| { | |
| "review_date": "2026-09-24", | |
| "status": "image_roles_and_mapping_reviewed", | |
| "reviewed_components": 12, | |
| "reviewed_images": 13, | |
| "record_ids": [ | |
| "llava_understanding:276270", | |
| "lvis:110338", | |
| "editing:156426", | |
| "generation:278530", | |
| "lrv:168710", | |
| "laion_gpt4v:5532", | |
| "sharegpt4v:57001", | |
| "docvqa:20801", | |
| "synthdog:259796", | |
| "chartqa:11849", | |
| "dvqa:71080", | |
| "ai2d:874773" | |
| ], | |
| "checks": [ | |
| "input/target placement", | |
| "generation caption-image subject consistency", | |
| "understanding image-topic correspondence", | |
| "OCR/chart/diagram presence" | |
| ], | |
| "findings": [ | |
| { | |
| "source": "lrv", | |
| "id": "lrv:168710", | |
| "observation": "The released question and answer in this sample are numbered sentence fragments. They are preserved verbatim; structural validity does not imply good instruction quality." | |
| } | |
| ], | |
| "limitations": "One sample per image-bearing component, plus editing target. This is a small image/role sanity check, not a semantic accuracy audit of the full dataset." | |
| } | |