Pin paper mapping downloads to the complete uploaded data revision

TASK_ALIGNMENT.md

@@ -207,7 +207,7 @@ Every training/test root has `annotations/paper_subtasks.json`. All four trainin
 
 For the two desktop wording aliases, mapped `paper_definition_en` uses the paper's **drawer** wording while `source_definition_en` retains **storage box**. Other source fields and every interval endpoint are preserved. The new canonical strings for all 30 existing test definitions match the paper. The ten simulation MEM reference definitions have no inferred temporal labels. Press By Number retains the printed S1-S4 list and its explicit protocol-conflict flag.
 
-Download the mapping files at revision `main`; use that revision when reading these additional files. The original training/evaluation download examples remain pinned to their prior data releases, so older snapshots do not contain the new mapping files. These annotation files are read separately from model image/instruction inputs.
+Download the mapping files at revision `5b494e84028aa6212b060de2f6014a00d569e8cd`; use that revision when reading these additional files. The original training/evaluation download examples remain pinned to their prior data releases, so older snapshots do not contain the new mapping files. These annotation files are read separately from model image/instruction inputs.
 
 ### Read a test segment's paper definition
 
@@ -222,7 +222,7 @@ prefix = "evaluation/real-world/test-30"
 mapping_snapshot = snapshot_download(
     repo_id="wzlllong/value-model-benchmark-training-data",
     repo_type="dataset",
-    revision="main",
+    revision="5b494e84028aa6212b060de2f6014a00d569e8cd",
     allow_patterns=[f"{prefix}/annotations/paper_*"],
     local_dir="./robovalue-paper-mapping",
 )
@@ -248,7 +248,7 @@ prefix = "training-data/simulation/one-shot"
 mapping_snapshot = snapshot_download(
     repo_id="wzlllong/value-model-benchmark-training-data",
     repo_type="dataset",
-    revision="main",
+    revision="5b494e84028aa6212b060de2f6014a00d569e8cd",
     allow_patterns=[f"{prefix}/annotations/subtasks/paper_*.parquet",
                     f"{prefix}/annotations/paper_subtasks.json"],
     local_dir="./robovalue-paper-mapping",

TASK_ALIGNMENT_zh.md

@@ -207,6 +207,6 @@
 
 书桌两处用词映射的 `paper_definition_en` 使用论文的 **drawer**，`source_definition_en` 保留原 **storage box**。其他源字段与全部区间端点保留。已有测试词条的30项对应论文文字均与论文一致；仿真MEM的10项参考定义不带推断的时间标签。“依数字按按钮”继续按印刷S1-S4对应，同时注明第32页协议内部冲突。
 
-新增映射文件固定下载版本为 `main`。训练／评测原下载示例仍固定到各自既有数据版本；旧快照没有这些新增对应文件，读取时请使用映射版本重新下载目标子集。标注独立于模型图像／指令输入。
+新增映射文件固定下载版本为 `5b494e84028aa6212b060de2f6014a00d569e8cd`。训练／评测原下载示例仍固定到各自既有数据版本；旧快照没有这些新增对应文件，读取时请使用映射版本重新下载目标子集。标注独立于模型图像／指令输入。
 
 英文对照说明的 [读取示例](TASK_ALIGNMENT.md#paper-linked-annotations)展示如何按测试轨迹编号和 `[start_frame,end_frame_exclusive)` 查询论文词条，以及如何按仿真One-Shot的轨迹／帧／累计索引读取论文定义。没有源区间覆盖的查询保持无标签；指标归属及仿真物体／动作／左右臂语义覆盖层仍从原标注读取。

evaluation/real-world/test-30/README.md

@@ -184,4 +184,4 @@ shasum -a 256 -c checksums.sha256
 
 ## Paper subtask correspondence
 
-This root provides `annotations/paper_subtasks.json` with source-to-paper task IDs, S-numbered definitions, and pages. Companion tables map existing temporal annotations while preserving their endpoints; reference-only entries add no temporal labels. See [mapping details and separate loading examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/TASK_ALIGNMENT.md#paper-linked-annotations). Mapping revision: `main`.
+This root provides `annotations/paper_subtasks.json` with source-to-paper task IDs, S-numbered definitions, and pages. Companion tables map existing temporal annotations while preserving their endpoints; reference-only entries add no temporal labels. See [mapping details and separate loading examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/TASK_ALIGNMENT.md#paper-linked-annotations). Mapping revision: `5b494e84028aa6212b060de2f6014a00d569e8cd`.

