fix(egohos): finalize review fixes and shared LFS rules

Use the same additive LFS rules as main to resolve the shared-file conflict; preserve raw/data content and pending human review. Regression-tested fixes and truthful evidence provenance.

EgoHOS/shared-v1.1/doc/REVIEW_FIXES.md

@@ -0,0 +1,32 @@
+# PR #7 review fixes — 2026-10-17
+
+Reviewed parent: `671194f1c18121fe9f415f813665ab0c99c58e4a` in `ParityLab/Hand-Datasets-Raw`.
+This maintenance change updates tools, tests, documentation and evidence bookkeeping only.
+The `data/` subtree and raw data are unchanged from that immutable parent.
+The merge replaces the old release paths atomically and preserves other datasets.
+
+## Evidence scope
+
+No full real-media decode, full numerical comparison or human visual acceptance was rerun
+for this maintenance change. Historical validation results remain historical, not new
+sign-off. Human review is still pending; this document does not grant approval.
+Git tree/blob identities establish unchanged tracked content, not annotation correctness.
+The maintenance audit records exact before/after Git trees and path changes outside the
+release; Git blob SHA1 identities must not be confused with file-content SHA256 hashes.
+
+## Fix and tests
+
+The v1.1 build/upload entries and both validators now reject non-EgoHOS data before
+processing it. Legacy OakInk helpers are retained only as historical code, not a supported
+v1.1 numeric converter. README/runbook and synthetic tests match this restriction.
+
+`python -B script/test_self.py` passes synthetic JPEG-copy, FFV1/x264 pixel equality,
+mask labels, provenance, directory/version, human-review gates, unsupported build/view/upload,
+and inventory/path/SE3 rejection tests. This is not a real-data acceptance run.
+
+`doc/build.json` and `doc/validation/report.json` record the current shipped recipe.
+Their maintenance metadata preserves the previous recipe and identifies the inherited
+full-data evidence. The report's file_hashes are refreshed for modified/new files and
+retain previously independently verified SHA256 values only for unchanged Git blobs.
+The inventory excludes the report itself and human_review.json, as the validator requires.
+No stale human approval is restored.

EgoHOS/shared-v1.1/doc/RUNBOOK.md

@@ -1,147 +1,42 @@
-# EgoHOS / OakInk-v1 预处理脚本
+# EgoHOS v1.1 工具
 
-以随包 CONTRACT.md 与 schema.json 为规范，输出 `<数据集>/shared-v1.1/{doc,script,data}`。版本串由 `script/schema.json` 声明，生成、校验与上传脚本均从它取值；当前为 `hand-warehouse/1.1` / `shared-v1.1`。默认只跑试点。**人工查看试点 HTML 后才能全量；全量结束后再次抽样人工确认才能上传。** 本包不是全量处理结果。
+本发布的转换、校验和上传入口只支持 **EgoHOS**（RGB、语义分割及各自独立 bool valid）。
+保留的 OakInk 辅助代码/来源文档是历史材料；数值 bundle 未在此实现，入口明确拒绝 OakInk，不能用它生成或上传“v1.1”几何数据。
+版本与目录名取自 `script/schema.json`。媒体与 JSONL 不参与数值合并。
 
-## 当前覆盖范围
+## 环境与回归
 
-| 数据集 | 可用路径 | 待补充 |
-|---|---|---|
-| EgoHOS | RGB + 9 类语义分割；四个官方 split；真实源样本可直接 Range 下载试点 | raw 包没有独立 contact 图，无标定/几何 |
-| OakInk-v1 / OakInk-Image | RGB-only 明确选项；可在服务器处理图像分卷 | 当前 HF raw 不含 anno_v2_1.zip；几何分支须补齐后真实试点 |
-| OakInk 几何可选分支 | K、外参、相机系右手 xyz/派生 uv、物体位姿、居中 OBJ 与依赖 | 无真实标注验收前，仅视为待验证适配器；MANO 未实现，不伪造参数 |
-
-OakInk-Shape/OakBase 不作为图像帧转换；具体排除原因见 DATA_MAPPING 文档。FreiHAND 参考链接本次匿名访问返回 401，未以未读内容作实现依据；目录和字段按用户提供的两个约定文件实现。
-
-## 环境
-
-Python 3.10+、FFmpeg（须有 ffv1 与 libx264rgb）、OakInk 分卷解压需要 7z/7zz。CPU 即可。
+Python 3.10+、FFmpeg，以及 requirements.txt 中的依赖。
 
 ```bash
-python3 -m venv .venv
-. .venv/bin/activate
-python -m pip install -r requirements.txt
-python test_self.py
+export PYTHONDONTWRITEBYTECODE=1
+python -B test_self.py
 ```
 
-如只用 CPU，可先按 PyTorch 官方 CPU 索引安装 torch 再安装 requirements，避免下载 CUDA 依赖。HF 下载/上传使用本机 `hf auth login` 或 HF_TOKEN；不要把 token 写入脚本或发布目录。
-
-## 1. EgoHOS 真实小试点（不下载 4.5 GB 全包）
+测试只使用合成图片/标签，检查无损媒体、版本、拒绝旧目录名、人工复核门禁及未适配分支；不替代真实人工验收。
 
-以下命令在本脚本目录运行，所有输出路径可改到大容量磁盘。extracted 是工作区；不在 raw 内解压。
+## 试点与全量
 
 ```bash
 python acquire.py pilot-egohos --out /storage/work/egohos-pilot --per-split 4
-python preprocess.py build --dataset EgoHOS \
-  --input /storage/work/egohos-pilot --out /storage/pilot/EgoHOS/shared-v1.1
+python preprocess.py build --dataset EgoHOS --input /storage/work/egohos-pilot \
+  --out /storage/pilot/EgoHOS/shared-v1.1
 ```
 
