 공간감을 유지하며 요구된 클로즈업 구도와 피사체의 배치를 훌륭하게 구현했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "마이크 아이콘이 화면 내부가 아닌 손가락과 기기 위에 2D 그래픽처럼 덧붙여져 있어, 사실적인 재질감과 화면 구성 지침을 심각하게 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "화면 인터페이스가 다소 왜곡되었으나 물리적인 질감과 공간감을 유지하며 요구된 클로즈업 구도와 피사체의 배치를 훌륭하게 구현했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "마이크 아이콘이 화면 내부가 아닌 손가락과 기기 위에 2D 그래픽처럼 덧붙여져 있어, 사실적인 재질감과 화면 구성 지침을 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": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S26sh7"
  }
 },
 "S27sh1::variants": {
  "author_fp": "3d8fdeea5a0cf683",
  "author": {
   "variants": [
    {
     "approach_ko": "얕은 심도로 혜수의 굳은 표정과 휴대전화에 시선을 밀착시키는 폐쇄적 심리 초상.",
     "prompt_en": "Use a restrained, intimate lens feel with very shallow depth of field, holding crisp focus across 혜수's widened eyes, rigid facial muscles, and the phone beside her face while the wrecked room falls into soft, desaturated peripheral blur. Let the dusk illumination from the open window graze her face from the side, with gentle falloff into the room, creating a claustrophobic psychological portrait without losing the physical presence of the surrounding disorder."
    },
    {
     "approach_ko": "더 깊은 초점과 정면적 공간 구도로 난장판 속 혜수의 고립과 멈춘 순간을 강조.",
     "prompt_en": "Use a more observational lens feel and moderately deep focus, preserving readable layers of the disordered living room around the centered medium portrait while keeping every background element subordinate in scale and contrast. Shape the dusk ambience with stark, restrained tonal separation: 혜수's still face and phone remain the clearest visual anchor, while the open window, toppled chairs, footprints, and wind-disturbed room recede as a bleak spatial enclosure around her."
    }
   ]
  },
  "reused": false
 },
 "S27sh1": {
  "input_fingerprint": "9368087f78ecc0c3",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 난장판이 된 옥탑방 거실 한가운데 서서 휴대전화 스피커를 귓가에 댄 채 두 눈을 크게 뜬 혜수의 굳은 얼굴.\n\nLOCATION (lock): In the middle of the ransacked rooftop apartment living room at dusk, with the entrance door rattling in the wind. 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 eye-level dolly-in on the head-on axis, settling into a medium portrait with 혜수 centered and the phone pressed beside her face. Her widened eyes and arrested movement carry the frame, while the disordered room remains peripheral rather than competing with her expression.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 휴대전화 (Playing 수리영's recorded message) — Its screen face angles away from the camera while the speaker side is held to her ear; screen content is unreadable; used as Held beside 혜수's face as the secondary focal point and the source of the recorded message; 난장판이 된 옥탑방 거실 (Disordered); used as Peripheral clutter frames 혜수's rigid isolation without obscuring her silhouette.\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-to-mid-key ambient illumination appropriate to dusk renders the room desaturated and stark around 혜수's fixed 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\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same wrecked rooftop-room living area, dusk lighting, open window, footprints, and toppled chairs as persistent conditions. Preserve the bleak, wind-disturbed atmosphere. Exclude the young woman from the reference and place only the female officer in the room; do not restore any displaced furniture.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains ransacked, with the open window, overturned chairs, and other signs of searching still undisturbed; the red circle remains on the bedroom wall without the removed 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: 혜수 (한국인 여성, 40대 중반, 단정한 검은 머리, 성숙한 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 난장판이 된 옥탑방 거실 한가운데 서서 휴대전화 스피커를 귓가에 댄 채 두 눈을 크게 뜬 혜수의 굳은 얼굴.\n\nLOCATION (lock): In the middle of the ransacked rooftop apartment living room at dusk, with the entrance door rattling in the wind. 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 eye-level dolly-in on the head-on axis, settling into a medium portrait with 혜수 centered and the phone pressed beside her face. Her widened eyes and arrested movement carry the frame, while the disordered room remains peripheral rather than competing with her expression.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 휴대전화 (Playing 수리영's recorded message) — Its screen face angles away from the camera while the speaker side is held to her ear; screen content is unreadable; used as Held beside 혜수's face as the secondary focal point and the source of the recorded message; 난장판이 된 옥탑방 거실 (Disordered); used as Peripheral clutter frames 혜수's rigid isolation without obscuring her silhouette.\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-to-mid-key ambient illumination appropriate to dusk renders the room desaturated and stark around 혜수's fixed 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\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same wrecked rooftop-room living area, dusk lighting, open window, footprints, and toppled chairs as persistent conditions. Preserve the bleak, wind-disturbed atmosphere. Exclude the young woman from the reference and place only the female officer in the room; do not restore any displaced furniture.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains ransacked, with the open window, overturned chairs, and other signs of searching still undisturbed; the red circle remains on the bedroom wall without the removed 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: 혜수 (한국인 여성, 40대 중반, 단정한 검은 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained, intimate lens feel with very shallow depth of field, holding crisp focus across 혜수's widened eyes, rigid facial muscles, and the phone beside her face while the wrecked room falls into soft, desaturated peripheral blur. Let the dusk illumination from the open window graze her face from the side, with gentle falloff into the room, creating a claustrophobic psychological portrait without losing the physical presence of the surrounding disorder.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 난장판이 된 옥탑방 거실 한가운데 서서 휴대전화 스피커를 귓가에 댄 채 두 눈을 크게 뜬 혜수의 굳은 얼굴.\n\nLOCATION (lock): In the middle of the ransacked rooftop apartment living room at dusk, with the entrance door rattling in the wind. 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 eye-level dolly-in on the head-on axis, settling into a medium portrait with 혜수 centered and the phone pressed beside her face. Her widened eyes and arrested movement carry the frame, while the disordered room remains peripheral rather than competing with her expression.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 휴대전화 (Playing 수리영's recorded message) — Its screen face angles away from the camera while the speaker side is held to her ear; screen content is unreadable; used as Held beside 혜수's face as the secondary focal point and the source of the recorded message; 난장판이 된 옥탑방 거실 (Disordered); used as Peripheral clutter frames 혜수's rigid isolation without obscuring her silhouette.\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-to-mid-key ambient illumination appropriate to dusk renders the room desaturated and stark around 혜수's fixed 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\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same wrecked rooftop-room living area, dusk lighting, open window, footprints, and toppled chairs as persistent conditions. Preserve the bleak, wind-disturbed atmosphere. Exclude the young woman from the reference and place only the female officer in the room; do not restore any displaced furniture.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains ransacked, with the open window, overturned chairs, and other signs of searching still undisturbed; the red circle remains on the bedroom wall without the removed 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: 혜수 (한국인 여성, 40대 중반, 단정한 검은 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a more observational lens feel and moderately deep focus, preserving readable layers of the disordered living room around the centered medium portrait while keeping every background element subordinate in scale and contrast. Shape the dusk ambience with stark, restrained tonal separation: 혜수's still face and phone remain the clearest visual anchor, while the open window, toppled chairs, footprints, and wind-disturbed room recede as a bleak spatial enclosure around her.\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 미디엄 샷 프레이밍을 정확히 구현하고 벽의 붉은 원(Carried State)을 유지해 우선순위에서 앞서나, 가죽 재킷 의상이 누락되었습니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "레퍼런스의 의상은 잘 반영했으나, 지시된 미디엄 샷보다 프레이밍이 지나치게 넓고 이전 샷의 붉은 원이 배경에서 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S25sh1_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:735029>"
   },
   {
    "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.",
    "path": "<bytes:524464>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "벽에 그려진 빨간색 동그라미가 속이 빈 선 형태가 아니라 안이 꽉 채워진 원으로 렌더링되었습니다.",
     "fix_en": "Change the solid red disc on the wall into an open red ring of brush-width, showing the wallpaper texture inside it."
    },
    {
     "issue_ko": "캐릭터 레퍼런스 이미지에서 입고 있는 갈색 가죽 재킷이 누락되어 회색 티셔츠만 입고 있습니다.",
     "fix_en": "Dress the woman in the brown leather jacket from her character reference over her grey t-shirt."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Pull the camera back to a true medium shot, framing Hyesu from approximately the waist up while keeping her centered on the same eye-level head-on axis and preserving the room as peripheral context.",
     "issue_ko": "프레이밍이 허리 위의 미디엄 숏이 아니라 가슴 부근에서 잘린 미디엄 클로즈업으로 너무 타이트해, 혜수가 화면에서 계약보다 크게 보인다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the solid red disc on the wall into an open red ring of brush-width, showing the wallpaper texture inside it.\n- Dress the woman in the brown leather jacket from her character reference over her grey t-shirt.\n- Pull the camera back to a true medium shot, framing Hyesu from approximately the waist up while keeping her centered on the same eye-level head-on axis and preserving the room as peripheral context.\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": "캐릭터 레퍼런스의 가죽 재킷을 누락했으며, 벽에 그려진 빨간 동그라미를 지침에 어긋나는 꽉 찬 원으로 렌더링했습니다."
     },
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "요구된 가죽 재킷 의상을 정확히 반영했고, 벽의 빨간색 원을 속이 빈 링 형태로 훌륭하게 구현하며 프레이밍과 표정 지시를 모두 충족했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "캐릭터 레퍼런스의 가죽 재킷을 누락했으며, 벽에 그려진 빨간 동그라미를 지침에 어긋나는 꽉 찬 원으로 렌더링했습니다."