evaluation/real-world/test-30/README_zh.md

@@ -184,4 +184,4 @@ shasum -a 256 -c checksums.sha256
 
 ## 论文子任务对应
 
-本根目录提供 `annotations/paper_subtasks.json`，将源子任务对应到论文任务、S编号、英文原文和页码。已有时间标注另附对应副本，原区间保持不变；空标注只提供参考定义。 See [mapping details and separate loading examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/TASK_ALIGNMENT_zh.md#paper-linked-annotations). Mapping revision: `main`.
+本根目录提供 `annotations/paper_subtasks.json`，将源子任务对应到论文任务、S编号、英文原文和页码。已有时间标注另附对应副本，原区间保持不变；空标注只提供参考定义。 See [mapping details and separate loading examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/TASK_ALIGNMENT_zh.md#paper-linked-annotations). Mapping revision: `5b494e84028aa6212b060de2f6014a00d569e8cd`.

evaluation/real-world/test-30/checksums.sha256

@@ -1,5 +1,5 @@
-33c4d8e606134ed828265ad93b1341b92c7a5f58a58d9c6ccac9b93e1af0f4ae  README.md
-4cf936882c3da7fa8d5a8759fe71b39cd23de7d069a882c520821427576412e5  README_zh.md
+623c786a5cc5ab4d4d9cb985c181d1c7c2d5bb2f27a40aba517953477c718ba8  README.md
+e3a0ecf454a4113e0748d8983db04484fc3c8e530ae65eb6dcdcf689eb2bc66f  README_zh.md
 a377ffb50a1adf5354a8d65e9a90175b73cacb4a9800593cf1eaf89885f2f1df  TRAJECTORIES.md
 3cea0680a8709e160e87ec3c01f6b355a3e3956ee68c6b9ef5e628269c792e5a  annotations/csvc_solutions.json
 a3be4d8110975a3a3214b4446a5d31cf41b99a91dc104d5a8f45b3129635806b  annotations/episodes.jsonl

evaluation/simulation/test-30/README.md

@@ -186,4 +186,4 @@ shasum -a 256 -c checksums.sha256
 
 ## Paper subtask correspondence
 
-This root provides `annotations/paper_subtasks.json` with source-to-paper task IDs, S-numbered definitions, and pages. Companion tables map existing temporal annotations while preserving their endpoints; reference-only entries add no temporal labels. See [mapping details and separate loading examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/TASK_ALIGNMENT.md#paper-linked-annotations). Mapping revision: `main`.
+This root provides `annotations/paper_subtasks.json` with source-to-paper task IDs, S-numbered definitions, and pages. Companion tables map existing temporal annotations while preserving their endpoints; reference-only entries add no temporal labels. See [mapping details and separate loading examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/TASK_ALIGNMENT.md#paper-linked-annotations). Mapping revision: `5b494e84028aa6212b060de2f6014a00d569e8cd`.

evaluation/simulation/test-30/README_zh.md

@@ -186,4 +186,4 @@ shasum -a 256 -c checksums.sha256
 
 ## 论文子任务对应
 
-本根目录提供 `annotations/paper_subtasks.json`，将源子任务对应到论文任务、S编号、英文原文和页码。已有时间标注另附对应副本，原区间保持不变；空标注只提供参考定义。 See [mapping details and separate loading examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/TASK_ALIGNMENT_zh.md#paper-linked-annotations). Mapping revision: `main`.
+本根目录提供 `annotations/paper_subtasks.json`，将源子任务对应到论文任务、S编号、英文原文和页码。已有时间标注另附对应副本，原区间保持不变；空标注只提供参考定义。 See [mapping details and separate loading examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/TASK_ALIGNMENT_zh.md#paper-linked-annotations). Mapping revision: `5b494e84028aa6212b060de2f6014a00d569e8cd`.

