{ "benchmark": { "name": "SVGBench", "source": "https://github.com/johnbean393/SVGBench", "questions": { "source": "https://raw.githubusercontent.com/johnbean393/SVGBench/dd42e2caf71104ff1ae096b339321b67c52fc4a8/questions/questions.json", "commit": "dd42e2caf71104ff1ae096b339321b67c52fc4a8", "commit_date": "2025-07-05T14:24:50Z", "fetched": "2026-09-06", "question_count": 105 }, "scoring": "score = requirements met / requirements total, judged on the Chrome-rendered PNG; pair mean_score = mean over scored artifacts", "comparability": "subset of the 105 questions; artifacts came from coding harnesses, not the SVGBench direct-API run; every artifact was judged by one judge, claude-opus-5, which is itself a contestant here — 21 of the 84 scored artifacts are its own, so the claude/claude-opus-5 row is self-judged. Not comparable to the upstream leaderboard." }, "generated_at": "2026-09-15T19:10:26", "pairs": [ { "harness": "claude", "model": "fable-5-1", "artifacts_scored": 4, "mean_score": 0.7291666666666666, "by_question": { "0": 0.7291666666666666 } }, { "harness": "claude", "model": "claude-opus-4-8", "artifacts_scored": 1, "mean_score": 0.7142857142857143, "by_question": { "7": 0.7142857142857143 } }, { "harness": "claude", "model": "claude-opus-5", "artifacts_scored": 21, "mean_score": 0.6934996220710506, "by_question": { "0": 0.75, "4": 0.6944444444444444, "5": 0.475, "7": 0.8571428571428571, "13": 0.6499999999999999 } }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifacts_scored": 25, "mean_score": 0.5909523809523809, "by_question": { "0": 0.5625, "4": 0.5555555555555556, "6": 0.5, "7": 0.7857142857142857, "12": 0.25 } }, { "harness": "opencode", "model": "qwen3.8-27b-8bit", "artifacts_scored": 5, "mean_score": 0.4990476190476191, "by_question": { "0": 0.5833333333333334, "5": 0.5, "7": 0.42857142857142855, "13": 0.4 } }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifacts_scored": 28, "mean_score": 0.33624381570810147, "by_question": { "0": 0.45833333333333337, "4": 0.25, "5": 0.35, "6": 0.46875, "7": 0.5357142857142857, "12": 0.09090909090909091, "13": 0.2 } } ], "artifacts": [ { "harness": "claude", "model": "claude-opus-4-8", "artifact": "logs/claude/claude-opus-4-8/dolphin-hoop.svg", "question_index": 7, "animated": false, "slug": "dolphin-hoop", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 0.7142857142857143, "met": 5, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-4-8/dolphin-hoop.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": true, "note": "" }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": false, "note": "Water surface is present, but the splash is a small three-bump cluster with three bubbles sitting ~80px to the left of where the dolphin crosses the waterline; it is neither large nor at the exit point." }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": false, "note": "The tail stock and fluke are drawn below the water surface line, so the body is not entirely in mid-air; the pose is also a straight diagonal rather than an arc." }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": true, "note": "" }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": true, "note": "" } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/cow-plowing-animated.svg", "question_index": 0, "animated": true, "slug": "cow-plowing-animated", "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "score": 0.75, "met": 9, "total": 12, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/cow-plowing-animated.png", "requirements": [ { "text": "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "met": true, "note": "" }, { "text": "The cow must have black and white patches for its coloring.", "met": true, "note": "" }, { "text": "Add cow ears, eyes, and snout for facial recognition.", "met": true, "note": "" }, { "text": "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "met": true, "note": "" }, { "text": "The cow's hooves must be colored brown, as if covered in soil from the field.", "met": false, "note": "Hoof fill samples RGB (35,31,29) and (58,51,46) - a neutral near-black with a red-minus-blue spread of 6-12; the hooves are black, not brown." }, { "text": "Include a traditional wooden plow with a visible metal blade/share.", "met": true, "note": "" }, { "text": "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "met": false, "note": "A padded brown collar sits at the shoulders, but the two traces running back to the plow are plain solid leather straps with no links anywhere along them; no chains are visible." }, { "text": "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "met": true, "note": "" }, { "text": "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "met": true, "note": "" }, { "text": "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "met": true, "note": "" }, { "text": "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "met": false, "note": "The unplowed section is a tan/khaki stubble field, and its tufts are drawn in ochre brown, not green; no green grass, and none is shown being overturned at the share." }, { "text": "Add a simple background with a clear horizon line and a blue sky containing a yellow sun.", "met": true, "note": "" } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/cow-plowing-v2.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v2", "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "score": 0.75, "met": 9, "total": 12, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/cow-plowing-v2.png", "requirements": [ { "text": "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "met": true, "note": "" }, { "text": "The cow must have black and white patches for its coloring.", "met": true, "note": "" }, { "text": "Add cow ears, eyes, and snout for facial recognition.", "met": true, "note": "" }, { "text": "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "met": true, "note": "" }, { "text": "The cow's hooves must be colored brown, as if covered in soil from the field.", "met": false, "note": "Hoof fill samples RGB (35,31,29) and (58,51,46) - a neutral near-black with a red-minus-blue spread of 6-12; the hooves are black, not brown." }, { "text": "Include a traditional wooden plow with a visible metal blade/share.", "met": true, "note": "" }, { "text": "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "met": false, "note": "A padded brown collar sits at the shoulders, but the two traces running back to the plow are plain solid leather straps with no links anywhere along them; no chains are visible." }, { "text": "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "met": true, "note": "" }, { "text": "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "met": true, "note": "" }, { "text": "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "met": true, "note": "" }, { "text": "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "met": false, "note": "The unplowed section is a tan/khaki stubble field, and its tufts are drawn in ochre brown, not green; no green grass, and none is shown being overturned at the share." }, { "text": "Add a simple background with a clear horizon line and a blue sky containing a yellow sun.", "met": true, "note": "" } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/cow-plowing-v3.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v3", "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "score": 0.75, "met": 9, "total": 12, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/cow-plowing-v3.png", "requirements": [ { "text": "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "met": true, "note": "" }, { "text": "The cow must have black and white patches for its coloring.", "met": true, "note": "" }, { "text": "Add cow ears, eyes, and snout for facial recognition.", "met": true, "note": "" }, { "text": "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "met": true, "note": "" }, { "text": "The cow's hooves must be colored brown, as if covered in soil from the field.", "met": false, "note": "Hoof fill samples RGB (35,31,29) and (58,51,46) - a neutral near-black with a red-minus-blue spread of 6-12; the hooves are black, not brown." }, { "text": "Include a traditional wooden plow with a visible metal blade/share.", "met": true, "note": "" }, { "text": "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "met": false, "note": "A padded brown collar sits at the shoulders, but the two traces running back to the plow are plain solid leather straps with no links anywhere along them; no chains are visible." }, { "text": "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "met": true, "note": "" }, { "text": "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "met": true, "note": "" }, { "text": "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "met": true, "note": "" }, { "text": "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "met": false, "note": "The unplowed section is a tan/khaki stubble field, and its tufts are drawn in ochre brown, not green; no green grass, and none is shown being overturned at the share." }, { "text": "Add a simple background with a clear horizon line and a blue sky containing a yellow sun.", "met": true, "note": "" } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/cow-plowing.svg", "question_index": 0, "animated": false, "slug": "cow-plowing", "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "score": 0.75, "met": 9, "total": 12, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/cow-plowing.png", "requirements": [ { "text": "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "met": true, "note": "" }, { "text": "The cow must have black and white patches for its coloring.", "met": true, "note": "" }, { "text": "Add cow ears, eyes, and snout for facial recognition.", "met": true, "note": "" }, { "text": "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "met": true, "note": "" }, { "text": "The cow's hooves must be colored brown, as if covered in soil from the field.", "met": false, "note": "Hoof fill samples RGB (35,31,29) and (58,51,46) - a neutral near-black with a red-minus-blue spread of 6-12; the hooves are black, not brown." }, { "text": "Include a traditional wooden plow with a visible metal blade/share.", "met": true, "note": "" }, { "text": "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "met": false, "note": "A padded brown collar sits at the shoulders, but the two traces running back to the plow are plain solid leather straps with no links anywhere along them; no chains are visible." }, { "text": "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "met": true, "note": "" }, { "text": "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "met": true, "note": "" }, { "text": "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "met": true, "note": "" }, { "text": "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "met": false, "note": "The unplowed section is a tan/khaki stubble field, and its tufts are drawn in ochre brown, not green; no green grass, and none is shown being overturned at the share." }, { "text": "Add a simple background with a clear horizon line and a blue sky containing a yellow sun.", "met": true, "note": "" } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/dolphin-animated.svg", "question_index": 7, "animated": true, "slug": "dolphin-animated", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 1.0, "met": 7, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/dolphin-animated.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": true, "note": "" }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": true, "note": "" }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": true, "note": "" }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": true, "note": "" }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": true, "note": "" } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/dolphin-hoop-jump.svg", "question_index": 7, "animated": false, "slug": "dolphin-hoop-jump", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 0.5714285714285714, "met": 4, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/dolphin-hoop-jump.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": true, "note": "" }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": true, "note": "" }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": false, "note": "The tail stock and fluke are drawn well below the water surface line, inside the blue water, so the body is not entirely in mid-air." }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": false, "note": "The hoop is a single flat orange colour with no second colour anywhere on the ring, so it is not multi-coloured." }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": true, "note": "" }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": false, "note": "The snout is drawn closed - a smooth grey beak with a white underside and no gape - although the fish is held at its tip." } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/dolphin-v2.svg", "question_index": 7, "animated": false, "slug": "dolphin-v2", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 0.8571428571428571, "met": 6, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/dolphin-v2.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": true, "note": "" }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": true, "note": "" }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": true, "note": "" }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": true, "note": "" }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": false, "note": "The beak is parted only as a thin dark sliver (gape angle around 10 degrees) at the snout tip; that reads as a slightly open mouth, not a wide-open one." } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/dolphin-v3.svg", "question_index": 7, "animated": false, "slug": "dolphin-v3", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 1.0, "met": 7, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/dolphin-v3.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": true, "note": "" }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": true, "note": "" }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": true, "note": "" }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": true, "note": "" }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": true, "note": "" } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/dolphin.svg", "question_index": 7, "animated": false, "slug": "dolphin", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 0.8571428571428571, "met": 6, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/dolphin.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": true, "note": "" }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": true, "note": "" }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": true, "note": "" }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": true, "note": "" }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": false, "note": "The rostrum is drawn closed: a single dark jaw line with a thin white sliver beneath it and no gape, even though the fish touches the snout tip." } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/picnic-on-clouds-animated.svg", "question_index": 5, "animated": true, "slug": "picnic-on-clouds-animated", "prompt": "Write `svg` code for an image of a picnic on top of the clouds, where 2 parents and 2 children have landed with a hot air ballon, and are setting up a picnic with a tarp and food items.", "score": 0.5, "met": 5, "total": 10, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/picnic-on-clouds-animated.png", "requirements": [ { "text": "Create a scene set on top of a solid-looking cloudscape that serves as the ground.", "met": true, "note": "" }, { "text": "Include a hot air balloon with a large envelope featuring vertical red and white stripes and a brown wicker basket.", "met": true, "note": "" }, { "text": "The balloon's basket must be resting firmly on the surface of a large, flat-topped cloud.", "met": true, "note": "" }, { "text": "Show ropes connecting the balloon's envelope to the basket.", "met": true, "note": "" }, { "text": "Include four human figures styled as a family: two adults and two children.", "met": true, "note": "" }, { "text": "Position one adult figure holding a corner of a red and white checkered picnic blanket, while one child figure holds the opposite corner, as if they are spreading it out together on the cloud.", "met": false, "note": "The blanket is already spread flat on the cloud; the woman stands on it with her arms down and no child holds anything, so no adult-and-child pair grips opposite corners." }, { "text": "Place an open picnic basket on a corner of the blanket that is already spread out.", "met": false, "note": "The basket sits at the middle of the blanket's left edge - its contact footprint is about 17% across the blanket's width and 63% of the way from the far edge to the near edge, i.e. in a mid-edge cell rather than a corner cell." }, { "text": "A thermos, a bunch of grapes, and a sandwich must be visible emerging from the open picnic basket.", "met": false, "note": "Only a baguette emerges from the picnic basket; the teal thermos, the orange bottle, the purple grapes and the sandwiches are all drawn standing on the blanket beside the basket." }, { "text": "Position the second adult and second child near the landed hot air balloon, with the adult pointing up at the balloon's envelope.", "met": false, "note": "The man beside the balloon holds a hamper in both hands rather than pointing up at the envelope, and no child stands near the balloon." }, { "text": "The background must be a clear blue sky containing a bright yellow sun and two small, distant clouds.", "met": false, "note": "Six small distant clouds are drawn in the sky - at roughly (85,85), (405,120), (620,45), (540,240), (890,230) and (375,300) - not two." } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/picnic-on-clouds-v2.svg", "question_index": 5, "animated": false, "slug": "picnic-on-clouds-v2", "prompt": "Write `svg` code for an image of a picnic on top of the clouds, where 2 parents and 2 children have landed with a hot air ballon, and are setting up a picnic with a tarp and food items.", "score": 0.5, "met": 5, "total": 10, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/picnic-on-clouds-v2.png", "requirements": [ { "text": "Create a scene set on top of a solid-looking cloudscape that serves as the ground.", "met": true, "note": "" }, { "text": "Include a hot air balloon with a large envelope featuring vertical red and white stripes and a brown wicker basket.", "met": true, "note": "" }, { "text": "The balloon's basket must be resting firmly on the surface of a large, flat-topped cloud.", "met": true, "note": "" }, { "text": "Show ropes connecting the balloon's envelope to the basket.", "met": true, "note": "" }, { "text": "Include four human figures styled as a family: two adults and two children.", "met": true, "note": "" }, { "text": "Position one adult figure holding a corner of a red and white checkered picnic blanket, while one child figure holds the opposite corner, as if they are spreading it out together on the cloud.", "met": false, "note": "The blanket is already spread flat on the cloud. The woman stands on it with her arms hanging down and her hands resting on its surface, and no child holds anything; no figure grips a corner, so the two-person spreading pose is absent." }, { "text": "Place an open picnic basket on a corner of the blanket that is already spread out.", "met": false, "note": "The basket sits at the middle of the blanket's left edge - its contact footprint is about 11% across the blanket's width but 60% of the way from the far edge to the near edge, i.e. in a mid-edge cell rather than a corner cell." }, { "text": "A thermos, a bunch of grapes, and a sandwich must be visible emerging from the open picnic basket.", "met": false, "note": "Only a baguette emerges from the picnic basket. The orange thermos bottle, the bunch of purple grapes and the sandwich plate are all drawn separately on the blanket, none of them emerging from the basket." }, { "text": "Position the second adult and second child near the landed hot air balloon, with the adult pointing up at the balloon's envelope.", "met": false, "note": "No child stands near the balloon, and the adult who does (the man in green) holds a hamper with both hands rather than pointing up at the envelope. The second adult is at the blanket, about 350 px away." }, { "text": "The background must be a clear blue sky containing a bright yellow sun and two small, distant clouds.", "met": false, "note": "Five small distant clouds are drawn in the sky - at roughly (85,90), (420,125), (625,45), (545,245) and (890,225) - not two." } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/picnic-on-clouds-v3.svg", "question_index": 5, "animated": false, "slug": "picnic-on-clouds-v3", "prompt": "Write `svg` code for an image of a picnic on top of the clouds, where 2 parents and 2 children have landed with a hot air ballon, and are setting up a picnic with a tarp and food items.", "score": 0.5, "met": 5, "total": 10, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/picnic-on-clouds-v3.png", "requirements": [ { "text": "Create a scene set on top of a solid-looking cloudscape that serves as the ground.", "met": true, "note": "" }, { "text": "Include a hot air balloon with a large envelope featuring vertical red and white stripes and a brown wicker basket.", "met": true, "note": "" }, { "text": "The balloon's basket must be resting firmly on the surface of a large, flat-topped cloud.", "met": true, "note": "" }, { "text": "Show ropes connecting the balloon's envelope to the basket.", "met": true, "note": "" }, { "text": "Include four human figures styled as a family: two adults and two children.", "met": true, "note": "" }, { "text": "Position one adult figure holding a corner of a red and white checkered picnic blanket, while one child figure holds the opposite corner, as if they are spreading it out together on the cloud.", "met": false, "note": "The blanket is already spread flat on the cloud, the woman merely stands on it with her arms down, and no child holds anything; no figure grips a corner of the blanket." }, { "text": "Place an open picnic basket on a corner of the blanket that is already spread out.", "met": false, "note": "The basket sits at the middle of the blanket's left edge - its contact footprint is about 17% across the blanket's width and 63% of the way from the far edge to the near edge, i.e. in a mid-edge cell rather than a corner cell." }, { "text": "A thermos, a bunch of grapes, and a sandwich must be visible emerging from the open picnic basket.", "met": false, "note": "Only a baguette emerges from the picnic basket. The teal thermos, the orange bottle, the bunch of purple grapes and the sandwiches on the plate all stand separately on the blanket beside the basket, not emerging from it." }, { "text": "Position the second adult and second child near the landed hot air balloon, with the adult pointing up at the balloon's envelope.", "met": false, "note": "The man beside the balloon carries a hamper in both hands and does not point up at the envelope, and no child is near him; the second adult and both children are at the blanket." }, { "text": "The background must be a clear blue sky containing a bright yellow sun and two small, distant clouds.", "met": false, "note": "Six small distant clouds are drawn in the sky - at roughly (85,85), (405,120), (620,45), (540,240), (890,230) and (370,300) - not two." } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/picnic-on-clouds.svg", "question_index": 5, "animated": false, "slug": "picnic-on-clouds", "prompt": "Write `svg` code for an image of a picnic on top of the clouds, where 2 parents and 2 children have landed with a hot air ballon, and are setting up a picnic with a tarp and food items.", "score": 0.4, "met": 4, "total": 10, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/picnic-on-clouds.png", "requirements": [ { "text": "Create a scene set on top of a solid-looking cloudscape that serves as the ground.", "met": true, "note": "" }, { "text": "Include a hot air balloon with a large envelope featuring vertical red and white stripes and a brown wicker basket.", "met": false, "note": "The envelope's vertical stripes are gold, red and blue - sampled at (150,200) RGB 226,190,76 and (225,200) RGB 39,104,164. There is no white stripe anywhere on the envelope." }, { "text": "The balloon's basket must be resting firmly on the surface of a large, flat-topped cloud.", "met": true, "note": "" }, { "text": "Show ropes connecting the balloon's envelope to the basket.", "met": true, "note": "" }, { "text": "Include four human figures styled as a family: two adults and two children.", "met": true, "note": "" }, { "text": "Position one adult figure holding a corner of a red and white checkered picnic blanket, while one child figure holds the opposite corner, as if they are spreading it out together on the cloud.", "met": false, "note": "The checkered blanket lies flat on the cloud with nobody touching it. No adult and no child holds any corner, so the spreading-it-out pose is absent." }, { "text": "Place an open picnic basket on a corner of the blanket that is already spread out.", "met": false, "note": "The basket sits at the middle of the blanket's left edge - its contact footprint is about 26% across the blanket's width and 62% of the way from the far edge to the near edge, i.e. in a mid-edge cell rather than a corner cell." }, { "text": "A thermos, a bunch of grapes, and a sandwich must be visible emerging from the open picnic basket.", "met": false, "note": "Only a baguette emerges from the picnic basket. The orange bottle, the fruit bowl and the sandwich plate all sit separately on the blanket, and no bunch of grapes is drawn anywhere in the render." }, { "text": "Position the second adult and second child near the landed hot air balloon, with the adult pointing up at the balloon's envelope.", "met": false, "note": "The second adult (the woman) stands in the middle of the scene about 270 px from the balloon with both arms down, and the second child stands further right still. The only adult near the balloon holds a hamper with his arm extended sideways, not pointing up at the envelope." }, { "text": "The background must be a clear blue sky containing a bright yellow sun and two small, distant clouds.", "met": false, "note": "Four small distant clouds are drawn in the sky - at roughly (100,90), (430,155), (640,50) and (890,205) - not two." } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/rubber-ducky-animated.svg", "question_index": 4, "animated": true, "slug": "rubber-ducky-animated", "prompt": "Write `svg` code for an image of a rubber ducky floating on top of a soapy bathtub.", "score": 0.8888888888888888, "met": 8, "total": 9, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/rubber-ducky-animated.png", "requirements": [ { "text": "Create a classic rubber ducky shape with a distinct body and head, colored bright yellow.", "met": true, "note": "" }, { "text": "The duck must have an orange beak and a simple black dot for an eye.", "met": true, "note": "" }, { "text": "Draw a white, claw-foot bathtub shape, showing the inside view with a visible rim.", "met": true, "note": "" }, { "text": "Fill the lower portion of the bathtub with light blue water.", "met": true, "note": "" }, { "text": "Ensure the water line is clearly visible across the duck's body, showing the lower third of the duck submerged.", "met": true, "note": "" }, { "text": "Position the duck so it is floating on the water's surface, creating small, concentric circular ripples in the water around its base.", "met": true, "note": "" }, { "text": "Depict soap bubbles as clusters of overlapping circles with a slight iridescence, using semi-transparent white, light pink, and light blue fills.", "met": false, "note": "All the foam and floating bubbles are white with pale-blue shading; the only pink in the render is the soap bar. No light pink bubble fill." }, { "text": "Place a small cluster of bubbles on top of the duck's head.", "met": true, "note": "" }, { "text": "Place a large pile of bubbles against one side of the tub and a few floating on the water's surface around the duck.", "met": true, "note": "" } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/rubber-ducky-v2.