     },
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "요구된 가죽 재킷 의상을 정확히 반영했고, 벽의 빨간색 원을 속이 빈 링 형태로 훌륭하게 구현하며 프레이밍과 표정 지시를 모두 충족했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지시된 인물의 의상(가죽 재킷)과 텅 빈 원 형태의 붉은 자국을 정확히 묘사했으며, 요구된 구도와 얕은 피사계 심도도 잘 살렸습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "구도와 표정은 좋으나 인물의 가죽 재킷이 누락되었고, 벽의 붉은 원이 지시와 달리 속이 채워진 형태로 그려져 감점되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지시된 인물의 의상(가죽 재킷)과 텅 빈 원 형태의 붉은 자국을 정확히 묘사했으며, 요구된 구도와 얕은 피사계 심도도 잘 살렸습니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "구도와 표정은 좋으나 인물의 가죽 재킷이 누락되었고, 벽의 붉은 원이 지시와 달리 속이 채워진 형태로 그려져 감점되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 11,
     "B": 17
    },
    "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": "S25sh1"
  }
 },
 "S28sh3::bgfirst_bg": {
  "input_fingerprint": "644cbac1bdcfcc96",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 수리영의 손가락 끝이 낡은 흑백 사진을 쥔 채 인우의 눈앞으로 뻗어 있는 클로즈업.\n\nLOCATION (lock): On the open deck of an aging small fishing boat at sea, beside the wheelhouse.\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 인우's face height and within arm's length, place the lens just beside 수리영's extended forearm so her fingertips and the old photograph sit sharp in the center foreground while 인우's eyes remain close behind them. End the dolly-in before the shallow arc begins, using the arm as a diagonal that pins his recoiling attention to the photograph.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 낡은 흑백 사진 (Old black-and-white photograph) — The printed front face is visible at a slight angle, with 수리영's fingertips gripping its edge; used as Central foreground focal surface held directly before 인우's eyes; 갑판 (On a small fishing boat at sea); used as Provides limited environmental context behind 인우 without distracting from his 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: Restrained dusk ambience gives the photograph and faces a desaturated, low-to-mid-key naturalism.\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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 수리영의 손가락 끝이 낡은 흑백 사진을 쥔 채 인우의 눈앞으로 뻗어 있는 클로즈업.\n\nLOCATION (lock): On the open deck of an aging small fishing boat at sea, beside the wheelhouse.\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 인우's face height and within arm's length, place the lens just beside 수리영's extended forearm so her fingertips and the old photograph sit sharp in the center foreground while 인우's eyes remain close behind them. End the dolly-in before the shallow arc begins, using the arm as a diagonal that pins his recoiling attention to the photograph.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 낡은 흑백 사진 (Old black-and-white photograph) — The printed front face is visible at a slight angle, with 수리영's fingertips gripping its edge; used as Central foreground focal surface held directly before 인우's eyes; 갑판 (On a small fishing boat at sea); used as Provides limited environmental context behind 인우 without distracting from his 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: Restrained dusk ambience gives the photograph and faces a desaturated, low-to-mid-key naturalism.\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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S28sh3__bgfirst_bg.png",
  "asset_id": "1d5e7fc6-cb80-4b8d-be79-dd710a80173a",
  "input_asset_ids": [
   "012d6fc9-27f7-4a40-b257-b3c6c1e28ec7",
   "3c125004-6fd2-4620-bea8-4524553384c0"
  ]
 },
 "S28sh3": {
  "input_fingerprint": "3c4979e41ffe6b39",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손가락 끝이 낡은 흑백 사진을 쥔 채 인우의 눈앞으로 뻗어 있는 클로즈업.\n\nLOCATION (lock): On the open deck of an aging small fishing boat at sea, beside the wheelhouse. 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 인우's face height and within arm's length, place the lens just beside 수리영's extended forearm so her fingertips and the old photograph sit sharp in the center foreground while 인우's eyes remain close behind them. End the dolly-in before the shallow arc begins, using the arm as a diagonal that pins his recoiling attention to the photograph.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 낡은 흑백 사진 (Old black-and-white photograph) — The printed front face is visible at a slight angle, with 수리영's fingertips gripping its edge; used as Central foreground focal surface held directly before 인우's eyes; 갑판 (On a small fishing boat at sea); used as Provides limited environmental context behind 인우 without distracting from his 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: Restrained dusk ambience gives the photograph and faces a desaturated, low-to-mid-key naturalism.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young still carries her backpack of printed maps and holds the old photograph removed from the red wall circle, now extending it toward In-woo.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손가락 끝이 낡은 흑백 사진을 쥔 채 인우의 눈앞으로 뻗어 있는 클로즈업.\n\nLOCATION (lock): On the open deck of an aging small fishing boat at sea, beside the wheelhouse. 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 인우's face height and within arm's length, place the lens just beside 수리영's extended forearm so her fingertips and the old photograph sit sharp in the center foreground while 인우's eyes remain close behind them. End the dolly-in before the shallow arc begins, using the arm as a diagonal that pins his recoiling attention to the photograph.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 낡은 흑백 사진 (Old black-and-white photograph) — The printed front face is visible at a slight angle, with 수리영's fingertips gripping its edge; used as Central foreground focal surface held directly before 인우's eyes; 갑판 (On a small fishing boat at sea); used as Provides limited environmental context behind 인우 without distracting from his 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: Restrained dusk ambience gives the photograph and faces a desaturated, low-to-mid-key naturalism.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young still carries her backpack of printed maps and holds the old photograph removed from the red wall circle, now extending it toward In-woo.\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 초반, 긴 검은 머리, 타원형의 젊은 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손가락 끝이 낡은 흑백 사진을 쥔 채 인우의 눈앞으로 뻗어 있는 클로즈업.\n\nLOCATION (lock): On the open deck of an aging small fishing boat at sea, beside the wheelhouse. 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 인우's face height and within arm's length, place the lens just beside 수리영's extended forearm so her fingertips and the old photograph sit sharp in the center foreground while 인우's eyes remain close behind them. End the dolly-in before the shallow arc begins, using the arm as a diagonal that pins his recoiling attention to the photograph.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 낡은 흑백 사진 (Old black-and-white photograph) — The printed front face is visible at a slight angle, with 수리영's fingertips gripping its edge; used as Central foreground focal surface held directly before 인우's eyes; 갑판 (On a small fishing boat at sea); used as Provides limited environmental context behind 인우 without distracting from his 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: Restrained dusk ambience gives the photograph and faces a desaturated, low-to-mid-key naturalism.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young still carries her backpack of printed maps and holds the old photograph removed from the red wall circle, now extending it toward In-woo.\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 초반, 긴 검은 머리, 타원형의 젊은 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S28sh3__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S28sh3.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    },
    {
     "label": "PROP REFERENCE — 피 묻은 낡은 흑백사진: the exact object appearing in this shot; match its look, material and wear exactly.",
     "path": "<bytes:1430500>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L18B01.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    },
    {
     "label": "PROP REFERENCE — 피 묻은 낡은 흑백사진: the exact object appearing in this shot; match its look, material and wear exactly.",
     "path": "<bytes:1430500>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "사진이 눈앞에 위치하는 클로즈업 구도와 시선을 정확히 구현했으나, 배경 앵글을 레퍼런스 그대로 복사한 점이 아쉽습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "요구된 클로즈업 스케일을 벗어나 더 넓게 잡혔으며, 사진이 눈앞이 아닌 가슴 높이에 위치하여 핵심 연출을 놓쳤습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "사진이 눈앞에 위치하는 클로즈업 구도와 시선을 정확히 구현했으나, 배경 앵글을 레퍼런스 그대로 복사한 점이 아쉽습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "요구된 클로즈업 스케일을 벗어나 더 넓게 잡혔으며, 사진이 눈앞이 아닌 가슴 높이에 위치하여 핵심 연출을 놓쳤습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "클로즈업 샷 크기와 인물의 눈 바로 앞에 사진이 위치한 구도를 정확히 구현했으며, 수리영의 의상(회색 후드)도 잘 일치합니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "프레임이 너무 넓게 잡혀 클로즈업 지시를 어겼으며, 사진이 얼굴에서 다소 멀리 떨어져 있고 팔의 의상도 레퍼런스와 다릅니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "클로즈업 샷 크기와 인물의 눈 바로 앞에 사진이 위치한 구도를 정확히 구현했으며, 수리영의 의상(회색 후드)도 잘 일치합니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "프레임이 너무 넓게 잡혀 클로즈업 지시를 어겼으며, 사진이 얼굴에서 다소 멀리 떨어져 있고 팔의 의상도 레퍼런스와 다릅니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 9,
     "B": 15
    },
    "ranking": [
     "B",
     "A"
    ],
    "agreement": true,
    "policy": 1
   }
  },
  "totals": {
   "A": 9,
   "B": 15
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "사진이 눈앞에 위치하는 클로즈업 구도와 시선을 정확히 구현했으나, 배경 앵글을 레퍼런스 그대로 복사한 점이 아쉽습니다."
   },
   {
    "label": "A",
    "score": 4,
    "verdict_ko": "요구된 클로즈업 스케일을 벗어나 더 넓게 잡혔으며, 사진이 눈앞이 아닌 가슴 높이에 위치하여 핵심 연출을 놓쳤습니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L18B01.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   },
   {
    "label": "PROP REFERENCE — 피 묻은 낡은 흑백사진: the exact object appearing in this shot; match its look, material and wear exactly.",
    "path": "<bytes:1430500>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "사진을 쥔 손이 20대 초반 여성(수리영)의 프로필과 어긋나게 털이 많고 투박한 남성의 손으로 묘사되었으며, 엄지손가락 아래의 손가락 구조가 기형적으로 뭉개져 있습니다.",
     "fix_en": "Render the hand holding the photograph as a smooth, slender young woman's hand with anatomically correct fingers gripping the edge."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Dolly in to a tight close-up that fills the frame with the centered photograph, fingertips, and In-woo’s eyes immediately behind them, leaving only minimal boat context.",
     "issue_ko": "프레임이 너무 넓어 인우의 상반신과 갑판·조타실이 크게 보이며, 지정된 클로즈업보다 환경 비중이 과도하다."
    },
    {
     "fix_en": "Reposition the camera just beside the extended forearm so the arm recedes diagonally into the centered photograph rather than crossing the frame almost horizontally.",
     "issue_ko": "팔이 화면 왼쪽에서 거의 수평으로 가로질러, 렌즈를 뻗은 팔 바로 옆에 두고 팔을 대각선 축으로 사용하는 지정 구도와 다르다."
    },
    {
     "fix_en": "Redirect In-woo’s gaze onto the photograph directly in front of his eyes while preserving his recoiling posture.",
     "issue_ko": "인우의 시선이 사진이 아니라 화면 왼쪽을 향해 있어, 사진에 붙들린 반응 축이 어긋난다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Render the hand holding the photograph as a smooth, slender young woman's hand with anatomically correct fingers gripping the edge.\n- Dolly in to a tight close-up that fills the frame with the centered photograph, fingertips, and In-woo’s eyes immediately behind them, leaving only minimal boat context.\n- Reposition the camera just beside the extended forearm so the arm recedes diagonally into the centered photograph rather than crossing the frame almost horizontally.\n- Redirect In-woo’s gaze onto the photograph directly in front of his eyes while preserving his recoiling posture.\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": "클로즈업 샷의 구도, 인물과 소품의 배치, 배경과 수리영의 의상(소매)까지 지시사항을 매우 충실히 반영했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "화면 좌측에 후드 소매를 입은 팔과 사진을 쥔 맨팔이 동시에 존재하는 심각한 신체 오류(여분의 팔)가 발생했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "클로즈업 샷의 구도, 인물과 소품의 배치, 배경과 수리영의 의상(소매)까지 지시사항을 매우 충실히 반영했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "화면 좌측에 후드 소매를 입은 팔과 사진을 쥔 맨팔이 동시에 존재하는 심각한 신체 오류(여분의 팔)가 발생했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "왼쪽에 회색 소매가 떠 있고 사진을 쥔 맨팔이 아래에서 별개로 등장하는 심각한 신체 구조 오류(Hard Violation)가 있어 실격입니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "수리영의 회색 소매와 자연스럽게 이어진 손이 사진을 인우의 눈앞에 제시하는 프레이밍을 정확하게 구현했으며, 배경과 소품의 일치도도 뛰어납니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "왼쪽에 회색 소매가 떠 있고 사진을 쥔 맨팔이 아래에서 별개로 등장하는 심각한 신체 구조 오류(Hard Violation)가 있어 실격입니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "수리영의 회색 소매와 자연스럽게 이어진 손이 사진을 인우의 눈앞에 제시하는 프레이밍을 정확하게 구현했으며, 배경과 소품의 일치도도 뛰어납니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 16,
     "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S28sh3__bgfirst_bg.png",
   "bg_asset_id": "1d5e7fc6-cb80-4b8d-be79-dd710a80173a",
   "bg_record_key": "S28sh3::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": "클로즈업 샷으로 인물들의 손과 사진만 묘사되어 선박 내 특정 공간(갑판 혹은 조타실 등)을 명확히 구분할 수 있는 근거가 없으므로, 현재 할당된 후보를 유지합니다.",
   "kept": "L18B02"
  }
 },
 "S28sh4::variants": {
  "author_fp": "5dc583471583946a",
  "author": {
   "variants": [
    {
     "approach_ko": "얕은 심도로 인우의 커진 눈과 굳어지는 표정에 집중하고 사진과 조타실은 위협적인 흐림으로 남긴 밀착형 클로즈업.",
     "prompt_en": "Use an intimate, shallow-focus close-up that locks critical sharpness on In-woo’s widened eyes and startled facial tension as the dolly-out reveals his recoiling shoulders. Let the photograph remain a soft, partial intrusion at the lower edge and keep the wheelhouse entrance subdued behind his shoulder, barely legible through the falloff. Soft desaturated dusk ambience wraps his face with restrained contrast, preserving natural skin texture and minute changes in his brow, jaw, and gaze."
