Update training paths, pinned download versions, and paper-alignment documentation

README.md

@@ -88,6 +88,8 @@ The real-world subset contains **12 desk**, **12 flower-and-diffuser**, **3 drum
 
 See the [real-world metric coverage and trajectory index](evaluation/real-world/test-30/README.md#metric-coverage) and [simulation metric coverage and trajectory index](evaluation/simulation/test-30/README.md#metric-coverage). Simulation ED/MS terminal labels and semantic alignment are user-attested; TRR/MEM terminal-label evidence retains the source status of pending independent verification. The simulation README documents these annotation evidence states.
 
+**Task and subtask alignment:** [The paper comparison](TASK_ALIGNMENT.md) maps the eight public-test tasks to the 35 training tasks and documents annotation availability, matching definitions, and wording differences. All 58 supplementary Simulation One-Shot subtask definitions match the provided paper snapshot. Its Press By Number card has a confirmation-order inconsistency; the comparison records it while retaining the released annotations. A [machine-readable crosswalk](task_alignment.json) provides local task indices and paper page references.
+
 
 ## 🟦 2. Simulation Training <a name="simulation"></a>
 
@@ -201,7 +203,7 @@ Loaded images use **CHW (channels, height, width)** order.
 
 [**Browse files →**](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/tree/main/training-data/simulation/one-shot) · [Task–episode–row mapping](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/provenance/episode_mapping.json) · [Download and read](#download-and-reading)
 
-**Subtask annotations:** [58 paper-defined segments](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/annotations/subtasks/README.md) label all **8,080 frames**, with one subtask per frame. These annotations accompany the training demonstrations and are separate from benchmark test-set labels. [`dataset_labeler`-compatible JSON](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/tree/MIGRATED_TRAINING_REVISION/training-data/simulation/one-shot/annotations/subtasks/dataset_labeler) uses the same subtasks and boundaries as the frame labels.
+**Subtask annotations:** [58 paper-defined segments](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/annotations/subtasks/README.md) label all **8,080 frames**, with one subtask per frame. These annotations accompany the training demonstrations and are separate from benchmark test-set labels. [`dataset_labeler`-compatible JSON](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/tree/main/training-data/simulation/one-shot/annotations/subtasks/dataset_labeler) uses the same subtasks and boundaries as the frame labels.
 
 ## 🟧 3. Real-World Training <a name="real-world"></a>
 
@@ -368,6 +370,8 @@ evaluation/
 
 Each released training or public test subset directory is an independent **LeRobot** dataset root (`root`). Both test subsets have frame references, RGB videos, and English instructions, with evaluation labels alongside them; they have no robot state/action vectors. The primary view is the head camera for real-world and the front camera for simulation.
 
+The repository-level training directory is `training-data/`. Each LeRobot root retains its native `data/` frame-table directory; for example, Simulation One-Shot frame tables are stored under `training-data/simulation/one-shot/data/`.
+
 <details>
 <summary><strong>Training subset files and metadata</strong></summary>
 
@@ -422,7 +426,7 @@ from lerobot.datasets.lerobot_dataset import LeRobotDataset
 
 repo_id = "wzlllong/value-model-benchmark-training-data"
 prefix = "training-data/simulation/one-shot"
-revision = "MIGRATED_TRAINING_REVISION"
+revision = "main"
 local_dir = Path(os.environ.get("ROBOVALUE_LOCAL_DIR", "./robovalue-data"))
 
 snapshot = snapshot_download(
@@ -470,7 +474,7 @@ annotation_prefix = "training-data/simulation/one-shot/annotations/subtasks"
 annotation_snapshot = snapshot_download(
     repo_id=repo_id,
     repo_type="dataset",
-    revision="MIGRATED_TRAINING_REVISION",
+    revision="main",
     allow_patterns=[f"{annotation_prefix}/**"],
     local_dir=local_dir,
 )
@@ -525,7 +529,7 @@ shasum -a 256 -c checksums.sha256
 
