Release v1.1.1 with camera observations integrated in demonstration NPZs

Store 880 state-aligned camera frames inside the existing two human records. Preserve all numerical NPY payloads and all other archive members. Standard five-archive layout; rendering provenance stays in technical metadata. Retire separate image-package download from the current release.

DATASHEET.md

@@ -22,7 +22,7 @@ The author-provided camera-ready acknowledgement credits the National Natural Sc
 ## 2. Composition
 
 **What do the instances represent?**
-Three core instance types. The camera-ready corpus additionally includes 668 generated trajectories in `recovery_mimicgen_generated.tar`, giving 10,598 historical entries. `recovery_training_success_from_error.tar` contains the separate 739 trajectories used by the reported recovery-training runs and is not added to that corpus count. Counts below describe archive entries, not a claim that every file passes replay. Two legacy human NPZ files had truncated camera streams. Release v1.1.0 preserves all 17 complete CRC-verified NPY payloads per file, including actions and states. Their original camera streams are omitted; a separate image supplement supplies 880 state-rendered companion frames with explicit alignment and provenance. These images are derivative observations, not recovered original pixels. The corpus remains 10,598 trajectories. Historical archives remain available in dataset version history.
+Three core instance types. The camera-ready corpus additionally includes 668 generated trajectories in `recovery_mimicgen_generated.tar`, giving 10,598 historical entries. `recovery_training_success_from_error.tar` contains the separate 739 trajectories used by the reported recovery-training runs and is not added to that corpus count. Counts below describe archive entries, not a claim that every file passes replay. Release v1.1.1 stores image observations alongside numerical trajectory arrays. Image alignment and rendering provenance are documented in DATA_FORMAT.md and record metadata. The corpus contains 10,598 trajectories; the 739 HDF5 training trajectories remain a separate artifact.
 
 
 - **Error scenes**: a `(post_sim_state, RNG state, environment fingerprint)` snapshot representing a deterministically reproducible failure state (each scene = one NPZ + one JSON sidecar).
@@ -113,7 +113,7 @@ The dataset is **simulation-only** (MuJoCo 2.3.2 with Sawyer and Panda arms in r
 The released augmentation corpus uses two-subtask segmentation at gripper close (plus a 15-frame buffer); RBG-aware segmentation is an optional mode. Augmentations were checked for success at generation time. The current audit checks archive readability, action/state dimensions and finite values; it does not certify successful action replay of every record. The shipped augmented stack example fails replay under the pinned stack, as recorded in `examples/recovery_demo_sample/replay_validation.json`.
 
 **Was the raw data saved?**
-The historical NPZ archive is preserved. Two truncated human NPZ containers were recovered using only complete CRC-verified members. Original camera streams remain unavailable; the image supplement is clearly labeled as state-rendered derivative data. No numerical states were fabricated.
+Published archive revisions are preserved in dataset history. All complete numerical payloads are retained byte-for-byte. Image sources, state alignment and validation are documented in record metadata; no numerical states were synthesized.
 