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky-v2", "prompt": "Write `svg` code for an image of a rubber ducky floating on top of a soapy bathtub.", "score": 0.5555555555555556, "met": 5, "total": 9, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/rubber-ducky-v2.png", "requirements": [ { "text": "Create a classic rubber ducky shape with a distinct body and head, colored bright yellow.", "met": true, "note": "" }, { "text": "The duck must have an orange beak and a simple black dot for an eye.", "met": true, "note": "" }, { "text": "Draw a white, claw-foot bathtub shape, showing the inside view with a visible rim.", "met": true, "note": "" }, { "text": "Fill the lower portion of the bathtub with light blue water.", "met": false, "note": "The tub interior is solid white foam edge to edge; the only light blue water visible is an 8-pixel sliver at the right rim (1.2% of the interior area). No blue water fills the lower portion of the tub in the render." }, { "text": "Ensure the water line is clearly visible across the duck's body, showing the lower third of the duck submerged.", "met": false, "note": "No water line is drawn anywhere across the duck - its yellow body simply ends where the white foam begins, so nothing shows a lower third submerged." }, { "text": "Position the duck so it is floating on the water's surface, creating small, concentric circular ripples in the water around its base.", "met": false, "note": "No water surface is visible in the tub, and there are no ripple lines of any kind around the duck's base." }, { "text": "Depict soap bubbles as clusters of overlapping circles with a slight iridescence, using semi-transparent white, light pink, and light blue fills.", "met": false, "note": "The foam bubbles are white only. The pink on the airborne bubbles is a hairline iridescent crescent of 4-21 pixels, invisible at render scale, so there is no light pink fill." }, { "text": "Place a small cluster of bubbles on top of the duck's head.", "met": true, "note": "" }, { "text": "Place a large pile of bubbles against one side of the tub and a few floating on the water's surface around the duck.", "met": true, "note": "" } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/rubber-ducky-v3.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky-v3", "prompt": "Write `svg` code for an image of a rubber ducky floating on top of a soapy bathtub.", "score": 0.8888888888888888, "met": 8, "total": 9, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/rubber-ducky-v3.png", "requirements": [ { "text": "Create a classic rubber ducky shape with a distinct body and head, colored bright yellow.", "met": true, "note": "" }, { "text": "The duck must have an orange beak and a simple black dot for an eye.", "met": true, "note": "" }, { "text": "Draw a white, claw-foot bathtub shape, showing the inside view with a visible rim.", "met": true, "note": "" }, { "text": "Fill the lower portion of the bathtub with light blue water.", "met": true, "note": "" }, { "text": "Ensure the water line is clearly visible across the duck's body, showing the lower third of the duck submerged.", "met": true, "note": "" }, { "text": "Position the duck so it is floating on the water's surface, creating small, concentric circular ripples in the water around its base.", "met": true, "note": "" }, { "text": "Depict soap bubbles as clusters of overlapping circles with a slight iridescence, using semi-transparent white, light pink, and light blue fills.", "met": false, "note": "All the foam and floating bubbles are white with pale-blue shading; the only pink is the soap bar and a 50-pixel iridescent crescent on one bubble that is not visible at render scale. No light pink fill." }, { "text": "Place a small cluster of bubbles on top of the duck's head.", "met": true, "note": "" }, { "text": "Place a large pile of bubbles against one side of the tub and a few floating on the water's surface around the duck.", "met": true, "note": "" } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/rubber-ducky.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky", "prompt": "Write `svg` code for an image of a rubber ducky floating on top of a soapy bathtub.", "score": 0.4444444444444444, "met": 4, "total": 9, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/rubber-ducky.png", "requirements": [ { "text": "Create a classic rubber ducky shape with a distinct body and head, colored bright yellow.", "met": true, "note": "" }, { "text": "The duck must have an orange beak and a simple black dot for an eye.", "met": true, "note": "" }, { "text": "Draw a white, claw-foot bathtub shape, showing the inside view with a visible rim.", "met": false, "note": "The tub is white with a visible rim and an inside view, but its two feet are smooth flared pedestals ending in a plain rounded pad - no toes, claws or gripped ball, so it is a pedestal tub, not a claw-foot tub." }, { "text": "Fill the lower portion of the bathtub with light blue water.", "met": true, "note": "" }, { "text": "Ensure the water line is clearly visible across the duck's body, showing the lower third of the duck submerged.", "met": true, "note": "" }, { "text": "Position the duck so it is floating on the water's surface, creating small, concentric circular ripples in the water around its base.", "met": false, "note": "The water is a flat blue band with no ripple lines anywhere; the only shapes around the duck's base are the row of soap bubbles, so there are no small concentric circular ripples." }, { "text": "Depict soap bubbles as clusters of overlapping circles with a slight iridescence, using semi-transparent white, light pink, and light blue fills.", "met": false, "note": "Every bubble is white with pale-blue shading; the only pink in the render is the soap bar on the rim. No light pink bubble fill exists, so the three-colour condition fails." }, { "text": "Place a small cluster of bubbles on top of the duck's head.", "met": false, "note": "The duck's head is bare - no bubbles sit on top of it." }, { "text": "Place a large pile of bubbles against one side of the tub and a few floating on the water's surface around the duck.", "met": false, "note": "The bubbles form one evenly spaced single-file row of similar circles running the whole width of the tub; nothing is banked into a large pile against either side." } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/treasure-barrel-animated.svg", "question_index": 13, "animated": true, "slug": "treasure-barrel-animated", "prompt": "Write `svg` code for an image of half buried barrel of treasure.", "score": 0.7, "met": 7, "total": 10, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/treasure-barrel-animated.png", "requirements": [ { "text": "Create a wooden barrel with visible vertical planks and two horizontal metal hoops.", "met": false, "note": "Three grey metal hoops are drawn on the barrel body - one just under the rim and two lower down - not two." }, { "text": "The barrel must be buried in a mound of sand, so only the top half is visible.", "met": false, "note": "More than the top half shows: the barrel's silhouette reaches its maximum width of 305 px at y=370 and has tapered back to about 177 px (58% of maximum) where the sand cuts it at y=450, so the bulge and most of the lower taper are above the sand." }, { "text": "Show the sand mounded up slightly around the visible base of the barrel.", "met": true, "note": "" }, { "text": "The barrel must be open at the top and filled with treasure items.", "met": true, "note": "" }, { "text": "Show treasure overflowing from the top and spilling down one side of the barrel onto the sand.", "met": true, "note": "" }, { "text": "The treasure must include a large pile of shiny gold coins, both inside and outside the barrel.", "met": true, "note": "" }, { "text": "Add a variety of colorful gemstones (red rubies, green emeralds, blue sapphires) mixed in with the coins.", "met": true, "note": "" }, { "text": "A string of white pearls must be draped over the edge of the barrel, trailing down into the spilled coins.", "met": true, "note": "" }, { "text": "A golden goblet must be visible, partially buried in the coins inside the barrel.", "met": false, "note": "The goblet lies on the open sand to the right of the barrel, outside it; no goblet sits in the coins inside the barrel." }, { "text": "Use bright highlights and glint effects on the coins, gems, and goblet to make them look shiny.", "met": true, "note": "" } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/treasure-barrel-v2.svg", "question_index": 13, "animated": false, "slug": "treasure-barrel-v2", "prompt": "Write `svg` code for an image of half buried barrel of treasure.", "score": 0.6, "met": 6, "total": 10, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/treasure-barrel-v2.png", "requirements": [ { "text": "Create a wooden barrel with visible vertical planks and two horizontal metal hoops.", "met": false, "note": "Three grey metal hoops are drawn on the barrel body - one just under the rim and two lower down - not two." }, { "text": "The barrel must be buried in a mound of sand, so only the top half is visible.", "met": true, "note": "" }, { "text": "Show the sand mounded up slightly around the visible base of the barrel.", "met": true, "note": "" }, { "text": "The barrel must be open at the top and filled with treasure items.", "met": true, "note": "" }, { "text": "Show treasure overflowing from the top and spilling down one side of the barrel onto the sand.", "met": true, "note": "" }, { "text": "The treasure must include a large pile of shiny gold coins, both inside and outside the barrel.", "met": true, "note": "" }, { "text": "Add a variety of colorful gemstones (red rubies, green emeralds, blue sapphires) mixed in with the coins.", "met": true, "note": "" }, { "text": "A string of white pearls must be draped over the edge of the barrel, trailing down into the spilled coins.", "met": false, "note": "The pearl string is draped over the left rim but its last pearl stops on bare wood about two pearl-diameters short of the nearest coin, so it does not trail into the spilled coins." }, { "text": "A golden goblet must be visible, partially buried in the coins inside the barrel.", "met": false, "note": "No goblet appears anywhere in the render; the only vessel-like object is a crown lying on the sand at the lower left." }, { "text": "Use bright highlights and glint effects on the coins, gems, and goblet to make them look shiny.", "met": false, "note": "There is no goblet in the render, so the required glint on the goblet cannot be present; coins and gems are shaded." } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/treasure-barrel-v3.svg", "question_index": 13, "animated": false, "slug": "treasure-barrel-v3", "prompt": "Write `svg` code for an image of half buried barrel of treasure.", "score": 0.7, "met": 7, "total": 10, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/treasure-barrel-v3.png", "requirements": [ { "text": "Create a wooden barrel with visible vertical planks and two horizontal metal hoops.", "met": false, "note": "Three grey metal hoops are drawn on the barrel body - one just under the rim and two lower down - not two." }, { "text": "The barrel must be buried in a mound of sand, so only the top half is visible.", "met": false, "note": "More than the top half shows: the barrel's silhouette reaches its maximum width of 306 px at y=376 and has tapered back to 166 px (54% of maximum) where the sand cuts it at y=456, so the whole bulge and most of the lower taper are above the sand." }, { "text": "Show the sand mounded up slightly around the visible base of the barrel.", "met": true, "note": "" }, { "text": "The barrel must be open at the top and filled with treasure items.", "met": true, "note": "" }, { "text": "Show treasure overflowing from the top and spilling down one side of the barrel onto the sand.", "met": true, "note": "" }, { "text": "The treasure must include a large pile of shiny gold coins, both inside and outside the barrel.", "met": true, "note": "" }, { "text": "Add a variety of colorful gemstones (red rubies, green emeralds, blue sapphires) mixed in with the coins.", "met": true, "note": "" }, { "text": "A string of white pearls must be draped over the edge of the barrel, trailing down into the spilled coins.", "met": true, "note": "" }, { "text": "A golden goblet must be visible, partially buried in the coins inside the barrel.", "met": false, "note": "The goblet lies on the open sand to the right of the barrel, outside it; no goblet sits in the coins inside the barrel." }, { "text": "Use bright highlights and glint effects on the coins, gems, and goblet to make them look shiny.", "met": true, "note": "" } ] }, { "harness": "claude", "model": "claude-opus-5", "artifact": "logs/claude/claude-opus-5/treasure-barrel.svg", "question_index": 13, "animated": false, "slug": "treasure-barrel", "prompt": "Write `svg` code for an image of half buried barrel of treasure.", "score": 0.6, "met": 6, "total": 10, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/claude-opus-5/treasure-barrel.png", "requirements": [ { "text": "Create a wooden barrel with visible vertical planks and two horizontal metal hoops.", "met": false, "note": "Three grey metal hoops are drawn on the barrel body - one just under the rim and two lower down - not two." }, { "text": "The barrel must be buried in a mound of sand, so only the top half is visible.", "met": true, "note": "" }, { "text": "Show the sand mounded up slightly around the visible base of the barrel.", "met": true, "note": "" }, { "text": "The barrel must be open at the top and filled with treasure items.", "met": true, "note": "" }, { "text": "Show treasure overflowing from the top and spilling down one side of the barrel onto the sand.", "met": true, "note": "" }, { "text": "The treasure must include a large pile of shiny gold coins, both inside and outside the barrel.", "met": true, "note": "" }, { "text": "Add a variety of colorful gemstones (red rubies, green emeralds, blue sapphires) mixed in with the coins.", "met": true, "note": "" }, { "text": "A string of white pearls must be draped over the edge of the barrel, trailing down into the spilled coins.", "met": false, "note": "The pearl string is draped over the right rim but ends in bare wood halfway down the barrel; the spilled coins are all on the left side and on the sand, so it trails into nothing." }, { "text": "A golden goblet must be visible, partially buried in the coins inside the barrel.", "met": false, "note": "The goblet lies on the open sand at the lower right, outside the barrel; nothing goblet-shaped sits in the coins inside the barrel." }, { "text": "Use bright highlights and glint effects on the coins, gems, and goblet to make them look shiny.", "met": false, "note": "The gemstones are flat single-fill diamonds with no facet highlight, sparkle or gradient (the crimson gem samples one uniform colour, RGB 216,30,91, across its whole area); only the coins and goblet are shaded." } ] }, { "harness": "claude", "model": "fable-5-1", "artifact": "logs/claude/fable-5-1/cow-plowing-animated.svg", "question_index": 0, "animated": true, "slug": "cow-plowing-animated", "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "score": 0.75, "met": 9, "total": 12, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/fable-5-1/cow-plowing-animated.png", "requirements": [ { "text": "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "met": true, "note": "" }, { "text": "The cow must have black and white patches for its coloring.", "met": false, "note": "The cow is tan/caramel with off-white cream patches; there is no black anywhere on the body." }, { "text": "Add cow ears, eyes, and snout for facial recognition.", "met": true, "note": "" }, { "text": "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "met": true, "note": "" }, { "text": "The cow's hooves must be colored brown, as if covered in soil from the field.", "met": true, "note": "" }, { "text": "Include a traditional wooden plow with a visible metal blade/share.", "met": true, "note": "" }, { "text": "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "met": false, "note": "A dark brown collar/shoulder strap is present, but the line running back to the plow is a plain smooth strap with no links; no chains are visible." }, { "text": "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "met": true, "note": "" }, { "text": "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "met": true, "note": "" }, { "text": "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "met": true, "note": "" }, { "text": "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "met": false, "note": "Short green grass is present in the unplowed section, but nothing shows it being overturned - only bare brown soil is turned at the share, with no green sod lifted or flipped." }, { "text": "Add a simple background with a clear horizon line and a blue sky containing a yellow sun.", "met": true, "note": "" } ] }, { "harness": "claude", "model": "fable-5-1", "artifact": "logs/claude/fable-5-1/cow-plowing-v2.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v2", "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "score": 0.75, "met": 9, "total": 12, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/fable-5-1/cow-plowing-v2.png", "requirements": [ { "text": "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "met": true, "note": "" }, { "text": "The cow must have black and white patches for its coloring.", "met": false, "note": "The cow is tan/caramel with off-white cream patches; there is no black anywhere on the body." }, { "text": "Add cow ears, eyes, and snout for facial recognition.", "met": true, "note": "" }, { "text": "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "met": true, "note": "" }, { "text": "The cow's hooves must be colored brown, as if covered in soil from the field.", "met": true, "note": "" }, { "text": "Include a traditional wooden plow with a visible metal blade/share.", "met": true, "note": "" }, { "text": "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "met": false, "note": "A dark brown collar/shoulder strap is present, but the line running back to the plow is a plain smooth strap with no links; no chains are visible." }, { "text": "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "met": true, "note": "" }, { "text": "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "met": true, "note": "" }, { "text": "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "met": true, "note": "" }, { "text": "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "met": false, "note": "Short green grass is present in the unplowed section, but nothing shows it being overturned - only bare brown soil is turned at the share, with no green sod lifted or flipped." }, { "text": "Add a simple background with a clear horizon line and a blue sky containing a yellow sun.", "met": true, "note": "" } ] }, { "harness": "claude", "model": "fable-5-1", "artifact": "logs/claude/fable-5-1/cow-plowing-v3.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v3", "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "score": 0.75, "met": 9, "total": 12, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/fable-5-1/cow-plowing-v3.png", "requirements": [ { "text": "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "met": true, "note": "" }, { "text": "The cow must have black and white patches for its coloring.", "met": false, "note": "The cow is tan/caramel with off-white cream patches; there is no black anywhere on the body." }, { "text": "Add cow ears, eyes, and snout for facial recognition.", "met": true, "note": "" }, { "text": "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "met": true, "note": "" }, { "text": "The cow's hooves must be colored brown, as if covered in soil from the field.", "met": true, "note": "" }, { "text": "Include a traditional wooden plow with a visible metal blade/share.", "met": true, "note": "" }, { "text": "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "met": false, "note": "A dark brown collar/shoulder strap is present, but the line running back to the plow is a plain smooth strap with no links; no chains are visible." }, { "text": "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "met": true, "note": "" }, { "text": "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "met": true, "note": "" }, { "text": "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "met": true, "note": "" }, { "text": "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "met": false, "note": "Short green grass is present in the unplowed section, but nothing shows it being overturned - only bare brown soil is turned at the share, with no green sod lifted or flipped." }, { "text": "Add a simple background with a clear horizon line and a blue sky containing a yellow sun.", "met": true, "note": "" } ] }, { "harness": "claude", "model": "fable-5-1", "artifact": "logs/claude/fable-5-1/cow-plowing.svg", "question_index": 0, "animated": false, "slug": "cow-plowing", "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "score": 0.6666666666666666, "met": 8, "total": 12, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/claude/fable-5-1/cow-plowing.png", "requirements": [ { "text": "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "met": true, "note": "" }, { "text": "The cow must have black and white patches for its coloring.", "met": false, "note": "The cow is tan/caramel with off-white cream patches; there is no black anywhere on the body." }, { "text": "Add cow ears, eyes, and snout for facial recognition.", "met": true, "note": "" }, { "text": "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "met": true, "note": "" }, { "text": "The cow's hooves must be colored brown, as if covered in soil from the field.", "met": true, "note": "" }, { "text": "Include a traditional wooden plow with a visible metal blade/share.", "met": true, "note": "" }, { "text": "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "met": false, "note": "A dark brown collar/shoulder strap is present, but the line running back to the plow is a plain smooth strap with no links; no chains are visible." }, { "text": "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "met": false, "note": "The grey share is drawn fully visible on top of the flat brown field on a small wheeled carriage; no part of it is below the soil surface and no chunk of earth is being lifted or turned." }, { "text": "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "met": true, "note": "" }, { "text": "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "met": true, "note": "" }, { "text": "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "met": false, "note": "Short green grass is present in the unplowed section, but nothing is shown being overturned anywhere - the plow sits well inside the bare brown area and no sod is disturbed." }, { "text": "Add a simple background with a clear horizon line and a blue sky containing a yellow sun.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/cow-plowing-animated.svg", "question_index": 0, "animated": true, "slug": "cow-plowing-animated", "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "score": 0.5, "met": 6, "total": 12, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/cow-plowing-animated.png", "requirements": [ { "text": "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "met": true, "note": "" }, { "text": "The cow must have black and white patches for its coloring.", "met": false, "note": "The cow is red-brown with cream patches; no black patches anywhere on the body." }, { "text": "Add cow ears, eyes, and snout for facial recognition.", "met": true, "note": "" }, { "text": "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "met": true, "note": "" }, { "text": "The cow's hooves must be colored brown, as if covered in soil from the field.", "met": true, "note": "" }, { "text": "Include a traditional wooden plow with a visible metal blade/share.", "met": true, "note": "" }, { "text": "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "met": false, "note": "A brown wooden yoke does sit across the shoulders, but the two traces running back to the plow are smooth thin gold lines with no links; no chains are visible." }, { "text": "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "met": false, "note": "The share is drawn fully above the soil with its whole outline visible; nothing occludes it, there is no trench or mound at its edge, and it does not cross below any soil surface line." }, { "text": "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "met": false, "note": "The field's furrow lines are drawn LIGHTER than the surrounding soil; only about two bands read as darker than the field, short of the three dark furrows required." }, { "text": "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "met": false, "note": "The entire field is bare brown soil edge to edge; the green is distant hills beyond the horizon, not an unplowed section of the field, so there is no boundary for the cow and plow to stand at." }, { "text": "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "met": false, "note": "There is no green grass anywhere in the field - the whole field is bare soil - so nothing green is being overturned." }, { "text": "Add a simple background with a clear horizon line and a blue sky containing a yellow sun.