-浏览 `/storage/pilot/EgoHOS/shared-v1.1/doc/validation/index.html`（静态 HTML，可直接打开）。下拉框能逐帧查看全部试点样本、源 RGB、shared 解码 RGB、mask、叠加、valid、来源、元信息和缺失原因。请核对左右手、背景编码、细物体边界、四个 split。**只有人实际看过并接受后执行：**
-
-```bash
-python preprocess.py approve --release /storage/pilot/EgoHOS/shared-v1.1 --reviewer '审核者姓名'
-```
-
-approve 是人工声明记录，不会自行判断人工是否看过。程序不会自动调用 approve。
-
-## 2. 原始包下载与工作区解压
+人工查看试点 `doc/validation/index.html`，确认后才执行：
 
 ```bash
+python preprocess.py approve --release /storage/pilot/EgoHOS/shared-v1.1 --reviewer '真实审核者姓名'
 python acquire.py download --dataset EgoHOS --raw /storage/EgoHOS/raw
 python acquire.py extract --dataset EgoHOS --raw /storage/EgoHOS/raw --out /storage/work/egohos-full
-
-python acquire.py download --dataset OakInk-v1 --raw /storage/OakInk-v1/raw
-python acquire.py extract --dataset OakInk-v1 --raw /storage/OakInk-v1/raw --out /storage/work/oakink-full
-```
-
-固定 HF commit，并校验来源文档给出的每包大小/SHA256；原始包保持不变。OakInk `.z01`–`.z10` 与最终 `.zip` 是分卷 ZIP，由 7z 从最终 .zip 读取，不用 `cat` 伪拼接。下载约 124 GB，还需解压、编码候选与输出空间；预先准备数百 GB 空间，命令做磁盘预检但不保证压缩率估计准确。提取失败的目录不自动覆盖，诊断后选新目录重跑。
-
-## 3. OakInk 试点
-
-无需先下载全部 124 GB，也可做有限范围 RGB 小试点（只从首个分卷的前 128 个 PNG 条目中按相机抽样，不代表全数据集随机覆盖）：
-
-```bash
-python acquire.py pilot-oakink-rgb --out /storage/work/oakink-pilot --per-split 2
-python acquire.py fetch-oakink-objects --out /storage/work/oakink-pilot
-python preprocess.py build --dataset OakInk-v1 --input /storage/work/oakink-pilot \
-  --out /storage/pilot/OakInk-v1/shared-v1.1 --allow-rgb-only
-```
-
-完整工作区已解压时，当前 HF 没有 anno，若本次范围只接受现有 RGB：
-
-```bash
-python preprocess.py build --dataset OakInk-v1 \
-  --input /storage/work/oakink-full --out /storage/pilot/OakInk-v1/shared-v1.1 \
-  --allow-rgb-only --per-stratum 4
-```
-
-按 camera×split 固定种子抽样，unknown split 明确保留，不随意划分训练集。该结果没有手关键点或手/物逐帧位姿；如果工作区已有 image/obj，会保留静态网格和规范坐标系预览，仍不能作为完整几何交付。
-
-若已获可信官方 `anno_v2_1.zip`，先用以下命令校验每成员 CRC、记录文件和原包 SHA256，再解压到同一工作区 `image/anno`：
-
-```bash
-python acquire.py extract-annotations --archive /storage/OakInk-v1/raw/data/image/anno_v2_1.zip \
-  --out /storage/work/oakink-full
+python preprocess.py build --dataset EgoHOS --phase full --input /storage/work/egohos-full \
+  --out /storage/EgoHOS/shared-v1.1 --pilot /storage/pilot/EgoHOS/shared-v1.1
 ```
 
-该命令要求包内根目录为 anno/，发现不同布局会停止，避免猜错路径。然后运行：
+全量运行后再次人工查看，再 approve。上传需另行授权；上传命令拒绝已存在的远端发布树，不能用于自动覆盖既有发布：
 
 ```bash
-python preprocess.py build --dataset OakInk-v1 \
-  --input /storage/work/oakink-full --out /storage/pilot/OakInk-v1/shared-v1.1 \
-  --annotations /storage/work/oakink-full/image/anno --trust-pickle --split-scheme split0
-```
-
-`--trust-pickle` 仅用于可信官方标注，因为 pickle 可执行代码。每条来源记录格式（路径相对 --input；SHA256 填真实结果）：
-
-```json
-{"path":"image/anno/seq_all.json","sha256":"实际文件SHA256","source_locator":{"file":"OakInk-v1/raw/data/image/anno_v2_1.zip","archive_sha256":"实际包SHA256","member":"anno/seq_all.json"}}
-```
-
-几何路径不会强行把 sbj_flag 0/1 映射成左右手；两人的右手分别在 person0/person1 序列。检查 canonical 网格、相机系关键点/物体点云以及 2D 叠加。模型来源、量纲或关键点定义无法确认时不要 approve。MANO 需要另做模型配置与顶点重建对照，当前不导出。
-
-## 4. 试点批准后全量
-
-```bash
-python preprocess.py build --dataset EgoHOS --phase full \
-  --input /storage/work/egohos-full --out /storage/EgoHOS/shared-v1.1 \
-  --pilot /storage/pilot/EgoHOS/shared-v1.1
-
-python preprocess.py approve --release /storage/pilot/OakInk-v1/shared-v1.1 --reviewer '审核者姓名'
-python preprocess.py build --dataset OakInk-v1 --phase full \
-  --input /storage/work/oakink-full --out /storage/OakInk-v1/shared-v1.1 \
-  --pilot /storage/pilot/OakInk-v1/shared-v1.1 --allow-rgb-only
-```
-
-OakInk 若试点启用标注，全量也用相同 --annotations、--trust-pickle、--split-scheme，去掉 --allow-rgb-only。chunk、seed、代码、模态范围须与试点一致；变更后须重做试点。输出目录必须不存在。EgoHOS full 强制核对四组完整数量，防止把试点数据当全量。
-
-所有视图均逐帧校验视频/数组；全量 HTML 按 split×camera 随机抽取视图并显示所选视图的每一帧，绝不将人工抽检称为全量人工审核。若要浏览所有帧的所有要素（会生成大量 PNG/HTML）：
-
-```bash
-python validate.py /storage/EgoHOS/shared-v1.1 --all --source-root /storage/work/egohos-full
-```
-
-重新 validate 会使旧人工审核失效。输入、输出、每帧 provenance 与文件 SHA256 均可检查。自动检查只验证实现支持的字段，并不能自动证明数据集原生语义和真实几何正确。
-
-## 5. 最终抽检与上传
-
-打开全量结果的 doc/validation/index.html，再次人工抽检后：
-
-```bash
-python preprocess.py approve --release /storage/EgoHOS/shared-v1.1 --reviewer '审核者姓名'
 python preprocess.py upload --release /storage/EgoHOS/shared-v1.1
-
-python preprocess.py approve --release /storage/OakInk-v1/shared-v1.1 --reviewer '审核者姓名'
-python preprocess.py upload --release /storage/OakInk-v1/shared-v1.1
 ```
 