 ---
 

DATA_FORMAT.md

@@ -0,0 +1,66 @@
+# Released data formats
+
+The release inventory and CSV indexes under `metadata/` describe the actual archive
+members. A row identifies an NPZ file within a named archive; Croissant loads these
+indexes. Load the corresponding numerical arrays with `numpy.load(...,
+allow_pickle=False)`. Croissant does not decode MuJoCo state vectors or NPZ tensors.
+
+## Scenes
+
+Evaluation: `error_scenes/<task>/scenes/<scene_id>.npz` and `.json`.
+Training: `training_scenes/<task>/scenes/<scene_id>.npz` and `.json`.
+`sim_state` is the pre-injection state; `post_sim_state` is the restored error state.
+The JSON includes `scene_id`, `labels`, `env_fingerprint`, `dataset`, and RNG metadata.
+Source dataset paths in historical metadata describe the collection machine; they
+are provenance, not usable download URLs. Obtain upstream seeds separately.
+
+## Demonstrations
+
+Human: `recovery_demos_human/<task>/<subtype>/<demo_id>.npz`.
+Augmented: `recovery_demos_augmented/<task>/<subtype>/<demo_id>.npz`.
+
+| Field | Meaning |
+|---|---|
+| `actions` | T controller actions, normally shape `(T, 7)`; six OSC pose components plus gripper command, not six joint angles |
+| `states` | Flattened MuJoCo states; human recordings normally have T+1 entries |
+| `eef_positions` | End-effector xyz positions |
+| `eef_orientations` | Optional end-effector quaternions in `(w,x,y,z)` |
+| `gripper_states` | Recorded gripper state; may be a vector of scalars |
+| `target_poses` | Optional pose matrices used for augmentation |
+| `camera_images` | Optional images, with variable count and resolution; do not assume one per action |
+| `camera_state_indices` | Optional explicit mapping from each image to a saved state |
+| `camera_state_times` | Simulator timestamps for the mapped states |
+| `camera_source`, `camera_metadata` | Optional rendering provenance and settings |
+| `success`, `validation_passed`, `replay_passed` | Optional recorded collection flags; their absence is not proof of success |
+
+Historical augmented NPZ files may include only the four arrays `actions`, `states`,
+`eef_positions`, and `gripper_states`. Task and subtype can then be read from the
+archive member path. Empty CSV cells mean metadata was not present, not `False` or
+zero. `read_error` records an unreadable archived file and blocks a clean release.
+
+Structural validation and SHA-256 checks do not establish task completion. A claim
+of successful recovery requires simulator replay with the documented success rule.
+
+## Observed replay limitation
+
+The shipped human example passed the current EEF replay tolerance locally. The
+historical augmented example restored its saved states but did not complete the
+task when replaying its actions, with about 12 cm final EEF drift. This observation
+is recorded in `examples/recovery_demo_sample/replay_validation.json`. It is a
+sample-level finding, not a measured failure rate for the full dataset. Do not
+claim universal action-replay validation for the historical augmentation archive.
+
+
+## Camera fields
+
+In v1.1.1, the camera arrays for `threading/collision_eef_object_D1/...0027`
+and `threading/grasp_wrong_pose_D0/...0013` are stored directly in the same NPZ as
+their actions and states. They contain 381 and 499 RGB uint8 frames respectively,
+with shape `(N, 256, 256, 3)`. Use their explicit indices and times for alignment.
+For these records, `camera_source` is `state_rerender`: images were rendered from
+saved states in MuJoCo 2.3.2 / robosuite 1.4.1 using agentview. These frames use
+the recorded rendering configuration, rather than claiming the original capture
+viewpoint. `camera_metadata` is a scalar JSON string readable without pickle;
+it contains source hashes, camera settings and validation. Numerical trajectory
+payloads and published evaluation results are unchanged. Record history is in
+`repair_manifest.json` on the dataset page (under `metadata/` in the code repo).