evaluation/simulation/test-30/checksums.sha256

@@ -1,5 +1,5 @@
-0f31fa0174bb00ccf8933124fa493823396fa8f90227cbf232af446a25bda8aa  README.md
-4d89c0d6efc89bcbb2575e952ea26fe3629cedb9269e1463b5a98f2f0ce65f2d  README_zh.md
+9a927d43573cde8cd2435f470754290da8e237548ac8f86c114034f820e2d9a2  README.md
+81539f9c30b21d0e7f7d83ca5a2b9611038a09f347cb891036525ae8953004bb  README_zh.md
 36ef94a87991421de07478d7d6667ff1755a93e45a8b173372b18c5ce9bc99ca  TRAJECTORIES.md
 c6609d781c66a153962e5c5c9412ba6289c4f0372d23d5a8c2f84842fa657130  annotations/annotation_provenance.json
 272dbff320bc7531dc0977a02eaa0a439e70b574394df28ff1420bec0ed8317b  annotations/csvc_solutions.json

subset_catalog.json

@@ -117,13 +117,13 @@
         "清扫积木"
       ],
       "role": "training",
-      "size_bytes": 45468451331,
+      "size_bytes": 45468451367,
       "size_scope": "entire subset directory, including paper mappings, documentation and integrity files",
       "paper_subtasks": {
         "dataset_root": "training-data/simulation/full-shot",
         "definitions": "training-data/simulation/full-shot/annotations/paper_subtasks.json",
         "temporal_annotations_available": false,
-        "revision": "main",
+        "revision": "5b494e84028aa6212b060de2f6014a00d569e8cd",
         "reference_only_definitions": true,
         "source_temporal_intervals_preserved": true
       }
@@ -284,11 +284,11 @@
             "gap": 20
           },
           "source_revision": "7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145",
-          "directory_size_bytes": 228885
+          "directory_size_bytes": 228921
         }
       },
       "role": "training",
-      "size_bytes": 499853470,
+      "size_bytes": 499853542,
       "size_scope": "entire subset directory, including paper mappings, documentation and integrity files",
       "paper_subtasks": {
         "dataset_root": "training-data/simulation/one-shot",
@@ -297,7 +297,7 @@
         "mapped_segments": "training-data/simulation/one-shot/annotations/subtasks/paper_subtask_segments.parquet",
         "frames": 8080,
         "segments": 58,
-        "revision": "main",
+        "revision": "5b494e84028aa6212b060de2f6014a00d569e8cd",
         "reference_only_definitions": true,
         "source_temporal_intervals_preserved": true
       }
@@ -374,7 +374,7 @@
         "插花与摆放藤条香薰",
         "准备早餐"
       ],
-      "size_bytes": 23126412932,
+      "size_bytes": 23126412968,
       "duration_seconds": 51144.03333333333,
       "role": "training",
       "size_scope": "entire subset directory, including paper mappings, documentation and integrity files",
@@ -382,7 +382,7 @@
         "dataset_root": "training-data/real-world/full-shot",
         "definitions": "training-data/real-world/full-shot/annotations/paper_subtasks.json",
         "temporal_annotations_available": false,
-        "revision": "main",
+        "revision": "5b494e84028aa6212b060de2f6014a00d569e8cd",
         "reference_only_definitions": true,
         "source_temporal_intervals_preserved": true
       }
@@ -415,7 +415,7 @@
       "source_episode_selection": "minimum original episode_id among the 100 source episodes for 19 tasks; tea uses the minimum original episode_id from the 27-episode tea refresh batch",
       "videos": 60,
       "platform": "AgiBot Genie02",
-      "size_bytes": 215743376,
+      "size_bytes": 215743412,
       "duration_seconds": 545.4,
       "display_task_names": [
         "Transfer Blocks Using Plate",
@@ -478,7 +478,7 @@
         "dataset_root": "training-data/real-world/one-shot",
         "definitions": "training-data/real-world/one-shot/annotations/paper_subtasks.json",
         "temporal_annotations_available": false,
-        "revision": "main",
+        "revision": "5b494e84028aa6212b060de2f6014a00d569e8cd",