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/cow-plowing-v1.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v1", "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "score": 0.4166666666666667, "met": 5, "total": 12, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/cow-plowing-v1.png", "requirements": [ { "text": "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "met": false, "note": "No udder is drawn anywhere under the belly, and only two legs are present (two hooves in the render); the tail is drawn but the leg count and udder both fail." }, { "text": "The cow must have black and white patches for its coloring.", "met": false, "note": "The cow is red-brown with cream patches; no black patches anywhere on the body." }, { "text": "Add cow ears, eyes, and snout for facial recognition.", "met": true, "note": "" }, { "text": "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "met": true, "note": "" }, { "text": "The cow's hooves must be colored brown, as if covered in soil from the field.", "met": true, "note": "" }, { "text": "Include a traditional wooden plow with a visible metal blade/share.", "met": true, "note": "" }, { "text": "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "met": false, "note": "No yoke is drawn at all - the single strap runs from the cow's muzzle back to the plow - and that strap is a plain smooth line with no links, so there are no chains either." }, { "text": "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "met": false, "note": "The share is drawn fully above the soil with its whole outline visible; nothing occludes it, there is no trench or mound at its edge, and it does not cross below any soil surface line." }, { "text": "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "met": false, "note": "The field's furrow lines are drawn LIGHTER than the surrounding soil; only about two bands read as darker than the field, short of the three dark furrows required." }, { "text": "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "met": false, "note": "The entire field is bare brown soil edge to edge; the green is distant hills beyond the horizon, not an unplowed section of the field, so there is no boundary for the cow and plow to stand at." }, { "text": "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "met": false, "note": "There is no green grass anywhere in the field - the whole field is bare soil - so nothing green is being overturned." }, { "text": "Add a simple background with a clear horizon line and a blue sky containing a yellow sun.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/cow-plowing-v2.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v2", "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "score": 0.4166666666666667, "met": 5, "total": 12, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/cow-plowing-v2.png", "requirements": [ { "text": "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "met": false, "note": "No udder is drawn under the belly, and only three legs are present (three hooves in the render)." }, { "text": "The cow must have black and white patches for its coloring.", "met": false, "note": "The cow is red-brown with cream patches; no black patches anywhere on the body." }, { "text": "Add cow ears, eyes, and snout for facial recognition.", "met": true, "note": "" }, { "text": "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "met": true, "note": "" }, { "text": "The cow's hooves must be colored brown, as if covered in soil from the field.", "met": true, "note": "" }, { "text": "Include a traditional wooden plow with a visible metal blade/share.", "met": true, "note": "" }, { "text": "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "met": false, "note": "A brown wooden yoke does sit across the shoulders, but the two traces running back to the plow are smooth thin gold lines with no links; no chains are visible." }, { "text": "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "met": false, "note": "The share is drawn fully above the soil with its whole outline visible; nothing occludes it, there is no trench or mound at its edge, and it does not cross below any soil surface line." }, { "text": "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "met": false, "note": "The field's furrow lines are drawn LIGHTER than the surrounding soil; only about two bands read as darker than the field, short of the three dark furrows required." }, { "text": "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "met": false, "note": "The entire field is bare brown soil edge to edge; the green is distant hills beyond the horizon, not an unplowed section of the field, so there is no boundary for the cow and plow to stand at." }, { "text": "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "met": false, "note": "There is no green grass anywhere in the field - the whole field is bare soil - so nothing green is being overturned." }, { "text": "Add a simple background with a clear horizon line and a blue sky containing a yellow sun.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/cow-plowing-v3.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v3", "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "score": 0.5, "met": 6, "total": 12, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/cow-plowing-v3.png", "requirements": [ { "text": "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "met": true, "note": "" }, { "text": "The cow must have black and white patches for its coloring.", "met": false, "note": "The cow is red-brown with cream patches; no black patches anywhere on the body." }, { "text": "Add cow ears, eyes, and snout for facial recognition.", "met": true, "note": "" }, { "text": "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "met": true, "note": "" }, { "text": "The cow's hooves must be colored brown, as if covered in soil from the field.", "met": true, "note": "" }, { "text": "Include a traditional wooden plow with a visible metal blade/share.", "met": true, "note": "" }, { "text": "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "met": false, "note": "A brown wooden yoke does sit across the shoulders, but the two traces running back to the plow are smooth thin gold lines with no links; no chains are visible." }, { "text": "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "met": false, "note": "The share is drawn fully above the soil with its whole outline visible; nothing occludes it, there is no trench or mound at its edge, and it does not cross below any soil surface line." }, { "text": "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "met": false, "note": "The field's furrow lines are drawn LIGHTER than the surrounding soil; only about two bands read as darker than the field, short of the three dark furrows required." }, { "text": "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "met": false, "note": "The entire field is bare brown soil edge to edge; the green is distant hills beyond the horizon, not an unplowed section of the field, so there is no boundary for the cow and plow to stand at." }, { "text": "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "met": false, "note": "There is no green grass anywhere in the field - the whole field is bare soil - so nothing green is being overturned." }, { "text": "Add a simple background with a clear horizon line and a blue sky containing a yellow sun.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/dolphin-hula-hoop-fish-animated.svg", "question_index": 7, "animated": true, "slug": "dolphin-hula-hoop-fish-animated", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 0.5714285714285714, "met": 4, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/dolphin-hula-hoop-fish-animated.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": false, "note": "The dolphin is a saturated teal blue, not grey." }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": false, "note": "The splash is three small pale translucent shards; it is not a large splash." }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": false, "note": "The belly, lower flank and tail fluke are drawn below the top edge of the sea, so the body is not entirely in mid-air." }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": true, "note": "" }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/dolphin-hula-hoop-fish-v1.svg", "question_index": 7, "animated": false, "slug": "dolphin-hula-hoop-fish-v1", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 0.2857142857142857, "met": 2, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/dolphin-hula-hoop-fish-v1.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": false, "note": "The dolphin is a saturated teal blue (about RGB 124,196,215 on the flank, an 90-point red-to-blue spread), not grey." }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": false, "note": "A water surface is drawn, but there is no splash at all - only flat wave arcs elsewhere in the sea." }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": false, "note": "The tail stock and both fluke lobes are drawn deep inside the blue water below the surface line, so the body is not entirely in mid-air." }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": false, "note": "The forked orange tail fin is at the far left of the fish and the head (eye and mouth line) is at the right, where the trainer's hand grips it; the fish is held by the head, not the tail." }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": false, "note": "The dolphin's jaw is drawn closed, and the fish is held roughly 200px below and left of the snout rather than at the mouth." } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/dolphin-hula-hoop-fish-v2.svg", "question_index": 7, "animated": false, "slug": "dolphin-hula-hoop-fish-v2", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 0.7142857142857143, "met": 5, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/dolphin-hula-hoop-fish-v2.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": false, "note": "The dolphin is a saturated teal blue, not grey." }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": false, "note": "The only splash is a pair of small pale translucent shards straddling the waterline under the tail; it is not a large splash." }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": true, "note": "" }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": true, "note": "" }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/dolphin-hula-hoop-fish-v3.svg", "question_index": 7, "animated": false, "slug": "dolphin-hula-hoop-fish-v3", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 0.5714285714285714, "met": 4, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/dolphin-hula-hoop-fish-v3.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": false, "note": "The dolphin is a saturated teal blue, not grey." }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": false, "note": "The splash is three small pale translucent shards; it is not a large splash." }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": false, "note": "The belly, lower flank and tail fluke are drawn below the top edge of the sea, so the body is not entirely in mid-air." }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": true, "note": "" }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/fruit-stall-market-animated.svg", "question_index": 12, "animated": true, "slug": "fruit-stall-market-animated", "prompt": "Write `svg` code for an image of a fruit stall in the market.", "score": 0.09090909090909091, "met": 1, "total": 11, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/fruit-stall-market-animated.png", "requirements": [ { "text": "Create a main structure for a wooden fruit stall, including a counter.", "met": true, "note": "" }, { "text": "Add a red and white striped canopy over the stall, supported by two vertical wooden posts.", "met": false, "note": "The canopy's light stripe is a cream, not white: it measures RGB 250,234,191, a spread of 59, against the binding white test of all channels >= 200 with spread <= 40. Two vertical wooden posts are present." }, { "text": "Display exactly four different, recognizable fruits.", "met": false, "note": "Six fruit kinds are shown, not exactly four: oranges, red apples, green pears, yellow bananas, purple grapes and a watermelon." }, { "text": "First fruit: A pile of red apples in a wicker basket on the left side of the counter.", "met": false, "note": "The red apples sit in the top-CENTRE crate, not on the left side of the counter (the oranges occupy the left crate), and the container is a plain trapezoidal wooden crate with plank lines, not a wicker basket." }, { "text": "Second fruit: A bunch of yellow bananas placed next to the apples.", "met": false, "note": "The bananas are in the bottom-LEFT crate and the apples in the top-CENTRE crate, with the orange crate lying directly between them; measured centre to centre the bananas are the farthest of the five other fruit groups from the apples (about 330 px, against about 150 px for the grapes and 180 px for the pears), so they are not placed next to the apples." }, { "text": "Third fruit: A pile of oranges in another wicker basket on the right side of the counter.", "met": false, "note": "The oranges sit in the top-LEFT crate, not on the right side of the counter, and the container is a plain trapezoidal wooden crate with plank lines, not a wicker basket." }, { "text": "Fourth fruit: A single, large slice of watermelon resting directly on the counter in the center.", "met": false, "note": "The watermelon is a dome shape sitting inside the bottom-RIGHT crate; it is neither resting directly on the counter nor in the centre." }, { "text": "Include a character representing the stall vendor, a smiling man with a mustache, positioned behind the counter and between the baskets.", "met": false, "note": "No human figure appears anywhere in the render - the stall is empty." }, { "text": "Add a small, hanging chalkboard sign from the canopy that reads 'Fresh Fruit'.", "met": false, "note": "The sign is a cream board with dark brown lettering (FRESH FRUIT over MARKET MORNING) fixed flat against the awning valance with no cord, rope or hook - it is a painted board, not a hanging chalkboard." }, { "text": "Depict a cobblestone ground surface in front of the stall.", "met": false, "note": "The ground is a flat brown gradient carrying two thin light path lines and a shadow ellipse; no paving stones are drawn." }, { "text": "Include the silhouette of another market stall in the background to suggest a larger market setting.", "met": false, "note": "The two background shapes are tan buildings with red pitched roofs and a door/window panel; neither carries an awning, canopy or open counter, so neither reads as another market stall." } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/fruit-stall-market-v1.svg", "question_index": 12, "animated": false, "slug": "fruit-stall-market-v1", "prompt": "Write `svg` code for an image of a fruit stall in the market.", "score": 0.09090909090909091, "met": 1, "total": 11, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/fruit-stall-market-v1.png", "requirements": [ { "text": "Create a main structure for a wooden fruit stall, including a counter.", "met": true, "note": "" }, { "text": "Add a red and white striped canopy over the stall, supported by two vertical wooden posts.", "met": false, "note": "The canopy's light stripe is a cream, not white: its dominant fill measures RGB 247,229,170, a max-minus-min spread of 77, against the binding white test of all channels >= 200 with spread <= 40. Two vertical wooden posts are present." }, { "text": "Display exactly four different, recognizable fruits.", "met": false, "note": "Six fruit kinds are shown, not exactly four: oranges, red apples, green pears, a pair of flat yellow blobs, purple grapes and a watermelon." }, { "text": "First fruit: A pile of red apples in a wicker basket on the left side of the counter.", "met": false, "note": "The red apples sit in the top-CENTRE crate, not on the left side of the counter (the oranges occupy the left crate), and the container is a plain trapezoidal wooden crate with plank lines, not a wicker basket." }, { "text": "Second fruit: A bunch of yellow bananas placed next to the apples.", "met": false, "note": "The only yellow fruit is a pair of flat yellow blobs in the bottom-LEFT crate that do not read as a bunch of bananas, and they sit at the opposite corner of the 3x2 display from the apples: measured centre to centre they are the farthest of the five other fruit groups from the apples (about 305 px, against about 165 px for the grapes and 210 px for the pears), with the orange crate directly between them." }, { "text": "Third fruit: A pile of oranges in another wicker basket on the right side of the counter.", "met": false, "note": "The oranges sit in the top-LEFT crate, not on the right side of the counter, and the container is a plain trapezoidal wooden crate with plank lines, not a wicker basket." }, { "text": "Fourth fruit: A single, large slice of watermelon resting directly on the counter in the center.", "met": false, "note": "The watermelon is a dome shape sitting inside the bottom-RIGHT crate; it is neither resting directly on the counter nor in the centre." }, { "text": "Include a character representing the stall vendor, a smiling man with a mustache, positioned behind the counter and between the baskets.", "met": false, "note": "No human figure appears anywhere in the render - the stall is empty." }, { "text": "Add a small, hanging chalkboard sign from the canopy that reads 'Fresh Fruit'.", "met": false, "note": "The sign is a cream board (fill RGB 244,226,177) carrying dark brown lettering, fixed flat against the awning valance with no cord, rope or hook - it is a painted board, not a hanging chalkboard. The words FRESH FRUIT are legible but the outer glyphs overflow both ends of the board." }, { "text": "Depict a cobblestone ground surface in front of the stall.", "met": false, "note": "The ground is a flat tan fill with one soft shadow ellipse and a single thin light path line; no paving stones of any kind are drawn." }, { "text": "Include the silhouette of another market stall in the background to suggest a larger market setting.", "met": false, "note": "The two background shapes are tan buildings with red pitched roofs and a door/window panel; neither carries an awning, canopy or open counter, so neither reads as another market stall." } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/fruit-stall-market-v2.svg", "question_index": 12, "animated": false, "slug": "fruit-stall-market-v2", "prompt": "Write `svg` code for an image of a fruit stall in the market.", "score": 0.09090909090909091, "met": 1, "total": 11, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/fruit-stall-market-v2.png", "requirements": [ { "text": "Create a main structure for a wooden fruit stall, including a counter.", "met": true, "note": "" }, { "text": "Add a red and white striped canopy over the stall, supported by two vertical wooden posts.", "met": false, "note": "The canopy's light stripe is a cream, not white: it measures RGB 251,233,186, a spread of 65, against the binding white test of all channels >= 200 with spread <= 40. Two vertical wooden posts are present." }, { "text": "Display exactly four different, recognizable fruits.", "met": false, "note": "Six fruit kinds are shown, not exactly four: oranges, red apples, green pears, yellow bananas, purple grapes and a watermelon." }, { "text": "First fruit: A pile of red apples in a wicker basket on the left side of the counter.", "met": false, "note": "The red apples sit in the top-CENTRE crate, not on the left side of the counter (the oranges occupy the left crate), and the container is a plain trapezoidal wooden crate with plank lines, not a wicker basket." }, { "text": "Second fruit: A bunch of yellow bananas placed next to the apples.", "met": false, "note": "The bananas are in the bottom-LEFT crate and the apples in the top-CENTRE crate, with the orange crate lying directly between them; measured centre to centre the bananas are the farthest of the five other fruit groups from the apples (about 330 px, against about 150 px for the grapes and 180 px for the pears), so they are not placed next to the apples." }, { "text": "Third fruit: A pile of oranges in another wicker basket on the right side of the counter.", "met": false, "note": "The oranges sit in the top-LEFT crate, not on the right side of the counter, and the container is a plain trapezoidal wooden crate with plank lines, not a wicker basket." }, { "text": "Fourth fruit: A single, large slice of watermelon resting directly on the counter in the center.", "met": false, "note": "The watermelon is a dome shape sitting inside the bottom-RIGHT crate; it is neither resting directly on the counter nor in the centre." }, { "text": "Include a character representing the stall vendor, a smiling man with a mustache, positioned behind the counter and between the baskets.", "met": false, "note": "No human figure appears anywhere in the render - the stall is empty." }, { "text": "Add a small, hanging chalkboard sign from the canopy that reads 'Fresh Fruit'.", "met": false, "note": "The sign is a cream board with dark brown lettering (FRESH FRUIT over PICKED THIS MORNING) fixed flat against the awning valance with no cord, rope or hook - it is a painted board, not a hanging chalkboard." }, { "text": "Depict a cobblestone ground surface in front of the stall.", "met": false, "note": "The ground is a brown fill with two soft wave bands and a shadow ellipse; no paving stones are drawn." }, { "text": "Include the silhouette of another market stall in the background to suggest a larger market setting.", "met": false, "note": "The two background shapes are tan buildings with red pitched roofs and a door/window panel; neither carries an awning, canopy or open counter, so neither reads as another market stall." } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/fruit-stall-market-v3.svg", "question_index": 12, "animated": false, "slug": "fruit-stall-market-v3", "prompt": "Write `svg` code for an image of a fruit stall in the market.", "score": 0.09090909090909091, "met": 1, "total": 11, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/fruit-stall-market-v3.png", "requirements": [ { "text": "Create a main structure for a wooden fruit stall, including a counter.", "met": true, "note": "" }, { "text": "Add a red and white striped canopy over the stall, supported by two vertical wooden posts.", "met": false, "note": "The canopy's light stripe is a cream, not white: it measures RGB 249,233,192 with a spread of 57 (dominant fill 249,227,162, spread 87), against the binding white test of all channels >= 200 with spread <= 40. Two vertical wooden posts are present." }, { "text": "Display exactly four different, recognizable fruits.", "met": false, "note": "Six fruit kinds are shown, not exactly four: oranges, red apples, green pears, yellow bananas, purple grapes and a watermelon." }, { "text": "First fruit: A pile of red apples in a wicker basket on the left side of the counter.", "met": false, "note": "The red apples sit in the top-CENTRE crate, not on the left side of the counter (the oranges occupy the left crate), and the container is a plain trapezoidal wooden crate with plank lines, not a wicker basket." }, { "text": "Second fruit: A bunch of yellow bananas placed next to the apples.", "met": false, "note": "The bananas are in the bottom-LEFT crate and the apples in the top-CENTRE crate, with the orange crate lying directly between them; measured centre to centre the bananas are the farthest of the five other fruit groups from the apples (about 330 px, against about 150 px for the grapes and 180 px for the pears), so they are not placed next to the apples." }, { "text": "Third fruit: A pile of oranges in another wicker basket on the right side of the counter.", "met": false, "note": "The oranges sit in the top-LEFT crate, not on the right side of the counter, and the container is a plain trapezoidal wooden crate with plank lines, not a wicker basket." }, { "text": "Fourth fruit: A single, large slice of watermelon resting directly on the counter in the center.", "met": false, "note": "The watermelon is a dome shape sitting inside the bottom-RIGHT crate; it is neither resting directly on the counter nor in the centre." }, { "text": "Include a character representing the stall vendor, a smiling man with a mustache, positioned behind the counter and between the baskets.", "met": false, "note": "No human figure appears anywhere in the render - the stall is empty." }, { "text": "Add a small, hanging chalkboard sign from the canopy that reads 'Fresh Fruit'.", "met": false, "note": "The sign is a cream board with dark brown lettering (FRESH FRUIT over MARKET MORNING) fixed flat against the awning valance with no cord, rope or hook - it is a painted board, not a hanging chalkboard." }, { "text": "Depict a cobblestone ground surface in front of the stall.", "met": false, "note": "The ground is a flat brown gradient carrying two thin light path lines and a shadow ellipse; no paving stones are drawn." }, { "text": "Include the silhouette of another