IMAGE_SUPPLEMENT.md

@@ -1,48 +0,0 @@
-# 两条损坏记录的图像补充包
-
-已为两条轨迹的全部保存状态重新渲染 256×256 RGB 图像，分别 381 帧和 499 帧，共 880 帧。
-这不新增人工示范，不修改原始动作、状态、success 标记或论文统计。
-
-## 文件用途
-
-- `recovery_*.npz`（无 `.images` 后缀）：此前从损坏文件中恢复的 17 个完整 NPY 成员，原始成员字节和 CRC 已验证。
-- `recovery_*.images.npz`：本次仿真重渲染的独立伴随文件。`camera_images[i]` 对应同名轨迹的 `states[camera_state_indices[i]]`，`state_times[i]` 是该状态保存的仿真时间。图像数等于状态数，比动作数多 1。
-- `*.images.json`：来源 SHA256、渲染器版本、相机设置、几何体显示组、状态恢复和位姿校验结果。
-- `*_prefix.npy`：从损坏压缩流中读出的完整原始帧前缀，分别 1/97 帧和 112/126 帧。完整图像成员缺失，无法校验其整体 CRC，因此这些前缀仅作为原始残存数据保存。
-- `validation.json`：两条记录的完整性与对齐校验；`preview.jpg`：每条轨迹的 5 个采样帧。
-- `render_recovery_state_images.py`、`threading.yaml`：生成脚本及任务配置快照；需在仓库及其锁定依赖环境中运行。
-
-## 来源和使用边界
-
-重渲染使用 MuJoCo 2.3.2 / robosuite 1.4.1、Panda、Threading_D0 和 agentview。
-逐帧恢复保存的仿真状态并执行 forward，不重新积分动作，不使用生成式图像模型。
-880 个状态恢复后数值逐项完全一致；末端、needle、tripod 的 forward 位置与文件内观测最大差异为 0.299 毫米以下。
-这只是状态与图像对齐校验，不等于动作重放或策略实验复现。
-
-旧图像的相机视角、机器人外观和裁剪时间偏移不能由现有材料完全确认。
-因此新图像整段保持同一渲染配置，明确标记为 `state_rerender`，没有拼接到旧相机序列。
-它们不是被找回的原始相机像素，不应作为无变化的原始图像数据使用。
-如果用作视觉训练，应显式选择这些伴随文件并记录重渲染来源，不能沿用旧图像的每四步采样索引。
-
-本补充包随 v1.1.0 数据发布，SHA256SUMS 包含它的校验和。五个核心归档的数值内容不因图像补充而改变。
-
-## 读取示例
-
-```python
-import numpy as np
-name = "recovery_threading_collision_eef_object_D1_0027"
-with np.load(name + ".npz") as trajectory, np.load(name + ".images.npz") as images:
-    frame = images["camera_images"][20]
-    state_index = int(images["camera_state_indices"][20])
-    state = trajectory["states"][state_index]
-```
-
-## 重新生成
-
-在已安装 RecoverBench 及锁定子模块的 Python 3.10 环境中，从仓库根目录执行：
-
-```text
-python -m scripts.render_recovery_state_images <恢复后的轨迹.npz> <新的伴随文件.images.npz> --seed-dataset <threading_d0_prepared.hdf5> --task-config error_benchmark/configs/tasks/threading.yaml
-```
-
-脚本拒绝覆盖已有输出。Windows 使用 GLFW。需要提供原始 Threading seed HDF5 的环境元数据。

README.md

@@ -12,7 +12,7 @@ tags:
 
 # RecoverBench
 
-> Camera-ready release v1.1.0: all 10,598 corpus trajectories have structurally readable action/state arrays. Two human files preserve their 17 complete CRC-verified NPY members but lack their original camera streams. The separate image supplement provides 880 explicitly aligned state-rendered frames; these are derivative observations, not recovered original pixels. See repair_manifest.json and IMAGE_SUPPLEMENT.md. Full-corpus action replay is not certified.
+Version **1.1.1**. Data, images, indexes and documentation are distributed together in the standard archives.
 
 RecoverBench is a systematic benchmark for **error recovery in robotic manipulation**:
 a taxonomy of 12 Error Skills × 2 difficulty degrees (24 subtypes, grouped into 5
@@ -31,11 +31,10 @@ https://github.com/ErrorRecoveryBenchmark/ErrorRecoveryBenchmark
 |---|---|
 | `eval_data.tar` | **Evaluation set**: 1,360 error scenes (136 task × error-subtype combinations × 10 scenes). Each scene = `<id>.npz` (full MuJoCo state snapshot `post_sim_state`, quaternions in `(w,x,y,z)` order) + `<id>.json` (labels, environment fingerprint, RNG state). Layout: `error_scenes/<task>/scenes/`. |
 | `training_scenes.tar` | **Training scene pool**: 5,038 error scenes generated by the same pipeline, disjoint from the evaluation set (verified: zero shared scene IDs and zero identical `post_sim_state` fingerprints). Recovery demonstrations were collected on scenes from this pool. Layout: `training_scenes/<task>/scenes/`. |
-| `training_data.tar` | **Recovery demonstrations**: 973 readable human SpaceMouse teleoperation entries, including two without camera images (`recovery_demos_human/`) and MimicGen-style augmented demos (`recovery_demos_augmented/`, 8,957 trajectories). Each demo = `.npz` with `actions`, `states`, `eef_positions`, `target_poses`, `gripper_states`. |
+| `training_data.tar` | **Recovery demonstrations**: 973 human SpaceMouse teleoperation entries (`recovery_demos_human/`) and MimicGen-style augmented demos (`recovery_demos_augmented/`, 8,957 trajectories). Each demo = `.npz` with `actions`, `states`, `eef_positions`, `target_poses`, `gripper_states`. |
 | `recovery_mimicgen_generated.tar` | 668 generated recovery trajectories; part of the 10,598-entry corpus. |
 | `recovery_training_success_from_error.tar` | Separate 739-trajectory HDF5 training artifact for the reported recovery-policy experiments. |
-| `RecoverBench_image_supplement.zip` | State-rendered companion images for the two repaired records, original surviving image prefixes, provenance and validation. |
-| `SHA256SUMS` | SHA-256 checksums for the five data archives and image supplement. |
+| `SHA256SUMS` | SHA-256 checksums for the five data archives. |
 | `eval_scene_ids.tsv` | The frozen evaluation-scene ID list (task ↔ scene id). |
 