         "reference_only_definitions": true,
         "source_temporal_intervals_preserved": true
       }
@@ -499,7 +499,7 @@
       "fps": 30,
       "camera_streams_per_episode": 3,
       "videos": 90,
-      "size_bytes": 382134689,
+      "size_bytes": 382134761,
       "duration_seconds": 916.6,
       "splits": {
         "train": "0:30"
@@ -572,7 +572,7 @@
         "definitions": "evaluation/real-world/test-30/annotations/paper_subtasks.json",
         "mapped_segments": "evaluation/real-world/test-30/annotations/paper_subtask_segments.parquet",
         "segments": 153,
-        "revision": "main",
+        "revision": "5b494e84028aa6212b060de2f6014a00d569e8cd",
         "reference_only_definitions": true,
         "source_temporal_intervals_preserved": true
       }
@@ -658,7 +658,7 @@
       "primary_camera": "observation.images.front",
       "primary_video_column": "video_primary",
       "annotation_evidence_note": "ED/MS terminal labels and semantic alignment are user-attested; TRR/MEM terminal-label evidence remains pending independent verification.",
-      "size_bytes": 594651915,
+      "size_bytes": 594651987,
       "data_revision": "578593ef28ee7a2d861ebf38162c928291b0d4a9",
       "size_scope": "entire subset directory, including paper mappings, documentation and integrity files",
       "paper_subtasks": {
@@ -666,7 +666,7 @@
         "definitions": "evaluation/simulation/test-30/annotations/paper_subtasks.json",
         "mapped_segments": "evaluation/simulation/test-30/annotations/paper_subtask_segments.parquet",
         "segments": 54,
-        "revision": "main",
+        "revision": "5b494e84028aa6212b060de2f6014a00d569e8cd",
         "reference_only_definitions": true,
         "source_temporal_intervals_preserved": true
       }
@@ -693,6 +693,6 @@
     "shared_test_tasks": 8,
     "paper_definition_catalog": "task_alignment.json#paper_definition_catalog",
     "distinct_paper_definitions": 155,
-    "paper_mapping_revision": "main"
+    "paper_mapping_revision": "5b494e84028aa6212b060de2f6014a00d569e8cd"
   }
 }

task_alignment.json

@@ -1181,7 +1181,7 @@
     "public_release": "These 60 trajectories are partial diagnostic coverage. Simulation keeps 10 formal metric entries, excluding a formal FPL cohort while retaining original FPL fields; real-world keeps 11."
   },
   "data_preservation": "This alignment index does not modify native training payloads or public test annotations, videos, frame indices, timestamps, interval endpoints, object assignments, or arm roles.",
-  "paper_mapping_revision": "main",
+  "paper_mapping_revision": "5b494e84028aa6212b060de2f6014a00d569e8cd",
   "paper_definition_catalog": [
     {
       "paper_definition_id": "simulation/arrange_largest_number/S1",

training-data/real-world/full-shot/README.md

@@ -75,4 +75,4 @@ Task indices match `meta/tasks.parquet`. Source episode identifiers and task cor
 
 ## Paper subtask correspondence
 
-This root provides `annotations/paper_subtasks.json` with source-to-paper task IDs, S-numbered definitions, and pages. Companion tables map existing temporal annotations while preserving their endpoints; reference-only entries add no temporal labels. See [mapping details and separate loading examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/TASK_ALIGNMENT.md#paper-linked-annotations). Mapping revision: `main`.
+This root provides `annotations/paper_subtasks.json` with source-to-paper task IDs, S-numbered definitions, and pages. Companion tables map existing temporal annotations while preserving their endpoints; reference-only entries add no temporal labels. See [mapping details and separate loading examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/TASK_ALIGNMENT.md#paper-linked-annotations). Mapping revision: `5b494e84028aa6212b060de2f6014a00d569e8cd`.