 ### 5.5 Download and read a public test subset
 
-Choose `evaluation/real-world/test-30` or `evaluation/simulation/test-30` to download one public test subset. The example is pinned to `578593ef28ee7a2d861ebf38162c928291b0d4a9`, which contains both test domains; downloading and loading use the same revision. The original training-data revision shown above predates this test release.
+Choose `evaluation/real-world/test-30` or `evaluation/simulation/test-30` to download one public test subset. The example is pinned to `578593ef28ee7a2d861ebf38162c928291b0d4a9`, which contains both test domains; downloading and loading use the same revision. Training examples use the separately pinned revision containing the `training-data/` directory.
 
 ```python
 from pathlib import Path

README_zh.md

@@ -59,6 +59,8 @@ RoboValue 提供面向**机器人价值模型**的专家示例，覆盖 **15 个
 
 请查看[真机指标覆盖与逐轨迹索引](evaluation/real-world/test-30/README_zh.md#metric-coverage)和[仿真指标覆盖与逐轨迹索引](evaluation/simulation/test-30/README_zh.md#metric-coverage)。仿真 ED/MS 的终局标签和语义对齐来自用户确认；TRR/MEM 的终局标签保留源标注中“待独立验证”的证据状态，详情见仿真说明。
 
+**任务与子任务一致性：**[论文对照说明](TASK_ALIGNMENT_zh.md)给出8个公开测试任务与35个训练任务的对应关系，列明标注是否存在、定义一致项及措辞差异。仿真 One-Shot 补充标注的58个子任务定义均与提供的PDF版本一致。其中“依数字按按钮”的论文卡片存在确认顺序矛盾，对照说明记录该问题并保留已有标注。[机器可读映射](task_alignment.json)提供局部任务编号和论文页码。
+
 
 ## 🟦 2. 仿真训练数据 <a name="simulation"></a>
 
@@ -172,7 +174,7 @@ Full-Shot 与 One-Shot 中，每条 **episode** 都是一条完整的任务示
 
 [**浏览文件 →**](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/tree/main/training-data/simulation/one-shot) · [任务—轨迹—行范围映射](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/provenance/episode_mapping.json) · [下载与读取](#download-and-reading)
 
-**子任务标注：**[58 个论文子任务区间](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/annotations/subtasks/README.md)覆盖全部 **8,080 帧**，每帧对应一个子任务。标注用于这些训练示例，与评测集标签分开。 另提供 [`dataset_labeler` 兼容 JSON](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/tree/MIGRATED_TRAINING_REVISION/training-data/simulation/one-shot/annotations/subtasks/dataset_labeler)，与逐帧标签使用相同的子任务和边界。
+**子任务标注：**[58 个论文子任务区间](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/annotations/subtasks/README.md)覆盖全部 **8,080 帧**，每帧对应一个子任务。标注用于这些训练示例，与评测集标签分开。 另提供 [`dataset_labeler` 兼容 JSON](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/tree/main/training-data/simulation/one-shot/annotations/subtasks/dataset_labeler)，与逐帧标签使用相同的子任务和边界。
 
 ## 🟧 3. 真实世界训练数据 <a name="real-world"></a>
 
@@ -339,6 +341,8 @@ evaluation/
 
 已发布的训练及公开测试子集目录均为独立的 **LeRobot** 数据根目录（`root`）。两个测试子集保留逐帧引用、RGB视频和英文指令，评测标注另行存放，不含机器人状态／动作向量。真机主视角是头部，仿真主视角是正面。
 
+仓库根目录的训练数据位于 `training-data/`。每个 LeRobot 子集内部保留原生逐帧表目录 `data/`，例如仿真 One-Shot 的帧表位于 `training-data/simulation/one-shot/data/`。
+
 <details>
 <summary><strong>训练子集文件结构与元数据</strong></summary>
 
@@ -393,7 +397,7 @@ from lerobot.datasets.lerobot_dataset import LeRobotDataset
 
 repo_id = "wzlllong/value-model-benchmark-training-data"
 prefix = "training-data/simulation/one-shot"
-revision = "MIGRATED_TRAINING_REVISION"
+revision = "main"
 local_dir = Path(os.environ.get("ROBOVALUE_LOCAL_DIR", "./robovalue-data"))
 
 snapshot = snapshot_download(
@@ -441,7 +445,7 @@ annotation_prefix = "training-data/simulation/one-shot/annotations/subtasks"
 annotation_snapshot = snapshot_download(
     repo_id=repo_id,
     repo_type="dataset",
-    revision="MIGRATED_TRAINING_REVISION",
+    revision="main",
     allow_patterns=[f"{annotation_prefix}/**"],
     local_dir=local_dir,
 )
@@ -496,7 +500,7 @@ shasum -a 256 -c checksums.sha256
 
 ### 5.5 下载并读取公开测试子集
 
-选择 `evaluation/real-world/test-30` 或 `evaluation/simulation/test-30`，仅下载其中一个公开测试子集。下方示例固定到包含两个测试集的版本 `578593ef28ee7a2d861ebf38162c928291b0d4a9`，下载和加载使用同一提交。上方训练示例的旧版本早于本测试集发布。
+选择 `evaluation/real-world/test-30` 或 `evaluation/simulation/test-30`，仅下载其中一个公开测试子集。下方示例固定到包含两个测试集的版本 `578593ef28ee7a2d861ebf38162c928291b0d4a9`，下载和加载使用同一提交。训练示例另行固定到包含 `training-data/` 新目录的提交。
 
 ```python
 from pathlib import Path

evaluation/simulation/README.md

@@ -2,7 +2,7 @@
 
 [简体中文](README_zh.md) · [Repository overview](../../README.md)
 
-Public simulation diagnostic subsets are stored here, separately from training demonstrations under `data/simulation/`.
+Public simulation diagnostic subsets are stored here, separately from training demonstrations under `training-data/simulation/`.
 
 | Subset | Tasks | Trajectories | Time steps | Videos | Format |
 | --- | ---: | ---: | ---: | ---: | --- |

evaluation/simulation/README_zh.md

@@ -2,7 +2,7 @@
 
 [English](README.md) · [仓库总览](../../README_zh.md)
 
-本目录存放仿真公开诊断测试子集，与 `data/simulation/` 下的训练示范分开。
+本目录存放仿真公开诊断测试子集，与 `training-data/simulation/` 下的训练示范分开。
 