-目标按用户要求固定为 **ParityLab/Hand-Datasets-Raw**，分别是 `EgoHOS/shared-v1.1` 和 **大小写准确的** `OakInk-v1/shared-v1.1`。需要写权限；任何 pilot 都禁止作为正式 shared-v1.1 上传。上传前验证文件哈希，按最多 128 个文件分批提交、文档最后上传；拒绝合并到已经存在的同版本目录，防止留下旧文件。中途失败会保留已提交批次，脚本不会自动覆盖/删除，须核验远端后恢复。未自动上传、未代替用户签署人工审核。
-
-## 输出文件
-
-- `doc/DATA_MAPPING.md` 与兼容约定文件名 `doc/RAW_TO_SHARED.md`：raw 映射、未纳入内容和原因。
-- `doc/validation/index.html`、预览 PNG、`report.json`：视觉与自动检查。
-- `doc/build.json`：完整命令、源 inventory 摘要、选择策略、处理数。
-- `script/`：本次处理脚��和全部必需支持文件；不包含 venv、token 或 MANO 权重。
-- `data/<sequence>/<camera>/`：metainfo、frames、MKV、CPU tensor、已选物体资产。
-
-测试状态与真实试点范围见随包 `TEST_RESULTS.md`。没有真实标注包时，对几何分支的检查只能算合成端到端检查。
+`validate.py` 与 `resample_validation.py` 均校验版本、视图集合与工具摘要；现有验收记录的历史范围和本次工具修复见发布 `doc/REVIEW_FIXES.md`。

EgoHOS/shared-v1.1/doc/build.json

@@ -8,9 +8,9 @@
     "annotations": false,
     "split_scheme": "split0",
     "code_sha256": {
-      "preprocess.py": "c278753f8ff5f141bfa19ffbac1b8073db2fc8c7035efa74b8af495bdc7b1bc3",
+      "preprocess.py": "dd31d901599b5841f1305197a546e548695afad368737ffd5b63192c671e600f",
       "media.py": "ace183f298d1073ee9d453f13464d58e0be685fe09a3ddd99a78c67b4c7337ec",
-      "validate.py": "14d650390dc2011c5d645a620b63250a22cb109f2e6bb78d8c19854461a7065d",
+      "validate.py": "dafe3082b4d1f328e53267aafcfb3d056317daf42024493ec8a982000b3e7341",
       "acquire.py": "1f010b87f6cc13d41058ba5b6b4ac89b4becc9ef4b5a15f887f00842dfdc6993",
       "schema.json": "9f3769f950b36f75f3d571c77b5a661ca8e72afae20c9f1ddd1deef140e6cca8",
       "CONTRACT.md": "4a61d6efb6daccdb3ff14841586953f776c7e3059521e42a423abcf1b67db7b9"
@@ -202,5 +202,28 @@
     ],
     "validation": "见 doc/validation/report.json（本地全量：106 个视图、11743 帧）",
     "registration": "发布树外登记与发布清单见 tmp/egohos-tools/MIGRATION_REGISTRATION.json、PUBLISH_MANIFEST.json"
+  },
+  "maintenance": {
+    "date": "2026-10-17",
+    "reviewed_parent": "671194f1c18121fe9f415f813665ab0c99c58e4a",
+    "scope": "unsupported-entry guards, docs and synthetic regression; no data modification",
+    "previous_recipe": {
+      "dataset": "EgoHOS",
+      "chunk": 128,
+      "seed": 20260928,
+      "annotations": false,
+      "split_scheme": "split0",
+      "code_sha256": {
+        "preprocess.py": "c278753f8ff5f141bfa19ffbac1b8073db2fc8c7035efa74b8af495bdc7b1bc3",
+        "media.py": "ace183f298d1073ee9d453f13464d58e0be685fe09a3ddd99a78c67b4c7337ec",
+        "validate.py": "14d650390dc2011c5d645a620b63250a22cb109f2e6bb78d8c19854461a7065d",
+        "acquire.py": "1f010b87f6cc13d41058ba5b6b4ac89b4becc9ef4b5a15f887f00842dfdc6993",
+        "schema.json": "9f3769f950b36f75f3d571c77b5a661ca8e72afae20c9f1ddd1deef140e6cca8",
+        "CONTRACT.md": "4a61d6efb6daccdb3ff14841586953f776c7e3059521e42a423abcf1b67db7b9"
+      }
+    },
+    "full_real_data_validation": "inherited from reviewed parent; not rerun with current recipe",
+    "evidence": "doc/REVIEW_FIXES.md",
+    "human_review": "pending"
   }
 }