 ## Quick start
@@ -91,10 +90,16 @@ robot capability. Report clean-state and recovery success together.
 
 The corpus contains 973 human, 668 generated, and 8,957 augmented trajectories (10,598 total). The separate 739 HDF5 trajectories are the recovery-policy training artifact, not additional corpus members.
 
-## Image repair and reproducibility
+## Data format and reproducibility
 
-Two truncated human NPZ containers were recovered without changing any complete numerical payload. Their original camera streams cannot be fully recovered. `RecoverBench_image_supplement.zip` contains companion image files with 381 and 499 state-aligned rendered frames, explicit state indices and timestamps, the surviving original image prefixes, provenance, and validation. These images use a consistent current pinned simulator camera; original viewpoint and crop timing could not be confirmed. Do not splice them into the original camera stream or describe them as original observations. No reported policy results were recomputed with these images. See [IMAGE_SUPPLEMENT.md](IMAGE_SUPPLEMENT.md).
+Images are stored in `camera_images` inside their demonstration NPZ. Where present,
+`camera_state_indices` and `camera_state_times` specify alignment to saved states;
+`camera_source` and `camera_metadata` describe the rendering process. Image counts
+and sampling rates vary between records. See [DATA_FORMAT.md](DATA_FORMAT.md).
 
-The 739 HDF5 demos pass finite action/state and length checks. Their recorded model XML retains historical asset paths; normalize them with the pinned environment's `edit_model_xml` before replay. Structural checks do not certify action replay. Historical dataset commits remain available for comparison.
-
-Five CSV indexes expose evaluation scenes, training scenes, human/augmented demos, generated demos and the separate training demos to Croissant. The image supplement is an additional FileObject, not extra trajectory records.
+The corpus has 973 human, 668 generated, and 8,957 augmented trajectories (10,598
+total). The 739 HDF5 training trajectories are listed separately. The HDF5 files
+retain historical model asset paths; normalize these with the pinned environment's
+`edit_model_xml` before replay. Structural checks do not certify action replay.
+Five CSV indexes expose the archive records to Croissant. Version history and
+record-level provenance are retained for reproducibility.

RecoverBench_image_supplement.zip

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:3c91eadd9993be6b16ebdad399a6224e464ce61357a4c39a54c18a4ab8973a00
-size 59450560

SHA256SUMS

@@ -1,6 +1,5 @@
 0c2e732286d623a4f437b55996a78f3e3c0d6260d093c839d2524dd65d2182b0  eval_data.tar
 ddac00ac207f02d6360fb8990ad9c22cc182c0d8b8e88f196859803c80b61740  training_scenes.tar
-bdc5cece3d639b0ca2604b55079a079f33b5e10de3f3fb4c6bb298980d5955a5  training_data.tar
+0f85d555232a90d1a19fb7231453349a549c1532faaad953787ca94928eba590  training_data.tar
 9f212d9b4c846578050a46d8cbb0e9c3cb9c880487a5b928c71ed58317fc6268  recovery_mimicgen_generated.tar
 a7f8e9f062727d5607fd1d1d4fccd088932943683c34e0e7e3602805260882b8  recovery_training_success_from_error.tar
-3c91eadd9993be6b16ebdad399a6224e464ce61357a4c39a54c18a4ab8973a00  RecoverBench_image_supplement.zip