 | 子集 | 任务 | 轨迹 | 时间步 | 视频 | 格式 |
 | --- | ---: | ---: | ---: | ---: | --- |

subset_catalog.json

@@ -116,7 +116,9 @@
         "交换积木位置",
         "清扫积木"
       ],
-      "role": "training"
+      "role": "training",
+      "size_bytes": 45468403617,
+      "size_scope": "entire subset directory, including documentation and integrity files"
     },
     {
       "subset_id": "simulation-one-shot-data",
@@ -201,7 +203,7 @@
       "annotations": {
         "subtasks": {
           "path_in_repo": "training-data/simulation/one-shot/annotations/subtasks",
-          "revision": "MIGRATED_TRAINING_REVISION",
+          "revision": "main",
           "status": "user_confirmed_release",
           "target": "active_semantic_subtask",
           "scope": "training_demonstrations",
@@ -273,10 +275,13 @@
             "group": "one-shot",
             "gap": 20
           },
-          "source_revision": "7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145"
+          "source_revision": "7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145",
+          "directory_size_bytes": 158390
         }
       },
-      "role": "training"
+      "role": "training",
+      "size_bytes": 499726133,
+      "size_scope": "entire subset directory, including documentation and integrity files"
     },
     {
       "subset_id": "real-world-full-shot-data",
@@ -350,9 +355,10 @@
         "插花与摆放藤条香薰",
         "准备早餐"
       ],
-      "size_bytes": 23125415964,
+      "size_bytes": 23126335625,
       "duration_seconds": 51144.03333333333,
-      "role": "training"
+      "role": "training",
+      "size_scope": "entire subset directory, including documentation and integrity files"
     },
     {
       "subset_id": "real-world-one-shot-data",
@@ -382,7 +388,7 @@
       "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": 215642018,
+      "size_bytes": 215666070,
       "duration_seconds": 545.4,
       "display_task_names": [
         "Transfer Blocks Using Plate",
@@ -439,7 +445,8 @@
           "selection": "minimum original episode_id in the 27-episode tea refresh batch"
         }
       ],
-      "role": "training"
+      "role": "training",
+      "size_scope": "entire subset directory, including documentation and integrity files"
     },
     {
       "subset_id": "real-world-test-30",
@@ -523,7 +530,8 @@
       ],
       "primary_camera": "observation.images.top_head",
       "primary_video_column": "video_primary",
-      "data_revision": "578593ef28ee7a2d861ebf38162c928291b0d4a9"
+      "data_revision": "578593ef28ee7a2d861ebf38162c928291b0d4a9",
+      "size_scope": "entire subset directory, including documentation and integrity files"
     },
     {
       "subset_id": "simulation-test-30",
@@ -607,15 +615,17 @@
       "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": 594582001,
-      "data_revision": "578593ef28ee7a2d861ebf38162c928291b0d4a9"
+      "data_revision": "578593ef28ee7a2d861ebf38162c928291b0d4a9",
+      "size_scope": "entire subset directory, including documentation and integrity files"
     }
   ],
   "planned_subsets": [],
-  "data_revision": "MIGRATED_TRAINING_REVISION",
+  "data_revision": "main",
   "publication_authorization": "All data were authorized before public release.",
   "pretty_name": "RoboValue Training and Small-Scale Evaluation Data",
-  "training_data_revision": "MIGRATED_TRAINING_REVISION",
+  "training_data_revision": "main",
   "revision_note": "training_data_revision identifies the unchanged training payloads under training-data/. source_training_data_revision records the original release under data/. evaluation_data_revision identifies the separate public test payloads.",
   "evaluation_data_revision": "578593ef28ee7a2d861ebf38162c928291b0d4a9",
-  "source_training_data_revision": "7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145"
+  "source_training_data_revision": "7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145",
+  "training_path_revision": "main"
 }

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

@@ -19,9 +19,9 @@ Expert demonstrations collected with the AgiBot Genie02 robot.
 | RGB cameras | Head: 640 × 400; left/right wrist: 640 × 528 each |
 | Camera videos | 6,000 · HEVC |
 
-This directory is an independent LeRobot dataset root. Set `root` to `data/real-world/full-shot` inside the downloaded snapshot. The [download and training examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data#download-and-reading) use LeRobot **0.4.2** and PyAV **15.1.0** and return states, actions, task instructions and three camera observations as PyTorch batches. Value or progress supervision is constructed by the training method.
+This directory is an independent LeRobot dataset root. Set `root` to `training-data/real-world/full-shot` inside the downloaded snapshot. The [download and training examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data#download-and-reading) use LeRobot **0.4.2** and PyAV **15.1.0** and return states, actions, task instructions and three camera observations as PyTorch batches. Value or progress supervision is constructed by the training method.
 
-[Feature definitions](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/data/real-world/full-shot/meta/info.json) · [Shared trajectory contents and task descriptions](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data#real-world)
+[Feature definitions](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/real-world/full-shot/meta/info.json) · [Shared trajectory contents and task descriptions](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data#real-world)
 
 <details>
 <summary><strong>Per-task episode allocation</strong></summary>
@@ -55,7 +55,7 @@ Task indices match `meta/tasks.parquet`. Source episode identifiers and task cor
 
 **Publication authorization:** All data were authorized before public release.
 
-**Data revision:** `3bcf64ae6d2a39e0795f413b65bfd08a3a8c6f95`. Use this fixed revision with the download examples.
+**Data revision:** `main`. Use this fixed revision with the download examples.
 
 ## 中文
 
@@ -65,10 +65,10 @@ Task indices match `meta/tasks.parquet`. Source episode identifiers and task cor
 
 平台为 **AgiBot Genie02**，采用 **LeRobot Dataset v3.0**，采样率为 **30 FPS**。状态为 `float32[290]`，动作为 `float32[52]`；三路 RGB 相机为头部：640 × 400；左腕、右腕：均为 640 × 528，共 **6,000 个原编码视频（HEVC）**。
 
-下载后，将 LeRobot 的 `root` 设置为快照中的 `data/real-world/full-shot`。主卡片的[下载与训练示例](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/README_zh.md#download-and-reading)提供状态、动作、任务指令与三路相机观测的 PyTorch batch；value 或 progress 监督由训练方法构造。
+下载后，将 LeRobot 的 `root` 设置为快照中的 `training-data/real-world/full-shot`。主卡片的[下载与训练示例](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/README_zh.md#download-and-reading)提供状态、动作、任务指令与三路相机观测的 PyTorch batch；value 或 progress 监督由训练方法构造。
 
 折叠表列出任务索引、轨迹分配及数据行范围；原始来源标识与对应关系保留在 `provenance/`。
 
-**数据版本：**`3bcf64ae6d2a39e0795f413b65bfd08a3a8c6f95`，下载与读取时使用该固定版本。
+**数据版本：**`main`，下载与读取时使用该固定版本。
 