EgoHOS/shared-v1.1/doc/validation/report.json

@@ -8,9 +8,9 @@
     "annotations": false,
     "split_scheme": "split0",
     "code_sha256": {
-      "preprocess.py": "c278753f8ff5f141bfa19ffbac1b8073db2fc8c7035efa74b8af495bdc7b1bc3",
+      "preprocess.py": "dd31d901599b5841f1305197a546e548695afad368737ffd5b63192c671e600f",
       "media.py": "ace183f298d1073ee9d453f13464d58e0be685fe09a3ddd99a78c67b4c7337ec",
-      "validate.py": "14d650390dc2011c5d645a620b63250a22cb109f2e6bb78d8c19854461a7065d",
+      "validate.py": "dafe3082b4d1f328e53267aafcfb3d056317daf42024493ec8a982000b3e7341",
       "acquire.py": "1f010b87f6cc13d41058ba5b6b4ac89b4becc9ef4b5a15f887f00842dfdc6993",
       "schema.json": "9f3769f950b36f75f3d571c77b5a661ca8e72afae20c9f1ddd1deef140e6cca8",
       "CONTRACT.md": "4a61d6efb6daccdb3ff14841586953f776c7e3059521e42a423abcf1b67db7b9"
@@ -433,25 +433,25 @@
   "file_hashes": {
     "doc/RAW_TO_SHARED.md": "cfaa0f3befc619c88931642fde9cbf34236dacd2cfe73210aed8de4caab2ecb1",
     "doc/schema.json": "9f3769f950b36f75f3d571c77b5a661ca8e72afae20c9f1ddd1deef140e6cca8",
-    "doc/RUNBOOK.md": "74d1401bde78e850781394bbd683b12d4621a86ad31e85b2f48695b11a591af8",
+    "doc/RUNBOOK.md": "ed72227f294ffe5e6735e1f4e59ea45553f6fc91a6325f9d50f0c8505c6bd863",
     "doc/DATA_MAPPING.md": "cfaa0f3befc619c88931642fde9cbf34236dacd2cfe73210aed8de4caab2ecb1",
     "doc/CONTRACT.md": "4a61d6efb6daccdb3ff14841586953f776c7e3059521e42a423abcf1b67db7b9",
     "doc/MIGRATION.md": "773f5146d49f5060bfc13ef92c143c59b683a6dc633cad44897414d5299088a0",
-    "doc/build.json": "7c4f3b01a9cf4d04131cd49c49db8b601df2087b65c480bdd77b8c57f32a81a4",
+    "doc/build.json": "82d0fbbcaafdd94b92d113891976ba667569ae5973d55e0b7b41d86b3db4abd8",
     "doc/V1_0_SEMANTICS.md": "c34cdf6645f91233cc28d86f8abc82f019f78767d863e225f75722984d9adb19",
     "doc/pilot_human_review.json": "1baa24dc868f248f6f2dae9fa34cee1374c72d4cf949e8d80d14b1fd4afd7b07",
-    "script/test_self.py": "c01c94063659b290ec72385dbce8ae7bc4dc24f940055f2ce07635c179a03488",
+    "script/test_self.py": "5cb8813f9686110aeb978e8f7d864782d2c27d03f31750b2af95aab4591292ca",
     "script/schema.json": "9f3769f950b36f75f3d571c77b5a661ca8e72afae20c9f1ddd1deef140e6cca8",
-    "script/README.md": "74d1401bde78e850781394bbd683b12d4621a86ad31e85b2f48695b11a591af8",
-    "script/validate.py": "14d650390dc2011c5d645a620b63250a22cb109f2e6bb78d8c19854461a7065d",
+    "script/README.md": "ed72227f294ffe5e6735e1f4e59ea45553f6fc91a6325f9d50f0c8505c6bd863",
+    "script/validate.py": "dafe3082b4d1f328e53267aafcfb3d056317daf42024493ec8a982000b3e7341",
     "script/acquire.py": "1f010b87f6cc13d41058ba5b6b4ac89b4becc9ef4b5a15f887f00842dfdc6993",
     "script/CONTRACT.md": "4a61d6efb6daccdb3ff14841586953f776c7e3059521e42a423abcf1b67db7b9",
     "script/requirements.txt": "b2ac75f2c8739e2bc1c18cca6d6df04c829b3cedb0897c178ab4a3da4f639ffc",
     "script/media.py": "ace183f298d1073ee9d453f13464d58e0be685fe09a3ddd99a78c67b4c7337ec",
     "script/EgoHOS_DATA_MAPPING.md": "cfaa0f3befc619c88931642fde9cbf34236dacd2cfe73210aed8de4caab2ecb1",
-    "script/OakInk-v1_DATA_MAPPING.md": "ec61d0b5129a49eb23af6733c98b12278d9f8f161953cf8ec48fe98e1c833212",
-    "script/preprocess.py": "c278753f8ff5f141bfa19ffbac1b8073db2fc8c7035efa74b8af495bdc7b1bc3",
-    "script/resample_validation.py": "1573339d0a6db3656394266c67b6a6c5f633ea2283800840c1b04f4ea7cde854",
+    "script/OakInk-v1_DATA_MAPPING.md": "f06dbb1fbf474122b1e200130b768321716140c5b745d7e614a38e55c931198a",
+    "script/preprocess.py": "dd31d901599b5841f1305197a546e548695afad368737ffd5b63192c671e600f",
+    "script/resample_validation.py": "9d8e1a00bc16d09c445eb35b0b433558475f7815eb459666d06558e7b28d471d",
     "data/train_1440x1080_000020/ego/metainfo.json": "67ee3587ae318d2d8c9ea69fe3f2eb40c420cd3f212215bb1a8f7eb159748cce",
     "data/train_1440x1080_000020/ego/encoding_validation.json": "510d2a00a62fe94ca007893111e005cba391df9319e6943dc4c5c0890eed18cc",
     "data/train_1440x1080_000020/ego/observation/rgb_valid.pt": "aa7fff018778d52c47e0824b1887b4cbff83ea2cb8242afd95a4ea76ee36bca0",
@@ -2000,11 +2000,35 @@
     "script/source_evidence/OakInk-v1-archive_checksums.json": "fbf23d0a9169596002cdc2a1a8810b95c651419de0e615c9b0624580a12c3b3f",
     "script/source_evidence/EgoHOS-README.md": "6e9d4c1a70e60a916057b00fa4c9b4a3c3aa1a434a8f70b1e69b5c975c2770fb",
     "script/source_evidence/EgoHOS-archive_checksums.json": "7f8e0f76461a08823fd8651e2261865eac177f84d15792eb29b1e08c916b31ef",
-    "script/source_evidence/OakInk-v1-README.md": "a0ed267db7e0870ce5edd56dc477f49ac3815641f491ef5ed74d1362a93e2afb"
+    "script/source_evidence/OakInk-v1-README.md": "a0ed267db7e0870ce5edd56dc477f49ac3815641f491ef5ed74d1362a93e2afb",
+    "doc/REVIEW_FIXES.md": "085d3bcba24d669ab4251d81726992a740d10f6a84f6e0a996c33ab46cfe5d21"
   },
   "limitations": [
     "OakInk MANO conversion not implemented until model-profile and translation parity verified",
     "OakInk-Shape/OakBase excluded: object-space grasp/knowledge assets are not synchronized camera observations",
     "No source visibility/contact guessed from geometry or masks"
-  ]
+  ],
+  "maintenance": {
+    "date": "2026-10-17",
+    "reviewed_parent": "671194f1c18121fe9f415f813665ab0c99c58e4a",
+    "scope": "unsupported-entry guards, docs and synthetic regression; no data modification",
+    "previous_recipe": {
+      "dataset": "EgoHOS",
+      "chunk": 128,
+      "seed": 20260928,
+      "annotations": false,
+      "split_scheme": "split0",
+      "code_sha256": {
+        "preprocess.py": "c278753f8ff5f141bfa19ffbac1b8073db2fc8c7035efa74b8af495bdc7b1bc3",
+        "media.py": "ace183f298d1073ee9d453f13464d58e0be685fe09a3ddd99a78c67b4c7337ec",
+        "validate.py": "14d650390dc2011c5d645a620b63250a22cb109f2e6bb78d8c19854461a7065d",
+        "acquire.py": "1f010b87f6cc13d41058ba5b6b4ac89b4becc9ef4b5a15f887f00842dfdc6993",
+        "schema.json": "9f3769f950b36f75f3d571c77b5a661ca8e72afae20c9f1ddd1deef140e6cca8",
+        "CONTRACT.md": "4a61d6efb6daccdb3ff14841586953f776c7e3059521e42a423abcf1b67db7b9"
+      }
+    },
+    "full_real_data_validation": "inherited from reviewed parent; not rerun with current recipe",
+    "evidence": "doc/REVIEW_FIXES.md",
+    "human_review": "pending"
+  }
 }