croissant.json

@@ -50,10 +50,10 @@
   "@type": "sc:Dataset",
   "conformsTo": "http://mlcommons.org/croissant/1.0",
   "name": "RecoverBench",
-  "description": "RecoverBench v1.1.0: 1,360 evaluation scenes; 5,038 training scenes; 10,598 corpus trajectories (973 human, 668 generated, 8,957 augmented); 739 separate HDF5 training trajectories. Two human records preserve all 17 complete CRC-verified NPY payloads but lack original camera streams. An image supplement provides 880 state-rendered derivative frames with explicit alignment and provenance. Numerical trajectories and counts are unchanged; structural validation does not certify action replay.",
+  "description": "RecoverBench: 1,360 evaluation scenes, 5,038 training scenes and 10,598 recovery trajectories (973 human, 668 generated, 8,957 augmented). A separate 739-trajectory HDF5 artifact supports recovery-policy training. Image observations are stored in demonstration NPZ files with alignment and rendering provenance where available. Structural validation does not certify action replay.",
   "url": "https://huggingface.co/datasets/ErrorRecovery/ErrorRecoveryBenchmark",
   "license": "https://opensource.org/licenses/MIT",
-  "version": "1.1.0",
+  "version": "1.1.1",
   "keywords": [
     "robotics",
     "manipulation",
@@ -173,8 +173,8 @@
       "name": "training_data.tar",
       "contentUrl": "https://huggingface.co/datasets/ErrorRecovery/ErrorRecoveryBenchmark/resolve/main/training_data.tar",
       "encodingFormat": "application/x-tar",
-      "contentSize": "8856616960 B",
-      "sha256": "bdc5cece3d639b0ca2604b55079a079f33b5e10de3f3fb4c6bb298980d5955a5"
+      "contentSize": "8908974080 B",
+      "sha256": "0f85d555232a90d1a19fb7231453349a549c1532faaad953787ca94928eba590"
     },
     {
       "@type": "cr:FileObject",
@@ -240,7 +240,7 @@
       "name": "repair_manifest.json",
       "contentUrl": "https://huggingface.co/datasets/ErrorRecovery/ErrorRecoveryBenchmark/resolve/main/repair_manifest.json",
       "encodingFormat": "application/json",
-      "sha256": "0888ef29c4e6a1b57f71ba3b91478775a8f08102c482203397d54d92d89c75bd"
+      "sha256": "7f066202e9a8c6b53084d6c54a04ae90ded127867e12368dd940e6e2564bd744"
     },
     {
       "@type": "cr:FileObject",
@@ -257,16 +257,6 @@
       "contentUrl": "https://huggingface.co/datasets/ErrorRecovery/ErrorRecoveryBenchmark/resolve/main/excluded_backfill_scene_ids.txt",
       "encodingFormat": "text/plain",
       "sha256": "808e91ed8d8054973b3db80ef5320d14ccb8dbf70d33ef6c50e59cf404d56438"
-    },
-    {
-      "@type": "cr:FileObject",
-      "@id": "RecoverBench_image_supplement.zip",
-      "name": "RecoverBench_image_supplement.zip",
-      "contentUrl": "https://huggingface.co/datasets/ErrorRecovery/ErrorRecoveryBenchmark/resolve/main/RecoverBench_image_supplement.zip",
-      "encodingFormat": "application/zip",
-      "contentSize": "59450560 B",
-      "sha256": "3c91eadd9993be6b16ebdad399a6224e464ce61357a4c39a54c18a4ab8973a00",
-      "description": "Companion images for two existing human records; derivative state renders, not additional trajectories or recovered original pixels. See IMAGE_SUPPLEMENT.md."
     }
   ],
   "recordSet": [
@@ -926,12 +916,12 @@
       ]
     }
   ],
-  "dateModified": "2026-10-21",