market stall in the background to suggest a larger market setting.", "met": false, "note": "The two background shapes are tan buildings with red pitched roofs and a door/window panel; neither carries an awning, canopy or open counter, so neither reads as another market stall." } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/half-buried-treasure-barrel-animated.svg", "question_index": 13, "animated": true, "slug": "half-buried-treasure-barrel-animated", "prompt": "Write `svg` code for an image of half buried barrel of treasure.", "score": 0.2, "met": 2, "total": 10, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/half-buried-treasure-barrel-animated.png", "requirements": [ { "text": "Create a wooden barrel with visible vertical planks and two horizontal metal hoops.", "met": false, "note": "Three metal hoops are strapped across the barrel body, not two." }, { "text": "The barrel must be buried in a mound of sand, so only the top half is visible.", "met": false, "note": "More than the top half shows: the barrel stands about 425 px tall above the dune against a drawn width of about 437 px (ratio 0.97, where half a barrel is roughly 0.7 of its width), and its lower-left base curve is drawn above the sand." }, { "text": "Show the sand mounded up slightly around the visible base of the barrel.", "met": true, "note": "" }, { "text": "The barrel must be open at the top and filled with treasure items.", "met": true, "note": "" }, { "text": "Show treasure overflowing from the top and spilling down one side of the barrel onto the sand.", "met": false, "note": "Nothing overflows the rim and nothing spills down any side of the barrel; the three coins and two pearls on the dune are isolated objects with no trail connecting them to the barrel." }, { "text": "The treasure must include a large pile of shiny gold coins, both inside and outside the barrel.", "met": false, "note": "Outside the barrel there are only three isolated coins scattered on the dune, which is not a large pile." }, { "text": "Add a variety of colorful gemstones (red rubies, green emeralds, blue sapphires) mixed in with the coins.", "met": false, "note": "Neither green nor blue is present: the two non-red gems are teal (46,188,163, hue 169 degrees) and the other two are crimson, so there is no green emerald and no blue sapphire." }, { "text": "A string of white pearls must be draped over the edge of the barrel, trailing down into the spilled coins.", "met": false, "note": "The pearls are three loose beads lying among the coins inside the barrel and two more sitting on the dune; there is no string, and nothing is draped over the barrel's edge." }, { "text": "A golden goblet must be visible, partially buried in the coins inside the barrel.", "met": false, "note": "No goblet appears anywhere in the render." }, { "text": "Use bright highlights and glint effects on the coins, gems, and goblet to make them look shiny.", "met": false, "note": "There is no goblet in the render, so the required glint on the goblet cannot be present." } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/half-buried-treasure-barrel-v1.svg", "question_index": 13, "animated": false, "slug": "half-buried-treasure-barrel-v1", "prompt": "Write `svg` code for an image of half buried barrel of treasure.", "score": 0.2, "met": 2, "total": 10, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/half-buried-treasure-barrel-v1.png", "requirements": [ { "text": "Create a wooden barrel with visible vertical planks and two horizontal metal hoops.", "met": false, "note": "No metal hoops: the barrel's horizontal bands are all the same brown wood palette as the staves (sampled 100,44,21 and 141,69,28 against wood at 92,38,22), with no grey or metallic band anywhere, and there are four or five such bands rather than two." }, { "text": "The barrel must be buried in a mound of sand, so only the top half is visible.", "met": false, "note": "The barrel is not buried at all - its rounded bottom is drawn in full above the sand and it stands on the surface with a cast shadow; the entire barrel is visible." }, { "text": "Show the sand mounded up slightly around the visible base of the barrel.", "met": true, "note": "" }, { "text": "The barrel must be open at the top and filled with treasure items.", "met": true, "note": "" }, { "text": "Show treasure overflowing from the top and spilling down one side of the barrel onto the sand.", "met": false, "note": "Nothing overflows the rim - the coins sit level inside the barrel mouth - and there is no treasure on the barrel's side or on the sand." }, { "text": "The treasure must include a large pile of shiny gold coins, both inside and outside the barrel.", "met": false, "note": "There are no coins at all outside the barrel; the only coins are the layer inside the mouth." }, { "text": "Add a variety of colorful gemstones (red rubies, green emeralds, blue sapphires) mixed in with the coins.", "met": false, "note": "No blue sapphire: the gems are a cyan diamond (56,197,210, hue 185 degrees), a violet circle (167,86,202, hue 282 degrees), a green circle and a red diamond, so the blue is missing." }, { "text": "A string of white pearls must be draped over the edge of the barrel, trailing down into the spilled coins.", "met": false, "note": "No pearls appear anywhere in the render." }, { "text": "A golden goblet must be visible, partially buried in the coins inside the barrel.", "met": false, "note": "No goblet appears anywhere in the render." }, { "text": "Use bright highlights and glint effects on the coins, gems, and goblet to make them look shiny.", "met": false, "note": "There is no goblet in the render, so the required glint on the goblet cannot be present, and the gems are flat single-fill shapes with no highlight." } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/half-buried-treasure-barrel-v2.svg", "question_index": 13, "animated": false, "slug": "half-buried-treasure-barrel-v2", "prompt": "Write `svg` code for an image of half buried barrel of treasure.", "score": 0.2, "met": 2, "total": 10, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/half-buried-treasure-barrel-v2.png", "requirements": [ { "text": "Create a wooden barrel with visible vertical planks and two horizontal metal hoops.", "met": false, "note": "Three metal hoops are strapped across the barrel body, not two." }, { "text": "The barrel must be buried in a mound of sand, so only the top half is visible.", "met": false, "note": "More than the top half shows: the barrel stands about 428 px tall above the dune against a drawn width of about 415 px (ratio 1.03, where half a barrel is roughly 0.7 of its width), and its lower-left base curve is drawn above the sand." }, { "text": "Show the sand mounded up slightly around the visible base of the barrel.", "met": true, "note": "" }, { "text": "The barrel must be open at the top and filled with treasure items.", "met": true, "note": "" }, { "text": "Show treasure overflowing from the top and spilling down one side of the barrel onto the sand.", "met": false, "note": "Nothing overflows the rim - the coins lie flat inside the mouth - and nothing spills down any side of the barrel; the three coins on the dune are isolated and unconnected to the barrel." }, { "text": "The treasure must include a large pile of shiny gold coins, both inside and outside the barrel.", "met": false, "note": "Outside the barrel there are only three isolated coins scattered on the dune, which is not a large pile." }, { "text": "Add a variety of colorful gemstones (red rubies, green emeralds, blue sapphires) mixed in with the coins.", "met": false, "note": "Neither green nor blue is present: the two non-red gems are teal (48,197,182, hue 174 degrees) and the other two are crimson, so there is no green emerald and no blue sapphire." }, { "text": "A string of white pearls must be draped over the edge of the barrel, trailing down into the spilled coins.", "met": false, "note": "No pearls appear anywhere in the render." }, { "text": "A golden goblet must be visible, partially buried in the coins inside the barrel.", "met": false, "note": "No goblet appears anywhere in the render." }, { "text": "Use bright highlights and glint effects on the coins, gems, and goblet to make them look shiny.", "met": false, "note": "There is no goblet in the render, so the required glint on the goblet cannot be present." } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/half-buried-treasure-barrel-v3.svg", "question_index": 13, "animated": false, "slug": "half-buried-treasure-barrel-v3", "prompt": "Write `svg` code for an image of half buried barrel of treasure.", "score": 0.2, "met": 2, "total": 10, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/half-buried-treasure-barrel-v3.png", "requirements": [ { "text": "Create a wooden barrel with visible vertical planks and two horizontal metal hoops.", "met": false, "note": "Three metal hoops are strapped across the barrel body, not two." }, { "text": "The barrel must be buried in a mound of sand, so only the top half is visible.", "met": false, "note": "More than the top half shows: the barrel stands about 425 px tall above the dune against a drawn width of about 437 px (ratio 0.97, where half a barrel is roughly 0.7 of its width), and its lower-left base curve is drawn above the sand." }, { "text": "Show the sand mounded up slightly around the visible base of the barrel.", "met": true, "note": "" }, { "text": "The barrel must be open at the top and filled with treasure items.", "met": true, "note": "" }, { "text": "Show treasure overflowing from the top and spilling down one side of the barrel onto the sand.", "met": false, "note": "Nothing overflows the rim and nothing spills down any side of the barrel; the three coins and two pearls on the dune are isolated objects with no trail connecting them to the barrel." }, { "text": "The treasure must include a large pile of shiny gold coins, both inside and outside the barrel.", "met": false, "note": "Outside the barrel there are only three isolated coins scattered on the dune, which is not a large pile." }, { "text": "Add a variety of colorful gemstones (red rubies, green emeralds, blue sapphires) mixed in with the coins.", "met": false, "note": "Neither green nor blue is present: the two non-red gems are teal (43,189,163, hue 169 degrees) and the other two are crimson, so there is no green emerald and no blue sapphire." }, { "text": "A string of white pearls must be draped over the edge of the barrel, trailing down into the spilled coins.", "met": false, "note": "The pearls are three loose beads lying among the coins inside the barrel and two more sitting on the dune; there is no string, and nothing is draped over the barrel's edge." }, { "text": "A golden goblet must be visible, partially buried in the coins inside the barrel.", "met": false, "note": "No goblet appears anywhere in the render." }, { "text": "Use bright highlights and glint effects on the coins, gems, and goblet to make them look shiny.", "met": false, "note": "There is no goblet in the render, so the required glint on the goblet cannot be present." } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/picnic-clouds-hot-air-balloon-animated.svg", "question_index": 5, "animated": true, "slug": "picnic-clouds-hot-air-balloon-animated", "prompt": "Write `svg` code for an image of a picnic on top of the clouds, where 2 parents and 2 children have landed with a hot air ballon, and are setting up a picnic with a tarp and food items.", "score": 0.3, "met": 3, "total": 10, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/picnic-clouds-hot-air-balloon-animated.png", "requirements": [ { "text": "Create a scene set on top of a solid-looking cloudscape that serves as the ground.", "met": true, "note": "" }, { "text": "Include a hot air balloon with a large envelope featuring vertical red and white stripes and a brown wicker basket.", "met": false, "note": "The envelope is orange-red with pale yellow vertical stripes and no white stripe, so it is not red-and-white striped." }, { "text": "The balloon's basket must be resting firmly on the surface of a large, flat-topped cloud.", "met": false, "note": "The balloon floats in the upper-left sky with its basket in clear blue sky and a mooring rope hanging loose beneath it; it rests on no cloud." }, { "text": "Show ropes connecting the balloon's envelope to the basket.", "met": true, "note": "" }, { "text": "Include four human figures styled as a family: two adults and two children.", "met": true, "note": "" }, { "text": "Position one adult figure holding a corner of a red and white checkered picnic blanket, while one child figure holds the opposite corner, as if they are spreading it out together on the cloud.", "met": false, "note": "The checkered blanket lies flat and untouched; no adult and child pair holds opposite corners." }, { "text": "Place an open picnic basket on a corner of the blanket that is already spread out.", "met": false, "note": "The wicker basket stands on the cloud outside the blanket's left corner rather than on the blanket." }, { "text": "A thermos, a bunch of grapes, and a sandwich must be visible emerging from the open picnic basket.", "met": false, "note": "The basket holds only a small yellow slab; no thermos, bunch of grapes or sandwich emerges from it." }, { "text": "Position the second adult and second child near the landed hot air balloon, with the adult pointing up at the balloon's envelope.", "met": false, "note": "All four figures are grouped at the blanket far from the balloon, and no adult points up at the envelope." }, { "text": "The background must be a clear blue sky containing a bright yellow sun and two small, distant clouds.", "met": false, "note": "Only one small distant cloud is drawn - a faint wisp beside the sun at about (860,155)." } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/picnic-clouds-hot-air-balloon-v1.svg", "question_index": 5, "animated": false, "slug": "picnic-clouds-hot-air-balloon-v1", "prompt": "Write `svg` code for an image of a picnic on top of the clouds, where 2 parents and 2 children have landed with a hot air ballon, and are setting up a picnic with a tarp and food items.", "score": 0.4, "met": 4, "total": 10, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/picnic-clouds-hot-air-balloon-v1.png", "requirements": [ { "text": "Create a scene set on top of a solid-looking cloudscape that serves as the ground.", "met": true, "note": "" }, { "text": "Include a hot air balloon with a large envelope featuring vertical red and white stripes and a brown wicker basket.", "met": false, "note": "The envelope's vertical stripes are yellow on an orange-red body with a maroon centre stripe - sampled at (175,250) RGB 254,207,101. There is no white stripe and no red-and-white striping." }, { "text": "The balloon's basket must be resting firmly on the surface of a large, flat-topped cloud.", "met": false, "note": "The balloon hangs in mid-air in the upper-left sky, its basket about 50 px above the top of the cloudscape with clear blue sky beneath it; it rests on nothing." }, { "text": "Show ropes connecting the balloon's envelope to the basket.", "met": true, "note": "" }, { "text": "Include four human figures styled as a family: two adults and two children.", "met": true, "note": "" }, { "text": "Position one adult figure holding a corner of a red and white checkered picnic blanket, while one child figure holds the opposite corner, as if they are spreading it out together on the cloud.", "met": false, "note": "The blanket is a plain yellow-tan mat with no red and white checks, it is already lying flat, and none of the four figures holds any part of it." }, { "text": "Place an open picnic basket on a corner of the blanket that is already spread out.", "met": false, "note": "The only basket-like object is a small brown box near the middle of the blanket's near edge, about two thirds across the width and 85% of the way forward - a mid-edge position, not a corner." }, { "text": "A thermos, a bunch of grapes, and a sandwich must be visible emerging from the open picnic basket.", "met": false, "note": "Nothing emerges from a basket: the food in the scene is three coloured dots on a white platter, so no thermos, no bunch of grapes and no sandwich is visible coming out of any basket." }, { "text": "Position the second adult and second child near the landed hot air balloon, with the adult pointing up at the balloon's envelope.", "met": false, "note": "All four figures are seated together around the blanket, roughly 500 px from the balloon, and no adult raises an arm toward the envelope." }, { "text": "The background must be a clear blue sky containing a bright yellow sun and two small, distant clouds.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/picnic-clouds-hot-air-balloon-v2.svg", "question_index": 5, "animated": false, "slug": "picnic-clouds-hot-air-balloon-v2", "prompt": "Write `svg` code for an image of a picnic on top of the clouds, where 2 parents and 2 children have landed with a hot air ballon, and are setting up a picnic with a tarp and food items.", "score": 0.4, "met": 4, "total": 10, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/picnic-clouds-hot-air-balloon-v2.png", "requirements": [ { "text": "Create a scene set on top of a solid-looking cloudscape that serves as the ground.", "met": true, "note": "" }, { "text": "Include a hot air balloon with a large envelope featuring vertical red and white stripes and a brown wicker basket.", "met": false, "note": "The envelope is orange-red with yellow vertical stripes; there is no white stripe, so the striping is not red and white." }, { "text": "The balloon's basket must be resting firmly on the surface of a large, flat-topped cloud.", "met": false, "note": "The balloon floats in the sky at upper left with its basket in clear blue sky and a mooring rope dangling free below it; the basket touches no cloud." }, { "text": "Show ropes connecting the balloon's envelope to the basket.", "met": true, "note": "" }, { "text": "Include four human figures styled as a family: two adults and two children.", "met": true, "note": "" }, { "text": "Position one adult figure holding a corner of a red and white checkered picnic blanket, while one child figure holds the opposite corner, as if they are spreading it out together on the cloud.", "met": false, "note": "The checkered blanket is already lying flat and none of the four figures touches it; no adult and child hold opposite corners." }, { "text": "Place an open picnic basket on a corner of the blanket that is already spread out.", "met": false, "note": "The brown box that serves as the basket sits on the blanket's right edge at about 75% of the width but only 53% of the depth - the middle of an edge rather than a corner." }, { "text": "A thermos, a bunch of grapes, and a sandwich must be visible emerging from the open picnic basket.", "met": false, "note": "No thermos, bunch of grapes or sandwich emerges from the box: it holds a single cream slab, and the other food (a platter of wavy orange food, a cherry, a blue tumbler) is drawn loose on the blanket." }, { "text": "Position the second adult and second child near the landed hot air balloon, with the adult pointing up at the balloon's envelope.", "met": false, "note": "All four figures sit at the blanket, roughly 600 px from the balloon, and no adult points up at the envelope." }, { "text": "The background must be a clear blue sky containing a bright yellow sun and two small, distant clouds.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/picnic-clouds-hot-air-balloon-v3.svg", "question_index": 5, "animated": false, "slug": "picnic-clouds-hot-air-balloon-v3", "prompt": "Write `svg` code for an image of a picnic on top of the clouds, where 2 parents and 2 children have landed with a hot air ballon, and are setting up a picnic with a tarp and food items.", "score": 0.3, "met": 3, "total": 10, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/picnic-clouds-hot-air-balloon-v3.png", "requirements": [ { "text": "Create a scene set on top of a solid-looking cloudscape that serves as the ground.", "met": true, "note": "" }, { "text": "Include a hot air balloon with a large envelope featuring vertical red and white stripes and a brown wicker basket.", "met": false, "note": "The envelope is orange-red with pale yellow vertical stripes - sampled at (160,200) RGB 255,223,151, a yellow rather than a white. No red-and-white striping." }, { "text": "The balloon's basket must be resting firmly on the surface of a large, flat-topped cloud.", "met": false, "note": "The balloon hangs in the sky at upper left with its basket surrounded by blue sky and an anchor rope dangling below; nothing supports it." }, { "text": "Show ropes connecting the balloon's envelope to the basket.", "met": true, "note": "" }, { "text": "Include four human figures styled as a family: two adults and two children.", "met": true, "note": "" }, { "text": "Position one adult figure holding a corner of a red and white checkered picnic blanket, while one child figure holds the opposite corner, as if they are spreading it out together on the cloud.", "met": false, "note": "The red and white checkered blanket is already lying flat on the cloud and none of the four figures holds a corner of it." }, { "text": "Place an open picnic basket on a corner of the blanket that is already spread out.", "met": false, "note": "The wicker basket is not on the blanket at all - it stands on the cloud outside the blanket's left corner, roughly 20 px clear of the checkered edge." }, { "text": "A thermos, a bunch of grapes, and a sandwich must be visible emerging from the open picnic basket.", "met": false, "note": "The basket holds only a small yellow slab. The bread on the platter, the yellow carton and the green pail of round fruit are all separate objects on the blanket, and no thermos or bunch of grapes appears anywhere." }, { "text": "Position the second adult and second child near the landed hot air balloon, with the adult pointing up at the balloon's envelope.", "met": false, "note": "All four figures are at the blanket, hundreds of pixels from the balloon, and no adult points up at the envelope." }, { "text": "The background must be a clear blue sky containing a bright yellow sun and two small, distant clouds.", "met": false, "note": "Only one small distant cloud is drawn - a faint wisp beside the sun at about (870,140). The other pale marks are 5 px round dots that read as sparkles, not clouds." } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/rubber-ducky-soapy-bathtub-animated.svg", "question_index": 4, "animated": true, "slug": "rubber-ducky-soapy-bathtub-animated", "prompt": "Write `svg` code for an image of a rubber ducky floating on top of a soapy bathtub.", "score": 0.1111111111111111, "met": 1, "total": 9, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/rubber-ducky-soapy-bathtub-animated.png", "requirements": [ { "text": "Create a classic rubber ducky shape with a distinct body and head, colored bright yellow.", "met": false, "note": "The duck's head is sheared off flat by the tub's inner rim, leaving a yellow lozenge with no head bump or neck notch; body and head are not distinct." }, { "text": "The duck must have an orange beak and a simple black dot for an eye.", "met": false, "note": "The beak is orange, but the eye is clipped by the tub rim into a flat-topped dark half-disc, not a simple black dot." }, { "text": "Draw a white, claw-foot bathtub shape, showing the inside view with a visible rim.", "met": false, "note": "The tub rests on two smooth flared pedestal stubs with no toes, claws or gripped ball." }, { "text": "Fill the lower portion of the bathtub with light blue water.", "met": true, "note": "" }, { "text": "Ensure the water line is clearly visible across the duck's body, showing the lower third of the duck submerged.", "met": false, "note": "Only a 9-pixel sliver of the 229-pixel-tall duck lies below the water line; the duck effectively sits on the surface rather than showing its lower third submerged." }, { "text": "Position the duck so it is floating on the water's surface, creating small, concentric circular ripples in the water around its base.", "met": false, "note": "The water shows only long horizontal wavy strokes across the tub; no concentric circular ripples around the duck's base." }, { "text": "Depict soap bubbles as clusters of overlapping circles with a slight iridescence, using semi-transparent white, light pink, and light blue fills.", "met": false, "note": "The only bubbles are a white clump in the top-left corner of the canvas; they are white and pale blue with no pink anywhere in the render." }, { "text": "Place a small cluster of bubbles on top of the duck's head.", "met": false, "note": "No bubbles are drawn on the duck's head." }, { "text": "Place a large pile of bubbles against one side of the tub and a few floating on the water's surface around the duck.", "met": false, "note": "Every bubble has drifted outside the tub to the top-left corner of the canvas; there is no pile against any side of the tub and nothing floating on the water around the duck." } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/rubber-ducky-soapy-bathtub-v1.