    },
    {
     "approach_ko": "더 깊은 초점과 압축된 공간감으로 인우의 후퇴, 전경 사진, 뒤편 조타실 입구를 하나의 긴장된 깊이 축에 묶은 환경적 클로즈업.",
     "prompt_en": "Photograph the close-up with a restrained compressed perspective and moderately deep focus, organizing the partial photograph, In-woo’s centered startled face and recoiling shoulders, and the distant wheelhouse entrance along one tense depth axis. Preserve the completed front three-quarter eye-level view while allowing the controlled dolly-out to make his backward movement read through changing spatial separation rather than exaggerated blur. Keep the dusk light flat, cool, and desaturated, with understated tonal distinction between his face, the foreground print, and the worn background."
    }
   ]
  },
  "reused": false
 },
 "S28sh4": {
  "input_fingerprint": "db2a5b9f26705d96",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 사진을 보고 두 눈이 커진 채 어깨를 움츠리며 뒤로 주춤 물러선 인우의 당황한 얼굴.\n\nLOCATION (lock): On the open deck of the small fishing boat at dusk, in front of the seated passenger and the wheelhouse. 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 the completed front three-quarter arc at 인우's eye height, begin a controlled dolly-out while holding his startled face near center and allowing his recoiling shoulders to enter the lower frame. Keep the photograph only as a partial near-edge threat, so the changed axis from the preceding shot is camera distance while his gaze and orientation remain continuous.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 낡은 흑백 사진 (Held close to 인우) — Only part of the printed front face enters from the lower edge, still directed toward 인우; used as A partially visible foreground cue motivating 인우's recoil; 조타실 입구 (Accessible from the deck) — The entrance faces obliquely toward the camera behind one of 인우's shoulders; used as The distant route of his impending escape, kept subdued behind his shoulder.\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: Desaturated dusk ambience and restrained contrast isolate 인우's widened eyes without introducing 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\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young continues holding the old photograph in front of In-woo, and her backpack of maps remains with her.\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, 24세, 검은 머리, 수려한 이목구비) — 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 사진을 보고 두 눈이 커진 채 어깨를 움츠리며 뒤로 주춤 물러선 인우의 당황한 얼굴.\n\nLOCATION (lock): On the open deck of the small fishing boat at dusk, in front of the seated passenger and the wheelhouse. 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 the completed front three-quarter arc at 인우's eye height, begin a controlled dolly-out while holding his startled face near center and allowing his recoiling shoulders to enter the lower frame. Keep the photograph only as a partial near-edge threat, so the changed axis from the preceding shot is camera distance while his gaze and orientation remain continuous.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 낡은 흑백 사진 (Held close to 인우) — Only part of the printed front face enters from the lower edge, still directed toward 인우; used as A partially visible foreground cue motivating 인우's recoil; 조타실 입구 (Accessible from the deck) — The entrance faces obliquely toward the camera behind one of 인우's shoulders; used as The distant route of his impending escape, kept subdued behind his shoulder.\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: Desaturated dusk ambience and restrained contrast isolate 인우's widened eyes without introducing 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\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young continues holding the old photograph in front of In-woo, and her backpack of maps remains with her.\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, 24세, 검은 머리, 수려한 이목구비) — 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, shallow-focus close-up that locks critical sharpness on In-woo’s widened eyes and startled facial tension as the dolly-out reveals his recoiling shoulders. Let the photograph remain a soft, partial intrusion at the lower edge and keep the wheelhouse entrance subdued behind his shoulder, barely legible through the falloff. Soft desaturated dusk ambience wraps his face with restrained contrast, preserving natural skin texture and minute changes in his brow, jaw, and gaze.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 사진을 보고 두 눈이 커진 채 어깨를 움츠리며 뒤로 주춤 물러선 인우의 당황한 얼굴.\n\nLOCATION (lock): On the open deck of the small fishing boat at dusk, in front of the seated passenger and the wheelhouse. 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 the completed front three-quarter arc at 인우's eye height, begin a controlled dolly-out while holding his startled face near center and allowing his recoiling shoulders to enter the lower frame. Keep the photograph only as a partial near-edge threat, so the changed axis from the preceding shot is camera distance while his gaze and orientation remain continuous.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 낡은 흑백 사진 (Held close to 인우) — Only part of the printed front face enters from the lower edge, still directed toward 인우; used as A partially visible foreground cue motivating 인우's recoil; 조타실 입구 (Accessible from the deck) — The entrance faces obliquely toward the camera behind one of 인우's shoulders; used as The distant route of his impending escape, kept subdued behind his shoulder.\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: Desaturated dusk ambience and restrained contrast isolate 인우's widened eyes without introducing 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\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young continues holding the old photograph in front of In-woo, and her backpack of maps remains with her.\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, 24세, 검은 머리, 수려한 이목구비) — 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-up with a restrained compressed perspective and moderately deep focus, organizing the partial photograph, In-woo’s centered startled face and recoiling shoulders, and the distant wheelhouse entrance along one tense depth axis. Preserve the completed front three-quarter eye-level view while allowing the controlled dolly-out to make his backward movement read through changing spatial separation rather than exaggerated blur. Keep the dusk light flat, cool, and desaturated, with understated tonal distinction between his face, the foreground print, and the worn 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": 7,
   "B": 4
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "지시된 3/4 측면 카메라 앵글과 어깨를 움츠린 당황한 표정을 잘 구현했으며, 사진을 화면 가장자리에 적절히 배치했습니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "카메라 앵글이 지시된 측면 구도가 아닌 정면을 유지하고 있으며, 사진이 화면 중앙에 배치되어 지시문과 어긋납니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S28sh3_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:420475>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "이전 샷(레퍼런스)에서는 손이 사진의 왼쪽 가장자리를 잡고 있었으나, 생성된 이미지에서는 사진의 오른쪽 가장자리를 잡고 있어 소품 유지(carried state) 연속성 규칙을 위반했습니다.",
     "fix_en": "Modify the hand holding the photograph so it enters from the left and holds the left edge of the photo, matching the continuity of the previous shot."
    },
    {
     "issue_ko": "프롬프트에 '텍스트 금지(No text)' 규칙이 명시되어 있으나, 인우의 어깨 너머 조타실 외벽 스티커와 좌측 배경 건물에 의미를 알 수 없는 형태의 문자가 생성되었습니다.",
     "fix_en": "Remove the garbled text from the sticker on the wheelhouse behind the character and from the building in the background."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Move the camera to In-woo’s eye-level front three-quarter axis so the wheelhouse entrance appears obliquely behind one shoulder, while keeping his face near center.",
     "issue_ko": "카메라가 정면 축에 가깝고 조타실 입구도 거의 정면으로 보여, 지정된 눈높이 전방 3/4 카메라 축이 구현되지 않았다."
    },
    {
     "fix_en": "Stage In-woo’s upper body recoiling backward with both shoulders visibly drawn up, preserving his centered close-up placement.",
     "issue_ko": "인우의 상체가 사진 쪽으로 앞으로 기울어 보여, 어깨를 움츠리며 뒤로 물러서는 동작 축과 반대다."
    },
    {
     "fix_en": "Keep In-woo’s gaze orientation continuous with the preceding shot and reposition the photograph along that same sightline.",
     "issue_ko": "인우의 시선과 사진의 화면 방향이 이전 쇼트와 반대로 바뀌어, 요구된 시선·방향 연속성이 깨졌다."
    },
    {
     "fix_en": "Reduce the photograph to a narrow, soft partial intrusion from the lower edge without changing the rest of the framing.",
     "issue_ko": "사진이 화면 오른쪽 아래를 크게 차지해, 하단 가장자리에서 일부만 위협적으로 들어와야 한다는 크기와 배치보다 지나치게 두드러진다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Modify the hand holding the photograph so it enters from the left and holds the left edge of the photo, matching the continuity of the previous shot.\n- Remove the garbled text from the sticker on the wheelhouse behind the character and from the building in the background.\n- Move the camera to In-woo’s eye-level front three-quarter axis so the wheelhouse entrance appears obliquely behind one shoulder, while keeping his face near center.\n- Stage In-woo’s upper body recoiling backward with both shoulders visibly drawn up, preserving his centered close-up placement.\n- Keep In-woo’s gaze orientation continuous with the preceding shot and reposition the photograph along that same sightline.\n- Reduce the photograph to a narrow, soft partial intrusion from the lower edge without changing the rest of the framing.\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": "이전 샷의 연속성을 유지하며 사진 소품의 위치와 어깨를 움츠린 당황한 얼굴의 클로즈업 구도를 충실히 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "화면 왼쪽 하단에 지시문과 이전 샷에 존재하지 않는 정체불명의 손과 종이가 추가되어 심각한 오류가 발생했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "이전 샷의 연속성을 유지하며 사진 소품의 위치와 어깨를 움츠린 당황한 얼굴의 클로즈업 구도를 충실히 구현했습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "화면 왼쪽 하단에 지시문과 이전 샷에 존재하지 않는 정체불명의 손과 종이가 추가되어 심각한 오류가 발생했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "지시된 클로즈업 프레이밍과 카메라 축을 잘 따랐으며, 인우의 당황한 표정과 하단 가장자리로 밀려난 사진의 배치가 요구사항을 정확히 충족합니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "프레임 왼쪽에 프롬프트에 없는 정체불명의 손과 종이가 나타나는 치명적인 오류가 있습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지시된 클로즈업 프레이밍과 카메라 축을 잘 따랐으며, 인우의 당황한 표정과 하단 가장자리로 밀려난 사진의 배치가 요구사항을 정확히 충족합니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "프레임 왼쪽에 프롬프트에 없는 정체불명의 손과 종이가 나타나는 치명적인 오류가 있습니다."
