Remove episode outcome metadata fields and overlays

README.md

@@ -137,30 +137,6 @@ The archive members contain the exported video and state/action bytes. Episode m
 reduced to public data fields. `dataset_info.json` summarizes the release and `shards.jsonl`
 lists archive sizes, episode counts, and SHA-256 checksums.
 
-## Episode outcomes
-
-`metadata/episode_outcomes.jsonl` provides one metadata record per episode, joined by
-`episode_id`. The fields are `episode_outcome` (`success`, `failure`, or `unknown`)
-and `episode_outcome_source` (label provenance). A schema is provided alongside it.
-
-All **134,537** retained episodes have `episode_outcome: "success"`, with provenance
-`source_directory:success_episodes`. Every retained episode maps to an upstream
-`success_episodes` directory; no source-labelled failure episodes occur in this subset.
-This preserves the source task-outcome label. It is not independent visual verification
-or a certification of pose, action, or video quality. Processing failures and alignment
-review flags are separate from task outcomes.
-
-The existing TAR episode JSON members predate this metadata addition. Merge the overlay
-into each episode's metadata when loading; video and binary payloads are unchanged:
-
-```python
-with open("metadata/episode_outcomes.jsonl", encoding="utf-8") as stream:
-    outcome_by_id = {row["episode_id"]: row for row in map(json.loads, stream)}
-metadata.update(outcome_by_id[metadata["episode_id"]])
-```
-
-`dataset_info.json` records the overlay counts and SHA-256 checksum.
-
 ## Source and license
 
 Upstream sources are [X-Humanoid/RoboMIND2.0-Agilex](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0-Agilex),

dataset_info.json

@@ -47,21 +47,5 @@
       "hours": 33.24141068607515
     }
   ],
-  "format": "WebDataset tar shards; one .mp4, .bin and .json per episode",
-  "episode_outcomes": {
-    "file": "metadata/episode_outcomes.jsonl",
-    "schema": "metadata/episode_outcomes.schema.json",
-    "join_key": "episode_id",
-    "field": "episode_outcome",
-    "provenance_field": "episode_outcome_source",
-    "episodes": 134537,
-    "counts": {
-      "success": 134537,
-      "failure": 0,
-      "unknown": 0
-    },
-    "sha256": "3a996aab4c28ca0dfc82a2ad5806bee01ff378adac38ebf9bb323109cde4edf2",
-    "note": "All retained episodes are labelled successful by the upstream success_episodes directory. This preserves the source label and is not independent visual verification.",
-    "archive_members_updated": false
-  }
+  "format": "WebDataset tar shards; one .mp4, .bin and .json per episode"
 }

metadata/episode_outcomes.jsonl

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:3a996aab4c28ca0dfc82a2ad5806bee01ff378adac38ebf9bb323109cde4edf2
-size 24082123

metadata/episode_outcomes.schema.json

@@ -1,29 +0,0 @@
-{
-  "$schema": "https://json-schema.org/draft/2020-12/schema",
-  "title": "Per-episode task outcome metadata overlay",
-  "type": "object",
-  "additionalProperties": false,
-  "required": [
-    "episode_id",
-    "episode_outcome",
-    "episode_outcome_source"
-  ],
-  "properties": {
-    "episode_id": {
-      "type": "string",
-      "pattern": "^episode-[0-9a-f]{64}$"
-    },
-    "episode_outcome": {
-      "type": "string",
-      "enum": [
-        "success",
-        "failure",
-        "unknown"
-      ]
-    },
-    "episode_outcome_source": {
-      "type": "string"
-    }
-  },
-  "description": "Join to the archive episode JSON by episode_id. Source labels describe task outcomes, not processing or action-quality checks. Unknown means no explicit source outcome was provided; do not treat it as success or failure."
-}