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky-soapy-bathtub-v1", "prompt": "Write `svg` code for an image of a rubber ducky floating on top of a soapy bathtub.", "score": 0.3333333333333333, "met": 3, "total": 9, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/rubber-ducky-soapy-bathtub-v1.png", "requirements": [ { "text": "Create a classic rubber ducky shape with a distinct body and head, colored bright yellow.", "met": true, "note": "" }, { "text": "The duck must have an orange beak and a simple black dot for an eye.", "met": true, "note": "" }, { "text": "Draw a white, claw-foot bathtub shape, showing the inside view with a visible rim.", "met": false, "note": "The tub is a white rounded-rectangle basin resting flat on the floor with only a soft drop shadow beneath it. It has no feet at all, let alone claw feet." }, { "text": "Fill the lower portion of the bathtub with light blue water.", "met": true, "note": "" }, { "text": "Ensure the water line is clearly visible across the duck's body, showing the lower third of the duck submerged.", "met": false, "note": "The duck's whole silhouette is drawn complete and unbroken above the surface, with a dark shadow ellipse beneath it; the water line does not cross the body anywhere and nothing is submerged." }, { "text": "Position the duck so it is floating on the water's surface, creating small, concentric circular ripples in the water around its base.", "met": false, "note": "There are two long horizontal wavy strokes spanning the entire tub and a dark shadow ellipse under the duck, but no concentric circular ripples radiating from its base." }, { "text": "Depict soap bubbles as clusters of overlapping circles with a slight iridescence, using semi-transparent white, light pink, and light blue fills.", "met": false, "note": "All bubbles are white circles with a white outer glow; there is not a single pink pixel anywhere in the render, so the light pink fill is absent." }, { "text": "Place a small cluster of bubbles on top of the duck's head.", "met": false, "note": "No bubbles are drawn on the duck's head." }, { "text": "Place a large pile of bubbles against one side of the tub and a few floating on the water's surface around the duck.", "met": false, "note": "The bubbles are four similar small clusters of about six circles each, spread symmetrically around the basin; the largest spans 19% of the tub width and none is banked into a large pile against one side." } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/rubber-ducky-soapy-bathtub-v2.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky-soapy-bathtub-v2", "prompt": "Write `svg` code for an image of a rubber ducky floating on top of a soapy bathtub.", "score": 0.4444444444444444, "met": 4, "total": 9, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/rubber-ducky-soapy-bathtub-v2.png", "requirements": [ { "text": "Create a classic rubber ducky shape with a distinct body and head, colored bright yellow.", "met": true, "note": "" }, { "text": "The duck must have an orange beak and a simple black dot for an eye.", "met": true, "note": "" }, { "text": "Draw a white, claw-foot bathtub shape, showing the inside view with a visible rim.", "met": false, "note": "The tub is a white rounded-rectangle basin sitting flat on the floor with only a drop shadow; no feet of any kind are drawn." }, { "text": "Fill the lower portion of the bathtub with light blue water.", "met": true, "note": "" }, { "text": "Ensure the water line is clearly visible across the duck's body, showing the lower third of the duck submerged.", "met": true, "note": "" }, { "text": "Position the duck so it is floating on the water's surface, creating small, concentric circular ripples in the water around its base.", "met": false, "note": "The ripple marks are three long horizontal wavy strokes running the full width of the tub plus a dark shadow ellipse; none of them is a concentric circle around the duck's base." }, { "text": "Depict soap bubbles as clusters of overlapping circles with a slight iridescence, using semi-transparent white, light pink, and light blue fills.", "met": false, "note": "Every bubble is a plain white circle with a white glow; the render contains no pink pixels at all, so the light pink fill is missing." }, { "text": "Place a small cluster of bubbles on top of the duck's head.", "met": false, "note": "The duck's head carries no bubbles." }, { "text": "Place a large pile of bubbles against one side of the tub and a few floating on the water's surface around the duck.", "met": false, "note": "Four small clusters of roughly six circles sit symmetrically at the two ends of the tub; none is a large pile banked against one side, and the clusters sit at the tub ends rather than floating around the duck." } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/rubber-ducky-soapy-bathtub-v3.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky-soapy-bathtub-v3", "prompt": "Write `svg` code for an image of a rubber ducky floating on top of a soapy bathtub.", "score": 0.1111111111111111, "met": 1, "total": 9, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/rubber-ducky-soapy-bathtub-v3.png", "requirements": [ { "text": "Create a classic rubber ducky shape with a distinct body and head, colored bright yellow.", "met": false, "note": "The duck is clipped by the tub's inner rim: the top of its head is sheared off flat, leaving a yellow lozenge with no head bump or neck notch, so body and head are not distinct." }, { "text": "The duck must have an orange beak and a simple black dot for an eye.", "met": false, "note": "The beak is orange, but the eye is cut in half by the tub-rim clip and renders as a flat-topped dark half-disc rather than a simple black dot." }, { "text": "Draw a white, claw-foot bathtub shape, showing the inside view with a visible rim.", "met": false, "note": "The tub stands on two smooth trumpet-flared pedestal stubs with no toes, claws or gripped ball, so it is not a claw-foot tub." }, { "text": "Fill the lower portion of the bathtub with light blue water.", "met": true, "note": "" }, { "text": "Ensure the water line is clearly visible across the duck's body, showing the lower third of the duck submerged.", "met": false, "note": "The water line crosses the duck, but only a 29-pixel olive sliver of a 235-pixel-tall duck lies below it - about 17%, far short of the lower third submerged." }, { "text": "Position the duck so it is floating on the water's surface, creating small, concentric circular ripples in the water around its base.", "met": false, "note": "The water carries long horizontal wavy strokes across the whole tub; there are no concentric circular ripples around the duck's base." }, { "text": "Depict soap bubbles as clusters of overlapping circles with a slight iridescence, using semi-transparent white, light pink, and light blue fills.", "met": false, "note": "All bubbles are white circles with a white glow and there is no pink anywhere in the render, so the light pink fill is missing." }, { "text": "Place a small cluster of bubbles on top of the duck's head.", "met": false, "note": "No bubbles are drawn on the duck's head." }, { "text": "Place a large pile of bubbles against one side of the tub and a few floating on the water's surface around the duck.", "met": false, "note": "Two similar small clusters sit at the left and right ends of the tub, each about 18% of the tub width and neither rising above the rim; there is no large pile against one side and no bubbles floating around the duck." } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/stunt-car-jumping-circle-fire-animated.svg", "question_index": 6, "animated": true, "slug": "stunt-car-jumping-circle-fire-animated", "prompt": "Write `svg` code for an image of a stunt car jumping through a circle of fire.", "score": 0.5, "met": 4, "total": 8, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/stunt-car-jumping-circle-fire-animated.png", "requirements": [ { "text": "Create a red, sporty car with a body, wheels, and windows.", "met": true, "note": "" }, { "text": "Draw a distinct circular hoop that is completely surrounded by jagged, irregular flames colored with reds, oranges, and yellows.", "met": true, "note": "" }, { "text": "Include a take-off ramp on the left side of the fiery ring and a landing ramp on the right side.", "met": false, "note": "The take-off ramp on the left is drawn, but there is no landing ramp on the right of the fiery ring. The pale blue-grey bar at the right (x 1110-1290, y ~490) is a flat floating gantry with lamp dots hanging under it, well above the ground line, mirrored by an identical one on the left of the canvas." }, { "text": "Position the car in mid-air, with its front half having passed through the ring and its back half still inside the ring.", "met": false, "note": "Nothing has passed through the ring. Against the hoop's fitted mid-band (centre 977,414, r=259), 0.0% of the car's red body is beyond the ring on the forward side; 36.6% is inside the opening and the remaining 63.4% is still outside behind it, so the car is short of the ring rather than straddling it." }, { "text": "The car's rear wheels must be depicted as just having left the edge of the take-off ramp.", "met": true, "note": "" }, { "text": "Add a ground surface below the entire jump setup.", "met": true, "note": "" }, { "text": "Incorporate gray speed lines trailing behind the car to convey high speed.", "met": false, "note": "No gray speed lines trailing behind the car. The pale strokes behind the rear wheel are the ramp's own outline: each one runs unbroken from the grey-outlined base plate at the far left, horizontally along the ground, through the bend at the ramp's foot and up the incline to the wheel, which is the ramp structure and its rail, not streaks in the air behind the car." }, { "text": "Add orange sparks where the car's tires last touched the take-off ramp.", "met": false, "note": "No sparks at the ramp lip. A 4x crop of the rear tyre at the ramp's top edge shows only dark ramp face and sky. The two orange shapes floating at the upper left are detached hoop flames, roughly 900 px from the ramp lip." } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/stunt-car-jumping-circle-fire-v1.svg", "question_index": 6, "animated": false, "slug": "stunt-car-jumping-circle-fire-v1", "prompt": "Write `svg` code for an image of a stunt car jumping through a circle of fire.", "score": 0.375, "met": 3, "total": 8, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/stunt-car-jumping-circle-fire-v1.png", "requirements": [ { "text": "Create a red, sporty car with a body, wheels, and windows.", "met": true, "note": "" }, { "text": "Draw a distinct circular hoop that is completely surrounded by jagged, irregular flames colored with reds, oranges, and yellows.", "met": true, "note": "" }, { "text": "Include a take-off ramp on the left side of the fiery ring and a landing ramp on the right side.", "met": false, "note": "Neither ramp is drawn. The only structures under the hoop are two thin grey posts with small flat feet that hold the ring up; there is no take-off ramp on the left and no landing ramp on the right." }, { "text": "Position the car in mid-air, with its front half having passed through the ring and its back half still inside the ring.", "met": false, "note": "The car is essentially all the way through. Fitting the hoop's mid-band as a circle at centre (444,396) with r=186 and splitting the car's red body against it, 86.7% of the body lies beyond the ring on the forward side and only 13.2% is still inside the opening, so the back half is not inside the ring." }, { "text": "The car's rear wheels must be depicted as just having left the edge of the take-off ramp.", "met": false, "note": "There is no take-off ramp for the rear wheels to have left; both wheels are in open sky with nothing beneath them." }, { "text": "Add a ground surface below the entire jump setup.", "met": true, "note": "" }, { "text": "Incorporate gray speed lines trailing behind the car to convey high speed.", "met": false, "note": "No speed lines behind the car. The only linear marks in that part of the canvas are two long brown-maroon horizon strokes (sampled at RGB 120,74,74 and 92,62,70) that belong to the background hills, and they run below and well clear of the car rather than trailing from it." }, { "text": "Add orange sparks where the car's tires last touched the take-off ramp.", "met": false, "note": "No sparks anywhere, and no take-off ramp for a last tyre contact point to be on. The scattered orange dots near the hoop are ember specks strewn across the sky, not a spark burst at a ramp edge." } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/stunt-car-jumping-circle-fire-v2.svg", "question_index": 6, "animated": false, "slug": "stunt-car-jumping-circle-fire-v2", "prompt": "Write `svg` code for an image of a stunt car jumping through a circle of fire.", "score": 0.5, "met": 4, "total": 8, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/stunt-car-jumping-circle-fire-v2.png", "requirements": [ { "text": "Create a red, sporty car with a body, wheels, and windows.", "met": true, "note": "" }, { "text": "Draw a distinct circular hoop that is completely surrounded by jagged, irregular flames colored with reds, oranges, and yellows.", "met": true, "note": "" }, { "text": "Include a take-off ramp on the left side of the fiery ring and a landing ramp on the right side.", "met": false, "note": "The take-off ramp on the left is drawn, but there is no landing ramp on the right of the fiery ring: the only things right of the hoop are the two grey posts and flat feet that support it." }, { "text": "Position the car in mid-air, with its front half having passed through the ring and its back half still inside the ring.", "met": false, "note": "Nothing has passed through the ring. Against the hoop's fitted mid-band (centre 980,421, r=241), 0.0% of the car's red body is beyond the ring on the forward side; 34.0% is inside the opening and the remaining 66.0% is still outside behind it, so the car is short of the ring rather than straddling it." }, { "text": "The car's rear wheels must be depicted as just having left the edge of the take-off ramp.", "met": true, "note": "" }, { "text": "Add a ground surface below the entire jump setup.", "met": true, "note": "" }, { "text": "Incorporate gray speed lines trailing behind the car to convey high speed.", "met": false, "note": "No gray speed lines trailing behind the car. The pale strokes behind the rear wheel are the ramp's own outline: each one runs unbroken from the grey-outlined base plate at the far left, horizontally along the ground, through the bend at the ramp's foot and up the incline to the wheel, which is the ramp structure and its rail, not streaks in the air behind the car." }, { "text": "Add orange sparks where the car's tires last touched the take-off ramp.", "met": false, "note": "No sparks at the ramp lip. Crops at 4x of the point where the rear tyre meets the ramp's top edge show only the dark ramp face and night sky; the nearest orange shapes are hoop flames over 300 px away." } ] }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/stunt-car-jumping-circle-fire-v3.svg", "question_index": 6, "animated": false, "slug": "stunt-car-jumping-circle-fire-v3", "prompt": "Write `svg` code for an image of a stunt car jumping through a circle of fire.", "score": 0.5, "met": 4, "total": 8, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/gpt-5.6-terra/stunt-car-jumping-circle-fire-v3.png", "requirements": [ { "text": "Create a red, sporty car with a body, wheels, and windows.", "met": true, "note": "" }, { "text": "Draw a distinct circular hoop that is completely surrounded by jagged, irregular flames colored with reds, oranges, and yellows.", "met": true, "note": "" }, { "text": "Include a take-off ramp on the left side of the fiery ring and a landing ramp on the right side.", "met": false, "note": "The take-off ramp on the left is drawn, but there is no landing ramp on the right of the fiery ring. The pale blue-grey bar at the right (x 1085-1250, y ~490) is not a ramp: it is a flat floating gantry with two lamp dots hanging under it, 140 px above the ground line, and an identical one is mirrored on the left of the canvas." }, { "text": "Position the car in mid-air, with its front half having passed through the ring and its back half still inside the ring.", "met": false, "note": "Nothing has passed through the ring. Against the hoop's fitted mid-band (centre 972,413, r=263), 0.0% of the car's red body is beyond the ring on the forward side; 36.0% is inside the opening and the remaining 64.0% is still outside behind it, so the car is short of the ring rather than straddling it." }, { "text": "The car's rear wheels must be depicted as just having left the edge of the take-off ramp.", "met": true, "note": "" }, { "text": "Add a ground surface below the entire jump setup.", "met": true, "note": "" }, { "text": "Incorporate gray speed lines trailing behind the car to convey high speed.", "met": false, "note": "No gray speed lines trailing behind the car. The pale strokes behind the rear wheel are the ramp's own outline: each one runs unbroken from the grey-outlined base plate at the far left, horizontally along the ground, through the bend at the ramp's foot and up the incline to the wheel, which is the ramp structure and its rail, not streaks in the air behind the car." }, { "text": "Add orange sparks where the car's tires last touched the take-off ramp.", "met": false, "note": "No sparks at the ramp lip. A 4x crop of the point where the rear tyre meets the ramp's top edge shows only dark ramp face and sky; the one orange stroke in the scene is a stripe painted inside the ramp's base plate at the far left, on the ground about 600 px from the lip." } ] }, { "harness": "opencode", "model": "qwen3.8-27b-8bit", "artifact": "logs/opencode/qwen3.8-27b-8bit/cow-plowing-animated.svg", "question_index": 0, "animated": true, "slug": "cow-plowing-animated", "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "score": 0.5833333333333334, "met": 7, "total": 12, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-27b-8bit/cow-plowing-animated.png", "requirements": [ { "text": "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "met": false, "note": "No udder is drawn anywhere under the belly; body, head, four legs and tail are all present, but the udder is absent." }, { "text": "The cow must have black and white patches for its coloring.", "met": false, "note": "The cow is mid-brown with two off-white patches; there is no black on the body." }, { "text": "Add cow ears, eyes, and snout for facial recognition.", "met": true, "note": "" }, { "text": "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "met": true, "note": "" }, { "text": "The cow's hooves must be colored brown, as if covered in soil from the field.", "met": true, "note": "" }, { "text": "Include a traditional wooden plow with a visible metal blade/share.", "met": true, "note": "" }, { "text": "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "met": false, "note": "There is no yoke across the shoulders - the plow beam runs straight to a brown boss at the cow's chest - and no chains are drawn anywhere." }, { "text": "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "met": false, "note": "The pale grey triangular share stands on top of the green turf, fully visible, with no soil over any part of it and no chunk of earth lifted or turned." }, { "text": "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "met": true, "note": "" }, { "text": "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "met": true, "note": "" }, { "text": "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "met": false, "note": "The field is flat green, but no grass is shown being overturned at the share - the blade sits on undisturbed turf and nothing green is lifted." }, { "text": "Add a simple background with a clear horizon line and a blue sky containing a yellow sun.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "qwen3.8-27b-8bit", "artifact": "logs/opencode/qwen3.8-27b-8bit/cow-plowing.svg", "question_index": 0, "animated": false, "slug": "cow-plowing", "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "score": 0.5833333333333334, "met": 7, "total": 12, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-27b-8bit/cow-plowing.png", "requirements": [ { "text": "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "met": false, "note": "No udder is drawn anywhere under the belly; body, head, four legs and tail are all present, but the udder is absent." }, { "text": "The cow must have black and white patches for its coloring.", "met": false, "note": "The cow is mid-brown with two off-white patches; there is no black on the body." }, { "text": "Add cow ears, eyes, and snout for facial recognition.", "met": true, "note": "" }, { "text": "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "met": true, "note": "" }, { "text": "The cow's hooves must be colored brown, as if covered in soil from the field.", "met": true, "note": "" }, { "text": "Include a traditional wooden plow with a visible metal blade/share.", "met": true, "note": "" }, { "text": "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "met": false, "note": "There is no yoke across the shoulders - the plow beam runs straight to a brown boss at the cow's chest - and no chains are drawn anywhere." }, { "text": "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "met": false, "note": "The pale grey triangular share stands on top of the green turf, fully visible, with no soil over any part of it and no chunk of earth lifted or turned." }, { "text": "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "met": true, "note": "" }, { "text": "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "met": true, "note": "" }, { "text": "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "met": false, "note": "The field is flat green, but no grass is shown being overturned at the share - the blade sits on undisturbed turf and nothing green is lifted." }, { "text": "Add a simple background with a clear horizon line and a blue sky containing a yellow sun.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "qwen3.8-27b-8bit", "artifact": "logs/opencode/qwen3.8-27b-8bit/dolphin.svg", "question_index": 7, "animated": false, "slug": "dolphin", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 0.42857142857142855, "met": 3, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-27b-8bit/dolphin.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": false, "note": "The dolphin's body is a light steel blue (about RGB 134,183,219, an 85-point red-to-blue spread), not grey." }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": false, "note": "A water surface is drawn, but there is no splash at the exit point - only a faint pale ellipse and three small bubbles." }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": false, "note": "The tail fluke is drawn below the water surface line, inside the blue water, so the body is not entirely in mid-air." }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": true, "note": "" }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": false, "note": "The snout is drawn closed: a single thin dark line runs to the tip where the fish is held, with no gape." } ] }, { "harness": "opencode", "model": "qwen3.8-27b-8bit", "artifact": "logs/opencode/qwen3.8-27b-8bit/picnic-on-clouds.svg", "question_index": 5, "animated": true, "slug": "picnic-on-clouds", "prompt": "Write `svg` code for an image of a picnic on top of the clouds, where 2 parents and 2 children have landed with a hot air ballon, and are setting up a picnic with a tarp and food items.", "score": 0.5, "met": 5, "total": 10, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-27b-8bit/picnic-on-clouds.png", "requirements": [ { "text": "Create a scene set on top of a solid-looking cloudscape that serves as the ground.", "met": true, "note": "" }, { "text": "Include a hot air balloon with a large envelope featuring vertical red and white stripes and a brown wicker basket.", "met": false, "note": "The envelope's vertical stripes are blue, yellow and red - sampled at (870,320) RGB 52,152,219 and (935,320) RGB 249,202,36. No white stripe is present." }, { "text": "The balloon's basket must be resting firmly on the surface of a large, flat-topped cloud.", "met": true, "note": "" }, { "text": "Show ropes connecting the balloon's envelope to the basket.", "met": true, "note": "" }, { "text": "Include four human figures styled as a family: two adults and two children.", "met": true, "note": "" }, { "text": "Position one adult figure holding a corner of a red and white checkered picnic blanket, while one child figure holds the opposite corner, as if they are spreading it out together on the cloud.", "met": false, "note": "The red and white checkered blanket is already spread flat on the cloud; the man stands on it with one arm raised in the air and no figure holds any corner of it." }, { "text": "Place an open picnic basket on a corner of the blanket that is already spread out.", "met": true, "note": "" }, { "text": "A thermos, a bunch of grapes, and a sandwich must be visible emerging from the open picnic basket.", "met": false, "note": "The only thing in the basket is a rounded cream loaf. The blue jug, the watermelon slice, the cheese wedge, the bread and the apple all lie loose on the blanket, and no bunch of grapes and no sandwich appears anywhere in the render." }, { "text": "Position the second adult and second child near the landed hot air balloon, with the adult pointing up at the balloon's envelope.", "met": false, "note": "The second adult crouches at the blanket about 400 px from the balloon; the child with both arms raised stands between them but no adult is beside the balloon and none points up at the envelope." }, { "text": "The background must be a clear blue sky containing a bright yellow sun and two small, distant clouds.", "met": false, "note": "Three distant clouds are drawn in the sky - beneath the sun at about (150,180), at (625,110) and at (1090,195) - not two." } ] }, { "harness": "opencode", "model": "qwen3.8-27b-8bit", "artifact": "logs/opencode/qwen3.8-27b-8bit/treasure-barrel.svg", "question_index": 13, "animated": false, "slug": "treasure-barrel", "prompt": "Write `svg` code for an image of half buried barrel of treasure.", "score": 0.4, "met": 4, "total": 10, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-27b-8bit/treasure-barrel.png", "requirements": [ { "text": "Create a wooden barrel with visible vertical planks and two horizontal metal hoops.", "met": true, "note": "" }, { "text": "The barrel must be buried in a mound of sand, so only the top half is visible.", "met": true, "note": "" }, { "text": "Show the sand mounded up slightly around the visible base of the barrel.", "met": true, "note": "" }, { "text": "The barrel must be open at the top and filled with treasure items.", "met": true, "note": "" }, { "text": "Show treasure overflowing from the top and spilling down one side of the barrel onto the sand.", "met": false, "note": "The coin heap does mound above the rim, but nothing spills down a side onto the sand: the four coins on the barrel face are symmetric studs on the two hoops, and the six coins on the sand are scattered evenly to both sides with no connected trail." }, { "text": "The treasure must include a large pile of shiny gold coins, both inside and outside the barrel.", "met": false, "note": "Outside the barrel there are only six flat coins spread singly across the whole width of the sand, which is not a large pile." }, { "text": "Add a variety of colorful gemstones (red rubies, green emeralds, blue sapphires) mixed in with the coins.", "met": false, "note": "Neither green nor blue is present: the gems are a red diamond, a yellow diamond and a teal diamond (35,165,154, hue 175 degrees), so there is no green emerald and no blue sapphire." }, { "text": "A string of white pearls must be draped over the edge of the barrel, trailing down into the spilled coins.", "met": false, "note": "No pearls appear anywhere in the render." }, { "text": "A golden goblet must be visible, partially buried in the coins inside the barrel.", "met": false, "note": "No goblet appears anywhere in the render; the gold object above the coin heap is a crown." }, { "text": "Use bright highlights and glint effects on the coins, gems, and goblet to make them look shiny.", "met": false, "note": "There is no goblet in the render, so the required glint on the goblet cannot be present." } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/cow-plowing-animated.svg", "question_index": 0, "animated": true, "slug": "cow-plowing-animated", "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "score": 0.5, "met": 6, "total": 12, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/cow-plowing-animated.png", "requirements": [ { "text": "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "met": false, "note": "No udder is drawn under the belly - the belly line is unbroken white with no pink sac - although body, head, four legs and tail are all present." }, { "text": "The cow must have black and white patches for its coloring.", "met": true, "note": "" }, { "text": "Add cow ears, eyes, and snout for facial recognition.", "met": true, "note": "" }, { "text": "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "met": true, "note": "" }, { "text": "The cow's hooves must be colored brown, as if covered in soil from the field.", "met": false, "note": "Hoof fill samples RGB (43,43,43) - a perfectly neutral dark grey with zero red-minus-blue spread; the hooves are black, not brown." }, { "text": "Include a traditional wooden plow with a visible metal blade/share.", "met": true, "note": "" }, { "text": "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "met": false, "note": "A brown neck strap and two smooth trace lines run back to the plow; there is no yoke beam across the shoulders and the traces carry no links, so no chains are visible." }, { "text": "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "met": false, "note": "Not partially buried: the share hangs entirely above the soil with its outline complete and unbroken. The two brown clods drawn across the blade are detached debris, which the convention excludes from the buried clause, and the top-down field gives no drawn soil-surface line for the silhouette to cross. Earth is visibly lifted at the share, but the blade itself is never in the soil." }, { "text": "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "met": true, "note": "" }, { "text": "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "met": false, "note": "The field is bare brown soil edge to edge with furrow lines throughout; there is no unplowed section, so there is no boundary for the cow and plow to be positioned at." }, { "text": "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "met": false, "note": "There is no green grass anywhere in the field - the whole field is bare tilled soil - so nothing green is being overturned." }, { "text": "Add a simple background with a clear horizon line and a blue sky containing a yellow sun.