 **发布授权：**全部数据均经授权后公开发布。

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

@@ -1998,10 +1998,10 @@ bd535062dbe63ba098e1f52a744a48739dcf233de1bce12d1eaf9a1212d88f25  data/chunk-001
 ad6956e579b020509c9582bb2a15cbe1c7b39803789f77a2a51caae28a4150c9  data/chunk-001/file-997.parquet
 0ebcdbd1b72526cdb240f78ae757623847d6dd86d264932e7a8d26adb4644892  data/chunk-001/file-998.parquet
 994b33933358cfabdaf2def445762bb06e9e941e0d865188b2e6980168d901e9  data/chunk-001/file-999.parquet
-a1c5174b88db6a6b65fc5f42d6c4888555c9cb344ea44bc3f6cd1c65ace34295  meta/episodes/chunk-000/file-000.parquet
+c5c40c5f86263f440c34c7374a29211bfd8fbc64a7c946a7fad96c0459495c38  meta/episodes/chunk-000/file-000.parquet
 faaf8dbabeb5222b1fdffab5e3d6122ed52417e2435e40ef3ba9fe1d4526c2f7  meta/info.json
 ba82dd5d05ac9583522d84ebe0483659ae9d3b2abc6310c4df4d4fe2e7a97445  meta/stats.json
-fe80e4c17ad7b530ffc48b089fe461d1dbb11d77b5507d0c921832f3d337d855  meta/tasks.parquet
+2610f566973107a58bcf8dd2876f97fd49020d815ddba54262060f372bf3e023  meta/tasks.parquet
 f859e3c477b87aee893f5dc59818a44816d2b640096eb548c148d4af5a68f1bb  provenance/conversion.json
 46f0481b6b00eb8e512bedf9ab6c5b0836582a24cbfc31069663e456c3e1d20c  provenance/episode_mapping.json
 35bed3d2a6cfc18c8ce0787664ab79b361d8521a883b8ea40f22a8ea32e9d764  provenance/episodes.parquet

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

@@ -19,9 +19,9 @@ Expert demonstrations collected with the **AgiBot Genie02** robot. For 19 tasks,
 | RGB cameras | Head: 640 × 400; left/right wrist: 640 × 528 each |
 | Camera videos | 60 · HEVC |
 
-This directory is an independent LeRobot dataset root. Set `root` to `data/real-world/one-shot` inside the downloaded snapshot. The [download and training examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data#download-and-reading) use LeRobot **0.4.2** and PyAV **15.1.0** and return states, actions, task instructions and three camera observations as PyTorch batches. Value or progress supervision is constructed by the training method.
+This directory is an independent LeRobot dataset root. Set `root` to `training-data/real-world/one-shot` inside the downloaded snapshot. The [download and training examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data#download-and-reading) use LeRobot **0.4.2** and PyAV **15.1.0** and return states, actions, task instructions and three camera observations as PyTorch batches. Value or progress supervision is constructed by the training method.
 
-[Feature definitions](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/data/real-world/one-shot/meta/info.json) · [Shared trajectory contents and task descriptions](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data#real-world) · [Machine-readable source IDs and row ranges](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/data/real-world/one-shot/provenance/episode_mapping.json)
+[Feature definitions](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/real-world/one-shot/meta/info.json) · [Shared trajectory contents and task descriptions](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data#real-world) · [Machine-readable source IDs and row ranges](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/real-world/one-shot/provenance/episode_mapping.json)
 
 <details>
 <summary><strong>Original IDs and task–episode–row mapping</strong></summary>
@@ -55,7 +55,7 @@ Task indices match `meta/tasks.parquet`. Video streams retain their original byt
 
 **Publication authorization:** All data were authorized before public release.
 
-**Data revision:** `3bcf64ae6d2a39e0795f413b65bfd08a3a8c6f95`. Use this fixed revision with the download examples.
+**Data revision:** `main`. Use this fixed revision with the download examples.
 
 ## 中文
 
@@ -65,7 +65,7 @@ Task indices match `meta/tasks.parquet`. Video streams retain their original byt
 
 采用 **LeRobot Dataset v3.0**，采样率为 **30 FPS**，总时长 **545.4 秒（约 9.09 分钟）**。训练数据大小为 **215,642,018 字节（约 0.22 GB）**。状态为 `float32[290]`，动作为 `float32[52]`；三路 RGB 相机为头部：640 × 400；左腕、右腕：均为 640 × 528，共 **60 个原编码视频（HEVC）**。
 
-下载后，将 LeRobot 的 `root` 设置为快照中的 `data/real-world/one-shot`。主卡片的[下载与训练示例](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/README_zh.md#download-and-reading)提供状态、动作、任务指令与三路相机观测的 PyTorch batch；value 或 progress 监督由训练方法构造。
+下载后，将 LeRobot 的 `root` 设置为快照中的 `training-data/real-world/one-shot`。主卡片的[下载与训练示例](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/README_zh.md#download-and-reading)提供状态、动作、任务指令与三路相机观测的 PyTorch batch；value 或 progress 监督由训练方法构造。
 
 <details>
 <summary><strong>原始 ID 与任务—轨迹—行范围映射</strong></summary>
@@ -97,6 +97,6 @@ Task indices match `meta/tasks.parquet`. Video streams retain their original byt
 
 视频保留原始字节。训练数据大小包含 `data/`、`videos/`、`meta/` 和 `provenance/`，不含本 README 与 `checksums.sha256`；时间步与时长按每条轨迹统计一次。
 
-**数据版本：**`3bcf64ae6d2a39e0795f413b65bfd08a3a8c6f95`，下载与读取时使用该固定版本。
+**数据版本：**`main`，下载与读取时使用该固定版本。
 
 **发布授权：**全部数据均经授权后公开发布。

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

@@ -18,10 +18,10 @@ ee36dda4be5b37f6532051e5f5b16f1b66cec84ef41d020b1ecf6f6c51bea80c  data/chunk-000
 ea24340f139735b3d7be3157ada8d6df3c4f8161a1926494f8c46d32fa8e9020  data/chunk-000/file-017.parquet
 3921f5f64140ed586efbd8a22b6ada9ba0088280f7af2f2b24c00c29e83225c1  data/chunk-000/file-018.parquet
 a40d7888df49540b10ea13047fb0fe823d71b95ed70398a4a3adaec6717438a6  data/chunk-000/file-019.parquet
-051287b040a04d554dad935e66a4b6ff6560804e31374e29b033a495857ea3ed  meta/episodes/chunk-000/file-000.parquet
+ddd99c9cf92cd8d09b1cd05eb4eb8ddf81185d76408ffe7133c83df3afcffcf1  meta/episodes/chunk-000/file-000.parquet
 cb686e58fb6c22b6aff18709b64b6e9fc803b7df29343aca910cf817697dee75  meta/info.json
 ca9667639e593294ac2d294c20740d6434678bb6cd0fd87c764eaafe501b8c2f  meta/stats.json
-08716881ff9ca56a8904d32fb8b1e80ac0f8eb6081e825b747d657ae4ca539e5  meta/tasks.parquet
+2ab53e9fbea8cb4d77ab2189d4ee695bb39b4b28b59823706767266aad1be945  meta/tasks.parquet
 f2e92aa3eb2eccb945eb3aa774ace22f17056838120330d6b1ebeac1dbed8807  provenance/conversion.json
 12d5cda0d51272727cb8b6c5a8dc761e223e727a6f8be3e709ebe7920e12d13d  provenance/episode_mapping.json
 c33db864188d6e5f0e327a2a202cbf01ecbbd77c34fb7ade0d93e82f0c538613  provenance/episodes.parquet

