Datasets:
Sanity checks and leakage guards
This file is intentionally aligned with docs/QA_AND_LEAKAGE.md for GWAM_Data v1.2 sparse-v2.
Normative current shapes:
node_state: float32[T, 256, 32]
view evidence + RLE masks: sparse visible node/view pairs
dynamic_edges: sparse JSONL contact + visibility_change
Phase 2: sparse visible-only feature files plus phase2_summary.json
Historical/stale shapes that must not be used for v1.2 sparse-v2 loaders:
317-D / 323-D node_feats
dense rel_state / rel_prior
dense visual_views
distractor family id 1
state_on/state_off events.jsonl vocab
Minimum package gate:
_PHASE1_DONE.jsonfor every packaged episode.audit/phase2_summary.jsonfor every packaged Phase 2 episode.- No invisible Phase 2 features.
- Phase 2 pairs partition Phase 1 visible pairs.
- No NaN/Inf in
node_state. - Padded rows all-zero; loaders gate on active channel.
- Goal/instruction/object-role fields remain supervision-only.
- Known drop
target/OpenDrawer/episode_000459documented and not packaged as a half episode.