     }
    ]
   },
   "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": "S28sh3"
  }
 },
 "S28sh7::variants": {
  "author_fp": "2b6f760ed544cde4",
  "author": {
   "variants": [
    {
     "approach_ko": "깊은 초점과 엄격한 측면 구도로 출입구의 경계와 두 사람의 힘겨루기를 냉정한 공간적 tableau로 포착한다.",
     "prompt_en": "Use a restrained, rectilinear wide-lens feel with deep focus, turning the wheelhouse threshold into a rigid architectural frame around the struggle. Preserve clear separation among the braced lower-left figure, the opposing diagonal reach, and the helm beyond, with the open deck providing sober negative space. Let soft natural dusk fall broadly across worn surfaces, retaining subdued shadow detail and an observational, tactile realism."
    },
    {
     "approach_ko": "압축된 원근과 팔의 대각선을 강조해 맞잡은 손에서 조타기까지 이어지는 긴장을 한 축으로 응축한다.",
     "prompt_en": "Use a mildly compressed lens feel from the prescribed static profile position, visually tightening the distance between the locked hands, the fully extended arm, and the helm while keeping both complete bodies and the blocked route readable. Build the composition around the taut diagonal from the lower-left brace toward the right-side controls, with the threshold layers closely stacked behind it. Shape the natural dusk as restrained side light, allowing firmer shadow and delicate edge separation to emphasize muscular strain without losing realistic texture."
    }
   ]
  },
  "reused": false
 },
 "S28sh7": {
  "input_fingerprint": "36fd1bd93ee1c891",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 조타실 입구에서 조타기를 향해 뻗은 인우의 팔을 두 손으로 꽉 움켜쥔 수리영의 정지된 전신.\n\nLOCATION (lock): At the wheelhouse entrance on the fishing boat's open deck, immediately beside the controls inside. 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: Hold a static wide profile from waist height beside the wheelhouse entrance, several meters away, so both full bodies and the blocked path to the helm remain legible. 수리영 anchors the lower-left in a braced pull while 인우 stretches diagonally toward the right-side helm, freezing the struggle at maximum extension.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 조타실 입구 (Open) — Its open side faces the camera in profile, with both figures caught across the threshold; used as Frames the collision and marks the boundary between deck and confined interior; 조타기 (Within reach but not yet grasped) — Its operating side faces 인우 and is seen obliquely by the camera; used as The destination of 인우's extended arm, visible inside the wheelhouse without becoming oversized; 갑판 (Occupied by 수리영 and 인우); used as Provides the wider exterior plane behind the locked struggle.\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 dusk ambience remains desaturated and low-to-mid-key, preserving tactile realism in the physical struggle.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The wheelhouse already contains numerous circular motifs, including the red circle revealed moments later. Suri-young still has her backpack and the old photograph as she grabs In-woo's arm.\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 초반, 긴 검은 머리, 타원형의 젊은 얼굴); 인우 (한국인 남성, 24세, 검은 머리, 수려한 이목구비) — 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 조타실 입구에서 조타기를 향해 뻗은 인우의 팔을 두 손으로 꽉 움켜쥔 수리영의 정지된 전신.\n\nLOCATION (lock): At the wheelhouse entrance on the fishing boat's open deck, immediately beside the controls inside. 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: Hold a static wide profile from waist height beside the wheelhouse entrance, several meters away, so both full bodies and the blocked path to the helm remain legible. 수리영 anchors the lower-left in a braced pull while 인우 stretches diagonally toward the right-side helm, freezing the struggle at maximum extension.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 조타실 입구 (Open) — Its open side faces the camera in profile, with both figures caught across the threshold; used as Frames the collision and marks the boundary between deck and confined interior; 조타기 (Within reach but not yet grasped) — Its operating side faces 인우 and is seen obliquely by the camera; used as The destination of 인우's extended arm, visible inside the wheelhouse without becoming oversized; 갑판 (Occupied by 수리영 and 인우); used as Provides the wider exterior plane behind the locked struggle.\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 dusk ambience remains desaturated and low-to-mid-key, preserving tactile realism in the physical struggle.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The wheelhouse already contains numerous circular motifs, including the red circle revealed moments later. Suri-young still has her backpack and the old photograph as she grabs In-woo's arm.\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 초반, 긴 검은 머리, 타원형의 젊은 얼굴); 인우 (한국인 남성, 24세, 검은 머리, 수려한 이목구비) — 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, turning the wheelhouse threshold into a rigid architectural frame around the struggle. Preserve clear separation among the braced lower-left figure, the opposing diagonal reach, and the helm beyond, with the open deck providing sober negative space. Let soft natural dusk fall broadly across worn surfaces, retaining subdued shadow detail and an observational, tactile realism.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 조타실 입구에서 조타기를 향해 뻗은 인우의 팔을 두 손으로 꽉 움켜쥔 수리영의 정지된 전신.\n\nLOCATION (lock): At the wheelhouse entrance on the fishing boat's open deck, immediately beside the controls inside. 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: Hold a static wide profile from waist height beside the wheelhouse entrance, several meters away, so both full bodies and the blocked path to the helm remain legible. 수리영 anchors the lower-left in a braced pull while 인우 stretches diagonally toward the right-side helm, freezing the struggle at maximum extension.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 조타실 입구 (Open) — Its open side faces the camera in profile, with both figures caught across the threshold; used as Frames the collision and marks the boundary between deck and confined interior; 조타기 (Within reach but not yet grasped) — Its operating side faces 인우 and is seen obliquely by the camera; used as The destination of 인우's extended arm, visible inside the wheelhouse without becoming oversized; 갑판 (Occupied by 수리영 and 인우); used as Provides the wider exterior plane behind the locked struggle.\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 dusk ambience remains desaturated and low-to-mid-key, preserving tactile realism in the physical struggle.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The wheelhouse already contains numerous circular motifs, including the red circle revealed moments later. Suri-young still has her backpack and the old photograph as she grabs In-woo's arm.\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 초반, 긴 검은 머리, 타원형의 젊은 얼굴); 인우 (한국인 남성, 24세, 검은 머리, 수려한 이목구비) — 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 mildly compressed lens feel from the prescribed static profile position, visually tightening the distance between the locked hands, the fully extended arm, and the helm while keeping both complete bodies and the blocked route readable. Build the composition around the taut diagonal from the lower-left brace toward the right-side controls, with the threshold layers closely stacked behind it. Shape the natural dusk as restrained side light, allowing firmer shadow and delicate edge separation to emphasize muscular strain without losing realistic texture.\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": "수리영이 두 손으로 팔을 움켜쥐고 사진을 들고 있는 모습이 정확히 묘사되었으며, 인물의 해부학적 구조가 안정적입니다."
   },
   {
    "label": "A",
    "score": 3,
    "verdict_ko": "수리영이 사진을 쥔 손의 구조가 무너졌고 인우의 오른팔이 보이지 않는 심각한 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S28sh4_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   },
   {
    "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:420475>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "카메라 앵글이 조타실 입구의 측면(profile)을 보여주지 않고 정면을 비추고 있으며, 인물들이 문지방을 가로지르지 않고 갑판 바깥에 서 있습니다.",
     "fix_en": "Re-render the background to show the wheelhouse entrance in profile, and position the figures so they are caught across the threshold."
    },
    {
     "issue_ko": "인우가 조타기를 향해 팔을 뻗고 있지 않으며, 수리영이 잡은 인우의 팔 역시 조타기 방향과 무관하게 뻗어 있습니다.",
     "fix_en": "Change In-woo's pose so he is stretching his arm diagonally toward the helm, with Suri-young tightly gripping that reaching arm."
    },
    {
     "issue_ko": "지문에서 의도적으로 카메라를 향해 보여주라는 지시가 없었음에도, 수리영이 든 사진의 앞면이 부자연스럽게 카메라를 정면으로 향하고 있습니다.",
     "fix_en": "Angle the photograph in Suri-young's hand so its front side does not perfectly face the camera."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Move the camera to a waist-height side-profile position beside the wheelhouse entrance, several meters away, while preserving the wide full-body framing.",
     "issue_ko": "카메라가 조타실 입구를 정면으로 바라보고 있어, 입구 옆 허리 높이의 측면 프로필 구도와 다르다."
    },
    {
     "fix_en": "Restage In-woo across the threshold with his body and extended arm reaching diagonally toward the visible right-side helm, while Suri-young remains braced in the lower-left pulling him back.",
     "issue_ko": "인우가 문턱을 가로질러 우측 조타기로 뻗는 대신 갑판 우측에서 조타기와 반대 방향으로 팔을 뻗고 있어, 조타기로 향하는 동선과 저지된 경로가 읽히지 않는다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Re-render the background to show the wheelhouse entrance in profile, and position the figures so they are caught across the threshold.\n- Change In-woo's pose so he is stretching his arm diagonally toward the helm, with Suri-young tightly gripping that reaching arm.\n- Angle the photograph in Suri-young's hand so its front side does not perfectly face the camera.\n- Move the camera to a waist-height side-profile position beside the wheelhouse entrance, several meters away, while preserving the wide full-body framing.\n- Restage In-woo across the threshold with his body and extended arm reaching diagonally toward the visible right-side helm, while Suri-young remains braced in the lower-left pulling him 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": "요구된 앵글과 피사체 배치를 잘 따랐으며, 수리영이 쥔 낡은 사진의 디테일과 두 사람의 신체 표현이 더 자연스럽습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "전체적인 구도는 일치하나, 수리영이 쥔 손과 사진의 형태가 뭉개졌고 바닥에 불필요한 구조물이 생성되었습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "요구된 앵글과 피사체 배치를 잘 따랐으며, 수리영이 쥔 낡은 사진의 디테일과 두 사람의 신체 표현이 더 자연스럽습니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "전체적인 구도는 일치하나, 수리영이 쥔 손과 사진의 형태가 뭉개졌고 바닥에 불필요한 구조물이 생성되었습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "두 후보가 동일한 이미지입니다. 조타실 측면 앵글 지시를 어기고 정면을 향했으며, 인우가 조타기가 아닌 허공으로 팔을 뻗어 핵심 액션과 프레이밍에서 실패했습니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "후보 A와 동일한 이미지로, 조타실 정면 구도를 취하고 인우의 팔 방향이 조타실 내부가 아닌 바깥을 향하고 있어 공간 및 동선 지시를 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "두 후보가 동일한 이미지입니다. 조타실 측면 앵글 지시를 어기고 정면을 향했으며, 인우가 조타기가 아닌 허공으로 팔을 뻗어 핵심 액션과 프레이밍에서 실패했습니다."