EgoHOS/shared-v1.1/script/OakInk-v1_DATA_MAPPING.md

@@ -1,3 +1,5 @@
+> 历史材料：本 EgoHOS v1.1 工具不再支持 OakInk 转换或上传，以下说明不是当前入口承诺。
+
 # OakInk-v1 raw → shared-v1.1
 
 本脚本的图像范围是 **OakInk-Image**。仓库名称 OakInk-v1 下包含 OakInk-Image v2 图像包，不应误当成 OakInk2。

EgoHOS/shared-v1.1/script/README.md

@@ -1,147 +1,42 @@
-# EgoHOS / OakInk-v1 预处理脚本
+# EgoHOS v1.1 工具
 
-以随包 CONTRACT.md 与 schema.json 为规范，输出 `<数据集>/shared-v1.1/{doc,script,data}`。版本串由 `script/schema.json` 声明，生成、校验与上传脚本均从它取值；当前为 `hand-warehouse/1.1` / `shared-v1.1`。默认只跑试点。**人工查看试点 HTML 后才能全量；全量结束后再次抽样人工确认才能上传。** 本包不是全量处理结果。
+本发布的转换、校验和上传入口只支持 **EgoHOS**（RGB、语义分割及各自独立 bool valid）。
+保留的 OakInk 辅助代码/来源文档是历史材料；数值 bundle 未在此实现，入口明确拒绝 OakInk，不能用它生成或上传“v1.1”几何数据。
+版本与目录名取自 `script/schema.json`。媒体与 JSONL 不参与数值合并。
 
-## 当前覆盖范围
+## 环境与回归
 
-| 数据集 | 可用路径 | 待补充 |
-|---|---|---|
-| EgoHOS | RGB + 9 类语义分割；四个官方 split；真实源样本可直接 Range 下载试点 | raw 包没有独立 contact 图，无标定/几何 |
-| OakInk-v1 / OakInk-Image | RGB-only 明确选项；可在服务器处理图像分卷 | 当前 HF raw 不含 anno_v2_1.zip；几何分支须补齐后真实试点 |
-| OakInk 几何可选分支 | K、外参、相机系右手 xyz/派生 uv、物体位姿、居中 OBJ 与依赖 | 无真实标注验收前，仅视为待验证适配器；MANO 未实现，不伪造参数 |
-
-OakInk-Shape/OakBase 不作为图像帧转换；具体排除原因见 DATA_MAPPING 文档。FreiHAND 参考链接本次匿名访问返回 401，未以未读内容作实现依据；目录和字段按用户提供的两个约定文件实现。
-
-## 环境
-
-Python 3.10+、FFmpeg（须有 ffv1 与 libx264rgb）、OakInk 分卷解压需要 7z/7zz。CPU 即可。
+Python 3.10+、FFmpeg，以及 requirements.txt 中的依赖。
 
 ```bash
-python3 -m venv .venv
-. .venv/bin/activate
-python -m pip install -r requirements.txt
-python test_self.py
+export PYTHONDONTWRITEBYTECODE=1
+python -B test_self.py
 ```
 
-如只用 CPU，可先按 PyTorch 官方 CPU 索引安装 torch 再安装 requirements，避免下载 CUDA 依赖。HF 下载/上传使用本机 `hf auth login` 或 HF_TOKEN；不要把 token 写入脚本或发布目录。
-
-## 1. EgoHOS 真实小试点（不下载 4.5 GB 全包）
+测试只使用合成图片/标签，检查无损媒体、版本、拒绝旧目录名、人工复核门禁及未适配分支；不替代真实人工验收。
 
-以下命令在本脚本目录运行，所有输出路径可改到大容量磁盘。extracted 是工作区；不在 raw 内解压。
+## 试点与全量
 
 ```bash
 python acquire.py pilot-egohos --out /storage/work/egohos-pilot --per-split 4
-python preprocess.py build --dataset EgoHOS \
-  --input /storage/work/egohos-pilot --out /storage/pilot/EgoHOS/shared-v1.1
+python preprocess.py build --dataset EgoHOS --input /storage/work/egohos-pilot \
+  --out /storage/pilot/EgoHOS/shared-v1.1
 ```
 