training-data/real-world/full-shot/checksums.sha256

@@ -1,4 +1,4 @@
-dc93a5ed4473fee34760a97aa516f4b314820bcb6b0c3aba0d55547254f8e933  README.md
+fb146776bd3bc32fddb3bd38e5936016dc86a36b80e835080afec470d489f508  README.md
 605f88771d0b2e91c58b33b327e2fbc079846ba31592db609367b26519d032e4  annotations/paper_subtasks.json
 80e785f6a7f703e48e25f00ee08ccb7a9d1698960fecbfc0f383d031df184959  data/chunk-000/file-000.parquet
 4a1370a74adc71fa8725f7b796b09c488d1a9894377f4e2b0c59d84460c9b147  data/chunk-000/file-001.parquet

training-data/real-world/one-shot/README.md

@@ -103,4 +103,4 @@ Task indices match `meta/tasks.parquet`. Video streams retain their original byt
 
 ## Paper subtask correspondence
 
-This root provides `annotations/paper_subtasks.json` with source-to-paper task IDs, S-numbered definitions, and pages. Companion tables map existing temporal annotations while preserving their endpoints; reference-only entries add no temporal labels. See [mapping details and separate loading examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/TASK_ALIGNMENT.md#paper-linked-annotations). Mapping revision: `main`.
+This root provides `annotations/paper_subtasks.json` with source-to-paper task IDs, S-numbered definitions, and pages. Companion tables map existing temporal annotations while preserving their endpoints; reference-only entries add no temporal labels. See [mapping details and separate loading examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/TASK_ALIGNMENT.md#paper-linked-annotations). Mapping revision: `5b494e84028aa6212b060de2f6014a00d569e8cd`.

training-data/real-world/one-shot/checksums.sha256

@@ -1,4 +1,4 @@
-c642f9b3981ec8d7df473fe26c44c3517e64d80c0691ac3eddd4753647477434  README.md
+838577b6548c8634a3b231c705aeb5aa0765e7d1cd5f11f3c8c48a3a019232b4  README.md
 e5c0ec48ac1da23236656635db9fb20abbacf5c4b1db4ee904b8eaa0ef19b042  annotations/paper_subtasks.json
 80e785f6a7f703e48e25f00ee08ccb7a9d1698960fecbfc0f383d031df184959  data/chunk-000/file-000.parquet
 83542f7c77692a0de24c9254a2114da68394d08ea59be373d78c458e1b6ef259  data/chunk-000/file-001.parquet

training-data/simulation/full-shot/README.md

@@ -64,4 +64,4 @@ Task indices match `meta/tasks.parquet`. Source episode identifiers and task cor
 
 ## Paper subtask correspondence
 
-This root provides `annotations/paper_subtasks.json` with source-to-paper task IDs, S-numbered definitions, and pages. Companion tables map existing temporal annotations while preserving their endpoints; reference-only entries add no temporal labels. See [mapping details and separate loading examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/TASK_ALIGNMENT.md#paper-linked-annotations). Mapping revision: `main`.
+This root provides `annotations/paper_subtasks.json` with source-to-paper task IDs, S-numbered definitions, and pages. Companion tables map existing temporal annotations while preserving their endpoints; reference-only entries add no temporal labels. See [mapping details and separate loading examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/TASK_ALIGNMENT.md#paper-linked-annotations). Mapping revision: `5b494e84028aa6212b060de2f6014a00d569e8cd`.

