Nov 5 Yesterday 15:02 Add State repeated-pouring anomaly S20 (35 trajectories) Add the 2026-11-05 anomalous repeated-pouring variant of S17: 35 unannotated real-robot trajectories and 105 frame-exact videos; State S20, task_variant repeated_pouring, not recovery. Keep nominal task instruction and describe the anomaly separately. Add 28 train / 7 val, preserve all existing files and frozen assignments. Dataset/split v1.4.0: 21 tasks, 1011 trajectories, 806 train / 205 val. No stage/progress or success/failure labels inferred. Nov 3 3 days ago 10:07 Consolidate all State tasks (941 trajectories) Copy S01–S10 from the pinned MINT-SJTU/ProPriMem State collection: 618 annotated simulation trajectories, 1536 videos and 618 annotation files. Preserve the existing S11–S19 real-robot recordings, source hashes, domain labels, stage annotations and every frozen split assignment. Dataset/split v1.2.0 totals: 19 tasks, 941 trajectories, 2505 videos, 750 train / 191 val. Source repository remains unchanged; no provenance directory is added. 09:44 Add four State tasks (137 trajectories) Add S16–S19 from the 2026-11-03 PiperX collection: 137 unannotated trajectories and 411 frame-exact three-view videos. Exclude the reviewed tea #0/#1 and upside-down-cup #0 recordings; correct the bottle instruction to three bottles. Add 109 train / 28 val samples and preserve all existing S11–S15 files and frozen splits. Dataset/split v1.1.0: 9 tasks, 323 trajectories, 257 train / 66 val. No provenance files. 03:09 Add five State tasks (186 trajectories) 186 user-reviewed PiperX trajectories, 558 frame-exact three-view videos. Six abnormal piggy-bank episodes excluded; raw recordings retained. 148 train / 38 val, frozen splits. No stage or success labels inferred. 