     },
     {
      "label": "A",
      "score": 4,
      "verdict_ko": "후보 A와 동일한 이미지로, 조타실 정면 구도를 취하고 인우의 팔 방향이 조타실 내부가 아닌 바깥을 향하고 있어 공간 및 동선 지시를 위반했습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 11,
     "B": 9
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": false,
    "policy": 1
   },
   "winner": "A",
   "fix_won": false,
   "policy": "fix_rejudge_v1_flip"
  },
  "ref_mode": "prev+엔티티",
  "share_plan": {
   "ref_plan": "prev",
   "prev_anchor_tag": "S28sh4"
  },
  "lane_policy": "ab_select_bypass:prev"
 },
 "S29sh1::bgfirst_bg": {
  "input_fingerprint": "7ad269d35080b125",
  "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 해 질 무렵 좁은 조타실 안, 조타기를 꽉 쥔 인우와 그를 온몸으로 가로막은 수리영이 밀착해 뒤엉킨 구도.\n\nLOCATION (lock): Inside the cramped wheelhouse of the small fishing boat at dusk, directly at the helm.\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 chest height near the confined wheelhouse entrance, hold a static wide composition down the open line toward the helm, with 인우 compressed into the right midground and 수리영 blocking him from the left-center. Their overlapping bodies form a tight diagonal around his hands on the helm, while enough of the cramped interior remains visible to establish the trap-like space.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 조타기 (Firmly gripped by 인우) — Its operating side faces 인우, with his hands visibly gripping it from the right midground; used as Central physical anchor around which both bodies are compressed; 조타실 입구 (Open) — The entrance faces into the confined room, with its edges enclosing the pair; used as Creates a narrow foreground frame and the camera's line of approach; 좁은 조타실 내부 (Occupied by the struggling pair); used as The enclosing background that limits lateral escape and intensifies the confrontation.\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-to-mid-key ambient illumination appropriate to dusk keeps the cramped wheelhouse desaturated and oppressive.\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.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\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.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 해 질 무렵 좁은 조타실 안, 조타기를 꽉 쥔 인우와 그를 온몸으로 가로막은 수리영이 밀착해 뒤엉킨 구도.\n\nLOCATION (lock): Inside the cramped wheelhouse of the small fishing boat at dusk, directly at the helm.\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 chest height near the confined wheelhouse entrance, hold a static wide composition down the open line toward the helm, with 인우 compressed into the right midground and 수리영 blocking him from the left-center. Their overlapping bodies form a tight diagonal around his hands on the helm, while enough of the cramped interior remains visible to establish the trap-like space.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 조타기 (Firmly gripped by 인우) — Its operating side faces 인우, with his hands visibly gripping it from the right midground; used as Central physical anchor around which both bodies are compressed; 조타실 입구 (Open) — The entrance faces into the confined room, with its edges enclosing the pair; used as Creates a narrow foreground frame and the camera's line of approach; 좁은 조타실 내부 (Occupied by the struggling pair); used as The enclosing background that limits lateral escape and intensifies the confrontation.\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-to-mid-key ambient illumination appropriate to dusk keeps the cramped wheelhouse desaturated and oppressive.\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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S29sh1__bgfirst_bg.png",
  "asset_id": "b2fe2fe9-31a7-4727-bd7d-c418f1d9d4f6",
  "input_asset_ids": [
   "7cc59904-27a7-41d0-a4f8-fd077beec978",
   "7339d4be-9495-44bd-9e80-f8bcb5b1b2f6"
  ]
 },
 "S29sh1": {
  "input_fingerprint": "4e731c95b330cb01",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 해 질 무렵 좁은 조타실 안, 조타기를 꽉 쥔 인우와 그를 온몸으로 가로막은 수리영이 밀착해 뒤엉킨 구도.\n\nLOCATION (lock): Inside the cramped wheelhouse of the small fishing boat at dusk, directly at the helm. 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 near the confined wheelhouse entrance, hold a static wide composition down the open line toward the helm, with 인우 compressed into the right midground and 수리영 blocking him from the left-center. Their overlapping bodies form a tight diagonal around his hands on the helm, while enough of the cramped interior remains visible to establish the trap-like space.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 조타기 (Firmly gripped by 인우) — Its operating side faces 인우, with his hands visibly gripping it from the right midground; used as Central physical anchor around which both bodies are compressed; 조타실 입구 (Open) — The entrance faces into the confined room, with its edges enclosing the pair; used as Creates a narrow foreground frame and the camera's line of approach; 좁은 조타실 내부 (Occupied by the struggling pair); used as The enclosing background that limits lateral escape and intensifies the confrontation.\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-to-mid-key ambient illumination appropriate to dusk keeps the cramped wheelhouse desaturated and oppressive.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The wheelhouse's numerous circles, including the red one, remain fixed around the struggling pair. Suri-young retains her backpack and the old 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 초반, 긴 검은 머리, 타원형의 젊은 얼굴); 인우 (한국인 남성, 24세, 검은 머리, 수려한 이목구비) — 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 해 질 무렵 좁은 조타실 안, 조타기를 꽉 쥔 인우와 그를 온몸으로 가로막은 수리영이 밀착해 뒤엉킨 구도.\n\nLOCATION (lock): Inside the cramped wheelhouse of the small fishing boat at dusk, directly at the helm. 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 near the confined wheelhouse entrance, hold a static wide composition down the open line toward the helm, with 인우 compressed into the right midground and 수리영 blocking him from the left-center. Their overlapping bodies form a tight diagonal around his hands on the helm, while enough of the cramped interior remains visible to establish the trap-like space.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 조타기 (Firmly gripped by 인우) — Its operating side faces 인우, with his hands visibly gripping it from the right midground; used as Central physical anchor around which both bodies are compressed; 조타실 입구 (Open) — The entrance faces into the confined room, with its edges enclosing the pair; used as Creates a narrow foreground frame and the camera's line of approach; 좁은 조타실 내부 (Occupied by the struggling pair); used as The enclosing background that limits lateral escape and intensifies the confrontation.\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-to-mid-key ambient illumination appropriate to dusk keeps the cramped wheelhouse desaturated and oppressive.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The wheelhouse's numerous circles, including the red one, remain fixed around the struggling pair. Suri-young retains her backpack and the old 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 초반, 긴 검은 머리, 타원형의 젊은 얼굴); 인우 (한국인 남성, 24세, 검은 머리, 수려한 이목구비) — 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 해 질 무렵 좁은 조타실 안, 조타기를 꽉 쥔 인우와 그를 온몸으로 가로막은 수리영이 밀착해 뒤엉킨 구도.\n\nLOCATION (lock): Inside the cramped wheelhouse of the small fishing boat at dusk, directly at the helm. 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 near the confined wheelhouse entrance, hold a static wide composition down the open line toward the helm, with 인우 compressed into the right midground and 수리영 blocking him from the left-center. Their overlapping bodies form a tight diagonal around his hands on the helm, while enough of the cramped interior remains visible to establish the trap-like space.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 조타기 (Firmly gripped by 인우) — Its operating side faces 인우, with his hands visibly gripping it from the right midground; used as Central physical anchor around which both bodies are compressed; 조타실 입구 (Open) — The entrance faces into the confined room, with its edges enclosing the pair; used as Creates a narrow foreground frame and the camera's line of approach; 좁은 조타실 내부 (Occupied by the struggling pair); used as The enclosing background that limits lateral escape and intensifies the confrontation.\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-to-mid-key ambient illumination appropriate to dusk keeps the cramped wheelhouse desaturated and oppressive.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The wheelhouse's numerous circles, including the red one, remain fixed around the struggling pair. Suri-young retains her backpack and the old 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 초반, 긴 검은 머리, 타원형의 젊은 얼굴); 인우 (한국인 남성, 24세, 검은 머리, 수려한 이목구비) — 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S29sh1__bgfirst_bg.png"
    },
    {
     "label": "LAYOUT SKETCH (people placement only)",
     "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/conti/conti_S29sh1.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    },
    {
     "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:420475>"
    }
   ],
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L19B01.png"
    },
    {
     "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:905420>"
    },
    {
     "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.",
     "path": "<bytes:420475>"
    }
   ]
  },
  "judge_flip": {
   "forward_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "조타실 입구를 전경 프레임으로 사용하는 카메라 구도 지침을 누락했으며, 유지되어야 할 붉은 원 요소가 화면에 없습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "조타실 입구를 좁은 전경 프레임으로 활용하라는 핵심 구도 지침을 잘 구현했으며, 명시된 붉은 원 요소도 충실히 반영했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "조타실 입구를 전경 프레임으로 사용하는 카메라 구도 지침을 누락했으며, 유지되어야 할 붉은 원 요소가 화면에 없습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "조타실 입구를 좁은 전경 프레임으로 활용하라는 핵심 구도 지침을 잘 구현했으며, 명시된 붉은 원 요소도 충실히 반영했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "지정된 전경의 입구 프레이밍과 원형 표식은 구현했으나, 인우가 조타기를 쥐고 있지 않아 샷 텍스트의 핵심 액션을 완전히 누락했습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "조타기를 쥔 인우의 액션과 주머니 속 낡은 사진 소품을 정확히 반영했으나, 요구된 입구 프레임과 붉은 원 표식이 누락되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "지정된 전경의 입구 프레이밍과 원형 표식은 구현했으나, 인우가 조타기를 쥐고 있지 않아 샷 텍스트의 핵심 액션을 완전히 누락했습니다."
     },
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "조타기를 쥔 인우의 액션과 주머니 속 낡은 사진 소품을 정확히 반영했으나, 요구된 입구 프레임과 붉은 원 표식이 누락되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 12,
     "B": 12
    },
    "ranking": [
     "A",
     "B"
    ],
    "agreement": false,
    "policy": 1
   }
  },
  "totals": {
   "A": 12,
   "B": 12
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 5,
    "verdict_ko": "조타실 입구를 전경 프레임으로 사용하는 카메라 구도 지침을 누락했으며, 유지되어야 할 붉은 원 요소가 화면에 없습니다."
   },
   {
    "label": "B",
    "score": 7,
    "verdict_ko": "조타실 입구를 좁은 전경 프레임으로 활용하라는 핵심 구도 지침을 잘 구현했으며, 명시된 붉은 원 요소도 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/3d61971a-7158-40f4-b742-8fd5e757dbbf/episodes/132fe629-d451-4f2f-ac9c-c33beb62bb0f/images/background_chain/L19B01.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:905420>"
   },
   {
    "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:420475>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "프롬프트와 스케치의 지시와 달리 인우의 오른손이 조타기를 잡지 않고 다리 위에 놓여 있습니다.",
     "fix_en": "Pose In-woo's right arm upward so his hand firmly grips the upper part of the steering wheel."
    },
    {
     "issue_ko": "그대로 유지해야 할 배경 요소인 조타기의 윗부분 절반이 완전히 지워졌습니다.",
     "fix_en": "Restore the missing upper half of the wooden steering wheel in front of In-woo."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Repose Inwoo so both hands visibly grip the helm from the right midground, and compress Suri-young’s overlapping body around this hand-and-helm anchor while preserving the camera and all other framing.",
     "issue_ko": "인우의 두 손이 조타기를 잡고 있는 모습이 보이지 않아, 두 인물의 밀착된 대각선 구도를 형성해야 할 중심 앵커가 사라졌다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Pose In-woo's right arm upward so his hand firmly grips the upper part of the steering wheel.\n- Restore the missing upper half of the wooden steering wheel in front of In-woo.\n- Repose Inwoo so both hands visibly grip the helm from the right midground, and compress Suri-young’s overlapping body around this hand-and-helm anchor while preserving the camera and all other framing.\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": "배경 원본에 있던 조타기가 완전히 삭제되어 프롬프트의 지시 사항을 심각하게 위반했습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "스케치의 인물 배치와 포즈를 충실히 따르며, 배경 원본의 핵심 요소인 조타기를 온전히 유지했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "배경 원본에 있던 조타기가 완전히 삭제되어 프롬프트의 지시 사항을 심각하게 위반했습니다."