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

@@ -17,9 +17,9 @@ Based on open-source [RoboDojo](https://huggingface.co/datasets/RoboDojo-Benchma
 | RGB cameras | High, left wrist, right wrist: 640 × 480 each |
 | Camera videos | 4,500 · AV1 / H.264 |
 
-This directory is an independent LeRobot dataset root. Set `root` to `data/simulation/full-shot` inside the downloaded snapshot. The [download and training examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data#download-and-reading) use LeRobot **0.4.2** and PyAV **15.1.0** and return states, actions, task instructions and three camera observations as PyTorch batches. Value or progress supervision is constructed by the training method.
+This directory is an independent LeRobot dataset root. Set `root` to `training-data/simulation/full-shot` inside the downloaded snapshot. The [download and training examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data#download-and-reading) use LeRobot **0.4.2** and PyAV **15.1.0** and return states, actions, task instructions and three camera observations as PyTorch batches. Value or progress supervision is constructed by the training method.
 
-[Feature definitions](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/data/simulation/full-shot/meta/info.json) · [Shared trajectory contents and task descriptions](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data#simulation)
+[Feature definitions](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/full-shot/meta/info.json) · [Shared trajectory contents and task descriptions](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data#simulation)
 
 <details>
 <summary><strong>Per-task episode allocation</strong></summary>
@@ -56,7 +56,7 @@ Task indices match `meta/tasks.parquet`. Source episode identifiers and task cor
 
 平台为 **ARX**，采用 **LeRobot Dataset v3.0**，采样率为 **25 FPS**。状态为 `float32[14]`，动作为 `float32[14]`；三路 RGB 相机为高位、左腕、右腕：均为 640 × 480，共 **4,500 个原编码视频（AV1 / H.264）**。
 
-下载后，将 LeRobot 的 `root` 设置为快照中的 `data/simulation/full-shot`。主卡片的[下载与训练示例](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/README_zh.md#download-and-reading)提供状态、动作、任务指令与三路相机观测的 PyTorch batch；value 或 progress 监督由训练方法构造。
+下载后，将 LeRobot 的 `root` 设置为快照中的 `training-data/simulation/full-shot`。主卡片的[下载与训练示例](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/README_zh.md#download-and-reading)提供状态、动作、任务指令与三路相机观测的 PyTorch batch；value 或 progress 监督由训练方法构造。
 
 折叠表列出任务索引、轨迹分配及数据行范围；原始来源标识与对应关系保留在 `provenance/`。
 

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

@@ -1,4 +1,4 @@
-c5642effb1b5be5b69bcb7ebaf8f03b3711f2215283991a72740491b104976aa  README.md
+f01d7d9e47e13dd3d65a7d6555f42292d9810b9c24a2eac1b36388ac3bc0d47d  README.md
 d0c983813f6b55338e4ae09508d96b92f312916c96db315e75e5a9e3c970dc63  data/chunk-000/file-000.parquet
 d04d0b7163f1b4be174fd56e3590b686a6f90e1981c68f94f52bdfc06a4a6574  data/chunk-000/file-001.parquet
 72853d5cd08dfe0e77dfe81a3c5103f656f52e3f5c62c16191e06d9c0358e757  data/chunk-000/file-002.parquet
@@ -1499,10 +1499,10 @@ dcd6f64c09074c6cb5cef8749415f9dc5f6851222ed53da67b1de8ab3e8aca28  data/chunk-001
 bf0b7a82161e3df3ff7e6f0b61bdfdb4b86f42665c6b7a0bf8b76ea74b561d6e  data/chunk-001/file-497.parquet
 9ab3bf4381def37bc7eb6f01ce2bbf930a2095ca076c9961f2d6ad3f31694a4c  data/chunk-001/file-498.parquet
 478c82310a7637ce150cda6c58ba3aea1c651b3eb3d96b5a9a67795d800ea376  data/chunk-001/file-499.parquet
-f80b172cf05d7b01b9dc30c94f39359d47649b46ed771d262a9fd591c7aaf7bc  meta/episodes/chunk-000/file-000.parquet
+634bd42f8fea19283ae4c0f3e201dd6d295cc2bd6de63446485d81b98acf409e  meta/episodes/chunk-000/file-000.parquet
 b9e9f19e26be620861ce39eb7b21392057b704c913638813435f5134bd7143ed  meta/info.json
 1f5321a8e7958ff87d6df3ea434370086da7b38674043ccffdc790bcb8266609  meta/stats.json
-263ddbdc7997ba49bd70baf1efdf1a9e4c3923081f654706eda937cb0330fd69  meta/tasks.parquet
+5928f46fdab0b4365502939078e0961b5ad23b131cd28871652e1aea0f11e22b  meta/tasks.parquet
 faa1639571bd1e998b7430ceb0986b0bcc38a02512d414055f32c67f8b05ab51  provenance/conversion.json
 80981b27909b83e858f3d8d778598cb2650f408122feee678a276c63f2ef160e  provenance/episode_mapping.json
 d243793fca5f1eaa457fb404349fba23b2ea695352622cc9bbfb6d53ae78e9ad  provenance/episodes.parquet