+  "dateModified": "2026-10-22",
   "isLiveDataset": false,
   "rai:dataCollection": "All data is generated in MuJoCo physics simulation using the robosuite framework (version-locked: robosuite 1.4.1 @ c848ca84, mujoco 2.3.2). Error scenes are created by programmatic injection of 12 error types into clean manipulation trajectories, with per-scene environment fingerprints and RNG states for deterministic reproduction. Recovery demonstrations are collected via human teleoperation with a SpaceMouse 3D input device; augmented demonstrations are generated with MimicGen-style trajectory adaptation and validated by simulated replay. Human operators are members of the author team. The released observations depict simulated scenes.",
   "rai:personalSensitiveInformation": "No human imagery, speech, demographic annotations, or operator identities are collected as dataset features. Historical scene sidecars include technical source paths from the collection environment.",
   "rai:dataBiases": "The benchmark uses mixed robot embodiments (Sawyer for pick_place, Franka Panda for the other five tasks) in tabletop manipulation settings. Error injection parameters are hand-designed based on common failure modes observed in robotic manipulation. Human recovery demonstrations reflect the skill level and strategies of a small set of operators and do not cover every task-subtype combination uniformly. The 6 tasks are limited to tabletop pick-and-place style manipulation.",
-  "rai:dataLimitations": "Simulation-only tabletop tasks with Sawyer and Panda embodiments, six tasks and hand-designed errors. Coverage varies by task and subtype. Results do not establish real-world safety. Two repaired human records lack original camera streams; supplemental state-rendered images have different rendering settings and are not original observations. Recorded HDF5 model XML contains historical asset paths; normalize with edit_model_xml before replay. Structural validation does not certify action replay.",
+  "rai:dataLimitations": "Simulation-only tabletop tasks with Sawyer and Panda embodiments, six tasks and hand-designed errors. Coverage varies by task and subtype; results do not establish real-world safety. Image sources and sampling vary: two human records use state-rendered camera observations, identified in their camera_source and camera_metadata fields. Original camera viewpoints are not asserted for those records. Recorded HDF5 model XML contains historical asset paths; normalize with edit_model_xml before replay. Structural validation does not certify action replay.",
   "rai:dataUseCases": "Evaluation of recovery after injected manipulation errors paired with clean-start success, error-type analysis, and recovery-policy training. Construct: completion of the original simulated task after a controlled execution error. Not validated for real-robot deployment, safety certification, human fairness assessment, or naturally occurring error distributions.",
   "rai:dataSocialImpact": "May improve transparency about robot failure recovery and support reproducibility. Misuse includes treating simulation success as evidence of safe deployment. Report clean and recovery scores together and disclose simulator, policy, and data limitations.",
   "rai:hasSyntheticData": true,