-浏览 `/storage/pilot/EgoHOS/shared-v1.1/doc/validation/index.html`（静态 HTML，可直接打开）。下拉框能逐帧查看全部试点样本、源 RGB、shared 解码 RGB、mask、叠加、valid、来源、元信息和缺失原因。请核对左右手、背景编码、细物体边界、四个 split。**只有人实际看过并接受后执行：**
-
-```bash
-python preprocess.py approve --release /storage/pilot/EgoHOS/shared-v1.1 --reviewer '审核者姓名'
-```
-
-approve 是人工声明记录，不会自行判断人工是否看过。程序不会自动调用 approve。
-
-## 2. 原始包下载与工作区解压
+人工查看试点 `doc/validation/index.html`，确认后才执行：
 
 ```bash
+python preprocess.py approve --release /storage/pilot/EgoHOS/shared-v1.1 --reviewer '真实审核者姓名'
 python acquire.py download --dataset EgoHOS --raw /storage/EgoHOS/raw
 python acquire.py extract --dataset EgoHOS --raw /storage/EgoHOS/raw --out /storage/work/egohos-full
-
-python acquire.py download --dataset OakInk-v1 --raw /storage/OakInk-v1/raw
-python acquire.py extract --dataset OakInk-v1 --raw /storage/OakInk-v1/raw --out /storage/work/oakink-full
-```
-
-固定 HF commit，并校验来源文档给出的每包大小/SHA256；原始包保持不变。OakInk `.z01`–`.z10` 与最终 `.zip` 是分卷 ZIP，由 7z 从最终 .zip 读取，不用 `cat` 伪拼接。下载约 124 GB，还需解压、编码候选与输出空间；预先准备数百 GB 空间，命令做磁盘预检但不保证压缩率估计准确。提取失败的目录不自动覆盖，诊断后选新目录重跑。
-
-## 3. OakInk 试点
-
-无需先下载全部 124 GB，也可做有限范围 RGB 小试点（只从首个分卷的前 128 个 PNG 条目中按相机抽样，不代表全数据集随机覆盖）：
-
-```bash
-python acquire.py pilot-oakink-rgb --out /storage/work/oakink-pilot --per-split 2
-python acquire.py fetch-oakink-objects --out /storage/work/oakink-pilot
-python preprocess.py build --dataset OakInk-v1 --input /storage/work/oakink-pilot \
-  --out /storage/pilot/OakInk-v1/shared-v1.1 --allow-rgb-only
-```
-
-完整工作区已解压时，当前 HF 没有 anno，若本次范围只接受现有 RGB：
-
-```bash
-python preprocess.py build --dataset OakInk-v1 \
-  --input /storage/work/oakink-full --out /storage/pilot/OakInk-v1/shared-v1.1 \
-  --allow-rgb-only --per-stratum 4
-```
-
-按 camera×split 固定种子抽样，unknown split 明确保留，不随意划分训练集。该结果没有手关键点或手/物逐帧位姿；如果工作区已有 image/obj，会保留静态网格和规范坐标系预览，仍不能作为完整几何交付。
-
-若已获可信官方 `anno_v2_1.zip`，先用以下命令校验每成员 CRC、记录文件和原包 SHA256，再解压到同一工作区 `image/anno`：
-
-```bash
-python acquire.py extract-annotations --archive /storage/OakInk-v1/raw/data/image/anno_v2_1.zip \
-  --out /storage/work/oakink-full
+python preprocess.py build --dataset EgoHOS --phase full --input /storage/work/egohos-full \
+  --out /storage/EgoHOS/shared-v1.1 --pilot /storage/pilot/EgoHOS/shared-v1.1
 ```
 
-该命令要求包内根目录为 anno/，发现不同布局会停止，避免猜错路径。然后运行：
+全量运行后再次人工查看，再 approve。上传需另行授权；上传命令拒绝已存在的远端发布树，不能用于自动覆盖既有发布：
 
 ```bash
-python preprocess.py build --dataset OakInk-v1 \
-  --input /storage/work/oakink-full --out /storage/pilot/OakInk-v1/shared-v1.1 \
-  --annotations /storage/work/oakink-full/image/anno --trust-pickle --split-scheme split0
-```
-
-`--trust-pickle` 仅用于可信官方标注，因为 pickle 可执行代码。每条来源记录格式（路径相对 --input；SHA256 填真实结果）：
-
-```json
-{"path":"image/anno/seq_all.json","sha256":"实际文件SHA256","source_locator":{"file":"OakInk-v1/raw/data/image/anno_v2_1.zip","archive_sha256":"实际包SHA256","member":"anno/seq_all.json"}}
-```
-
-几何路径不会强行把 sbj_flag 0/1 映射成左右手；两人的右手分别在 person0/person1 序列。检查 canonical 网格、相机系关键点/物体点云以及 2D 叠加。模型来源、量纲或关键点定义无法确认时不要 approve。MANO 需要另做模型配置与顶点重建对照，当前不导出。
-
-## 4. 试点批准后全量
-
-```bash
-python preprocess.py build --dataset EgoHOS --phase full \
-  --input /storage/work/egohos-full --out /storage/EgoHOS/shared-v1.1 \
-  --pilot /storage/pilot/EgoHOS/shared-v1.1
-
-python preprocess.py approve --release /storage/pilot/OakInk-v1/shared-v1.1 --reviewer '审核者姓名'
-python preprocess.py build --dataset OakInk-v1 --phase full \
-  --input /storage/work/oakink-full --out /storage/OakInk-v1/shared-v1.1 \
-  --pilot /storage/pilot/OakInk-v1/shared-v1.1 --allow-rgb-only
-```
-
-OakInk 若试点启用标注，全量也用相同 --annotations、--trust-pickle、--split-scheme，去掉 --allow-rgb-only。chunk、seed、代码、模态范围须与试点一致；变更后须重做试点。输出目录必须不存在。EgoHOS full 强制核对四组完整数量，防止把试点数据当全量。
-
-所有视图均逐帧校验视频/数组；全量 HTML 按 split×camera 随机抽取视图并显示所选视图的每一帧，绝不将人工抽检称为全量人工审核。若要浏览所有帧的所有要素（会生成大量 PNG/HTML）：
-
-```bash
-python validate.py /storage/EgoHOS/shared-v1.1 --all --source-root /storage/work/egohos-full
-```
-
-重新 validate 会使旧人工审核失效。输入、输出、每帧 provenance 与文件 SHA256 均可检查。自动检查只验证实现支持的字段，并不能自动证明数据集原生语义和真实几何正确。
-
-## 5. 最终抽检与上传
-
-打开全量结果的 doc/validation/index.html，再次人工抽检后：
-
-```bash
-python preprocess.py approve --release /storage/EgoHOS/shared-v1.1 --reviewer '审核者姓名'
 python preprocess.py upload --release /storage/EgoHOS/shared-v1.1
-
-python preprocess.py approve --release /storage/OakInk-v1/shared-v1.1 --reviewer '审核者姓名'
-python preprocess.py upload --release /storage/OakInk-v1/shared-v1.1
 ```
 