     },
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "스케치의 인물 배치와 포즈를 충실히 따르며, 배경 원본의 핵심 요소인 조타기를 온전히 유지했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "지정된 배경과 인물의 배치를 훌륭하게 유지했으나, 인우가 조타기를 꽉 쥐고 있는 핵심 동작이 누락되었습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "원본 조타기 앞에 거대한 조타기를 추가로 생성하여 배경을 그대로 유지하라는 최우선 조건을 심각하게 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "지정된 배경과 인물의 배치를 훌륭하게 유지했으나, 인우가 조타기를 꽉 쥐고 있는 핵심 동작이 누락되었습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "원본 조타기 앞에 거대한 조타기를 추가로 생성하여 배경을 그대로 유지하라는 최우선 조건을 심각하게 위반했습니다."
     }
    ]
   },
   "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S29sh1__bgfirst_bg.png",
   "bg_asset_id": "b2fe2fe9-31a7-4727-bd7d-c418f1d9d4f6",
   "bg_record_key": "S29sh1::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": "L19B01",
    "B": "L19B02",
    "C": "L19B03"
   },
   "assigned": "L19B01",
   "choice": "A",
   "confident": true,
   "reason_ko": "\"조타기를 꽉 쥔 인우와 그를 온몸으로 가로막은 수리영이 밀착해 뒤엉킨 구도\"라는 묘사를 볼 때, 조타기 주변으로 두 인물이 몸싸움을 벌일 수 있는 공간감이 확보된 A가 적합합니다.",
   "chosen": "L19B01"
  }
 },
 "S29sh3::variants": {
  "author_fp": "7868a5db5115e355",
  "author": {
   "variants": [
    {
     "approach_ko": "정중앙의 대칭적 구도와 절제된 명암으로 인물 식별의 냉정한 확신을 강조한다.",
     "prompt_en": "Treat the insert as a stark, evidentiary image: rigidly centered, nearly symmetrical, and visually still. Use soft, even dusk illumination with restrained contrast so the monochrome facial detail remains legible through the point of contact, while gentle edge falloff isolates the central identification without losing the photograph’s physical wear."
    },
    {
     "approach_ko": "비스듬히 들어오는 손가락과 낮게 스치는 빛으로 눌림의 촉감과 감정적 긴장을 강조한다.",
     "prompt_en": "Build the insert around tactile pressure, letting the finger enter on a strong diagonal toward the centered face. Shape the subdued dusk light at a shallow side angle so minute relief, contact shadow, paper grain, and the fingertip’s compression become palpable; allow the narrow outer edges to fall softly away while keeping fingertip and printed face equally sharp."
    }
   ]
  },
  "reused": false
 },
 "S29sh3": {
  "input_fingerprint": "8bf69de4c0b7b8a9",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 수리영의 검지손가락이 낡은 흑백 사진 속 어린 여자아이의 얼굴을 꾹 짚고 있는 클로즈업.\n\nLOCATION (lock): Inside the cramped fishing-boat wheelhouse, beside the helm where the old photograph is being examined. 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: Directly above the photograph at hand height, hold an extreme insert in which 수리영's fingertip presses into the pictured girl's face at center, with only a narrow surrounding portion of the print retained for context. Keep both the human fingertip and the printed face in the same shallow focal plane, excluding the wider confrontation entirely.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 낡은 흑백 사진 (Old black-and-white photograph) — The printed front face is square to the overhead camera, showing the young girl's face directly beneath 수리영's index fingertip; used as The isolated focal 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: Neutral, subdued dusk-appropriate illumination preserves the photograph's monochrome detail and the tactile pressure of the fingertip.\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 cramped wheelhouse, dusk illumination, worn interior materials, controls, and circular markings. Preserve the clothing of both people and the old photograph used in the confrontation. Exclude the preceding full-body struggle and steering-wheel emphasis from this tight finger-and-photograph insert.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young still holds the same old photograph while pointing to the child identified as Minsuk; her backpack of maps remains with her, and the wheelhouse circles remain unchanged.\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 초반, 긴 검은 머리, 타원형의 젊은 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": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 수리영의 검지손가락이 낡은 흑백 사진 속 어린 여자아이의 얼굴을 꾹 짚고 있는 클로즈업.\n\nLOCATION (lock): Inside the cramped fishing-boat wheelhouse, beside the helm where the old photograph is being examined. 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: Directly above the photograph at hand height, hold an extreme insert in which 수리영's fingertip presses into the pictured girl's face at center, with only a narrow surrounding portion of the print retained for context. Keep both the human fingertip and the printed face in the same shallow focal plane, excluding the wider confrontation entirely.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 낡은 흑백 사진 (Old black-and-white photograph) — The printed front face is square to the overhead camera, showing the young girl's face directly beneath 수리영's index fingertip; used as The isolated focal 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: Neutral, subdued dusk-appropriate illumination preserves the photograph's monochrome detail and the tactile pressure of the fingertip.\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 cramped wheelhouse, dusk illumination, worn interior materials, controls, and circular markings. Preserve the clothing of both people and the old photograph used in the confrontation. Exclude the preceding full-body struggle and steering-wheel emphasis from this tight finger-and-photograph insert.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young still holds the same old photograph while pointing to the child identified as Minsuk; her backpack of maps remains with her, and the wheelhouse circles remain unchanged.\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 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nTreat the insert as a stark, evidentiary image: rigidly centered, nearly symmetrical, and visually still. Use soft, even dusk illumination with restrained contrast so the monochrome facial detail remains legible through the point of contact, while gentle edge falloff isolates the central identification without losing the photograph’s physical wear.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 수리영의 검지손가락이 낡은 흑백 사진 속 어린 여자아이의 얼굴을 꾹 짚고 있는 클로즈업.\n\nLOCATION (lock): Inside the cramped fishing-boat wheelhouse, beside the helm where the old photograph is being examined. 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: Directly above the photograph at hand height, hold an extreme insert in which 수리영's fingertip presses into the pictured girl's face at center, with only a narrow surrounding portion of the print retained for context. Keep both the human fingertip and the printed face in the same shallow focal plane, excluding the wider confrontation entirely.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 낡은 흑백 사진 (Old black-and-white photograph) — The printed front face is square to the overhead camera, showing the young girl's face directly beneath 수리영's index fingertip; used as The isolated focal 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: Neutral, subdued dusk-appropriate illumination preserves the photograph's monochrome detail and the tactile pressure of the fingertip.\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 cramped wheelhouse, dusk illumination, worn interior materials, controls, and circular markings. Preserve the clothing of both people and the old photograph used in the confrontation. Exclude the preceding full-body struggle and steering-wheel emphasis from this tight finger-and-photograph insert.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young still holds the same old photograph while pointing to the child identified as Minsuk; her backpack of maps remains with her, and the wheelhouse circles remain unchanged.\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 초반, 긴 검은 머리, 타원형의 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nBuild the insert around tactile pressure, letting the finger enter on a strong diagonal toward the centered face. Shape the subdued dusk light at a shallow side angle so minute relief, contact shadow, paper grain, and the fingertip’s compression become palpable; allow the narrow outer edges to fall softly away while keeping fingertip and printed face equally sharp.\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": "지정된 소품 레퍼런스(두 소녀가 있는 낡은 사진)를 무시하고 캐릭터 레퍼런스의 얼굴을 사진 속에 잘못 배치하여 감점됨."
   },
   {
    "label": "B",
    "score": 8,
    "verdict_ko": "요구된 익스트림 인서트 클로즈업 앵글을 잘 구현했으며, 소품 레퍼런스 속 소녀의 얼굴과 의상 디테일을 사진에 정확히 반영함."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S29sh1_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:1166989>"
   },
   {
    "label": "PROP REFERENCE — 피 묻은 낡은 흑백사진: the exact object appearing in this shot; match its look, material and wear exactly.",
    "path": "<bytes:1430500>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "흑백 사진 속 이미지가 제공된 소품 레퍼런스(건물 앞 벤치에 앉은 두 소녀)와 일치하지 않고 완전히 다른 인물 초상화로 묘사되었습니다.",
     "fix_en": "Replace the printed image on the photograph with the exact scene from the prop reference showing two girls sitting in front of a building."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Move the camera directly above the photograph and align the print square to the image plane, preserving the same tight insert framing.",
     "issue_ko": "카메라가 사진을 수직으로 내려다보지 않아 인화지가 화면면과 평행하지 않고 비스듬한 원근으로 보인다."
    },
    {
     "fix_en": "Reorient the entering index finger along a strong diagonal toward the centered girl's face while preserving its contact point and everything else.",
     "issue_ko": "검지가 화면 위쪽에서 거의 수직 축으로 들어와, 중심 얼굴을 향한 강한 대각선 구도라는 지시와 다르다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Replace the printed image on the photograph with the exact scene from the prop reference showing two girls sitting in front of a building.\n- Move the camera directly above the photograph and align the print square to the image plane, preserving the same tight insert framing.\n- Reorient the entering index finger along a strong diagonal toward the centered girl's face while preserving its contact point and everything else.\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": "지시된 익스트림 인서트 클로즈업 프레이밍을 정확히 구현하여 사진의 좁은 여백만을 화면에 남긴 점이 프롬프트 우선순위에 부합합니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "흑백 사진의 전체 형태와 그 아래 배경까지 화면에 포함시켜, '좁은 여백만 남기라'는 익스트림 인서트 프레이밍 지시를 명백히 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 8,
      "verdict_ko": "지시된 익스트림 인서트 클로즈업 프레이밍을 정확히 구현하여 사진의 좁은 여백만을 화면에 남긴 점이 프롬프트 우선순위에 부합합니다."
     },
     {
      "label": "B",
      "score": 4,
      "verdict_ko": "흑백 사진의 전체 형태와 그 아래 배경까지 화면에 포함시켜, '좁은 여백만 남기라'는 익스트림 인서트 프레이밍 지시를 명백히 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 9,
      "verdict_ko": "요청된 익스트림 인서트 클로즈업 앵글을 완벽하게 구현했으며, 레퍼런스 사진 속 인물의 디테일과 종이 질감을 훌륭하게 살렸습니다."
     },
     {
      "label": "A",
      "score": 3,
      "verdict_ko": "프레이밍이 너무 넓어 인서트 샷의 지시를 어겼으며, 사진 속 배경과 인물이 제공된 소품 레퍼런스와 전혀 다르게 왜곡되었습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 9,
      "verdict_ko": "요청된 익스트림 인서트 클로즈업 앵글을 완벽하게 구현했으며, 레퍼런스 사진 속 인물의 디테일과 종이 질감을 훌륭하게 살렸습니다."
     },
     {
      "label": "B",
      "score": 3,
      "verdict_ko": "프레이밍이 너무 넓어 인서트 샷의 지시를 어겼으며, 사진 속 배경과 인물이 제공된 소품 레퍼런스와 전혀 다르게 왜곡되었습니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 17,
     "B": 7
    },
    "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": "S29sh1"
  }
 },
 "S29sh7::variants": {
  "author_fp": "ebe9569b8294cf58",
  "author": {
   "variants": [
    {
     "approach_ko": "얕은 심도로 두 얼굴과 어깨의 접촉을 고립시켜 수리영의 무력함과 인우의 냉랭한 시선을 대비한다.",
     "prompt_en": "Use a restrained, intimate lens feel with shallow depth of field, isolating Suriyoung’s slack left-center profile, the weight of her head on Inwoo’s shoulder, and his rigid three-quarter face at upper-right. Hold crisp focus across their point of contact and Inwoo’s cold gaze toward her, while the cramped wheelhouse and partly visible helm controls fall softly away. Shape the desaturated dusk ambience into gentle facial modeling, preserving natural skin texture and subdued shadow detail."