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

@@ -17,9 +17,9 @@ Based on open-source [RoboDojo](https://huggingface.co/datasets/RoboDojo-Benchma
 | RGB cameras | High, left wrist, right wrist: 640 × 480 each |
 | Camera videos | 45 · AV1 |
 
-This directory is an independent LeRobot dataset root. Set `root` to `data/simulation/one-shot` inside the downloaded snapshot. The [download and training examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data#download-and-reading) use LeRobot **0.4.2** and PyAV **15.1.0** and return states, actions, task instructions and three camera observations as PyTorch batches. Value or progress supervision is constructed by the training method.
+This directory is an independent LeRobot dataset root. Set `root` to `training-data/simulation/one-shot` inside the downloaded snapshot. The [download and training examples](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data#download-and-reading) use LeRobot **0.4.2** and PyAV **15.1.0** and return states, actions, task instructions and three camera observations as PyTorch batches. Value or progress supervision is constructed by the training method.
 
-[Feature definitions](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/data/simulation/one-shot/meta/info.json) · [Shared trajectory contents and task descriptions](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data#simulation)
+[Feature definitions](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/meta/info.json) · [Shared trajectory contents and task descriptions](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data#simulation)
 
 <details>
 <summary><strong>Task–episode–row mapping</strong></summary>
@@ -50,9 +50,9 @@ Task indices match `meta/tasks.parquet`. Source episode identifiers and task cor
 
 ## ☰ Subtask annotations
 
-[Supplementary subtask annotations](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/data/simulation/one-shot/annotations/subtasks/README.md) provide **58 paper-defined segments and 8,080 frame labels**, one active subtask per time step. Labels use `episode_index` and `frame_index`, with `global_index` matching the LeRobot `index`. Initial waiting is part of the first subtask, and the final subtask extends to the episode end. Segments use `[start_frame, end_frame)`; the included last frame is `end_frame - 1`.
+[Supplementary subtask annotations](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/annotations/subtasks/README.md) provide **58 paper-defined segments and 8,080 frame labels**, one active subtask per time step. Labels use `episode_index` and `frame_index`, with `global_index` matching the LeRobot `index`. Initial waiting is part of the first subtask, and the final subtask extends to the episode end. Segments use `[start_frame, end_frame)`; the included last frame is `end_frame - 1`.
 
-Read these files explicitly using PyArrow or `load_subtasks.py`; they are separate from the LeRobot samples and benchmark test-set labels. See the directory README for the fixed-revision download example and boundary adjustment mask. [`dataset_labeler`-compatible JSON](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/tree/7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145/data/simulation/one-shot/annotations/subtasks/dataset_labeler) uses the same subtasks and boundaries as the frame labels.
+Read these files explicitly using PyArrow or `load_subtasks.py`; they are separate from the LeRobot samples and benchmark test-set labels. See the directory README for the fixed-revision download example and boundary adjustment mask. [`dataset_labeler`-compatible JSON](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/tree/main/training-data/simulation/one-shot/annotations/subtasks/dataset_labeler) uses the same subtasks and boundaries as the frame labels.
 
 ## 中文
 
@@ -62,7 +62,7 @@ Read these files explicitly using PyArrow or `load_subtasks.py`; they are separa
 
 平台为 **ARX**，采用 **LeRobot Dataset v3.0**，采样率为 **25 FPS**。状态为 `float32[14]`，动作为 `float32[14]`；三路 RGB 相机为高位、左腕、右腕：均为 640 × 480，共 **45 个原编码视频（AV1）**。
 
-下载后，将 LeRobot 的 `root` 设置为快照中的 `data/simulation/one-shot`。主卡片的[下载与训练示例](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/README_zh.md#download-and-reading)提供状态、动作、任务指令与三路相机观测的 PyTorch batch；value 或 progress 监督由训练方法构造。
+下载后，将 LeRobot 的 `root` 设置为快照中的 `training-data/simulation/one-shot`。主卡片的[下载与训练示例](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/README_zh.md#download-and-reading)提供状态、动作、任务指令与三路相机���测的 PyTorch batch；value 或 progress 监督由训练方法构造。
 
 折叠表列出任务索引、轨迹分配及数据行范围；原始来源标识与对应关系保留在 `provenance/`。
 
@@ -70,7 +70,7 @@ Read these files explicitly using PyArrow or `load_subtasks.py`; they are separa
 
 ## ☰ 子任务标注
 
-[子任务标注](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/data/simulation/one-shot/annotations/subtasks/README.md)提供 **58 个论文子任务区间、8,080 条逐帧标签**，每个时间步一个当前子任务。使用 `episode_index` 与 `frame_index` 关联，并核对 `global_index` 与 LeRobot 的 `index` 一致。起始等待归入第一个子任务，最后一个子任务持续至轨迹末帧；区间为 `[start_frame, end_frame)`，包含式末帧为 `end_frame - 1`。
+[子任务标注](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/annotations/subtasks/README.md)提供 **58 个论文子任务区间、8,080 条逐帧标签**，每个时间步一个当前子任务。使用 `episode_index` 与 `frame_index` 关联，并核对 `global_index` 与 LeRobot 的 `index` 一致。起始等待归入第一个子任务，最后一个子任务持续至轨迹末帧；区间为 `[start_frame, end_frame)`，包含式末帧为 `end_frame - 1`。
 
-通过 PyArrow 或 `load_subtasks.py` 显式读取。标注独立于 LeRobot sample，与评测集标签分开；目录 README 提供固定版本的下载示例及边界调整标记说明。 另提供 [`dataset_labeler` 兼容 JSON](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/tree/7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145/data/simulation/one-shot/annotations/subtasks/dataset_labeler)，与逐帧标签使用相同的子任务和边界。
+通过 PyArrow 或 `load_subtasks.py` 显式读取。标注独立于 LeRobot sample，与评测集标签分开；目录 README 提供固定版本的下载示例及边界调整标记说明。 另提供 [`dataset_labeler` 兼容 JSON](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/tree/main/training-data/simulation/one-shot/annotations/subtasks/dataset_labeler)，与逐帧标签使用相同的子任务和边界。
 