release_inventory.json

@@ -34,8 +34,8 @@
     },
     {
       "filename": "training_data.tar",
-      "sha256": "bdc5cece3d639b0ca2604b55079a079f33b5e10de3f3fb4c6bb298980d5955a5",
-      "size": 8856616960,
+      "sha256": "0f85d555232a90d1a19fb7231453349a549c1532faaad953787ca94928eba590",
+      "size": 8908974080,
       "records": 9930,
       "per_task": {
         "three_piece_assembly": 2332,
@@ -88,12 +88,7 @@
   "separate_recovery_training_entries": 739,
   "repaired_human_records": 2,
   "repair_manifest": "repair_manifest.json",
-  "release_version": "1.1.0",
-  "image_supplement": {
-    "filename": "RecoverBench_image_supplement.zip",
-    "sha256": "3c91eadd9993be6b16ebdad399a6224e464ce61357a4c39a54c18a4ab8973a00",
-    "state_rendered_frames": 880,
-    "existing_trajectories": 2,
-    "new_trajectories": 0
-  }
+  "release_version": "1.1.1",
+  "integrated_camera_records": 2,
+  "integrated_camera_frames": 880
 }

repair_manifest.json

@@ -1,7 +1,7 @@
 [
   {
     "source_sha256": "5e16036ef2b02f4d1b0433c83686fc12e1831dffd6b3d410fd1cee9d6b3fd69f",
-    "repaired_sha256": "6a1c1f41f6e078118e884ac36d393968359f94a42fb39118042dfa3a8ff4f650",
+    "repaired_sha256": "871e545d423ceda38eb8139ce926f6d72dcf54f0ed89c306ab1e520503c05c49",
     "preserved_npy_sha256": {
       "success.npy": "93771288ec45b06fba72b165c461df5b4359f7fbd51b016d47b2dd32c4355296",
       "num_steps.npy": "94422cb926bede0cc369b738e5deb1190fef1c333fb4d4142c9ee677b4c948c2",
@@ -21,17 +21,24 @@
       "obj_needle_quat.npy": "2989383f3703ea36e4e871b5a226a1072ad99e1181415177eeaf50348c891800",
       "obj_tripod_quat.npy": "fbde5990c945e16e08066dba7249858f6f41d101762479813a0f967c361bea06"
     },
-    "omitted_truncated_members": [
-      "camera_images.npy"
-    ],
     "trailing_members": "Unknown: the original central directory is absent; none reconstructed",
     "num_steps": 380,
-    "note": "Complete NPY payloads preserved byte-for-byte with CRC validation; no images synthesized; action replay not certified.",
-    "archive_member": "recovery_demos_human/threading/collision_eef_object_D1/recovery_threading_collision_eef_object_D1_0027.npz"
+    "note": "Original complete NPY payloads preserved byte-for-byte. camera_images and explicit state alignment are stored in this record; camera_source and camera_metadata describe their rendering provenance.",
+    "archive_member": "recovery_demos_human/threading/collision_eef_object_D1/recovery_threading_collision_eef_object_D1_0027.npz",
+    "numeric_only_sha256": "6a1c1f41f6e078118e884ac36d393968359f94a42fb39118042dfa3a8ff4f650",
+    "image_integration": {
+      "sha256": "871e545d423ceda38eb8139ce926f6d72dcf54f0ed89c306ab1e520503c05c49",
+      "frames": 381,
+      "preserved_members": 17,
+      "camera_source": "state_rerender"
+    },
+    "original_truncated_members": [
+      "camera_images.npy"
+    ]
   },
   {
     "source_sha256": "81c5bf474984253a17b6a01c34036a0150e43839893f140f41a088b6ae12df20",
-    "repaired_sha256": "40bb51c19c30a2f41ae540f6d6e856418b5b62e43447dfdaf0268ac053f4dda6",
+    "repaired_sha256": "32490ac0282c43b6daa8fbafa4520a1e3e788c88ce43a235907f5681717cdf7a",
     "preserved_npy_sha256": {
       "success.npy": "93771288ec45b06fba72b165c461df5b4359f7fbd51b016d47b2dd32c4355296",
       "num_steps.npy": "d853eed0573863069fb8f51bda02385e94712a7ed4c4d4f59a64820ca326ac29",
@@ -51,12 +58,19 @@
       "obj_needle_quat.npy": "3e3b6a03fa88b2d295872b4068e4b3e66ff97a449ee32b1cc322af4c40336df9",
       "obj_tripod_quat.npy": "50a611f23b43d19e0f2ca5cd2c520dad73120d5ef2b0a025eae5d45209c35965"
     },
-    "omitted_truncated_members": [
-      "camera_images.npy"
-    ],
     "trailing_members": "Unknown: the original central directory is absent; none reconstructed",
     "num_steps": 498,
-    "note": "Complete NPY payloads preserved byte-for-byte with CRC validation; no images synthesized; action replay not certified.",
-    "archive_member": "recovery_demos_human/threading/grasp_wrong_pose_D0/recovery_threading_grasp_wrong_pose_D0_0013.npz"
+    "note": "Original complete NPY payloads preserved byte-for-byte. camera_images and explicit state alignment are stored in this record; camera_source and camera_metadata describe their rendering provenance.",
+    "archive_member": "recovery_demos_human/threading/grasp_wrong_pose_D0/recovery_threading_grasp_wrong_pose_D0_0013.npz",
+    "numeric_only_sha256": "40bb51c19c30a2f41ae540f6d6e856418b5b62e43447dfdaf0268ac053f4dda6",
+    "image_integration": {
+      "sha256": "32490ac0282c43b6daa8fbafa4520a1e3e788c88ce43a235907f5681717cdf7a",
+      "frames": 499,
+      "preserved_members": 17,
+      "camera_source": "state_rerender"
+    },
+    "original_truncated_members": [
+      "camera_images.npy"
+    ]
   }
 ]

training_data.tar

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bdc5cece3d639b0ca2604b55079a079f33b5e10de3f3fb4c6bb298980d5955a5
-size 8856616960
+oid sha256:0f85d555232a90d1a19fb7231453349a549c1532faaad953787ca94928eba590
+size 8908974080