[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"capsule:ab2985e4-da19-44fb-a372-49a6dfa3f484":3},["Reactive",4],{"id":5,"user_id":6,"title":7,"prompt_content":8,"prompt_type":9,"is_locked":10,"is_public":10,"fork_of":11,"created_at":12,"updated_at":13,"category":14,"brief":15,"attachments":16,"expected_markers":17,"user":19},"ab2985e4-da19-44fb-a372-49a6dfa3f484","695c8e6c-9654-4858-b2d3-925adefacc35","Mirror Torture Test","A wooden analog clock on a brick wall showing exactly 10:08, hanging directly \nabove a round mirror. In front of the mirror stands a woman with her back to \nthe camera, wearing a red glove on her right hand only (left hand bare), \nholding up 4 fingers on her bare hand. The mirror reflection must show her \nface and confirm the glove is on her LEFT hand as seen in the mirror. On a \ntable to the woman's right sit exactly 7 apples, one of them green, arranged \nin a single row — no apples stacked. A wooden sign on the wall reads \n\"NO PHOTOS ALLOWED\" in bold black letters. The room has no windows. \nLighting comes from a single lamp on the left side of the frame, casting \nall shadows to the right.","image",true,null,"2026-07-03T04:41:22.146651+00:00","2026-07-03T04:42:41.128251+00:00","content","## Why This Capsule Exists\n\nMost image benchmarks test one thing at a time: counting, text rendering, or spatial reasoning. Real-world image generation fails happen when **multiple constraints compete for the model's attention simultaneously** — which is exactly what happens in production use cases. This capsule stacks 8 independent, objectively verifiable constraints into a single prompt to reveal how models handle compounded instruction-following pressure.\n\n## What This Tests\n\n| Constraint | Capability Probed |\n|---|---|\n| Clock reads exactly 10:08 | Precise analog time rendering (a known weak spot across model generations) |\n| Glove on right hand only, left hand bare | Asymmetric instruction-following (resisting the default \"matching pair\" bias) |\n| Bare hand shows exactly 4 fingers | Exact small-number generation on body parts |\n| Mirror reflection matches reality (correct flip) | Geometric\u002Fspatial reasoning, not just pattern repetition |\n| Exactly 7 apples | Object counting accuracy at small scale |\n| Exactly 1 green apple | Basic attribute assignment (color) |\n| Sign reads \"NO PHOTOS ALLOWED\" | Legible text rendering without duplication\u002Fglitches |\n| No windows in the room | Negation handling (ignoring implied defaults) |\n\n## Expected Results\n\n- **No model is expected to pass all 8 criteria.** A perfect score would be a signal to add harder constraints, not evidence the benchmark has failed.\n- **Weakest points across the board:** exact clock time and exact object counts — these fail even on otherwise photorealistic, high-quality outputs, showing that visual fidelity and instruction accuracy are separate capabilities.\n- **The mirror consistency check is the strongest signal for genuine spatial reasoning.** Models that just repeat the face without correctly flipping the glove side are pattern-matching \"mirror = duplicate person,\" not simulating reflection physics.\n- **Asymmetric instructions (one gloved hand, one bare) are frequently collapsed into symmetric defaults** — this is a good proxy for how well a model resists its own training priors when they conflict with the prompt.\n\n## How to Read a Run\n\nScore each run against the 8-point checklist below. Partial credit applies where an element is present but incorrect (e.g., wrong finger count, right time but wrong minute). This capsule is designed to produce **score dispersion across models and over time**, not a pass\u002Ffail verdict — the goal is watching the score trend upward as models improve, not expecting any single run to \"win.\"\n\n### Scoring Checklist\n- [ ] Clock shows exactly 10:08\n- [ ] Glove on one hand only, other hand bare\n- [ ] Bare hand shows exactly 4 fingers\n- [ ] Mirror reflection is geometrically consistent (correct side flip)\n- [ ] Exactly 7 apples\n- [ ] Exactly 1 green apple\n- [ ] Sign text reads correctly, no duplication\u002Fglitches\n- [ ] No windows visible in the room",[],{"items":18,"matchAll":10},[],{"username":20,"avatar_url":21},"Ezarwebmaster","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F42354978?v=4"]