training-data/simulation/full-shot/checksums.sha256

@@ -1,4 +1,4 @@
-2179ddaa918e68cc7d148076775a91fe71d530bac09c51a7b44e913878f35ed4  README.md
+26cce342cddc916400f89e1ace3908f7ef1fac5ca2bbf68f7845ec2803e3ac67  README.md
 d90e99870c29d577d6d15b88e2e56b52913d79d92c53b27cb12801da0afdd1ff  annotations/paper_subtasks.json
 d0c983813f6b55338e4ae09508d96b92f312916c96db315e75e5a9e3c970dc63  data/chunk-000/file-000.parquet
 d04d0b7163f1b4be174fd56e3590b686a6f90e1981c68f94f52bdfc06a4a6574  data/chunk-000/file-001.parquet

training-data/simulation/one-shot/README.md

@@ -77,4 +77,4 @@ Read these files explicitly using PyArrow or `load_subtasks.py`; they are separa
 
 ## Paper subtask correspondence
 
-This root provides `annotations/paper_subtasks.json` with source-to-paper task IDs, S-numbered definitions, and pages. Companion tables map existing temporal annotations while preserving their endpoints; reference-only entries add no temporal labels. See [mapping details and separate loading examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/TASK_ALIGNMENT.md#paper-linked-annotations). Mapping revision: `main`.
+This root provides `annotations/paper_subtasks.json` with source-to-paper task IDs, S-numbered definitions, and pages. Companion tables map existing temporal annotations while preserving their endpoints; reference-only entries add no temporal labels. See [mapping details and separate loading examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/TASK_ALIGNMENT.md#paper-linked-annotations). Mapping revision: `5b494e84028aa6212b060de2f6014a00d569e8cd`.

training-data/simulation/one-shot/annotations/subtasks/README.md

@@ -121,4 +121,4 @@ print(record["episode_name"], video_path, intervals)
 
 ## Paper subtask correspondence
 
-The existing training supplement now has `paper_subtask_frame_index.parquet` and `paper_subtask_segments.parquet`. All original columns, endpoints, boundary masks, and labels are retained, with explicit paper definition IDs and page references added. / 原补标另提供逐帧和区间论文对应副本，保留原字段、端点、边界标记及标签。 See [mapping details and separate loading examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/TASK_ALIGNMENT.md#paper-linked-annotations). Mapping revision: `main`.
+The existing training supplement now has `paper_subtask_frame_index.parquet` and `paper_subtask_segments.parquet`. All original columns, endpoints, boundary masks, and labels are retained, with explicit paper definition IDs and page references added. / 原补标另提供逐帧和区间论文对应副本，保留原字段、端点、边界标记及标签。 See [mapping details and separate loading examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/TASK_ALIGNMENT.md#paper-linked-annotations). Mapping revision: `5b494e84028aa6212b060de2f6014a00d569e8cd`.

training-data/simulation/one-shot/checksums.sha256

@@ -1,6 +1,6 @@
-ccf68faa7b02e3e02d88c4de006f221adb6eb9b2abd0ceaefe193121e31990f2  README.md
+aebf2f59d28989e113f220bf2b57e3f5e5ff31b167c1f25b8cf103b0c0616433  README.md
 29866b8a39956ef240b0c0c02126227f327d36eabdcda0e084779bb349d8495b  annotations/paper_subtasks.json
-c929f08a5e917646addf5b01759673058874f3c87aae48a54896e7c2f19d5be3  annotations/subtasks/README.md
+f0d0ed030ddc944d641d281a5c470df7fc6cb1bc25176a3a23df90bdb034a1b7  annotations/subtasks/README.md
 b7454982fca24a6b5ef31483ee0244994b6fd92d644468351715c52e5d775dff  annotations/subtasks/dataset_labeler/arrange_largest_number_sia_cspc_annotations.json
 a9b8fe08f2f344c0c03a2093b47eca331ed9db20a3eb02c23ec26ac8d83fa133  annotations/subtasks/dataset_labeler/exceptional_intervals.json
 0f903cccbedb8123d5a640850732d1dfe37becaff034c9e7f1075bf305a5c5ef  annotations/subtasks/dataset_labeler/fill_pen_holder_sia_cspc_annotations.json