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

@@ -24,14 +24,14 @@ The accepted boundaries include frame **210** in Store Laptop And Headphones S2,
 
 | File / 文件 | Contents / 内容 |
 | --- | --- |
-| [taxonomy.json](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145/data/simulation/one-shot/annotations/subtasks/taxonomy.json) | Paper subtask definitions and task IDs / 论文子任务定义与任务编号 |
-| [segments.json](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145/data/simulation/one-shot/annotations/subtasks/segments.json), [segments.parquet](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145/data/simulation/one-shot/annotations/subtasks/segments.parquet) | 58 labeled intervals / 58 个标注区间 |
-| [frame_labels.parquet](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145/data/simulation/one-shot/annotations/subtasks/frame_labels.parquet) | 8,080 labels with integer join keys, text, and boundary mask / 逐帧标签、整数关联键、文本与边界标记 |
-| [metadata.json](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145/data/simulation/one-shot/annotations/subtasks/metadata.json) | Dataset identity and annotation conventions / 数据身份与标注约定 |
-| [validation.json](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145/data/simulation/one-shot/annotations/subtasks/validation.json) | Coverage, consistency, and accepted-boundary checks / 覆盖、一致性与已确认边界检查 |
-| [load_subtasks.py](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145/data/simulation/one-shot/annotations/subtasks/load_subtasks.py) | Explicit loading, SHA256 validation, and sample lookup / 显式读取、SHA256 校验与帧查询 |
-| [dataset_labeler/](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/tree/7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145/data/simulation/one-shot/annotations/subtasks/dataset_labeler) | 15 task-level SIA+CSPC JSON files and a Transfer all boundary export / 15 个任务级兼容 JSON 与 Transfer all 边界点导出 |
-| [manifest.json](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145/data/simulation/one-shot/annotations/subtasks/manifest.json) | SHA256 and byte size of the 23 annotation payload files; the manifest is not self-hashed / 23 个标注文件的校验值与字节数，清单不校验自身 |
+| [taxonomy.json](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/annotations/subtasks/taxonomy.json) | Paper subtask definitions and task IDs / 论文子任务定义与任务编号 |
+| [segments.json](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/annotations/subtasks/segments.json), [segments.parquet](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/annotations/subtasks/segments.parquet) | 58 labeled intervals / 58 个标注区间 |
+| [frame_labels.parquet](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/annotations/subtasks/frame_labels.parquet) | 8,080 labels with integer join keys, text, and boundary mask / 逐帧标签、整数关联键、文本与边界标记 |
+| [metadata.json](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/annotations/subtasks/metadata.json) | Dataset identity and annotation conventions / 数据身份与标注约定 |
+| [validation.json](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/annotations/subtasks/validation.json) | Coverage, consistency, and accepted-boundary checks / 覆盖、一致性与已确认边界检查 |
+| [load_subtasks.py](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/annotations/subtasks/load_subtasks.py) | Explicit loading, SHA256 validation, and sample lookup / 显式读取、SHA256 校验与帧查询 |
+| [dataset_labeler/](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/tree/main/training-data/simulation/one-shot/annotations/subtasks/dataset_labeler) | 15 task-level SIA+CSPC JSON files and a Transfer all boundary export / 15 个任务级兼容 JSON 与 Transfer all 边界点导出 |
+| [manifest.json](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/annotations/subtasks/manifest.json) | SHA256 and byte size of the 23 annotation payload files; the manifest is not self-hashed / 23 个标注文件的校验值与字节数，清单不校验自身 |
 
 ## Download and read / 下载与读取
 
@@ -47,8 +47,8 @@ from pathlib import Path
 from huggingface_hub import snapshot_download
 
 repo_id = "wzlllong/value-model-benchmark-training-data"
-prefix = "data/simulation/one-shot"
-revision = "7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145"
+prefix = "training-data/simulation/one-shot"
+revision = "main"
 local_dir = Path(os.environ.get("ROBOVALUE_LOCAL_DIR", "./robovalue-data"))
 snapshot = snapshot_download(
     repo_id=repo_id,
@@ -74,9 +74,9 @@ With the matching LeRobot simulation One-Shot dataset loaded as `dataset`, use `
 
 ## dataset_labeler-compatible JSON / 兼容标注格式
 
-The [15 task-level JSON files](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/tree/7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145/data/simulation/one-shot/annotations/subtasks/dataset_labeler) retain the `dataset_labeler` SIA+CSPC record format. Each `<task_key>_sia_cspc_annotations.json` maps a **dataset-root-relative head-camera video path** to one annotation record. `task_key` is the canonical simulation task name, including `sweep_blocks`. Resolve a video key against the local simulation One-Shot dataset root when relocating the files.
+The [15 task-level JSON files](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/tree/main/training-data/simulation/one-shot/annotations/subtasks/dataset_labeler) retain the `dataset_labeler` SIA+CSPC record format. Each `<task_key>_sia_cspc_annotations.json` maps a **dataset-root-relative head-camera video path** to one annotation record. `task_key` is the canonical simulation task name, including `sweep_blocks`. Resolve a video key against the local simulation One-Shot dataset root when relocating the files.
 