    },
    {
     "approach_ko": "깊은 초점과 조타실의 빽빽한 선들을 살려 두 사람을 빠져나갈 틈 없는 공간 속에 가둔다.",
     "prompt_en": "Favor a firmer, deeper-focus rendering that keeps both upper bodies and the nearby wheelhouse structure legible, using the confined interior lines to press tightly around the pair. Let Inwoo’s rigid posture form a hard vertical anchor at upper-right while Suriyoung’s gravity-led collapse cuts softly across the left-center; retain the operating side of the helm partly behind him as a quiet trace of the interrupted turn. Keep the dusk light flat, cool, and desaturated, with restrained contrast and tactile wear visible on the surrounding surfaces."
    }
   ]
  },
  "reused": false
 },
 "S29sh7": {
  "input_fingerprint": "d0f4d01ecff3a728",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 힘없이 무너져 내린 수리영의 머리가 자신을 서늘하게 노려보는 인우의 어깨 위에 기대어진 정지 순간.\n\nLOCATION (lock): Inside the fishing boat's narrow wheelhouse at dusk, immediately beside the helm. 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 the pair, dolly backward just enough to gather both upper bodies, then stop precisely as 수리영's head settles on 인우's shoulder. Place her slack profile across the left-center and his rigid three-quarter face at the upper-right, using their physical contact rather than a tighter cut to hold the menace.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 좁은 조타실 내부 (Occupied by 수리영 and 인우); used as Confines the held two-shot and leaves little open space around the bodies; 조타기 (No longer the immediate focus) — Its operating side is partly visible behind 인우's body; used as Remains partly visible behind 인우 as a reminder of the interrupted attempt to turn the boat.\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: Desaturated low-to-mid-key ambient light appropriate to dusk keeps 수리영's collapse tragic while 인우's fixed expression remains cold and 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\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 narrow wheelhouse, steering controls, worn surfaces, dusk light, and circular markings. Preserve both characters' clothing and their positions near the controls while changing them to the collapse described here. Exclude the old photograph and the earlier blocking struggle, showing only the unconscious woman resting on the boatman's shoulder.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): Suriyoung has collapsed against Inwoo with her body slack and her head resting on his shoulder.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suriyoung has collapsed against Inwoo with her body slack and her head resting on his shoulder, still carrying her backpack; absent any shown release, the photograph remains in her hand. The wheelhouse's many circles, including the red one, 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 초반, 긴 검은 머리, 타원형의 젊은 얼굴); 인우 (한국인 남성, 24세, 검은 머리, 수려한 이목구비) — 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 힘없이 무너져 내린 수리영의 머리가 자신을 서늘하게 노려보는 인우의 어깨 위에 기대어진 정지 순간.\n\nLOCATION (lock): Inside the fishing boat's narrow wheelhouse at dusk, immediately beside the helm. 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 the pair, dolly backward just enough to gather both upper bodies, then stop precisely as 수리영's head settles on 인우's shoulder. Place her slack profile across the left-center and his rigid three-quarter face at the upper-right, using their physical contact rather than a tighter cut to hold the menace.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 좁은 조타실 내부 (Occupied by 수리영 and 인우); used as Confines the held two-shot and leaves little open space around the bodies; 조타기 (No longer the immediate focus) — Its operating side is partly visible behind 인우's body; used as Remains partly visible behind 인우 as a reminder of the interrupted attempt to turn the boat.\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: Desaturated low-to-mid-key ambient light appropriate to dusk keeps 수리영's collapse tragic while 인우's fixed expression remains cold and 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\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 narrow wheelhouse, steering controls, worn surfaces, dusk light, and circular markings. Preserve both characters' clothing and their positions near the controls while changing them to the collapse described here. Exclude the old photograph and the earlier blocking struggle, showing only the unconscious woman resting on the boatman's shoulder.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): Suriyoung has collapsed against Inwoo with her body slack and her head resting on his shoulder.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suriyoung has collapsed against Inwoo with her body slack and her head resting on his shoulder, still carrying her backpack; absent any shown release, the photograph remains in her hand. The wheelhouse's many circles, including the red one, 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 초반, 긴 검은 머리, 타원형의 젊은 얼굴); 인우 (한국인 남성, 24세, 검은 머리, 수려한 이목구비) — 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, intimate lens feel with shallow depth of field, isolating Suriyoung’s slack left-center profile, the weight of her head on Inwoo’s shoulder, and his rigid three-quarter face at upper-right. Hold crisp focus across their point of contact and Inwoo’s cold gaze toward her, while the cramped wheelhouse and partly visible helm controls fall softly away. Shape the desaturated dusk ambience into gentle facial modeling, preserving natural skin texture and subdued 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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 힘없이 무너져 내린 수리영의 머리가 자신을 서늘하게 노려보는 인우의 어깨 위에 기대어진 정지 순간.\n\nLOCATION (lock): Inside the fishing boat's narrow wheelhouse at dusk, immediately beside the helm. 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 the pair, dolly backward just enough to gather both upper bodies, then stop precisely as 수리영's head settles on 인우's shoulder. Place her slack profile across the left-center and his rigid three-quarter face at the upper-right, using their physical contact rather than a tighter cut to hold the menace.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 좁은 조타실 내부 (Occupied by 수리영 and 인우); used as Confines the held two-shot and leaves little open space around the bodies; 조타기 (No longer the immediate focus) — Its operating side is partly visible behind 인우's body; used as Remains partly visible behind 인우 as a reminder of the interrupted attempt to turn the boat.\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: Desaturated low-to-mid-key ambient light appropriate to dusk keeps 수리영's collapse tragic while 인우's fixed expression remains cold and 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\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 narrow wheelhouse, steering controls, worn surfaces, dusk light, and circular markings. Preserve both characters' clothing and their positions near the controls while changing them to the collapse described here. Exclude the old photograph and the earlier blocking struggle, showing only the unconscious woman resting on the boatman's shoulder.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): Suriyoung has collapsed against Inwoo with her body slack and her head resting on his shoulder.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suriyoung has collapsed against Inwoo with her body slack and her head resting on his shoulder, still carrying her backpack; absent any shown release, the photograph remains in her hand. The wheelhouse's many circles, including the red one, 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 초반, 긴 검은 머리, 타원형의 젊은 얼굴); 인우 (한국인 남성, 24세, 검은 머리, 수려한 이목구비) — 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 a firmer, deeper-focus rendering that keeps both upper bodies and the nearby wheelhouse structure legible, using the confined interior lines to press tightly around the pair. Let Inwoo’s rigid posture form a hard vertical anchor at upper-right while Suriyoung’s gravity-led collapse cuts softly across the left-center; retain the operating side of the helm partly behind him as a quiet trace of the interrupted turn. Keep the dusk light flat, cool, and desaturated, with restrained contrast and tactile wear visible on the surrounding 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": 4
  },
  "selected": "A",
  "ranking": [
   "A",
   "B"
  ],
  "verdicts": [
   {
    "label": "A",
    "score": 7,
    "verdict_ko": "인우 뒤로 조타기를 배치한 배경 구도와 카메라를 노려보는 서늘한 시선, 요구된 미디엄 샷 프레이밍을 정확히 구현했습니다."
   },
   {
    "label": "B",
    "score": 4,
    "verdict_ko": "조타기가 인우 앞에 전경으로 배치되어 위치 지시를 위반했으며, 요구된 서늘한 시선 처리가 부족합니다."
   }
  ],
  "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/3d61971a-7158-40f4-b742-8fd5e757dbbf/images/132fe629-d451-4f2f-ac9c-c33beb62bb0f/scene/recipe/S29sh3_sel.png"
   },
   {
    "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:1254817>"
   },
   {
    "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.",
    "path": "<bytes:420475>"
   }
  ],
  "critique": {
   "issues": [
    {
     "issue_ko": "인우의 시선이 카메라 쪽(정면)을 향하고 있으며, 수리영을 서늘하게 노려본다는 지시와 어긋납니다.",
     "fix_en": "Redirect Inwoo's eyes to look down and to his right, fixing his cold gaze directly on Suriyoung."
    },
    {
     "issue_ko": "화면 우측 가장자리에 있는 흰색 구조물 표면에 검은색 텍스트가 생성되어 '텍스트 없음' 규칙을 위반했습니다.",
     "fix_en": "Remove the black text from the white panel on the right edge of the frame, leaving a blank surface."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Redirect Inwoo’s gaze downward toward Suriyoung while preserving the camera, framing, poses, and all other elements.",
     "issue_ko": "인우의 시선이 어깨에 기대어진 수리영을 향하지 않고 화면 앞쪽을 바라봐, 그녀를 서늘하게 노려보는 시선 축이 어긋나 있다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Redirect Inwoo's eyes to look down and to his right, fixing his cold gaze directly on Suriyoung.\n- Remove the black text from the white panel on the right edge of the frame, leaving a blank surface.\n- Redirect Inwoo’s gaze downward toward Suriyoung while preserving the camera, framing, poses, and all other elements.\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": "인우의 시선이 수리영을 향하도록 연출하여 프롬프트의 시선 지시를 정확히 이행했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "인우가 카메라 렌즈를 정면으로 응시하여 '수리영을 향한 시선'이라는 지시를 위반했습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 8,
      "verdict_ko": "인우의 시선이 수리영을 향하도록 연출하여 프롬프트의 시선 지시를 정확히 이행했습니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "인우가 카메라 렌즈를 정면으로 응시하여 '수리영을 향한 시선'이라는 지시를 위반했습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "인우의 시선이 수리영을 향하고 있어 '자신을 서늘하게 노려보는' 및 영문 지침(cold gaze toward her)을 정확히 충족하며 프레이밍도 우수합니다."
     },
     {
      "label": "B",
      "score": 5,
      "verdict_ko": "인우가 수리영이 아닌 카메라 렌즈를 정면으로 응시하고 있어 지정된 시선 방향 지침을 위반했습니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "인우의 시선이 수리영을 향하고 있어 '자신을 서늘하게 노려보는' 및 영문 지침(cold gaze toward her)을 정확히 충족하며 프레이밍도 우수합니다."
     },
     {
      "label": "A",
      "score": 5,
      "verdict_ko": "인우가 수리영이 아닌 카메라 렌즈를 정면으로 응시하고 있어 지정된 시선 방향 지침을 위반했습니다."
     }
    ]
   },
   "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": "S29sh3"
  }
 },
 "S30sh3::variants": {
  "author_fp": "064afe149c1d4267",
  "author": {
   "variants": [
    {
     "approach_ko": "절제된 망원 압축으로 어선과 안개 경계를 밀착시키고, 길게 뻗은 항적을 엄숙한 중심축으로 강조한다.",
     "prompt_en": "Use a restrained telephoto feel from the prescribed low trailing position, compressing the small upper-center boat against the dense fog boundary while preserving its environmental scale. Hold the composition in austere symmetry, with the continuous white wake forming a clean central axis through the darker sea. Keep the fog softly opaque and the muted golden reflection narrowly contained around the route, creating a solemn, compressed sense of disappearance."
    },
    {
     "approach_ko": "수면 가까운 광각 원근으로 전경 항적의 질감과 안개 속으로 빨려 들어가는 깊이를 몰입감 있게 살린다.",
     "prompt_en": "Use an immersive wide-angle feel at the specified low elevation, letting the wake occupy the foreground with strong natural perspective before tapering toward the small stern in the upper-center. Render the nearby water and churned white trail with crisp tactile detail, then allow contrast and definition to fall away progressively into the gray fog. Let the subdued sunset sheen spread across the disturbed surface without overpowering the desaturated, low-key atmosphere."
    }
   ]
  },
  "reused": false
 },
 "S30sh3": {
  "input_fingerprint": "0e39e26837d97f39",
  "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): sunset, foggy.\n\nSHOT TEXT (authoritative, Korean): 소형 어선의 뱃머리가 짙은 회색 안개 덩어리 속으로 절반쯤 들어간 채, 물살을 가르는 하얀 항적이 배 뒤편으로 길게 이어진 뒷모습.\n\nLOCATION (lock): On the open sea at dusk, where the small fishing boat is crossing into a dense bank of gray fog. 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 from a low elevation several boat lengths behind, centered precisely on the small fishing boat's stern and forward course as its bow disappears halfway into the fog bank. The boat remains in the upper-center at environmental scale, while the long white wake begins beneath it and runs toward the lower-center foreground as the principal depth line.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: dense fog bank in the upper-center of the frame, background; long white wake in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 소형 어선 (Bow partly inside the fog) — Its stern faces the camera while its bow points into the upper-center fog bank; used as Primary moving subject seen from behind as it crosses the visual boundary; 짙은 안개 띠 (Dense and erasing the distant boundary); used as Forms the upper-center boundary that obscures the boat's forward half; 하얀 항적 (Long and continuous behind the moving boat); used as Creates a long central perspective line from the stern toward the camera; 바다 (Cut by the boat and its wake) — The surface reflects the setting sun as a golden path beneath the gray fog boundary; used as Carries the setting sun's explicitly described golden reflection around the boat's route; 지는 해 (Slowly setting); used as Provides the distant source of the golden reflection and the scene's final temporal marker.\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: Muted golden sunset reflection meets the dense gray fog in a desaturated, solemn low-to-mid-key finish.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The unconscious Suri-young remains aboard In-woo's boat with her backpack and the old photograph as the vessel enters the fog; no recovery, unloading, or removal of her belongings is shown.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): sunset, foggy.\n\nSHOT TEXT (authoritative, Korean): 소형 어선의 뱃머리가 짙은 회색 안개 덩어리 속으로 절반쯤 들어간 채, 물살을 가르는 하얀 항적이 배 뒤편으로 길게 이어진 뒷모습.\n\nLOCATION (lock): On the open sea at dusk, where the small fishing boat is crossing into a dense bank of gray fog. 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 from a low elevation several boat lengths behind, centered precisely on the small fishing boat's stern and forward course as its bow disappears halfway into the fog bank. The boat remains in the upper-center at environmental scale, while the long white wake begins beneath it and runs toward the lower-center foreground as the principal depth line.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: dense fog bank in the upper-center of the frame, background; long white wake in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 소형 어선 (Bow partly inside the fog) — Its stern faces the camera while its bow points into the upper-center fog bank; used as Primary moving subject seen from behind as it crosses the visual boundary; 짙은 안개 띠 (Dense and erasing the distant boundary); used as Forms the upper-center boundary that obscures the boat's forward half; 하얀 항적 (Long and continuous behind the moving boat); used as Creates a long central perspective line from the stern toward the camera; 바다 (Cut by the boat and its wake) — The surface reflects the setting sun as a golden path beneath the gray fog boundary; used as Carries the setting sun's explicitly described golden reflection around the boat's route; 지는 해 (Slowly setting); used as Provides the distant source of the golden reflection and the scene's final temporal marker.\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: Muted golden sunset reflection meets the dense gray fog in a desaturated, solemn low-to-mid-key finish.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The unconscious Suri-young remains aboard In-woo's boat with her backpack and the old photograph as the vessel enters the fog; no recovery, unloading, or removal of her belongings is shown.\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 telephoto feel from the prescribed low trailing position, compressing the small upper-center boat against the dense fog boundary while preserving its environmental scale. Hold the composition in austere symmetry, with the continuous white wake forming a clean central axis through the darker sea. Keep the fog softly opaque and the muted golden reflection narrowly contained around the route, creating a solemn, compressed sense of disappearance.\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 — contemporary South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): sunset, foggy.\n\nSHOT TEXT (authoritative, Korean): 소형 어선의 뱃머리가 짙은 회색 안개 덩어리 속으로 절반쯤 들어간 채, 물살을 가르는 하얀 항적이 배 뒤편으로 길게 이어진 뒷모습.\n\nLOCATION (lock): On the open sea at dusk, where the small fishing boat is crossing into a dense bank of gray fog. 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 from a low elevation several boat lengths behind, centered precisely on the small fishing boat's stern and forward course as its bow disappears halfway into the fog bank. The boat remains in the upper-center at environmental scale, while the long white wake begins beneath it and runs toward the lower-center foreground as the principal depth line.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: dense fog bank in the upper-center of the frame, background; long white wake in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 소형 어선 (Bow partly inside the fog) — Its stern faces the camera while its bow points into the upper-center fog bank; used as Primary moving subject seen from behind as it crosses the visual boundary; 짙은 안개 띠 (Dense and erasing the distant boundary); used as Forms the upper-center boundary that obscures the boat's forward half; 하얀 항적 (Long and continuous behind the moving boat); used as Creates a long central perspective line from the stern toward the camera; 바다 (Cut by the boat and its wake) — The surface reflects the setting sun as a golden path beneath the gray fog boundary; used as Carries the setting sun's explicitly described golden reflection around the boat's route; 지는 해 (Slowly setting); used as Provides the distant source of the golden reflection and the scene's final temporal marker.\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: Muted golden sunset reflection meets the dense gray fog in a desaturated, solemn low-to-mid-key finish.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The unconscious Suri-young remains aboard In-woo's boat with her backpack and the old photograph as the vessel enters the fog; no recovery, unloading, or removal of her belongings is shown.\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 feel at the specified low elevation, letting the wake occupy the foreground with strong natural perspective before tapering toward the small stern in the upper-center. Render the nearby water and churned white trail with crisp tactile detail, then allow contrast and definition to fall away progressively into the gray fog. Let the subdued sunset sheen spread across the disturbed surface without overpowering the desaturated, low-key atmosphere.\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": 8,
   "A": 6
  },
  "selected": "B",
  "ranking": [
   "B",
   "A"
  ],
  "verdicts": [
   {
    "label": "B",
    "score": 8,
    "verdict_ko": "뱃머리가 안개 속으로 절반쯤 들어간 모습을 가장 정확하게 연출했으며 구도와 항적의 표현이 지시사항과 잘 일치합니다."
   },
   {
    "label": "A",
    "score": 6,
    "verdict_ko": "구도와 항적, 노을빛의 묘사는 우수하나, 배 전체가 선명하게 보여 '뱃머리가 안개 속으로 절반쯤 들어간' 핵심 연출을 누락했습니다."
   }
  ],
  "refs": [],
  "critique": {
   "issues": [
    {
     "issue_ko": "어선의 뱃머리가 안개 속으로 절반쯤 들어간 모습이 아니라 배 전체가 선명하게 보입니다.",
     "fix_en": "Obscure the forward half of the boat in the dense gray fog so only the stern remains clearly visible."
    }
   ]
  },
  "composition_critique": {
   "issues": [
    {
     "fix_en": "Lower the camera to a near-waterline tracking height while preserving the centered stern axis and wide framing.",
     "issue_ko": "카메라가 지정된 낮은 고도보다 높아 항적과 수면을 과도하게 내려다보고 있다."
    },
    {
     "fix_en": "Move the boat farther into the upper-center fog bank so its bow and forward half are obscured while the stern remains visible.",
     "issue_ko": "배 전체가 안개 앞에서 선명하게 보여, 뱃머리와 전반부가 안개 속에 절반쯤 사라져야 하는 깊이 배치가 구현되지 않았다."
    }
   ]
  },
  "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Obscure the forward half of the boat in the dense gray fog so only the stern remains clearly visible.\n- Lower the camera to a near-waterline tracking height while preserving the centered stern axis and wide framing.\n- Move the boat farther into the upper-center fog bank so its bow and forward half are obscured while the stern remains visible.\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": "짙은 안개 속으로 들어가는 어선의 뒷모습, 길게 이어진 하얀 항적, 해질녘의 황금빛 반사가 지시된 구도와 앵글에 완벽하게 부합합니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "요구된 모든 요소를 정확한 구도로 보여주나, 선미 부분까지 안개가 덮여 있어 '선수만 안개에 반쯤 들어간' 디테일과 선명도 면에서 A보다 약간 아쉽습니다."
     }
    ]
   },
   "forward_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "짙은 안개 속으로 들어가는 어선의 뒷모습, 길게 이어진 하얀 항적, 해질녘의 황금빛 반사가 지시된 구도와 앵글에 완벽하게 부합합니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "요구된 모든 요소를 정확한 구도로 보여주나, 선미 부분까지 안개가 덮여 있어 '선수만 안개에 반쯤 들어간' 디테일과 선명도 면에서 A보다 약간 아쉽습니다."
     }
    ]
   },
   "reverse_raw": {
    "winner": "B",
    "ranking": [
     "B",
     "A"
    ],
    "verdicts": [
     {
      "label": "B",
      "score": 7,
      "verdict_ko": "선미가 선명하게 보이고 선수가 안개 속으로 사라지는 모습을 잘 표현하여 프롬프트의 요구사항을 훌륭하게 충족합니다."
     },
     {
      "label": "A",
      "score": 6,
      "verdict_ko": "구도와 분위기는 좋으나 배 전체가 균일하게 안개에 가려져 있어 선수가 반쯤 들어간 묘사가 덜 명확합니다."
     }
    ]
   },
   "reverse_normalized": {
    "winner": "A",
    "ranking": [
     "A",
     "B"
    ],
    "verdicts": [
     {
      "label": "A",
      "score": 7,
      "verdict_ko": "선미가 선명하게 보이고 선수가 안개 속으로 사라지는 모습을 잘 표현하여 프롬프트의 요구사항을 훌륭하게 충족합니다."
     },
     {
      "label": "B",
      "score": 6,
      "verdict_ko": "구도와 분위기는 좋으나 배 전체가 균일하게 안개에 가려져 있어 선수가 반쯤 들어간 묘사가 덜 명확합니다."
     }
    ]
   },
   "combined": {
    "totals": {
     "A": 14,
     "B": 12
    },
    "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"
  }
 }
}