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/cow-plowing-v2.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v2", "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "score": 0.5833333333333334, "met": 7, "total": 12, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/cow-plowing-v2.png", "requirements": [ { "text": "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "met": true, "note": "" }, { "text": "The cow must have black and white patches for its coloring.", "met": true, "note": "" }, { "text": "Add cow ears, eyes, and snout for facial recognition.", "met": true, "note": "" }, { "text": "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "met": true, "note": "" }, { "text": "The cow's hooves must be colored brown, as if covered in soil from the field.", "met": false, "note": "Hoof fill samples RGB (43,43,43) - a perfectly neutral dark grey with zero red-minus-blue spread; the hooves are black, not brown." }, { "text": "Include a traditional wooden plow with a visible metal blade/share.", "met": true, "note": "" }, { "text": "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "met": false, "note": "A brown neck strap and two smooth trace lines run back to the plow; there is no yoke beam across the shoulders and the traces carry no links, so no chains are visible." }, { "text": "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "met": false, "note": "Not partially buried: the grey share is drawn on top of the tilled plane with its outline complete and unbroken, resting on the top edge of the raised tan slab of turned soil rather than crossing below it. No soil mass covers any part of the blade, and the top-down field gives no drawn soil-surface line for the silhouette to cross." }, { "text": "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "met": true, "note": "" }, { "text": "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "met": false, "note": "The field is bare brown soil edge to edge with furrow lines throughout; there is no unplowed section, so there is no boundary for the cow and plow to be positioned at." }, { "text": "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "met": false, "note": "There is no green grass anywhere in the field - the whole field is bare tilled soil - so nothing green is being overturned." }, { "text": "Add a simple background with a clear horizon line and a blue sky containing a yellow sun.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/cow-plowing-v3.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v3", "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "score": 0.5833333333333334, "met": 7, "total": 12, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/cow-plowing-v3.png", "requirements": [ { "text": "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "met": true, "note": "" }, { "text": "The cow must have black and white patches for its coloring.", "met": true, "note": "" }, { "text": "Add cow ears, eyes, and snout for facial recognition.", "met": true, "note": "" }, { "text": "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "met": true, "note": "" }, { "text": "The cow's hooves must be colored brown, as if covered in soil from the field.", "met": false, "note": "Hoof fill samples RGB (43,43,43) - a perfectly neutral dark grey with zero red-minus-blue spread; the hooves are black, not brown." }, { "text": "Include a traditional wooden plow with a visible metal blade/share.", "met": true, "note": "" }, { "text": "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "met": false, "note": "A brown neck strap and two smooth trace lines run back to the plow; there is no yoke beam across the shoulders and the traces carry no links, so no chains are visible." }, { "text": "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "met": false, "note": "Not partially buried: the grey share is drawn with a complete, unbroken outline on top of the tilled plane. The loose brown clods lie beside and below it, no soil mass covers any part of it, and the top-down field gives no drawn soil-surface line for the silhouette to cross." }, { "text": "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "met": true, "note": "" }, { "text": "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "met": false, "note": "The field is bare brown soil edge to edge with furrow lines throughout; there is no unplowed section, so there is no boundary for the cow and plow to be positioned at." }, { "text": "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "met": false, "note": "There is no green grass anywhere in the field - the whole field is bare tilled soil - so nothing green is being overturned." }, { "text": "Add a simple background with a clear horizon line and a blue sky containing a yellow sun.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/cow-plowing.svg", "question_index": 0, "animated": false, "slug": "cow-plowing", "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "score": 0.5833333333333334, "met": 7, "total": 12, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/cow-plowing.png", "requirements": [ { "text": "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "met": true, "note": "" }, { "text": "The cow must have black and white patches for its coloring.", "met": true, "note": "" }, { "text": "Add cow ears, eyes, and snout for facial recognition.", "met": true, "note": "" }, { "text": "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "met": true, "note": "" }, { "text": "The cow's hooves must be colored brown, as if covered in soil from the field.", "met": false, "note": "Hoof fill samples RGB (43,43,43) - a perfectly neutral dark grey with zero red-minus-blue spread; the hooves are black, not brown." }, { "text": "Include a traditional wooden plow with a visible metal blade/share.", "met": true, "note": "" }, { "text": "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "met": false, "note": "A brown neck strap and two smooth trace lines run back to the plow; there is no yoke beam across the shoulders and the traces carry no links, so no chains are visible." }, { "text": "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "met": false, "note": "Not partially buried: the grey share is drawn on top of the tilled plane with its outline complete and unbroken. No soil mass - no ground fill, mound, furrow wall or attached slice of earth - covers any part of it, and the field is an oblique top-down plane with no drawn soil-surface line for the silhouette to cross. The small dark blob at the share's tip is a detached clod, which the convention excludes from the buried clause." }, { "text": "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "met": true, "note": "" }, { "text": "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "met": false, "note": "The field is bare brown soil edge to edge with furrow lines throughout; there is no unplowed section, so there is no boundary for the cow and plow to be positioned at." }, { "text": "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "met": false, "note": "There is no green grass anywhere in the field - the whole field is bare tilled soil - so nothing green is being overturned." }, { "text": "Add a simple background with a clear horizon line and a blue sky containing a yellow sun.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/dolphin-animated.svg", "question_index": 7, "animated": true, "slug": "dolphin-animated", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 0.7142857142857143, "met": 5, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/dolphin-animated.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": true, "note": "" }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": true, "note": "" }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": false, "note": "The tail fluke is drawn below the waterline, inside the blue water, so the body is not entirely in mid-air." }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": false, "note": "The fish's eye and head are at the right end, inside the trainer's fist, and the tail fin is at the left end inside the dolphin's mouth; the fish is held by the head, not the tail." }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/dolphin-hoop-fish-animated.svg", "question_index": 7, "animated": true, "slug": "dolphin-hoop-fish-animated", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 0.8571428571428571, "met": 6, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/dolphin-hoop-fish-animated.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": true, "note": "" }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": true, "note": "" }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": false, "note": "The tail fluke is drawn below the waterline, inside the blue water, so the body is not entirely in mid-air." }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": true, "note": "" }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/dolphin-hoop-fish-v2.svg", "question_index": 7, "animated": false, "slug": "dolphin-hoop-fish-v2", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 0.7142857142857143, "met": 5, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/dolphin-hoop-fish-v2.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": true, "note": "" }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": false, "note": "The splash is a small cluster of low bumps roughly 55px wide with no vertical spray; it is not a large splash." }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": false, "note": "The tail fluke is drawn below the waterline, inside the blue water, so the body is not entirely in mid-air." }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": true, "note": "" }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/dolphin-hoop-fish-v3.svg", "question_index": 7, "animated": false, "slug": "dolphin-hoop-fish-v3", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 0.8571428571428571, "met": 6, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/dolphin-hoop-fish-v3.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": true, "note": "" }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": true, "note": "" }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": false, "note": "The tail fluke is drawn below the waterline, inside the blue water, so the body is not entirely in mid-air." }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": true, "note": "" }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/dolphin-hoop-fish.svg", "question_index": 7, "animated": false, "slug": "dolphin-hoop-fish", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 0.7142857142857143, "met": 5, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/dolphin-hoop-fish.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": true, "note": "" }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": false, "note": "The splash is a row of four small low semicircular bumps sitting on the surface with no vertical spray; it is not a large splash." }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": false, "note": "The tail fluke tip is drawn below the waterline, inside the blue water, so the body is not entirely in mid-air." }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": true, "note": "" }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/dolphin-hula-hoop-fish-animated.svg", "question_index": 7, "animated": true, "slug": "dolphin-hula-hoop-fish-animated", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 1.0, "met": 7, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/dolphin-hula-hoop-fish-animated.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": true, "note": "" }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": true, "note": "" }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": true, "note": "" }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": true, "note": "" }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/dolphin-hula-hoop-fish-v1.svg", "question_index": 7, "animated": false, "slug": "dolphin-hula-hoop-fish-v1", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 0.2857142857142857, "met": 2, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/dolphin-hula-hoop-fish-v1.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": true, "note": "" }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": false, "note": "There is no splash at the exit point - only two faint low translucent mounds and a few bubbles at the surface." }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": false, "note": "The tail stock and both fluke lobes are drawn well below the waterline, inside the blue water, so the body is not entirely in mid-air." }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": false, "note": "The hoop is not circular: it is two open, unjoined wavy ribbons that never close into a ring, and the dolphin does not pass through any enclosed centre." }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": false, "note": "The fish's tail fin is at the left and its eye and snout are at the right, inside the trainer's fingers; the fish is held by the head, not the tail." }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": false, "note": "The dolphin's mouth is almost entirely hidden behind the fish; all that is visible is a thin dark-red sliver, which does not read as a wide-open mouth." } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/dolphin-hula-hoop-fish-v2.svg", "question_index": 7, "animated": false, "slug": "dolphin-hula-hoop-fish-v2", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 1.0, "met": 7, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/dolphin-hula-hoop-fish-v2.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": true, "note": "" }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": true, "note": "" }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": true, "note": "" }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": true, "note": "" }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/dolphin-hula-hoop-fish-v3.svg", "question_index": 7, "animated": false, "slug": "dolphin-hula-hoop-fish-v3", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 1.0, "met": 7, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/dolphin-hula-hoop-fish-v3.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": true, "note": "" }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": true, "note": "" }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": true, "note": "" }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": true, "note": "" }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/dolphin.svg", "question_index": 7, "animated": false, "slug": "dolphin", "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "score": 0.7142857142857143, "met": 5, "total": 7, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/dolphin.png", "requirements": [ { "text": "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "met": true, "note": "" }, { "text": "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "met": true, "note": "" }, { "text": "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "met": false, "note": "The tail fluke is drawn below the waterline, inside the blue water, so the body is not entirely in mid-air." }, { "text": "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "met": true, "note": "" }, { "text": "Include a human trainer's arm and hand extending into the frame.", "met": true, "note": "" }, { "text": "The trainer's hand must be holding a small fish by the tail.", "met": false, "note": "The fish's eye and head are at the right end, inside the trainer's fist, and the tail fin is at the left end inside the dolphin's mouth; the fish is held by the head, not the tail." }, { "text": "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/fruit-stall-animated.svg", "question_index": 12, "animated": true, "slug": "fruit-stall-animated", "prompt": "Write `svg` code for an image of a fruit stall in the market.", "score": 0.2727272727272727, "met": 3, "total": 11, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/fruit-stall-animated.png", "requirements": [ { "text": "Create a main structure for a wooden fruit stall, including a counter.", "met": true, "note": "" }, { "text": "Add a red and white striped canopy over the stall, supported by two vertical wooden posts.", "met": true, "note": "" }, { "text": "Display exactly four different, recognizable fruits.", "met": false, "note": "Six fruit kinds are shown, not exactly four: oranges, red apples, purple grapes, yellow lemons, yellow bananas and watermelons." }, { "text": "First fruit: A pile of red apples in a wicker basket on the left side of the counter.", "met": false, "note": "The apple pile is on the left of the counter as required, but it stands in a plain rectangular wooden crate (flat brown fill with a border and one plank line), not a wicker basket." }, { "text": "Second fruit: A bunch of yellow bananas placed next to the apples.", "met": false, "note": "The banana bunch sits at the right end of the counter (around x=880, y=665) while the nearest apples are the shelf pile at about (730,510); the lemon crate at about (760,625) and the grape bowl at about (890,500) both lie between them and are both closer to the apples than the bananas are (about 119 px and 160 px against about 216 px), so the bananas are not placed next to the apples." }, { "text": "Third fruit: A pile of oranges in another wicker basket on the right side of the counter.", "met": false, "note": "The oranges are in the CENTRE crate of the counter and centre of the shelf, not on the right side of the counter; the only orange crate on the right stands on the ground beside the stall rather than on the counter, and both are plain wooden crates, not wicker baskets." }, { "text": "Fourth fruit: A single, large slice of watermelon resting directly on the counter in the center.", "met": false, "note": "The watermelons - two whole and one cut - lie on the GROUND in front of the counter and left of centre, not resting directly on the counter in the centre, and the cut one is a small round cross-section rather than a single large slice." }, { "text": "Include a character representing the stall vendor, a smiling man with a mustache, positioned behind the counter and between the baskets.", "met": false, "note": "The vendor is a smiling man behind the counter, but at 5x his face carries only two dot eyes and a smile arc - there is no moustache anywhere on it." }, { "text": "Add a small, hanging chalkboard sign from the canopy that reads 'Fresh Fruit'.", "met": false, "note": "The chalkboard reads 'FRESH FRUIT' over '$2 / KG' legibly in chalk lettering, but it is screwed flat to the counter's front panel (two screw dots at its top corners) below the counter top; nothing connects it to the canopy, so it does not hang from the canopy." }, { "text": "Depict a cobblestone ground surface in front of the stall.", "met": false, "note": "The ground is flat sand carrying sparse, widely spaced outlined ellipses with bare sand between them - scattered pebble outlines rather than a packed paving-stone surface; no joint lines or tiled stones are drawn." }, { "text": "Include the silhouette of another market stall in the background to suggest a larger market setting.