metadata.jsonl, the omission manifest and the dataset card

data/mega_liminal: 870 images with their Qwen3.5-9B captions

data/landscape: 766 images with their Qwen3.5-9B captions

data/suburban: 111 images with their Qwen3.5-9B captions

data/simulacrum: 36 images with their Qwen3.5-9B captions

data/vanishing_point: 21 images with their Qwen3.5-9B captions

data/empty_mall: 20 images with their Qwen3.5-9B captions

data/hand_sourced: 18 images with their Qwen3.5-9B captions

data/parking_garage: 13 images with their Qwen3.5-9B captions

data/cityscape: 10 images with their Qwen3.5-9B captions

data/movie_theatre: 8 images with their Qwen3.5-9B captions