-目标按用户要求固定为 **ParityLab/Hand-Datasets-Raw**，分别是 `EgoHOS/shared-v1.1` 和 **大小写准确的** `OakInk-v1/shared-v1.1`。需要写权限；任何 pilot 都禁止作为正式 shared-v1.1 上传。上传前验证文件哈希，按最多 128 个文件分批提交、文档最后上传；拒绝合并到已经存在的同版本目录，防止留下旧文件。中途失败会保留已提交批次，脚本不会自动覆盖/删除，须核验远端后恢复。未自动上传、未代替用户签署人工审核。
-
-## 输出文件
-
-- `doc/DATA_MAPPING.md` 与兼容约定文件名 `doc/RAW_TO_SHARED.md`：raw 映射、未纳入内容和原因。
-- `doc/validation/index.html`、预览 PNG、`report.json`：视觉与自动检查。
-- `doc/build.json`：完整命令、源 inventory 摘要、选择策略、处理数。
-- `script/`：本次处理脚��和全部必需支持文件；不包含 venv、token 或 MANO 权重。
-- `data/<sequence>/<camera>/`：metainfo、frames、MKV、CPU tensor、已选物体资产。
-
-测试状态与真实试点范围见随包 `TEST_RESULTS.md`。没有真实标注包时，对几何分支的检查只能算合成端到端检查。
+`validate.py` 与 `resample_validation.py` 均校验版本、视图集合与工具摘要；现有验收记录的历史范围和本次工具修复见发布 `doc/REVIEW_FIXES.md`。

EgoHOS/shared-v1.1/script/preprocess.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python3
-"""EgoHOS / OakInk-Image -> the hand-warehouse layout declared by schema.json
-(currently hand-warehouse/1.1). Full runs require pilot review."""
+"""EgoHOS -> the hand-warehouse layout declared by schema.json.
+Legacy OakInk helpers are not a supported v1.1 conversion path."""
 import argparse
 from collections import defaultdict
 from functools import lru_cache
@@ -441,6 +441,8 @@ def require_review(root):
 
 
 def run(args):
+    if args.dataset != "EgoHOS":
+        raise ValueError("This v1.1 builder supports EgoHOS only; OakInk numeric bundles are not adapted here")
     if args.dataset == 'OakInk-v1' and not args.annotations and not args.allow_rgb_only:
         raise ValueError('OakInk annotations absent. Supply --annotations + --trust-pickle, or explicitly select --allow-rgb-only (limited scope).')
     if args.annotations and (not args.trust_pickle or not args.annotations.resolve().is_relative_to(args.input.resolve())):
@@ -553,6 +555,9 @@ def main():
         approve(args.release, args.reviewer)
     else:
         from validate import check_inventory
+        d = load(args.release / 'doc/validation/report.json')
+        if d['dataset'] != 'EgoHOS':
+            raise ValueError('This v1.1 uploader supports EgoHOS only')
         d = require_review(args.release)
         if d['phase'] != 'full':
             raise ValueError(f'Pilot releases cannot be uploaded as {RELEASE_ID}')

EgoHOS/shared-v1.1/script/resample_validation.py

@@ -51,6 +51,8 @@ def check_inventory(root, expected):
 
 def validate_view(view, registry):
     m = read_json(view / 'metainfo.json')
+    if m.get('dataset_id') != 'EgoHOS':
+        raise ValueError('This v1.1 validator supports EgoHOS only')
     frames = read_jsonl(view / 'camera/frames.jsonl')
     t = m['num_frames']
     if t < 1 or m['hand_slots'] != ['left', 'right'] or len(frames) != t:

EgoHOS/shared-v1.1/script/test_self.py

@@ -1,7 +1,6 @@
 #!/usr/bin/env python3
 """One small runnable end-to-end check; fixtures are synthetic, not real validation."""
 import json
-import pickle
 import subprocess
 import sys
 import tempfile
@@ -13,7 +12,8 @@ from PIL import Image
 from acquire import sha, safe_path
 from media import mask, decode
 from preprocess import se3
-from validate import pt, check_inventory
+from validate import pt, check_inventory, validate_view
+from resample_validation import validate_view as resample_view
 
 HERE = Path(__file__).resolve().parent
 
@@ -39,12 +39,6 @@ def provenance(root):
     (root / 'sources.jsonl').write_text(''.join(json.dumps(r) + '\n' for r in entries))
 
 
-def save_pickle(path, value):
-    path.parent.mkdir(parents=True, exist_ok=True)
-    with path.open('wb') as f:
-        pickle.dump(value, f)
-
-
 def main():
     rng = np.random.default_rng(17)
     with tempfile.TemporaryDirectory() as directory:
@@ -93,56 +87,26 @@ def main():
         else:
             raise AssertionError('Mutation accepted')
 