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/fruit-stall-v2.svg", "question_index": 12, "animated": false, "slug": "fruit-stall-v2", "prompt": "Write `svg` code for an image of a fruit stall in the market.", "score": 0.2727272727272727, "met": 3, "total": 11, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/fruit-stall-v2.png", "requirements": [ { "text": "Create a main structure for a wooden fruit stall, including a counter.", "met": true, "note": "" }, { "text": "Add a red and white striped canopy over the stall, supported by two vertical wooden posts.", "met": true, "note": "" }, { "text": "Display exactly four different, recognizable fruits.", "met": false, "note": "Six fruit kinds are shown, not exactly four: oranges, red apples, purple grapes, yellow lemons, yellow bananas and watermelons." }, { "text": "First fruit: A pile of red apples in a wicker basket on the left side of the counter.", "met": false, "note": "The apple pile is on the left of the counter as required, but it stands in a plain rectangular wooden crate (flat brown fill with a border and one plank line), not a wicker basket." }, { "text": "Second fruit: A bunch of yellow bananas placed next to the apples.", "met": false, "note": "The banana bunch sits at the far right end of the counter (around x=935, y=675) while the nearest apples are the shelf pile at about (730,510); the lemon crate at (760,630) and the grape bowl at (890,505) both lie between them and are both closer to the apples than the bananas are (about 124 px and 145 px against about 271 px), so the bananas are not placed next to the apples." }, { "text": "Third fruit: A pile of oranges in another wicker basket on the right side of the counter.", "met": false, "note": "The oranges are in the CENTRE crate of the counter and centre of the shelf, not on the right side of the counter; the only orange crate on the right stands on the ground beside the stall rather than on the counter, and both are plain wooden crates, not wicker baskets." }, { "text": "Fourth fruit: A single, large slice of watermelon resting directly on the counter in the center.", "met": false, "note": "The watermelons - two whole and one cut - lie on the GROUND in front of the counter and left of centre, not resting directly on the counter in the centre, and the cut one is a small round cross-section rather than a single large slice." }, { "text": "Include a character representing the stall vendor, a smiling man with a mustache, positioned behind the counter and between the baskets.", "met": false, "note": "No human figure appears anywhere in the render." }, { "text": "Add a small, hanging chalkboard sign from the canopy that reads 'Fresh Fruit'.", "met": true, "note": "" }, { "text": "Depict a cobblestone ground surface in front of the stall.", "met": false, "note": "The ground is a flat tan fill carrying a few soft filled ellipse patches with bare ground between them; no paving stones or joint lines are drawn." }, { "text": "Include the silhouette of another market stall in the background to suggest a larger market setting.", "met": false, "note": "The only background object is a flat pale-blue building at the left with a flat roof band, one window and one door; it carries no awning, canopy or open counter, so it does not read as another market stall." } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/fruit-stall-v3.svg", "question_index": 12, "animated": false, "slug": "fruit-stall-v3", "prompt": "Write `svg` code for an image of a fruit stall in the market.", "score": 0.2727272727272727, "met": 3, "total": 11, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/fruit-stall-v3.png", "requirements": [ { "text": "Create a main structure for a wooden fruit stall, including a counter.", "met": true, "note": "" }, { "text": "Add a red and white striped canopy over the stall, supported by two vertical wooden posts.", "met": true, "note": "" }, { "text": "Display exactly four different, recognizable fruits.", "met": false, "note": "Six fruit kinds are shown, not exactly four: oranges, red apples, purple grapes, yellow lemons, yellow bananas and watermelons." }, { "text": "First fruit: A pile of red apples in a wicker basket on the left side of the counter.", "met": false, "note": "The apple pile is on the left of the counter as required, but it stands in a plain rectangular wooden crate (flat brown fill with a border and one plank line), not a wicker basket." }, { "text": "Second fruit: A bunch of yellow bananas placed next to the apples.", "met": false, "note": "The banana bunch sits at the far right end of the counter (around x=920, y=670) while the nearest apples are the shelf pile at about (730,510); the lemon crate and the grape bowl both lie between them and are both closer to the apples than the bananas are, so the bananas are not placed next to the apples." }, { "text": "Third fruit: A pile of oranges in another wicker basket on the right side of the counter.", "met": false, "note": "The oranges are in the CENTRE crate of the counter and centre of the shelf, not on the right side of the counter; the only orange crate on the right stands on the ground beside the stall rather than on the counter, and both are plain wooden crates, not wicker baskets." }, { "text": "Fourth fruit: A single, large slice of watermelon resting directly on the counter in the center.", "met": false, "note": "The watermelons - two whole and one cut - lie on the GROUND in front of the counter and left of centre, not resting directly on the counter in the centre, and the cut one is a small round cross-section rather than a single large slice." }, { "text": "Include a character representing the stall vendor, a smiling man with a mustache, positioned behind the counter and between the baskets.", "met": false, "note": "The figure behind the counter is reduced to a straw hat and two white sleeves ending in hands: at 4x the head, face and torso are entirely hidden behind the apple pile, so there is no visible smile and no moustache. The only complete figure is a customer standing in front of the stall on the right, back to the viewer, with a hair bun and no moustache." }, { "text": "Add a small, hanging chalkboard sign from the canopy that reads 'Fresh Fruit'.", "met": false, "note": "The chalkboard has moved to the counter's front panel below the counter top, so it does not hang from the canopy, and the watermelons occlude its first letter: at 4x the top line reads 'RESH FRUIT'. The other chalkboard, an A-frame on the ground at the left, reads 'SWEET & JUICY!' instead." }, { "text": "Depict a cobblestone ground surface in front of the stall.", "met": false, "note": "The ground is a flat tan fill carrying a few soft filled ellipse patches with bare ground between them; no paving stones or joint lines are drawn." }, { "text": "Include the silhouette of another market stall in the background to suggest a larger market setting.", "met": true, "note": "" } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/fruit-stall.svg", "question_index": 12, "animated": false, "slug": "fruit-stall", "prompt": "Write `svg` code for an image of a fruit stall in the market.", "score": 0.18181818181818182, "met": 2, "total": 11, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/fruit-stall.png", "requirements": [ { "text": "Create a main structure for a wooden fruit stall, including a counter.", "met": true, "note": "" }, { "text": "Add a red and white striped canopy over the stall, supported by two vertical wooden posts.", "met": true, "note": "" }, { "text": "Display exactly four different, recognizable fruits.", "met": false, "note": "Seven fruit groups are on display, not exactly four: orange circles, red apples, green ovals, purple grapes, a yellow lemon pile, a yellow banana bunch and three plain dark-green circles at the foot of the counter." }, { "text": "First fruit: A pile of red apples in a wicker basket on the left side of the counter.", "met": false, "note": "The apple pile is on the left of the counter as required, but it stands in a plain rectangular wooden crate (flat brown fill with a single plank line), not a wicker basket." }, { "text": "Second fruit: A bunch of yellow bananas placed next to the apples.", "met": false, "note": "The banana bunch hangs at the far RIGHT end of the counter (around x=920) while the apple pile is at the far LEFT (around x=360); the orange pile, the chalkboard and the lemon pile all lie between them, so the bananas are the farthest fruit group from the apples rather than next to them." }, { "text": "Third fruit: A pile of oranges in another wicker basket on the right side of the counter.", "met": false, "note": "The oranges sit in the CENTRE crate of the counter (and top-left on the shelf), not on the right side, and the container is a plain rectangular wooden crate, not a wicker basket." }, { "text": "Fourth fruit: A single, large slice of watermelon resting directly on the counter in the center.", "met": false, "note": "There is no watermelon slice: the only candidate is three plain dark-green circles with no rind stripes and no red flesh, and they sit on the ground at the foot of the counter front rather than resting directly on the counter in the centre." }, { "text": "Include a character representing the stall vendor, a smiling man with a mustache, positioned behind the counter and between the baskets.", "met": false, "note": "No human figure appears anywhere in the render." }, { "text": "Add a small, hanging chalkboard sign from the canopy that reads 'Fresh Fruit'.", "met": false, "note": "The chalkboard does read FRESH FRUIT legibly in white chalk lettering, but it stands against the back wall at counter level on a short post, with no cord, chain or hook of any kind connecting it to the canopy - it does not hang from the canopy." }, { "text": "Depict a cobblestone ground surface in front of the stall.", "met": false, "note": "The ground is a single flat tan band with no texture at all; no paving stones are drawn." }, { "text": "Include the silhouette of another market stall in the background to suggest a larger market setting.", "met": false, "note": "The background contains only sky, two clouds and a sun - there is no second stall, building or silhouette of any kind." } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/rubber-ducky-animated.svg", "question_index": 4, "animated": true, "slug": "rubber-ducky-animated", "prompt": "Write `svg` code for an image of a rubber ducky floating on top of a soapy bathtub.", "score": 0.5555555555555556, "met": 5, "total": 9, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/rubber-ducky-animated.png", "requirements": [ { "text": "Create a classic rubber ducky shape with a distinct body and head, colored bright yellow.", "met": true, "note": "" }, { "text": "The duck must have an orange beak and a simple black dot for an eye.", "met": true, "note": "" }, { "text": "Draw a white, claw-foot bathtub shape, showing the inside view with a visible rim.", "met": false, "note": "The tub stands on two plain rectangular stubs with rounded bottoms - simple blocks with no toes, claws or gripped ball, so it is not a claw-foot tub." }, { "text": "Fill the lower portion of the bathtub with light blue water.", "met": true, "note": "" }, { "text": "Ensure the water line is clearly visible across the duck's body, showing the lower third of the duck submerged.", "met": true, "note": "" }, { "text": "Position the duck so it is floating on the water's surface, creating small, concentric circular ripples in the water around its base.", "met": false, "note": "The only marks on the water are two short, nearly straight white dashes to the left and right of the duck at different heights; they are not nested concentric rings around the duck's base." }, { "text": "Depict soap bubbles as clusters of overlapping circles with a slight iridescence, using semi-transparent white, light pink, and light blue fills.", "met": false, "note": "The bubbles are clusters of overlapping white circles with pale blue shading, but the render contains no pink pixels at all, so the light pink fill is missing." }, { "text": "Place a small cluster of bubbles on top of the duck's head.", "met": true, "note": "" }, { "text": "Place a large pile of bubbles against one side of the tub and a few floating on the water's surface around the duck.", "met": false, "note": "The bubbles form two similar modest clusters at the left and right ends of the tub, each about 19% of the tub's interior width and neither rising above the rim; there is no large pile banked against one side." } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/rubber-ducky-v2.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky-v2", "prompt": "Write `svg` code for an image of a rubber ducky floating on top of a soapy bathtub.", "score": 0.5555555555555556, "met": 5, "total": 9, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/rubber-ducky-v2.png", "requirements": [ { "text": "Create a classic rubber ducky shape with a distinct body and head, colored bright yellow.", "met": true, "note": "" }, { "text": "The duck must have an orange beak and a simple black dot for an eye.", "met": true, "note": "" }, { "text": "Draw a white, claw-foot bathtub shape, showing the inside view with a visible rim.", "met": false, "note": "The tub stands on two plain rectangular stubs with rounded bottoms - simple blocks with no toes, claws or gripped ball, so it is not a claw-foot tub." }, { "text": "Fill the lower portion of the bathtub with light blue water.", "met": true, "note": "" }, { "text": "Ensure the water line is clearly visible across the duck's body, showing the lower third of the duck submerged.", "met": true, "note": "" }, { "text": "Position the duck so it is floating on the water's surface, creating small, concentric circular ripples in the water around its base.", "met": false, "note": "The only marks on the water are two short, nearly straight white dashes to the left and right of the duck at different heights; they are not nested concentric rings around the duck's base." }, { "text": "Depict soap bubbles as clusters of overlapping circles with a slight iridescence, using semi-transparent white, light pink, and light blue fills.", "met": false, "note": "The bubbles are clusters of overlapping white circles with pale blue shading, but the render contains no pink pixels at all, so the light pink fill is missing." }, { "text": "Place a small cluster of bubbles on top of the duck's head.", "met": true, "note": "" }, { "text": "Place a large pile of bubbles against one side of the tub and a few floating on the water's surface around the duck.", "met": false, "note": "The bubbles form two similar modest clusters at the left and right ends of the tub, each about 19% of the tub's interior width and neither rising above the rim; there is no large pile banked against one side." } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/rubber-ducky.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky", "prompt": "Write `svg` code for an image of a rubber ducky floating on top of a soapy bathtub.", "score": 0.5555555555555556, "met": 5, "total": 9, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/rubber-ducky.png", "requirements": [ { "text": "Create a classic rubber ducky shape with a distinct body and head, colored bright yellow.", "met": true, "note": "" }, { "text": "The duck must have an orange beak and a simple black dot for an eye.", "met": true, "note": "" }, { "text": "Draw a white, claw-foot bathtub shape, showing the inside view with a visible rim.", "met": false, "note": "The tub stands on two plain rectangular stubs with rounded bottoms - simple blocks with no toes, claws or gripped ball, so it is not a claw-foot tub." }, { "text": "Fill the lower portion of the bathtub with light blue water.", "met": true, "note": "" }, { "text": "Ensure the water line is clearly visible across the duck's body, showing the lower third of the duck submerged.", "met": true, "note": "" }, { "text": "Position the duck so it is floating on the water's surface, creating small, concentric circular ripples in the water around its base.", "met": false, "note": "The only marks on the water are two short, nearly straight white dashes to the left and right of the duck at different heights; they are not nested concentric rings around the duck's base." }, { "text": "Depict soap bubbles as clusters of overlapping circles with a slight iridescence, using semi-transparent white, light pink, and light blue fills.", "met": false, "note": "The bubbles are clusters of overlapping white circles with pale blue shading, but the render contains no pink pixels at all, so the light pink fill is missing." }, { "text": "Place a small cluster of bubbles on top of the duck's head.", "met": true, "note": "" }, { "text": "Place a large pile of bubbles against one side of the tub and a few floating on the water's surface around the duck.", "met": false, "note": "The bubbles form two similar modest clusters at the left and right ends of the tub, each about 19% of the tub's interior width and neither rising above the rim; there is no large pile banked against one side." } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/stunt-car-fire-ring-animated.svg", "question_index": 6, "animated": true, "slug": "stunt-car-fire-ring-animated", "prompt": "Write `svg` code for an image of a stunt car jumping through a circle of fire.", "score": 0.5, "met": 4, "total": 8, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/stunt-car-fire-ring-animated.png", "requirements": [ { "text": "Create a red, sporty car with a body, wheels, and windows.", "met": true, "note": "" }, { "text": "Draw a distinct circular hoop that is completely surrounded by jagged, irregular flames colored with reds, oranges, and yellows.", "met": true, "note": "" }, { "text": "Include a take-off ramp on the left side of the fiery ring and a landing ramp on the right side.", "met": true, "note": "" }, { "text": "Position the car in mid-air, with its front half having passed through the ring and its back half still inside the ring.", "met": false, "note": "The whole car is inside the ring; nothing has passed through it. Against the hoop's fitted mid-band (centre 399,233, r=151), 100.0% of the car's red body lies inside the opening and 0.0% beyond it, so the front half has not passed through." }, { "text": "The car's rear wheels must be depicted as just having left the edge of the take-off ramp.", "met": false, "note": "The rear wheel is nowhere near the take-off ramp's edge. The ramp's lip is at about (263,372) while the rear wheel sits at about (355,288) in the middle of the hoop - roughly 125 px away, about 2.8 wheel diameters, and well above the ramp." }, { "text": "Add a ground surface below the entire jump setup.", "met": true, "note": "" }, { "text": "Incorporate gray speed lines trailing behind the car to convey high speed.", "met": false, "note": "There are no gray speed lines. Behind the car there is a pink spoiler and a gold exhaust wisp, plus a few small tan debris dashes; nothing gray trails it." }, { "text": "Add orange sparks where the car's tires last touched the take-off ramp.", "met": false, "note": "No sparks at the take-off ramp's edge. A 4x crop of the striped ramp lip shows bare ramp and ground; the scattered tan dashes near the wheels are debris flecks inside the hoop, not orange sparks at the ramp." } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/stunt-car-fire-ring-v2.svg", "question_index": 6, "animated": false, "slug": "stunt-car-fire-ring-v2", "prompt": "Write `svg` code for an image of a stunt car jumping through a circle of fire.", "score": 0.5, "met": 4, "total": 8, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/stunt-car-fire-ring-v2.png", "requirements": [ { "text": "Create a red, sporty car with a body, wheels, and windows.", "met": true, "note": "" }, { "text": "Draw a distinct circular hoop that is completely surrounded by jagged, irregular flames colored with reds, oranges, and yellows.", "met": true, "note": "" }, { "text": "Include a take-off ramp on the left side of the fiery ring and a landing ramp on the right side.", "met": true, "note": "" }, { "text": "Position the car in mid-air, with its front half having passed through the ring and its back half still inside the ring.", "met": false, "note": "The whole car is inside the ring; nothing has passed through it. Against the hoop's fitted mid-band (centre 399,234, r=150), 100.0% of the car's red body lies inside the opening and 0.0% beyond it, so the front half has not passed through." }, { "text": "The car's rear wheels must be depicted as just having left the edge of the take-off ramp.", "met": false, "note": "The rear wheel is nowhere near the take-off ramp's edge. The ramp's lip is at about (263,372) while the rear wheel sits at about (355,288) in the middle of the hoop - roughly 125 px away, about 2.8 wheel diameters, and well above the ramp." }, { "text": "Add a ground surface below the entire jump setup.", "met": true, "note": "" }, { "text": "Incorporate gray speed lines trailing behind the car to convey high speed.", "met": false, "note": "The three streaks trailing the car are not gray: they are the same tan as the base variant (sampled about RGB 210,166,111, a max-minus-min channel spread near 99), well outside the near-neutral test (spread <= 40) used for the binding colour words. The dark navy shape behind the roof is a spoiler, not a speed line." }, { "text": "Add orange sparks where the car's tires last touched the take-off ramp.", "met": false, "note": "No sparks at the take-off ramp's edge. A 4x crop of the striped ramp lip shows bare ramp and ground; the only orange near the car is the exhaust wisp at its tail, inside the hoop." } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/stunt-car-fire-ring-v3.svg", "question_index": 6, "animated": false, "slug": "stunt-car-fire-ring-v3", "prompt": "Write `svg` code for an image of a stunt car jumping through a circle of fire.", "score": 0.5, "met": 4, "total": 8, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/stunt-car-fire-ring-v3.png", "requirements": [ { "text": "Create a red, sporty car with a body, wheels, and windows.", "met": true, "note": "" }, { "text": "Draw a distinct circular hoop that is completely surrounded by jagged, irregular flames colored with reds, oranges, and yellows.", "met": true, "note": "" }, { "text": "Include a take-off ramp on the left side of the fiery ring and a landing ramp on the right side.", "met": true, "note": "" }, { "text": "Position the car in mid-air, with its front half having passed through the ring and its back half still inside the ring.", "met": false, "note": "The whole car is inside the ring; nothing has passed through it. Against the hoop's fitted mid-band (centre 399,234, r=150), 100.0% of the car's red body lies inside the opening and 0.0% beyond it, so the front half has not passed through." }, { "text": "The car's rear wheels must be depicted as just having left the edge of the take-off ramp.", "met": false, "note": "The rear wheel is nowhere near the take-off ramp's edge. The ramp's lip is at about (263,372) while the rear wheel sits at about (355,288) in the middle of the hoop - roughly 125 px away, about 2.8 wheel diameters, and well above the ramp." }, { "text": "Add a ground surface below the entire jump setup.", "met": true, "note": "" }, { "text": "Incorporate gray speed lines trailing behind the car to convey high speed.", "met": false, "note": "There are no speed lines at all in this variant. The tan streaks of the base and v2 have been replaced by a pink rear spoiler and a gold exhaust wisp; nothing gray trails the car." }, { "text": "Add orange sparks where the car's tires last touched the take-off ramp.", "met": false, "note": "No sparks at the take-off ramp's edge. A 4x crop of the striped ramp lip shows bare ramp and ground; the gold wisp is at the car's tail in the middle of the hoop, not at the ramp." } ] }, { "harness": "opencode", "model": "qwen3.8-flash-next", "artifact": "logs/opencode/qwen3.8-flash-next/stunt-car-fire-ring.svg", "question_index": 6, "animated": false, "slug": "stunt-car-fire-ring", "prompt": "Write `svg` code for an image of a stunt car jumping through a circle of fire.", "score": 0.5, "met": 4, "total": 8, "judge": { "kind": "claude-vision", "model": "claude-opus-5", "date": "2026-09-07", "method": "viewed the Chrome render plus 2x crops of detail regions; strict per-requirement pass/fail" }, "render": "logs/renders/opencode/qwen3.8-flash-next/stunt-car-fire-ring.png", "requirements": [ { "text": "Create a red, sporty car with a body, wheels, and windows.", "met": true, "note": "" }, { "text": "Draw a distinct circular hoop that is completely surrounded by jagged, irregular flames colored with reds, oranges, and yellows.", "met": true, "note": "" }, { "text": "Include a take-off ramp on the left side of the fiery ring and a landing ramp on the right side.", "met": true, "note": "" }, { "text": "Position the car in mid-air, with its front half having passed through the ring and its back half still inside the ring.", "met": false, "note": "The whole car is inside the ring; nothing has passed through it. Against the hoop's fitted mid-band (centre 399,241, r=151), 100.0% of the car's red body lies inside the opening and 0.0% beyond it, so the front half has not passed through." }, { "text": "The car's rear wheels must be depicted as just having left the edge of the take-off ramp.", "met": false, "note": "The rear wheel is nowhere near the take-off ramp's edge. The ramp's lip is at about (263,372) while the rear wheel sits at about (355,288) in the middle of the hoop - roughly 125 px away, about 2.8 wheel diameters, and well above the ramp." }, { "text": "Add a ground surface below the entire jump setup.", "met": true, "note": "" }, { "text": "Incorporate gray speed lines trailing behind the car to convey high speed.", "met": false, "note": "The three streaks trailing the car are not gray: they sample RGB 210,166,111 / 207,164,111, a max-minus-min channel spread of about 99, which is a tan, well outside the near-neutral test (spread <= 40) used for the binding colour words." }, { "text": "Add orange sparks where the car's tires last touched the take-off ramp.", "met": false, "note": "No sparks at the take-off ramp's edge. A 4x crop of the ramp lip shows bare grey ramp and ground; the only orange near the car is a thin exhaust wisp at its tail, 130 px away inside the hoop." } ] } ], "questions": [ { "question_index": 0, "prompt": "Write `svg` code to draw an image of a cow plowing a field.", "n_requirements": 12, "requirements": [ "Create a cow with clearly recognizable bovine features, including a body, head, four legs, tail, and udder.", "The cow must have black and white patches for its coloring.", "Add cow ears, eyes, and snout for facial recognition.", "Position the cow in a realistic plowing stance, leaning forward as if pulling.", "The cow's hooves must be colored brown, as if covered in soil from the field.", "Include a traditional wooden plow with a visible metal blade/share.", "Depict a wooden yoke across the cow's shoulders, connected to the plow by visible chains.", "The plow's blade must be partially buried in the soil, actively turning over a chunk of earth.", "Show at least three distinct, dark furrows in the soil trailing directly behind the plow.", "The field must be split into a plowed section and an unplowed section, with the cow and plow positioned at the boundary between them.", "The unplowed section of the field must have short green grass, which is visibly being overturned by the plow.", "Add a simple background with a clear horizon line and a blue sky containing a yellow sun." ], "artifacts": [ "logs/claude/claude-opus-5/cow-plowing-animated.svg", "logs/claude/claude-opus-5/cow-plowing-v2.svg", "logs/claude/claude-opus-5/cow-plowing-v3.svg", "logs/claude/claude-opus-5/cow-plowing.svg", "logs/claude/fable-5-1/cow-plowing-animated.svg", "logs/claude/fable-5-1/cow-plowing-v2.svg", "logs/claude/fable-5-1/cow-plowing-v3.svg", "logs/claude/fable-5-1/cow-plowing.svg", "logs/opencode/gpt-5.6-terra/cow-plowing-animated.svg", "logs/opencode/gpt-5.6-terra/cow-plowing-v1.svg", "logs/opencode/gpt-5.6-terra/cow-plowing-v2.svg", "logs/opencode/gpt-5.6-terra/cow-plowing-v3.svg", "logs/opencode/qwen3.8-27b-8bit/cow-plowing-animated.svg", "logs/opencode/qwen3.8-27b-8bit/cow-plowing.svg", "logs/opencode/qwen3.8-flash-next/cow-plowing-animated.svg", "logs/opencode/qwen3.8-flash-next/cow-plowing-v2.svg", "logs/opencode/qwen3.8-flash-next/cow-plowing-v3.svg", "logs/opencode/qwen3.8-flash-next/cow-plowing.svg" ] }, { "question_index": 4, "prompt": "Write `svg` code for an image of a rubber ducky floating on top of a soapy bathtub.", "n_requirements": 9, "requirements": [ "Create a classic rubber ducky shape with a distinct body and head, colored bright yellow.", "The duck must have an orange beak and a simple black dot for an eye.", "Draw a white, claw-foot bathtub shape, showing the inside view with a visible rim.", "Fill the lower portion of the bathtub with light blue water.", "Ensure the water line is clearly visible across the duck's body, showing the lower third of the duck submerged.", "Position the duck so it is floating on the water's surface, creating small, concentric circular ripples in the water around its base.", "Depict soap bubbles as clusters of overlapping circles with a slight iridescence, using semi-transparent white, light pink, and light blue fills.", "Place a small cluster of bubbles on top of the duck's head.", "Place a large pile of bubbles against one side of the tub and a few floating on the water's surface around the duck." ], "artifacts": [ "logs/claude/claude-opus-5/rubber-ducky-animated.svg", "logs/claude/claude-opus-5/rubber-ducky-v2.svg", "logs/claude/claude-opus-5/rubber-ducky-v3.svg", "logs/claude/claude-opus-5/rubber-ducky.svg", "logs/opencode/gpt-5.6-terra/rubber-ducky-soapy-bathtub-animated.svg", "logs/opencode/gpt-5.6-terra/rubber-ducky-soapy-bathtub-v1.svg", "logs/opencode/gpt-5.6-terra/rubber-ducky-soapy-bathtub-v2.svg", "logs/opencode/gpt-5.6-terra/rubber-ducky-soapy-bathtub-v3.svg", "logs/opencode/qwen3.8-flash-next/rubber-ducky-animated.svg", "logs/opencode/qwen3.8-flash-next/rubber-ducky-v2.svg", "logs/opencode/qwen3.8-flash-next/rubber-ducky.svg" ] }, { "question_index": 5, "prompt": "Write `svg` code for an image of a picnic on top of the clouds, where 2 parents and 2 children have landed with a hot air ballon, and are setting up a picnic with a tarp and food items.", "n_requirements": 10, "requirements": [ "Create a scene set on top of a solid-looking cloudscape that serves as the ground.", "Include a hot air balloon with a large envelope featuring vertical red and white stripes and a brown wicker basket.", "The balloon's basket must be resting firmly on the surface of a large, flat-topped cloud.", "Show ropes connecting the balloon's envelope to the basket.", "Include four human figures styled as a family: two adults and two children.", "Position one adult figure holding a corner of a red and white checkered picnic blanket, while one child figure holds the opposite corner, as if they are spreading it out together on the cloud.", "Place an open picnic basket on a corner of the blanket that is already spread out.", "A thermos, a bunch of grapes, and a sandwich must be visible emerging from the open picnic basket.", "Position the second adult and second child near the landed hot air balloon, with the adult pointing up at the balloon's envelope.", "The background must be a clear blue sky containing a bright yellow sun and two small, distant clouds." ], "artifacts": [ "logs/claude/claude-opus-5/picnic-on-clouds-animated.svg", "logs/claude/claude-opus-5/picnic-on-clouds-v2.svg", "logs/claude/claude-opus-5/picnic-on-clouds-v3.svg", "logs/claude/claude-opus-5/picnic-on-clouds.svg", "logs/opencode/gpt-5.6-terra/picnic-clouds-hot-air-balloon-animated.svg", "logs/opencode/gpt-5.6-terra/picnic-clouds-hot-air-balloon-v1.svg", "logs/opencode/gpt-5.6-terra/picnic-clouds-hot-air-balloon-v2.svg", "logs/opencode/gpt-5.6-terra/picnic-clouds-hot-air-balloon-v3.svg", "logs/opencode/qwen3.8-27b-8bit/picnic-on-clouds.svg" ] }, { "question_index": 6, "prompt": "Write `svg` code for an image of a stunt car jumping through a circle of fire.", "n_requirements": 8, "requirements": [ "Create a red, sporty car with a body, wheels, and windows.", "Draw a distinct circular hoop that is completely surrounded by jagged, irregular flames colored with reds, oranges, and yellows.", "Include a take-off ramp on the left side of the fiery ring and a landing ramp on the right side.", "Position the car in mid-air, with its front half having passed through the ring and its back half still inside the ring.", "The car's rear wheels must be depicted as just having left the edge of the take-off ramp.", "Add a ground surface below the entire jump setup.", "Incorporate gray speed lines trailing behind the car to convey high speed.", "Add orange sparks where the car's tires last touched the take-off ramp." ], "artifacts": [ "logs/opencode/gpt-5.6-terra/stunt-car-jumping-circle-fire-animated.svg", "logs/opencode/gpt-5.6-terra/stunt-car-jumping-circle-fire-v1.svg", "logs/opencode/gpt-5.6-terra/stunt-car-jumping-circle-fire-v2.svg", "logs/opencode/gpt-5.6-terra/stunt-car-jumping-circle-fire-v3.svg", "logs/opencode/qwen3.8-flash-next/stunt-car-fire-ring-animated.svg", "logs/opencode/qwen3.8-flash-next/stunt-car-fire-ring-v2.svg", "logs/opencode/qwen3.8-flash-next/stunt-car-fire-ring-v3.svg", "logs/opencode/qwen3.8-flash-next/stunt-car-fire-ring.svg" ] }, { "question_index": 7, "prompt": "Write `svg` code for an image of a dolphin jumping out of the water and through a hula hoop to bite a fish out of its trainers hand.", "n_requirements": 7, "requirements": [ "Create a recognizable grey dolphin with a streamlined body, dorsal fin, and tail fluke.", "Show a water surface below the dolphin with a large splash effect at the point where the dolphin has exited the water.", "Position the dolphin in a dynamic jumping arc, with its entire body in mid-air.", "Draw a circular, multi-colored hula hoop, and position the dolphin so its mid-section is passing through the center of the hoop.", "Include a human trainer's arm and hand extending into the frame.", "The trainer's hand must be holding a small fish by the tail.", "The dolphin's mouth must be depicted as wide open, just about to bite the body of the fish being held by the trainer." ], "artifacts": [ "logs/claude/claude-opus-4-8/dolphin-hoop.svg", "logs/claude/claude-opus-5/dolphin-animated.svg", "logs/claude/claude-opus-5/dolphin-hoop-jump.svg", "logs/claude/claude-opus-5/dolphin-v2.svg", "logs/claude/claude-opus-5/dolphin-v3.svg", "logs/claude/claude-opus-5/dolphin.svg", "logs/opencode/gpt-5.6-terra/dolphin-hula-hoop-fish-animated.svg", "logs/opencode/gpt-5.6-terra/dolphin-hula-hoop-fish-v1.svg", "logs/opencode/gpt-5.6-terra/dolphin-hula-hoop-fish-v2.svg", "logs/opencode/gpt-5.6-terra/dolphin-hula-hoop-fish-v3.svg", "logs/opencode/qwen3.8-27b-8bit/dolphin.svg", "logs/opencode/qwen3.8-flash-next/dolphin-animated.svg", "logs/opencode/qwen3.8-flash-next/dolphin-hoop-fish-animated.svg", "logs/opencode/qwen3.8-flash-next/dolphin-hoop-fish-v2.svg", "logs/opencode/qwen3.8-flash-next/dolphin-hoop-fish-v3.svg", "logs/opencode/qwen3.8-flash-next/dolphin-hoop-fish.svg", "logs/opencode/qwen3.8-flash-next/dolphin-hula-hoop-fish-animated.svg", "logs/opencode/qwen3.8-flash-next/dolphin-hula-hoop-fish-v1.svg", "logs/opencode/qwen3.8-flash-next/dolphin-hula-hoop-fish-v2.svg", "logs/opencode/qwen3.8-flash-next/dolphin-hula-hoop-fish-v3.svg", "logs/opencode/qwen3.8-flash-next/dolphin.svg" ] }, { "question_index": 12, "prompt": "Write `svg` code for an image of a fruit stall in the market.", "n_requirements": 11, "requirements": [ "Create a main structure for a wooden fruit stall, including a counter.", "Add a red and white striped canopy over the stall, supported by two vertical wooden posts.", "Display exactly four different, recognizable fruits.", "First fruit: A pile of red apples in a wicker basket on the left side of the counter.", "Second fruit: A bunch of yellow bananas placed next to the apples.", "Third fruit: A pile of oranges in another wicker basket on the right side of the counter.", "Fourth fruit: A single, large slice of watermelon resting directly on the counter in the center.", "Include a character representing the stall vendor, a smiling man with a mustache, positioned behind the counter and between the baskets.", "Add a small, hanging chalkboard sign from the canopy that reads 'Fresh Fruit'.", "Depict a cobblestone ground surface in front of the stall.", "Include the silhouette of another market stall in the background to suggest a larger market setting." ], "artifacts": [ "logs/opencode/gpt-5.6-terra/fruit-stall-market-animated.svg", "logs/opencode/gpt-5.6-terra/fruit-stall-market-v1.svg", "logs/opencode/gpt-5.6-terra/fruit-stall-market-v2.svg", "logs/opencode/gpt-5.6-terra/fruit-stall-market-v3.svg", "logs/opencode/qwen3.8-flash-next/fruit-stall-animated.svg", "logs/opencode/qwen3.8-flash-next/fruit-stall-v2.svg", "logs/opencode/qwen3.8-flash-next/fruit-stall-v3.svg", "logs/opencode/qwen3.8-flash-next/fruit-stall.svg" ] }, { "question_index": 13, "prompt": "Write `svg` code for an image of half buried barrel of treasure.", "n_requirements": 10, "requirements": [ "Create a wooden barrel with visible vertical planks and two horizontal metal hoops.", "The barrel must be buried in a mound of sand, so only the top half is visible.", "Show the sand mounded up slightly around the visible base of the barrel.", "The barrel must be open at the top and filled with treasure items.", "Show treasure overflowing from the top and spilling down one side of the barrel onto the sand.", "The treasure must include a large pile of shiny gold coins, both inside and outside the barrel.", "Add a variety of colorful gemstones (red rubies, green emeralds, blue sapphires) mixed in with the coins.", "A string of white pearls must be draped over the edge of the barrel, trailing down into the spilled coins.", "A golden goblet must be visible, partially buried in the coins inside the barrel.", "Use bright highlights and glint effects on the coins, gems, and goblet to make them look shiny." ], "artifacts": [ "logs/claude/claude-opus-5/treasure-barrel-animated.svg", "logs/claude/claude-opus-5/treasure-barrel-v2.svg", "logs/claude/claude-opus-5/treasure-barrel-v3.svg", "logs/claude/claude-opus-5/treasure-barrel.svg", "logs/opencode/gpt-5.6-terra/half-buried-treasure-barrel-animated.svg", "logs/opencode/gpt-5.6-terra/half-buried-treasure-barrel-v1.svg", "logs/opencode/gpt-5.6-terra/half-buried-treasure-barrel-v2.svg", "logs/opencode/gpt-5.6-terra/half-buried-treasure-barrel-v3.svg", "logs/opencode/qwen3.8-27b-8bit/treasure-barrel.svg" ] } ], "unscored": [ { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/cow-plowing-animated.svg", "question_index": 0, "animated": true, "slug": "cow-plowing-animated", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/cow-plowing-v1.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v1", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/cow-plowing-v2.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v2", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/cow-plowing-v3.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v3", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/dolphin-animated.svg", "question_index": 7, "animated": true, "slug": "dolphin-animated", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/dolphin-v1.svg", "question_index": 7, "animated": false, "slug": "dolphin-v1", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/dolphin-v2.svg", "question_index": 7, "animated": false, "slug": "dolphin-v2", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/dolphin-v3.svg", "question_index": 7, "animated": false, "slug": "dolphin-v3", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/fruit-stall-animated.svg", "question_index": 12, "animated": true, "slug": "fruit-stall-animated", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/fruit-stall-v1.svg", "question_index": 12, "animated": false, "slug": "fruit-stall-v1", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/fruit-stall-v2.svg", "question_index": 12, "animated": false, "slug": "fruit-stall-v2", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/fruit-stall-v3.svg", "question_index": 12, "animated": false, "slug": "fruit-stall-v3", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/picnic-on-clouds-animated.svg", "question_index": 5, "animated": true, "slug": "picnic-on-clouds-animated", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/picnic-on-clouds-v1.svg", "question_index": 5, "animated": false, "slug": "picnic-on-clouds-v1", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/picnic-on-clouds-v2.svg", "question_index": 5, "animated": false, "slug": "picnic-on-clouds-v2", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/picnic-on-clouds-v3.svg", "question_index": 5, "animated": false, "slug": "picnic-on-clouds-v3", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/rubber-ducky-animated.svg", "question_index": 4, "animated": true, "slug": "rubber-ducky-animated", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/rubber-ducky-v1.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky-v1", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/rubber-ducky-v2.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky-v2", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/rubber-ducky-v3.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky-v3", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/stunt-car-animated.svg", "question_index": 6, "animated": true, "slug": "stunt-car-animated", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/stunt-car-v1.svg", "question_index": 6, "animated": false, "slug": "stunt-car-v1", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/stunt-car-v2.svg", "question_index": 6, "animated": false, "slug": "stunt-car-v2", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/stunt-car-v3.svg", "question_index": 6, "animated": false, "slug": "stunt-car-v3", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/treasure-barrel-animated.svg", "question_index": 13, "animated": true, "slug": "treasure-barrel-animated", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/treasure-barrel-v1.svg", "question_index": 13, "animated": false, "slug": "treasure-barrel-v1", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/treasure-barrel-v2.svg", "question_index": 13, "animated": false, "slug": "treasure-barrel-v2", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-5.6-sol", "artifact": "logs/codex/gpt-5.6-sol/treasure-barrel-v3.svg", "question_index": 13, "animated": false, "slug": "treasure-barrel-v3", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/cow-plowing-animated.svg", "question_index": 0, "animated": true, "slug": "cow-plowing-animated", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/cow-plowing-v1.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v1", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/cow-plowing-v2.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v2", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/cow-plowing-v3.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v3", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/dolphin-animated.svg", "question_index": 7, "animated": true, "slug": "dolphin-animated", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/dolphin-v1.svg", "question_index": 7, "animated": false, "slug": "dolphin-v1", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/dolphin-v2.svg", "question_index": 7, "animated": false, "slug": "dolphin-v2", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/dolphin-v3.svg", "question_index": 7, "animated": false, "slug": "dolphin-v3", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/fruit-stall-animated.svg", "question_index": 12, "animated": true, "slug": "fruit-stall-animated", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/fruit-stall-v1.svg", "question_index": 12, "animated": false, "slug": "fruit-stall-v1", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/fruit-stall-v2.svg", "question_index": 12, "animated": false, "slug": "fruit-stall-v2", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/fruit-stall-v3.svg", "question_index": 12, "animated": false, "slug": "fruit-stall-v3", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/picnic-on-clouds-animated.svg", "question_index": 5, "animated": true, "slug": "picnic-on-clouds-animated", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/picnic-on-clouds-v1.svg", "question_index": 5, "animated": false, "slug": "picnic-on-clouds-v1", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/picnic-on-clouds-v2.svg", "question_index": 5, "animated": false, "slug": "picnic-on-clouds-v2", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/picnic-on-clouds-v3.svg", "question_index": 5, "animated": false, "slug": "picnic-on-clouds-v3", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/rubber-ducky-animated.svg", "question_index": 4, "animated": true, "slug": "rubber-ducky-animated", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/rubber-ducky-v1.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky-v1", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/rubber-ducky-v2.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky-v2", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/rubber-ducky-v3.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky-v3", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/stunt-car-animated.svg", "question_index": 6, "animated": true, "slug": "stunt-car-animated", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/stunt-car-v1.svg", "question_index": 6, "animated": false, "slug": "stunt-car-v1", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/stunt-car-v2.svg", "question_index": 6, "animated": false, "slug": "stunt-car-v2", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/stunt-car-v3.svg", "question_index": 6, "animated": false, "slug": "stunt-car-v3", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/treasure-barrel-animated.svg", "question_index": 13, "animated": true, "slug": "treasure-barrel-animated", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/treasure-barrel-v1.svg", "question_index": 13, "animated": false, "slug": "treasure-barrel-v1", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/treasure-barrel-v2.svg", "question_index": 13, "animated": false, "slug": "treasure-barrel-v2", "reason": "no verdict file" }, { "harness": "codex", "model": "gpt-6-astra", "artifact": "logs/codex/gpt-6-astra/treasure-barrel-v3.svg", "question_index": 13, "animated": false, "slug": "treasure-barrel-v3", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/cow-plowing-animated.svg", "question_index": 0, "animated": true, "slug": "cow-plowing-animated", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/cow-plowing-v1.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v1", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/cow-plowing-v2.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v2", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/cow-plowing-v3.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v3", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/dolphin-animated.svg", "question_index": 7, "animated": true, "slug": "dolphin-animated", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/dolphin-v1.svg", "question_index": 7, "animated": false, "slug": "dolphin-v1", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/dolphin-v2.svg", "question_index": 7, "animated": false, "slug": "dolphin-v2", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/dolphin-v3.svg", "question_index": 7, "animated": false, "slug": "dolphin-v3", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/fruit-stall-animated.svg", "question_index": 12, "animated": true, "slug": "fruit-stall-animated", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/fruit-stall-v1.svg", "question_index": 12, "animated": false, "slug": "fruit-stall-v1", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/fruit-stall-v2.svg", "question_index": 12, "animated": false, "slug": "fruit-stall-v2", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/fruit-stall-v3.svg", "question_index": 12, "animated": false, "slug": "fruit-stall-v3", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/picnic-on-clouds-animated.svg", "question_index": 5, "animated": true, "slug": "picnic-on-clouds-animated", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/picnic-on-clouds-v1.svg", "question_index": 5, "animated": false, "slug": "picnic-on-clouds-v1", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/picnic-on-clouds-v2.svg", "question_index": 5, "animated": false, "slug": "picnic-on-clouds-v2", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/picnic-on-clouds-v3.svg", "question_index": 5, "animated": false, "slug": "picnic-on-clouds-v3", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/rubber-ducky-animated.svg", "question_index": 4, "animated": true, "slug": "rubber-ducky-animated", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/rubber-ducky-v1.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky-v1", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/rubber-ducky-v2.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky-v2", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/rubber-ducky-v3.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky-v3", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/stunt-car-animated.svg", "question_index": 6, "animated": true, "slug": "stunt-car-animated", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/stunt-car-v1.svg", "question_index": 6, "animated": false, "slug": "stunt-car-v1", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/stunt-car-v2.svg", "question_index": 6, "animated": false, "slug": "stunt-car-v2", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/stunt-car-v3.svg", "question_index": 6, "animated": false, "slug": "stunt-car-v3", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/treasure-barrel-animated.svg", "question_index": 13, "animated": true, "slug": "treasure-barrel-animated", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/treasure-barrel-v1.svg", "question_index": 13, "animated": false, "slug": "treasure-barrel-v1", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/treasure-barrel-v2.svg", "question_index": 13, "animated": false, "slug": "treasure-barrel-v2", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4-flash-0731-high", "artifact": "logs/opencode/deepseek-v4-flash-0731-high/treasure-barrel-v3.svg", "question_index": 13, "animated": false, "slug": "treasure-barrel-v3", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/cow-plowing-animated.svg", "question_index": 0, "animated": true, "slug": "cow-plowing-animated", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/cow-plowing-v1.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v1", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/cow-plowing-v2.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v2", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/cow-plowing-v3.svg", "question_index": 0, "animated": false, "slug": "cow-plowing-v3", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/dolphin-animated.svg", "question_index": 7, "animated": true, "slug": "dolphin-animated", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/dolphin-v1.svg", "question_index": 7, "animated": false, "slug": "dolphin-v1", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/dolphin-v2.svg", "question_index": 7, "animated": false, "slug": "dolphin-v2", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/dolphin-v3.svg", "question_index": 7, "animated": false, "slug": "dolphin-v3", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/fruit-stall-animated.svg", "question_index": 12, "animated": true, "slug": "fruit-stall-animated", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/fruit-stall-v1.svg", "question_index": 12, "animated": false, "slug": "fruit-stall-v1", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/fruit-stall-v2.svg", "question_index": 12, "animated": false, "slug": "fruit-stall-v2", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/fruit-stall-v3.svg", "question_index": 12, "animated": false, "slug": "fruit-stall-v3", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/picnic-on-clouds-animated.svg", "question_index": 5, "animated": true, "slug": "picnic-on-clouds-animated", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/picnic-on-clouds-v1.svg", "question_index": 5, "animated": false, "slug": "picnic-on-clouds-v1", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/picnic-on-clouds-v2.svg", "question_index": 5, "animated": false, "slug": "picnic-on-clouds-v2", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/picnic-on-clouds-v3.svg", "question_index": 5, "animated": false, "slug": "picnic-on-clouds-v3", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/rubber-ducky-animated.svg", "question_index": 4, "animated": true, "slug": "rubber-ducky-animated", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/rubber-ducky-v1.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky-v1", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/rubber-ducky-v2.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky-v2", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/rubber-ducky-v3.svg", "question_index": 4, "animated": false, "slug": "rubber-ducky-v3", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/stunt-car-animated.svg", "question_index": 6, "animated": true, "slug": "stunt-car-animated", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/stunt-car-v1.svg", "question_index": 6, "animated": false, "slug": "stunt-car-v1", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/stunt-car-v2.svg", "question_index": 6, "animated": false, "slug": "stunt-car-v2", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/stunt-car-v3.svg", "question_index": 6, "animated": false, "slug": "stunt-car-v3", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/treasure-barrel-animated.svg", "question_index": 13, "animated": true, "slug": "treasure-barrel-animated", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/treasure-barrel-v1.svg", "question_index": 13, "animated": false, "slug": "treasure-barrel-v1", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/treasure-barrel-v2.svg", "question_index": 13, "animated": false, "slug": "treasure-barrel-v2", "reason": "no verdict file" }, { "harness": "opencode", "model": "deepseek-v4.1-flash-high", "artifact": "logs/opencode/deepseek-v4.1-flash-high/treasure-barrel-v3.svg", "question_index": 13, "animated": false, "slug": "treasure-barrel-v3", "reason": "no verdict file" }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/dawn-beach-animated.svg", "question_index": null, "animated": true, "slug": "dawn-beach-animated", "reason": "prompt is not an SVGBench question" }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/dawn-beach-v1.svg", "question_index": null, "animated": true, "slug": "dawn-beach-v1", "reason": "prompt is not an SVGBench question" }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/dawn-beach-v2.svg", "question_index": null, "animated": true, "slug": "dawn-beach-v2", "reason": "prompt is not an SVGBench question" }, { "harness": "opencode", "model": "gpt-5.6-terra", "artifact": "logs/opencode/gpt-5.6-terra/dawn-beach-v3.svg", "question_index": null, "animated": true, "slug": "dawn-beach-v3", "reason": "prompt is not an SVGBench question" }, { "harness": "opencode", "model": "qwen3.8-27b-8bit", "artifact": "logs/opencode/qwen3.8-27b-8bit/dawn-beach.svg", "question_index": null, "animated": true, "slug": "dawn-beach", "reason": "prompt is not an SVGBench question" } ], "skipped": [] }