-[15 个任务级 JSON 文件](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/tree/7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145/data/simulation/one-shot/annotations/subtasks/dataset_labeler)保持 `dataset_labeler` 的 SIA+CSPC 记录格式；每个 `<task_key>_sia_cspc_annotations.json` 以**相对仿真 One-Shot 数据根目录的头部相机视频路径**为键，对应一条轨迹记录。任务名使用发布数据的统一名称，包括 `sweep_blocks`；移动数据后，由读取者将相对路径与本地数据根目录拼接。
+[15 个任务级 JSON 文件](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/tree/main/training-data/simulation/one-shot/annotations/subtasks/dataset_labeler)保持 `dataset_labeler` 的 SIA+CSPC 记录格式；每个 `<task_key>_sia_cspc_annotations.json` 以**相对仿真 One-Shot 数据根目录的头部相机视频路径**为键，对应一条轨迹记录。任务名使用发布数据的统一名称，包括 `sweep_blocks`；移动数据后，由读取者将相对路径与本地数据根目录拼接。
 
 | Field / 字段 | Meaning / 含义 |
 | --- | --- |
@@ -92,9 +92,9 @@ The [15 task-level JSON files](https://huggingface.co/datasets/wzlllong/value-mo
 
 `marks` 包含第 **0 帧**、各区间的半开结束点，以及等于轨迹帧数的末端点。**N** 个子任务对应 **N + 1** 个边界点，相邻点按 S1…Sn 顺序还原 `[start_frame, end_frame)` 区间；末端点仅闭合最后一个子任务，不增加“任务已完成”标签。`time = frame / 25`。已确认的标签和边界均保持不变。
 
-[`exceptional_intervals.json`](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145/data/simulation/one-shot/annotations/subtasks/dataset_labeler/exceptional_intervals.json) follows the labeler's **Transfer all** output: a list of **15** `{"episode": video_path, "frames": boundary_points}` records containing **73 boundary points**. Those points close the same **58 intervals**; they are not additional subtask labels. The JSON export, `segments.json`, and `frame_labels.parquet` describe the same **58 segments / 8,080 labels**. This is a portable format export for these training demonstrations; the integer-keyed Parquet files and helper remain available for explicit training-label joins.
+[`exceptional_intervals.json`](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/annotations/subtasks/dataset_labeler/exceptional_intervals.json) follows the labeler's **Transfer all** output: a list of **15** `{"episode": video_path, "frames": boundary_points}` records containing **73 boundary points**. Those points close the same **58 intervals**; they are not additional subtask labels. The JSON export, `segments.json`, and `frame_labels.parquet` describe the same **58 segments / 8,080 labels**. This is a portable format export for these training demonstrations; the integer-keyed Parquet files and helper remain available for explicit training-label joins.
 
-[`exceptional_intervals.json`](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/7fd21cb2519b4d59cbe9fe8b23cfdf477e3ad145/data/simulation/one-shot/annotations/subtasks/dataset_labeler/exceptional_intervals.json)保持标注工具 **Transfer all** 的列表格式，每条记录为 `{"episode": 视频路径, "frames": 边界点列表}`。共 **15 条轨迹、73 个边界点**，对应相同的 **58 个子任务区间、8,080 条逐帧标签**；边界点不增加标签类别。用于训练时仍可通过整数索引和原读取工具显式关联标签。
+[`exceptional_intervals.json`](https://huggingface.co/datasets/wzlllong/value-model-benchmark-training-data/blob/main/training-data/simulation/one-shot/annotations/subtasks/dataset_labeler/exceptional_intervals.json)保持标注工具 **Transfer all** 的列表格式，每条记录为 `{"episode": 视频路径, "frames": 边界点列表}`。共 **15 条轨迹、73 个边界点**，对应相同的 **58 个子任务区间、8,080 条逐帧标签**；边界点不增加标签类别。用于训练时仍可通过整数索引和原读取工具显式关联标签。
 
 After the annotation-only download example above, read the compatibility JSON locally without downloading videos:
 

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

@@ -1,4 +1,4 @@
-f79139a2d07106cdf272ab559bcfe1969b59befb7073694db0c0a99acf44866b  README.md
+584d9ab1a7e134a3a298c44bd5be581c1057330d1caf93c86b3f7d167d3137ef  README.md
 05c2937fa1434ae11f8293f721ed0157db5e378e9d91fd9e5b148d233a668de4  data/chunk-000/file-000.parquet
 14128a288e41913e838119f34fdbf754b472a06ec358a5c0053ed92b74741e82  data/chunk-000/file-001.parquet
 2c1209fd07a3953484c963490bc07f7713abbe44395bdc74e2e56687abbbcd51  data/chunk-000/file-002.parquet
@@ -14,10 +14,10 @@ b5f4366f2886d6ac430fc3a36fbc69c574e94a3d3633ff72790d3316ce19df19  data/chunk-000
 b3ccd903367faffd1c051e75f35880c264cb9edac16862e0f75090eda5cc77d6  data/chunk-000/file-012.parquet
 fa8a530cbd10b9ab4aa67f342f0eeda256edcd8d6bac5e8c55720e745a05a0bb  data/chunk-000/file-013.parquet
 230bff412e0c5c09b329fed79c8106e5700eadd1bbda9f0771cf0ff6781170a6  data/chunk-000/file-014.parquet
-920f7e897cd61c2132461618a089ce58c996ce5b1273a24ab6127c660be5f6c6  meta/episodes/chunk-000/file-000.parquet
+dc844b01dd43ad0e5694bb40eb5ae35f66ab264cb5ea9a499f91ca60266e1785  meta/episodes/chunk-000/file-000.parquet
 6324cc597a1a5af00020e7184cca3d288e8393f37d843bc6e09f9ff5d6476d25  meta/info.json
 8006c1e93cb28f5fc59bc8f6606e9becfb8d6b2d6c683d56d218a698f4bca534  meta/stats.json
-263ddbdc7997ba49bd70baf1efdf1a9e4c3923081f654706eda937cb0330fd69  meta/tasks.parquet
+5928f46fdab0b4365502939078e0961b5ad23b131cd28871652e1aea0f11e22b  meta/tasks.parquet
 5c828e227beeb8d90a39ccfe21e25f8bfa0e2237573ffdf9128fba07e36da96f  provenance/conversion.json
 9096a572a9de67f482e11c46b5352147d922db3c56fb1295dac4c8166a2ebba3  provenance/episode_mapping.json
 f3d6e344d7917c86d944522368ee4f42a1e69a8bba8ecce511639ab6f2e92b6c  videos/observation.images.cam_high/chunk-000/file-000.mp4