This 12,500-clip corpus preserves the 10,000 rows from osazuwa/2d_dungeon_flier_video and appends 2,500 deterministic clips over do(X/M/R=0/1). Together the corpus covers observational data and do(variable=0/1) for every binary variable D,Z,V,W,X,M,A,Y,R.
Clips and arrays ship as tar shards under shards/, listed with their clip ranges and sha256 digests in shards/index.json. Restore them with prepare_data.py --fetch-published… See the full description on the dataset page:
https://huggingface.co/datasets/osazuwa/2d_dungeon_flier_video_all_interventions.