-        oak = temp / 'oak'
-        seq = 'A00001_0004_0001_0002/2021-01-01-00-00-00'
-        image = oak / 'image/stream_release_v2' / seq / 'north_east_color_10.png'
-        image.parent.mkdir(parents=True)
-        Image.fromarray(rng.integers(0, 256, (24, 32, 3), dtype=np.uint8)).save(image)
-        anno = oak / 'image/anno'
-        anno.mkdir()
-        info = [[seq, person, 10, 0] for person in (0, 1)]
-        (anno / 'seq_all.json').write_text(json.dumps(info))
-        for rel, value in [('split_train_val/split0/example_split_train.json', info),
-                           ('split_train_val/split0/example_split_val.json', []), ('split/split0/seq_test.json', [])]:
-            p = anno / rel
-            p.parent.mkdir(parents=True, exist_ok=True)
-            p.write_text(json.dumps(value))
-        k = np.array([[20, 0, 16], [0, 20, 12], [0, 0, 1]], np.float32)
-        xyz = np.column_stack((np.linspace(-.1, .1, 21), np.zeros(21), np.ones(21))).astype(np.float32)
-        pose = np.eye(4, dtype=np.float32)
-        pose[2, 3] = 1
-        for x in info:
-            key = '__'.join(map(str, x)).replace('/', '__') + '.pkl'
-            for folder, value in [('cam_intr', k), ('hand_j', xyz), ('obj_transf', pose), ('general_info', {'cam_extr': np.eye(4, dtype=np.float32)})]:
-                save_pickle(anno / folder / key, value)
-        obj = oak / 'image/obj/A00001.obj'
-        obj.parent.mkdir()
-        obj.write_text('mtllib absent.mtl\nv 0 0 0\nv .2 0 0\nv 0 .2 0\nusemtl default\nf 1 2 3\n')
-        (oak / 'acquisition.json').write_text(json.dumps({'command': 'extract', 'dataset': 'OakInk-v1', 'note': 'SYNTHETIC TEST ONLY'}))
-        provenance(oak)
-        command('build', '--dataset', 'OakInk-v1', '--input', oak, '--out', release_dir(temp, 'oak-blocked'), succeeds=False)
-        command('build', '--dataset', 'OakInk-v1', '--input', oak, '--out', release_dir(temp, 'oak-rgb'), '--allow-rgb-only')
-        out = release_dir(temp, 'oak-geo')
-        command('build', '--dataset', 'OakInk-v1', '--input', oak, '--out', out,
-                '--annotations', anno, '--trust-pickle')
-        views = list((out / 'data').glob('*/*/metainfo.json'))
-        assert len(views) == 2 and any('person0' in str(p) for p in views) and any('person1' in str(p) for p in views)
-        for m in views:
-            v = m.parent
-            points = pt(v / 'annotation/keypoints/xyz.pt')
-            assert np.isnan(points[:, 0]).all() and np.array_equal(points[0, 1], xyz)
-            uv = pt(v / 'annotation/keypoints/uv.pt')
-            assert np.allclose(uv[0, 1, 0], [14, 12])
-            text = (v / 'annotation/object/A00001/assets/A00001.obj').read_text()
-            assert 'v -0.1 -0.1 0' in text
-            assert 'mtllib ' not in text and 'usemtl ' not in text
-            metadata = json.loads(m.read_text())
-            assert metadata['components']['object_mesh']['provenance']['source_missing_materials'] == ['absent.mtl']
-        # Test review records apply only to these temporary synthetic fixtures.
-        command('approve', '--release', out, '--reviewer', 'SYNTHETIC_TEST_ONLY')
-        command('upload', '--release', out, succeeds=False)  # Reviewed pilot still cannot publish.
-        command('build', '--dataset', 'OakInk-v1', '--input', oak, '--out', release_dir(temp, 'oak-full'),
-                '--annotations', anno, '--trust-pickle', '--phase', 'full', '--pilot', out)
+        # Reject the unadapted numeric branch before it reads inputs or creates output.
+        blocked = release_dir(temp, 'oak-blocked')
+        command('build', '--dataset', 'OakInk-v1', '--input', temp / 'absent-input',
+                '--out', blocked, '--allow-rgb-only', succeeds=False)
+        assert not blocked.exists()
+        unsupported = temp / 'unsupported-view'
+        unsupported.mkdir()
+        (unsupported / 'metainfo.json').write_text(json.dumps({'dataset_id': 'OakInk-v1'}))
+        for validator in (validate_view, resample_view):
+            try:
+                validator(unsupported, registry)
+            except ValueError as error:
+                assert 'EgoHOS only' in str(error)
+            else:
+                raise AssertionError('Unsupported numeric view accepted')
+        unsupported_release = temp / 'unsupported-release'
+        (unsupported_release / 'doc/validation').mkdir(parents=True)
+        (unsupported_release / 'doc/validation/report.json').write_text(json.dumps({'dataset': 'OakInk-v1'}))
+        result = command('upload', '--release', unsupported_release, succeeds=False)
+        assert 'EgoHOS only' in result.stderr
         for bad in ('../escape', '/absolute', 'a/../../escape'):
             try:
                 safe_path(temp, bad)
@@ -158,7 +122,7 @@ def main():
             pass
         else:
             raise AssertionError('Invalid SE3 accepted')
-    print('PASS: JPEG copy, FFV1/x264 RGB, labels, provenance, review gates, RGB-only, synthetic OakInk geometry, NaN slots, projection, mesh centering, mutation/path/SE3 rejection, registry version and release-directory name')
+    print('PASS: JPEG copy, FFV1/x264 RGB, labels, provenance, review gates, unsupported OakInk rejection, mutation/path/SE3 rejection, registry version and release-directory name')
 
 
 if __name__ == '__main__':

EgoHOS/shared-v1.1/script/validate.py

@@ -51,6 +51,8 @@ def check_inventory(root, expected):
 
 def validate_view(view, registry):
     m = read_json(view / 'metainfo.json')
+    if m.get('dataset_id') != 'EgoHOS':
+        raise ValueError('This v1.1 validator supports EgoHOS only')
     frames = read_jsonl(view / 'camera/frames.jsonl')
     t = m['num_frames']
     if t < 1 or m['hand_slots'] != ['left', 'right'] or len(frames) != t: