diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/0/000405.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/0/000405.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..aa68d432b527fd316abfe76ef26b5a6ee61945e6
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/0/000405.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d3dbb42cab72030ee142782a5da8f88bca3684670cfd8d12f0d4327b332a1a7
+size 246380
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/0/000540.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/0/000540.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..8bbe69d4d8e6ec1af54ebd3db75a6e7052bc56e2
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/0/000540.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1f71ee8d1778d2f28e5b2a09b2c108b032c9fdb5b0835b998008bf0b5d1a9ff
+size 212296
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000000.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000000.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..78419bb9468a9ebbeff7f8400727a198229e7f40
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000000.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:124b0756cc67ca7b456ccc96adc13038f431494c8ddba1319723bbb1b115ba08
+size 99997
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000001.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000001.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..cc0e957ff820e4b54c23eb3dcc2f842667482480
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000001.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61c7085c12d59ca2005b1b46ab27abc460a265b2c238bce12c6fb5eec2e24f83
+size 104526
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000026.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000026.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..4201ee4b5e40b3e7f6f83281eb67d61fb8935655
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000026.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f52bcee7754da7e0807fbe4482ec49318bc02ac1a0e13ac49d1c3ceb6b2f8b28
+size 205400
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000027.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000027.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..5b6e8b65e302703070052675a5d7f2ce9a70f763
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000027.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fac47ae8965a1abcb28937a7fda99c206b80f50c2caffd0883ac46778e7a380e
+size 206415
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000135.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000135.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..5a5ac50b0cddcad694d176f0d0a64e71ed9bd546
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000135.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5eba013d9ff8c2e1edf139324cca99fe285a17d2bf9a5cc5a1904e7cff119ef
+size 204712
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000179.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000179.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..f3b879f16fbdbf0d8c47385d6be291edf00de033
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000179.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10638e6c5f71cf6afc09a870aa7d679d284d0210df71b08ecaf653dab714e829
+size 204121
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000180.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000180.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..61e37c6dbcb07f734bc301eb375795096f9337aa
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000180.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4ef3bc5f57d76b8a46b927ab163fe53b51227797bbd61bac2d7025db904c5993
+size 204507
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000270.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000270.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..5166197263fbe7a5db5232ec669b4e154756510d
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000270.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:100d4575f4bd949d8a1d0f103766f4e453dc5d82fe54acd8f5e036355910f83a
+size 204516
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000405.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000405.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..0b34dd1ad58d9059324290e839d309a539131da4
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000405.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dfc5ca01e863585e21b440467fd085f7c8f9368c5820403e676d62078c770209
+size 203700
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000540.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000540.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..9e873de34f142df602c6f2a824b4dc31313e5486
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/1/000540.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc5ca088849a6dfa1062894a5c6982a9ae183b605fb03b24eee6beb1f56c31a5
+size 202536
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000000.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000000.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..82dd044057d3a9b238746b56c2afa929752b4d94
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000000.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f06a6e231ed4a325da1c266143af81d90f4e5d4d3d1b4ce1fc0a035f056cf65
+size 100091
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000001.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000001.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..0fbf38f806ddb55716a0a4a0d468ce3f592d76d2
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000001.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec91f7ad65a0e718769598345885168d8e6a80e676df58bb9d19d3cc8a19415d
+size 104601
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000026.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000026.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..38a396996fa621bf98174892a497557f1bce1063
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000026.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:640832be95df793db25916c5d302b54faf43cb71e6dcae762d1a71cf4e88a452
+size 181118
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000027.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000027.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..f7f524a80b32a170f24541aa64394963e3ea38d0
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000027.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa0c5bc82d74b09885a93354c2d7d319409f28bfafe6a2591ba795293d15b457
+size 181161
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000135.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000135.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..77a8a6725af1ab91ad9ede56feb232202c5ccbc4
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000135.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:82e21272082ed4db0c9e75f2221bac5179b008bb426bc0df92e15d0e73adae7a
+size 187901
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000179.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000179.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..1529e1056fa408f7314c42996366c5c32cd350c0
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000179.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4901b4143ed839862745a234dc5cd80013e9e0c0ccec93f4fd435c35d644c99
+size 189172
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000180.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000180.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..f6295a1555829878bfc705e0e0c1d3fef5f3fcc0
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000180.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf2b1a0cdc397d1ecdecf1e860516e1539099b6baaf7b8c5b7a981b297315f2e
+size 189671
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000270.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000270.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..eaa0a9c733ecd7f4fc9ee7aacab2bbc061be18aa
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000270.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c1df1946c649396e6c191b1b1eae16e75d8e0a67217ea5269f6e39c8b55bc55
+size 184070
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000405.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000405.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..b4463bf4c9cdab4b0afd0d1051683e4e5abaad48
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000405.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0764a1794ea6963658727433eab5c1cc8f240c2d182c8fb70ebb3c829d3ec47e
+size 186486
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000540.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000540.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..c0e3f041f5aa11d59fd5b41bf6c13602ccac46f5
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/2/000540.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6134348cf20cfa24eb22c240e4096f9540bd845a8144a56f93e1365b728a7468
+size 180981
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000000.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000000.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..bbc8e3005828c8c603626a2790ad645fe21539a9
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000000.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc8431511b066161d999013924132ec3bab0ce35749d900c725db7d680f94668
+size 99529
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000001.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000001.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..d24332f98b50cd07f2754aafebb20e91e1d873d6
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000001.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:718efb04aba8e7bcc55e3212e69249b3024a7bbde8f6f79180c3f965ccaf547a
+size 106816
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000026.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000026.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..140383e5c1b01157934af1a3550cbace2a0b26e3
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000026.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e2c4a31ab643fc0d590360a56d9ff7c45f2bc34ab4955c266f5c3e5cdf97b1b
+size 189717
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000027.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000027.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..c0c2ece8a9d39384bd2d4ade4333df70acda6aed
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000027.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7dfabde1f4cefca3d6e25edfe66483832b79ced844c4e9d2bb673677dfd94372
+size 189722
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000135.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000135.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..533d4d0d19d853c76106d1c522df510e19c54f29
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000135.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bedd383c87ac9c6887b7e00aa70437195f772cc2ed3fdbbb3bcb84fecade636d
+size 197935
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000179.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000179.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..087d32bc3b79d1e2111e9b790488ac80623d8e44
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000179.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c83fbd6b4baa3ffb9d5223cb0db52b2530897a34c8a01a4dad0d1bcfd95081bf
+size 199480
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000180.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000180.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..e8a28655f40826de16928cc0483761a24de9efb4
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000180.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:297c0589c490cf9a7daeafecf96c3e776c42eb5bdf1d2c5da3b6327ff779d482
+size 199663
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000270.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000270.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..2d7548b3b6de1b9d7ac553dfb58f555034d2c68c
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000270.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7111a25e4a1a74d9f4be7a193c6072d34128b6a35e38827cec7fd893ae1880a
+size 186269
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000405.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000405.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..d48969eb0c1d519b8b5d6a50053c0d1eb5966259
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000405.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c9b6e1966698317de506aad217dd77e32f61e7cb6325bebac201f9d0426a4a77
+size 197700
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000540.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000540.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..45092d9526c22ec15272f0723a60d6e28f65e3c8
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/3/000540.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:baf0f1d3daf7bf06113598aff7ef05418a1f6339ba4223d628c10751a30a5ca6
+size 180334
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000000.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000000.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..9416b988103f69b09582178ecf518247ddc62c95
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000000.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:65621b7003aa94945ee59356a0c0c11f74671072336f4c7426e3e9e7d3681a73
+size 97239
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000001.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000001.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..15508c49d12bfee8cef8c8252c74844d4748f725
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000001.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7d96ad0ff3eb33832566cbd1c3bb81f0fe9e3324e2f48e309a2d2d976cc7682
+size 103779
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000026.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000026.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..3d28c0ac971f2c2a2806d11ba431ced2977d3b9b
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000026.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d7e362ffa214b821fc5bd2bbb561b4a9b8d2d47b7e01419089da0843aed7d2b0
+size 195218
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000027.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000027.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..caf0c7e8570f3275d20008b72407089d38a7fde3
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000027.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ebebe87a0bc6a739c8a9c112df69fa34f931abd7a1b5f72c9cacabbbecb010fb
+size 195199
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000135.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000135.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..8d3dcd0e0bbb6454ac1124003ce601253c6a8eee
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000135.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e8d697b174acf0b770db9e84738322112f95ba3a3cc8d52adf7f2ba8bf2a16a
+size 192083
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000179.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000179.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..81e1ac052c2ded20c685849058f0e8d7b7596941
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000179.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:822ba7667754fb6167ba602f56dfd3e5090465ee09c102365d87a8b9e8a4a677
+size 190699
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000180.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000180.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..1460069ec8ddcb5c048ffa9ccfabbe6b29937f1b
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000180.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:899b77a9a3a61932737d06d0794c02c013a7440d122f0cd320fa4668eaff24d1
+size 190920
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000270.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000270.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..4e21493378ad838571ca5ea5b4e19849e0438096
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000270.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b2a270856dacaac872a6fa36fdb2c0e41c043fab9950566161fb101a8cb5cbb
+size 189958
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000405.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000405.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..689ac1a9b746ee7c99a5efef7eb71496f53e9897
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000405.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:703ece8f601336e052740c88ce7d05b7d43b0e640f2c98962bbf8eef84d6d65d
+size 192010
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000540.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000540.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..0c15e31b6381cb9f58877874351da43dcd2f75f2
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/4/000540.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:97332831e5343ba041b96bcf436ce25eb7572af95305807e70af346481ee2606
+size 189632
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000000.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000000.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..516f02c694c7696ff63f71e93cb7c6cac80f086b
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000000.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e44c1a6bbccf2f5946612c6fe29d55f285c902228dec24b917b4cac058a2b1ce
+size 97240
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000001.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000001.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..cb4184d36cbff2e5b457b85cc538122e9c235adb
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000001.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:75dc47a8b4f120b2e521aaf99feb797f64bfc1650e3c0d629bb68b9d6f5aae93
+size 103851
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000026.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000026.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..abdd0b913d8e883e9b898ac77341d7aacb596b09
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000026.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e3c3f3243060d46ad3db71d9d4b4c3b8f57826f45317bfc13fcdccbcd8c8652
+size 192242
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000027.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000027.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..5982251dd5a28b03606a72f803b154af80a1bb4a
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000027.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d899600406e77ce42e617a8d4cf429371e613844f26b08247ecb9f139fc23877
+size 192530
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000135.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000135.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..537ef9e5687d6a45d7f53a029bb33e7a5b823b3e
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000135.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61ff68f2b52e1c061a3082235e69d7fba83bb1df5497598717575133c9eab164
+size 188873
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000179.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000179.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..8abceb34642ea3583e722a2e77613a70a620c9d3
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000179.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:250a630c39cd3b85c3b6ab91fe424159d772217aa76f3b786f4d7b262e52e438
+size 186847
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000180.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000180.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..c45d2cb0b8be02e92c2c6ec675c92b8e66d7c9f6
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000180.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:45963e3bb7d8a33be0972708d7ff0646f698cf96559d150a4acd50a8cbeebc4d
+size 187074
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000270.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000270.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..37bf958f084ca88435109586ae866f382dbb6c15
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000270.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9052676b181f3d4081b967f383524c28ec40788baefaadfe2fc6064acbd0288a
+size 188857
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000405.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000405.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..7f1a6f137328b6dc6ae15cf730e94f202559af90
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000405.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f5ebada9fa27256b1f8e49699974a7fd678861af0327a3cdeef3a6e4461a2fc
+size 189112
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000540.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000540.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..ad9606fb49d46f2f4691089c1d8931dce9d22202
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/5/000540.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:63a4c4309214ee6ab22eba23c813b06400517d70ecdb34b8bbd0ecf484fbb116
+size 186775
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000000.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000000.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..5b0b33aa0e572a1d432d43a0bdc84fdc5db3e5ea
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000000.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:05e811e8406b38efedc9b8a1fbb7acd8657e067bc75eaee1d84c8853031d4050
+size 102793
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000001.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000001.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..e73b6639b9491848f9810fcba54ed0545db32b6e
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000001.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aaff6733d4028ebcc24225bc082a6317bbb4f746c660b7fb4128f2464c092f1e
+size 111730
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000026.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000026.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..6760c6e2cf20f82a105e96b53026be83f84d8aba
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000026.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:444ed3cf427238f8293418ae0a7340e8b9788feaa93a7cf4961400e78aebec69
+size 190233
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000027.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000027.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..a86abb4a800a4e644663d16b394001e4eff5b86d
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000027.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5354702b237a4a289f8978e3c82d0f65bb54957ab3539046f57a38cc0d5cd18a
+size 188953
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000135.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000135.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..ee3754b14477e8368a3aedf33253c06b695b4ef6
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000135.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:40022701accce2066151b7989c500d9a2cffa24cfad083465c2e277d8e0f17c2
+size 202918
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000179.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000179.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..ecd64f241161c19dfed69455ac8d173d667de589
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000179.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3fcace8e865297e14c77f50e406f2d34e4011bb345613845885df98fdbcf9455
+size 202549
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000180.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000180.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..5a7f65b25db6ace091002f54699b0a1a2c377be5
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000180.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26ca79133af5820f6272fc351491f042c3d7a535692d97c78c92d6110f49d181
+size 202501
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000270.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000270.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..2ed83514cf5d30123974df4414d4ac82f9c86965
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000270.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d0772b78d07f6eb0452e512edd9b27992c311f48db7ec7ba41e16cb0c35e469
+size 197568
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000405.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000405.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..61f0aa54105a904cb21c6b73ff86f48f2dd20151
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000405.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b175957070556c3eadd40117658f2ee8e427713544680a0fbd5d3ad71db091a2
+size 203750
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000540.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000540.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..94f4c2620be97165a1e6278abc08cd97664aff36
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/6/000540.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24913f939b4d590180318214d5c8e675cc0461bf99617d2d8c656abdddcf0617
+size 187782
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000000.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000000.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..9052664a376375d31dc9e96c6fc1775db2ae40ca
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000000.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:58390570665c370e353e8aa6868f909c9d75ff8f5b2e691d1a87c85a9e32c75a
+size 103924
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000001.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000001.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..41665e9d3709184fed0a41cbc2ba3dde11141e91
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000001.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e72d8543727bdb56d89af3448ffb04687d33e5a337566b713d53cd6469b08ee4
+size 108496
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000026.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000026.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..de1b7b2051b6209b1555b34c8f3e752cf1504f51
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000026.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18b35a5dc41b2cdfb6b782ef0a55fc47b6585b788b56e3be7908f167be69b660
+size 182050
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000027.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000027.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..dbd6094b58aa5ca849b0371595db6b980a203738
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000027.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:08ede355ee68ee918bb3d124e156c9150bbc10b65e8bdc13f5e02259c67930bc
+size 182525
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000135.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000135.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..4e7829aa84659a9d1a4992fb5a809e1888e7eb2c
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000135.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8eec29a79549be39bd2f0933b0a6caee33282fe14d36481f3c6fd37819f0ad3a
+size 188766
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000179.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000179.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..56cab5569a93039d4fd3845da638728abef230c1
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000179.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a7408e2cbc0bac13d3ffffb617b7a76ae7a792d6116aa5c05aaae7d54976bceb
+size 195612
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000180.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000180.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..119a69fdecb56e21b7a2f78500d33921e970a33f
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000180.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ab1b985f0e78434389b2b459e0feff87c175ceed4fb1be2fd8c0cace2bdf7d53
+size 195695
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000270.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000270.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..39dfcfecb6955693a8dbff874f6370e8472918d1
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000270.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37a5918c80533341f6ff793ea0021d1554d3f99635157218bbd673d31bed3cab
+size 187346
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000405.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000405.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..c5e63ea4f1360ba65a0a3c87720ee1c09a99b262
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000405.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca59ed2db80afa9afe651fa92284a7cf250c7098d41f5a69af3a68ef8c626f42
+size 188284
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000540.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000540.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..290e67d2148c6fd9c74c913eea9ad63c6dc67331
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/7/000540.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3343a56b287a710f030cbc0c0ead2e2101a64a34e087712120a903c5be28f97c
+size 183337
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000000.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000000.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..d37f59308fbdbae10e9a6a20a26cae1052944dcc
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000000.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:30e49fdf7ea1858892ed7cb0909c3a42c5e8dd28b0985464af9620035d9add65
+size 100551
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000001.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000001.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..3390e27baee64dc0915b872ec6e500e16157a147
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000001.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:423b5260e9679b1ce34a7588801f951fb15cbebd252a485e45124c41559695c7
+size 104687
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000026.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000026.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..b4a6aa1aaff5b90a40704a45c8d33bcd6eec271a
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000026.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1614990827cdc515a1bc39b48ed68e6a793bf59d9ec271373233b11ff8fd118
+size 199547
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000027.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000027.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..c7815b04ed424fa9ce440728d3e2d9c3759ce531
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000027.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e0b617c5a0de44745cfe6f51569d674fdfe55e57b3db6ab54949ddf807ba57e
+size 200253
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000135.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000135.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..e94c378377a385552c0026b3a4912b7e1b5bee79
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000135.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5ac944ece665b74d387d308f3636bdbaf9158cecaaade8b6131df034f1ee44a5
+size 199182
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000179.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000179.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..dc8956eda6f5d5bd4d16aa16425102da0d23b6c4
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000179.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:79e70d9ec4ddba2a917369e507170467bf03bd9b92a1553949d3293e16d54f3b
+size 197606
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000180.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000180.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..122b952c9c99a54f6534a468bbdd038ef83b855e
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000180.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:115109fede366082d546014ab3a256b8bc245e9e1dc4559f99c9019cbd8a01d6
+size 197444
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000270.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000270.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..92cf6884b875f1d28aeec56f013fd53c0009a466
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000270.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d35a062d61e110c055327e677a77f84e3c283c1a62f1e38c85e72f953e698a4b
+size 197946
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000405.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000405.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..07e9c561048cb353caf43750c42cadede79ac055
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000405.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca99354682b5773131cfbafbcd31d25fcd8330f131c6bccb5c4df5c8f835922e
+size 198085
diff --git a/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000540.jpg b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000540.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..8e9aebd036a742de67a8aae8d0898015faff1cda
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/video_snapshots/8/000540.jpg
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:274c6790b07f861b0968789d95ff4456c04572fca1a7bb3aa634a2b4a7d88214
+size 195918
diff --git a/ARCTIC/shared-v1.0/doc/validation/videos/0-full.mp4 b/ARCTIC/shared-v1.0/doc/validation/videos/0-full.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..a767ad4e69f50779074a47fdbbc983166bd95d42
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/videos/0-full.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8518eee01d4c19b7de8c9bda8e533cedfc2b83385d916a1811f8da1f0fcf1ea5
+size 20226206
diff --git a/ARCTIC/shared-v1.0/doc/validation/videos/1-full.mp4 b/ARCTIC/shared-v1.0/doc/validation/videos/1-full.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..f76d65eac94d0b24aa78d6758166a80bf65f6240
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/videos/1-full.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72c2a9ceb80e80ab169997ce80ad85590db918a8fde1dd2333332a9d2a91fdea
+size 3992332
diff --git a/ARCTIC/shared-v1.0/doc/validation/videos/2-full.mp4 b/ARCTIC/shared-v1.0/doc/validation/videos/2-full.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..d6850eea6e97939e0dcd82e007206ff33d819eac
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/videos/2-full.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1ad73fa0142ea2e2e1e711adc3a07cce7f47f14731ebe69da047cdd98fefbff9
+size 4102784
diff --git a/ARCTIC/shared-v1.0/doc/validation/videos/3-full.mp4 b/ARCTIC/shared-v1.0/doc/validation/videos/3-full.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..77190496c8b6eef7cae1dcbe420c138fc4faa26a
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/videos/3-full.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6c0c382e024e7cdcf9f5f558afbb7165bcbe0c191d8050e0c48c3196c5ab1cb
+size 6067451
diff --git a/ARCTIC/shared-v1.0/doc/validation/videos/4-full.mp4 b/ARCTIC/shared-v1.0/doc/validation/videos/4-full.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..c88857c3dd6d03d0bb36747bd86b1b4cdc94666a
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/videos/4-full.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1075de8dcabd33acd3d573dd2ec8aa99bcbd87b842091b6ff90f2db2b29cd911
+size 4928541
diff --git a/ARCTIC/shared-v1.0/doc/validation/videos/5-full.mp4 b/ARCTIC/shared-v1.0/doc/validation/videos/5-full.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..c8805ed367ee894eeedb1dbf8e50df12c843716f
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/videos/5-full.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c996198500c5bb70e549077b56af397cb2d2baec548c869caab41df28bbd5531
+size 4512296
diff --git a/ARCTIC/shared-v1.0/doc/validation/videos/6-full.mp4 b/ARCTIC/shared-v1.0/doc/validation/videos/6-full.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..139c6af28abe0de02afa391a96c4714039155378
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/videos/6-full.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:351a0254d2975ebcefbd2ccea6397f02caf3a67bd35182f23c6ccf369ab0b344
+size 6737732
diff --git a/ARCTIC/shared-v1.0/doc/validation/videos/7-full.mp4 b/ARCTIC/shared-v1.0/doc/validation/videos/7-full.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..f5f34d83bb382c99b8ffcabdafd4f9a813ec9588
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/videos/7-full.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ecd040b3647f7b40074b1533360527580299e7ccfcca5e4a33c3cd31cf683cbf
+size 4184096
diff --git a/ARCTIC/shared-v1.0/doc/validation/videos/8-full.mp4 b/ARCTIC/shared-v1.0/doc/validation/videos/8-full.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..12e03687b89e46f29b2cf5b1109d3650b00c7fae
--- /dev/null
+++ b/ARCTIC/shared-v1.0/doc/validation/videos/8-full.mp4
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:290455e814d21482bbe4dc3478adb8adab14120eadae0a0e9577360af5871877
+size 3837950
diff --git a/ARCTIC/shared-v1.0/script/README.md b/ARCTIC/shared-v1.0/script/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..1dfaee31c247eb3668e3153d3257066fc6587654
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/README.md
@@ -0,0 +1,2 @@
+运行条件与配置见 ../doc/RAW_TO_SHARED.md。
+模型权重不分发；日志与提取缓存不分发。
diff --git a/ARCTIC/shared-v1.0/script/accelerated_media.py b/ARCTIC/shared-v1.0/script/accelerated_media.py
new file mode 100644
index 0000000000000000000000000000000000000000..544cef82c46ec16a527ec56015cf30ab8c5b3dcc
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/accelerated_media.py
@@ -0,0 +1,128 @@
+"""All JPEG packets verified; deterministic RGB sample verification without re-encoding."""
+from pathlib import Path
+import hashlib,io,json,random,subprocess,time,zipfile
+import fresh_sequence_common as c
+
+POLICY='jpeg-packets-all-rgb-sample16/v1'
+
+def indices_for(sequence,camera,count):
+    assert count>0,'Empty media sequence'
+    seed=int.from_bytes(hashlib.sha256(f'{sequence}/{camera}/20260920'.encode()).digest()[:8],'big')
+    indices={0,count-1,count//4,count//2,3*count//4}
+    rng=random.Random(seed)
+    remaining=[i for i in range(count) if i not in indices]
+    indices.update(rng.sample(remaining,min(len(remaining),max(0,16-len(indices)))))
+    return sorted(indices),seed
+
+def rgb_hashes(args,width,height,count):
+    """Same FFmpeg rgb24 decode, accelerated SHA256 via hashlib/OpenSSL; bounded RAM."""
+    cmd=[str(c.FF),'-v','error','-threads','2','-noautorotate',*map(str,args),'-map','0:v:0','-pix_fmt','rgb24','-threads','2','-fps_mode','passthrough','-f','rawvideo','pipe:1']
+    p=subprocess.Popen(cmd,stdout=subprocess.PIPE,stderr=subprocess.PIPE)
+    size=width*height*3;rows=[]
+    try:
+        for i in range(count):
+            block=p.stdout.read(size);assert len(block)==size,('early RGB EOF',i,count)
+            rows.append((size,hashlib.sha256(block).hexdigest()))
+        assert not p.stdout.read(1),'Unexpected extra RGB frames'
+        error=p.stderr.read();assert p.wait()==0,error.decode(errors='replace')[-2000:]
+    finally:
+        if p.poll() is None:p.kill();p.wait()
+        p.stdout.close();p.stderr.close()
+    return rows
+
+def zip_members(path,names):
+    with zipfile.ZipFile(path) as z:
+        for name in names:yield name,z.read(name)
+
+def file_members(paths):
+    for path in paths:yield path.name,path.read_bytes()
+
+def view_members(p,m,frames):
+    source=m['components']['rgb']['source_locator'];base=c.DATA_ROOT/m['dataset_id']
+    if 'archive' in source:
+        return zip_members(base/source['archive'],[f['source_locator']['member'] for f in frames])
+    paths=sorted((base/'tmp'/c.RUN/m['sequence_id']/'images').glob(m['camera_id']+'-*.jpg'))
+    assert len(paths)==len(frames)
+    return file_members(paths)
+
+def verify(p,m,members,tmp,write=False,full_pixels=False):
+    start=time.perf_counter();tmp.mkdir(parents=True,exist_ok=True)
+    prefix=m['sequence_id']+'__'+m['camera_id'];total=m['num_frames'];W,H=m['camera']['image_size']
+    sample,seed=indices_for(m['sequence_id'],m['camera_id'],total)
+    if full_pixels:sample=list(range(total))
+    chosen=set(sample);spool=tmp/(prefix+'.mjpeg');sample_source=tmp/(prefix+'.sample-source.mjpeg');sample_output=tmp/(prefix+'.sample-output.mjpeg')
+    hashes=[];modes=set();created=[sample_source,sample_output]
+    sink=spool.open('wb') if write else None
+    if write:created.append(spool)
+    try:
+        with sample_source.open('wb') as selected:
+            for i,(name,content) in enumerate(members):
+                assert content[:2]==b'\xff\xd8',name
+                with c.Image.open(io.BytesIO(content)) as image:
+                    assert list(image.size)==[W,H],(name,image.size,[W,H])
+                    modes.add(image.mode)
+                hashes.append((len(content),hashlib.sha256(content).hexdigest()))
+                if sink:sink.write(content)
+                if i in chosen:selected.write(content)
+        if sink:sink.close();sink=None
+        assert len(hashes)==total
+        source_s=time.perf_counter()-start
+        dst=p/'observation/rgb.mkv';dst.parent.mkdir(parents=True,exist_ok=True)
+        if write:c.command([c.FF,'-v','error','-y','-f','image2pipe','-framerate','1','-c:v','mjpeg','-i',spool,'-map','0:v:0','-c:v','copy',dst])
+        mux_s=time.perf_counter()-start-source_s
+        packet=c.framehash(['-i',str(dst)],copy=True);assert packet==hashes,'JPEG count/order/payload changed'
+        packet_s=time.perf_counter()-start-source_s-mux_s
+        if full_pixels:
+            c.command([c.FF,'-v','error','-y','-i',dst,'-map','0:v:0','-c:v','copy','-f','image2pipe',sample_output])
+        else:
+            # n is the zero-based packet index. No bitstream noise is added.
+            keep='+'.join(f'eq(n,{i})' for i in sample)
+            c.command([c.FF,'-v','error','-y','-i',dst,'-map','0:v:0','-c:v','copy','-bsf:v',f"noise=amount=0:drop='not({keep})'",'-f','image2pipe',sample_output])
+        sample_args=['-f','image2pipe','-framerate','1','-c:v','mjpeg','-i',str(sample_output)]
+        selected_packets=c.framehash(sample_args,copy=True)
+        assert selected_packets==[hashes[i] for i in sample],'Sample extraction changed bytes or indices'
+        native_args=['-f','image2pipe','-framerate','1','-c:v','mjpeg','-i',str(sample_source)]
+        before=rgb_hashes(native_args,W,H,len(sample))
+        # Decode directly from the MKV so container-level color/rotation semantics are tested.
+        if full_pixels:
+            after=rgb_hashes(['-i',str(dst)],W,H,total)
+        else:
+            # MJPEG is all-intra. Seek by index-only 1fps PTS; validate the selected
+            # packet at each seek before RGB comparison to guard off-by-one seeking.
+            after=[]
+            for i in sample:
+                seek=['-ss',str(i),'-i',str(dst)]
+                check=c.command([c.FF,'-v','error',*seek,'-map','0:v:0','-frames:v','1','-c:v','copy','-f','framehash','-hash','sha256','pipe:1']).decode()
+                rows=[s for s in check.splitlines() if s and not s.startswith('#')]
+                assert len(rows)==1 and (int(rows[0].split(',')[-2]),rows[0].split(',')[-1].strip())==hashes[i]
+                after.extend(rgb_hashes([*seek,'-frames:v','1'],W,H,1))
+        assert before==after,'Decoded RGB differs'
+        info=subprocess.run([str(c.FF),'-hide_banner','-i',str(dst)],capture_output=True).stderr.decode(errors='replace')
+        stream=next(s.strip() for s in info.splitlines() if 'Video: mjpeg' in s)
+        record={'test':'all_source_JPEG_packets_exact','passed':True,'frames':total,'policy':POLICY,'packet_manifest_sha256':hashlib.sha256(json.dumps(hashes).encode()).hexdigest(),'all_JPEG_dimensions_checked':True,'source_JPEG_modes':sorted(modes),'output_bytes':dst.stat().st_size}
+        rgb={'test':'source_RGB_pixels_exact','passed':True,'scope':'all_frames' if full_pixels else 'deterministic_sample','frames':total,'sampled_frames':len(sample),'frame_indices':sample,'seed':seed,'seed_method':'first 8 bytes SHA256(sequence/camera/20260920), big-endian','pixel_manifest_sha256':hashlib.sha256(json.dumps(after).encode()).hexdigest(),'hash_implementation':'hashlib.sha256/OpenSSL','decoded_from_container':True}
+        timings={'source_read_hash_s':source_s,'mux_s':mux_s,'packet_check_s':packet_s,'RGB_check_s':time.perf_counter()-start-source_s-mux_s-packet_s,'total_s':time.perf_counter()-start}
+        return record,rgb,stream,hashes,after,timings
+    finally:
+        if sink:sink.close()
+        for path in created:
+            assert path.resolve().is_relative_to(tmp.resolve())
+            if path.exists():path.unlink()
+
+def write_video(view,members):
+    assert c.shutil.disk_usage(c.DATA_ROOT).free>400_000_000_000,'Disk reserve reached; raw retained'
+    record,rgb,stream,packets,pixels,timings=verify(view.p,view.m,members,view.tmp/'accelerated-media',write=True)
+    pixel=c.re.search(r'Video: mjpeg[^,]*,\s*(\w+)',stream).group(1);version=c.command([c.FF,'-version']).decode().splitlines()[0]
+    view.component('rgb','observation/rgb.mkv','rgb/uint8/v1',['T','height','width','channel'],c.np.ones(view.T,bool),source=view.source,
+        method='Original JPEG packets copied into MKV; every payload/order/dimension checked; declared deterministic RGB sample checked',
+        storage={'profile':'mkv/mjpeg_copy/v1','codec':'mjpeg','pixel_format':pixel,'encoded_size':view.m['camera']['image_size'],'source_lossy':True,
+            'decoder':{'name':'ffmpeg','version':version,'output_pixel_format':'rgb24','autorotate':False,'source_color_declaration':'unknown','actual_stream_description':stream},'encoder_parameters':{'mode':'stream_copy','version':version}},storage_fps=1,timebase_role='index_only')
+    view.checks.extend([record,rgb]);view.m['media_validation_policy']=POLICY
+    c.save(view.tmp/'media_checks'/f'{view.m["sequence_id"]}__{view.m["camera_id"]}.json',{'packets':packets,'pixel_sample':{'indices':rgb['frame_indices'],'values':pixels},'policy':POLICY,'timings':timings})
+    print('MEDIA_CHECK_COMPLETE',view.m['sequence_id'],view.m['camera_id'],round(timings['total_s'],2),flush=True)
+
+def revalidate(p,full_pixels=False):
+    m=c.load(p/'metainfo.json');frames=[json.loads(l) for l in (p/'camera/frames.jsonl').read_text().splitlines()]
+    tmp=c.DATA_ROOT/m['dataset_id']/'tmp'/c.RUN/'accelerated-revalidation'
+    rec,rgb,_,_,_,timings=verify(p,m,view_members(p,m,frames),tmp,full_pixels=full_pixels)
+    return {'sequence_id':m['sequence_id'],'camera_id':m['camera_id'],'checks':[rec,rgb],'timings':timings}
diff --git a/ARCTIC/shared-v1.0/script/acceleration-config.json b/ARCTIC/shared-v1.0/script/acceleration-config.json
new file mode 100644
index 0000000000000000000000000000000000000000..a5f795f4934528f59a1e9562b59907fdcb1432dc
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/acceleration-config.json
@@ -0,0 +1,12 @@
+{
+  "media_policy": "jpeg-packets-all-rgb-sample16/v1",
+  "workers_by_dataset": {
+    "ARCTIC": 3,
+    "HOT3D": 2
+  },
+  "torch_threads": 4,
+  "max_heavy_dataset_workers": 1,
+  "selection_method": "Median of comparable runs; exclude HOT3D initial cold-file read and use its repeated 1-worker trial; choose fewer workers if within 3% of fastest measured; >=1 GB observed available memory",
+  "authorized_by": "User requested stopping old production and switching to accelerated version with measured optimal parallelism",
+  "created_at": "2026-09-20T22:27:01.369591+08:00"
+}
\ No newline at end of file
diff --git a/ARCTIC/shared-v1.0/script/apply_hot3d_qa.py b/ARCTIC/shared-v1.0/script/apply_hot3d_qa.py
new file mode 100644
index 0000000000000000000000000000000000000000..942ced8b3d5b27195bd14b959a20d07e92c5ed9b
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/apply_hot3d_qa.py
@@ -0,0 +1,65 @@
+"""Apply native HOT3D manual QA to supervision validity, preserving raw diagnostics."""
+from fresh_sequence_common import *
+
+VIDEO_RUN='full-video-review-v2'
+SEQ='P0015_179e1b84'
+
+def csvrows(z,name):return list(csv.DictReader(io.StringIO(z.read(name).decode())))
+
+def apply_qa(sequence=None,publish=True):
+    seq=sequence or SEQ
+    base=DATA_ROOT/'HOT3D';root=base/RELEASE;src=base/'raw/data/Aria'/seq
+    gtfile=next(src.glob('*ground_truth.zip'));qa_report={}
+    with zipfile.ZipFile(gtfile) as z:
+        masks=csvrows(z,'masks/mask_qa_pass.csv');hands=csvrows(z,'box2d_hands.csv');objects=csvrows(z,'box2d_objects.csv')
+    for meta in sorted((root/'data'/seq).glob('*/metainfo.json')):
+        p=meta.parent;m=load(meta);cam=m['camera_id'];T=m['num_frames'];frames=[json.loads(x) for x in (p/'camera/frames.jsonl').read_text().splitlines()]
+        cache=base/'tmp'/VIDEO_RUN/'diagnostic'/(cam if seq==SEQ else seq+'/'+cam)
+        if (cache/'snapshot.json').exists():assert Path(load(cache/'snapshot.json')['source_view']).resolve()==p.resolve()
+        if not (cache/'snapshot.json').exists():
+            cache.mkdir(parents=True,exist_ok=True)
+            for file in p.rglob('*.pt'):
+                target=cache/file.relative_to(p);target.parent.mkdir(parents=True,exist_ok=True);shutil.copyfile(file,target)
+            save(cache/'metainfo.json',m);save(cache/'snapshot.json',{'purpose':'pre-QA native parameter diagnostics only; not shared usable labels','source_view':str(p),'frames':T})
+        lookup={int(r['timestamp[ns]']):r['mask']=='True' for r in masks if r['stream_id']==cam}
+        qa=np.array([lookup[f['timecode_ns']] for f in frames],bool)
+        source={'archive':gtfile.relative_to(base).as_posix(),'member':'masks/mask_qa_pass.csv','stream_id':cam,'time_index_ref':'camera/frames.jsonl:timecode_ns'}
+        policy={'semantics':'usable_annotation','formula':'source_annotation_present AND source_numerically_valid AND official_manual_qa_pass',
+            'qa_scope':'all visible hand/object poses in this camera frame; conservatively exclude all hand/object supervision if false',
+            'source_locator':source,'no_qa_source_rule':'use the component source-validity rules; do not invent QA'}
+        for key,c in m['components'].items():
+            # Camera calibration and RGB are not labels covered by this manual hand/object QA flag.
+            if key=='mano' or c['schema_id'] in ['object/rigid/v1','object/joint_state/v1','object_keypoints/camera/v1']:
+                values=torch.load(cache/c['data_ref'],weights_only=True).numpy();present=torch.load(cache/c['valid_ref'],weights_only=True).numpy()
+                valid=present & qa.reshape((T,)+(1,)*(present.ndim-1));values=values.copy();values[~valid]=np.nan
+                tensor(p/c['data_ref'],values);tensor(p/c['valid_ref'],valid);c['valid_policy']=policy
+            for attrname,a in c.get('attributes',{}).items():
+                if attrname!='visibility':continue
+                ishand=key=='mano';oid=c['source_locator'].get('object_uid');ids=[0,1] if ishand else [oid]
+                rows=hands if ishand else objects;identity='hand_index' if ishand else 'object_uid'
+                native={(int(r['timestamp[ns]']),str(r[identity])):float(r['visibility_ratio[%]']) for r in rows if r['stream_id']==cam and r['visibility_ratio[%]']!=''}
+                values=np.array([[native.get((f['timecode_ns'],str(k)),np.nan) for k in ids] for f in frames],np.float32)
+                if not ishand:values=values[:,0]
+                present=np.isfinite(values);assert ((values[present]>=0)&(values[present]<=1)).all()
+                # Correct the diagnostic copy as well: these values are fractions, despite the source header.
+                tensor(cache/a['data_ref'],values);tensor(cache/a['valid_ref'],present)
+                valid=present & qa.reshape((T,)+(1,)*(present.ndim-1));filtered=values.copy();filtered[~valid]=np.nan
+                tensor(p/a['data_ref'],filtered);tensor(p/a['valid_ref'],valid)
+                a['provenance']={'kind':'source','method':'native [0,1] ratio retained; official loader does not divide misleading [%] header by 100'}
+                a['valid_policy']={'semantics':'usable_visibility_annotation','formula':'source_ratio_present AND official_manual_qa_pass','source_locator':source}
+                a['source_locator']['match_method']='exact'
+        m['validity_revision']='usable-annotation-2026-09-20'
+        m['split']='train';m['split_scheme']='HOT3D_official_participant_split';m.pop('split_reason',None)
+        m['split_source_locator']={'url':'https://github.com/facebookresearch/hot3d/blob/146b34afef8c1a32adeef7e981c070109f225c87/README.md#testtrain-split','participant_id':seq.split('_')[0]}
+        assert seq.split('_')[0] not in {'P0004','P0005','P0006','P0008','P0016','P0020'}
+        save(meta,m)
+        qa_report[cam]={'frames':T,'qa_pass':qa.tolist(),'passed_count':int(qa.sum()),'source_locator':source,
+            'interpretation':'manual QA of visible hand/object poses; not RGB validity or individual error localization'}
+        print('QA_APPLIED',cam,'pass',int(qa.sum()),'fail',int((~qa).sum()),flush=True)
+    save(root/'doc/validation/source_qa.json' if publish else base/'tmp'/RUN/'qa'/f'{seq}.json',qa_report)
+    if publish:
+        for f in ['apply_hot3d_qa.py','fresh_sequence_common.py','convert_fresh_sequence_pilots.py']:
+            shutil.copyfile(WORK/'scripts'/f,root/'script'/f)
+    return qa_report
+
+if __name__=='__main__':apply_qa()
diff --git a/ARCTIC/shared-v1.0/script/arctic_object_review.py b/ARCTIC/shared-v1.0/script/arctic_object_review.py
new file mode 100644
index 0000000000000000000000000000000000000000..bbc6d1df5e8c58c30e2d689924b85af85962a822
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/arctic_object_review.py
@@ -0,0 +1,224 @@
+"""Eleven-object, complete-sequence review. No full-dataset conversion or upload.
+
+Raw archives stay immutable. Two bounded camera workers, sequence-scoped caches.
+"""
+from fresh_sequence_common import *
+from convert_fresh_sequence_pilots import arctic
+from diagnose_arctic_review import official_object, OFFICIAL
+from review_fresh_sequence_pilots import arr, validate_view
+from render_full_sequence_review import render_view, asset_preview
+import concurrent.futures, html
+
+BASE=DATA_ROOT/'ARCTIC';OUT=BASE/RELEASE
+TMP=BASE/'tmp/object-review-11';DOC=OUT/'doc/validation/objects'
+SELECTION=WORK/'tmp/arctic-object-selection.json'
+torch.set_num_threads(3)
+
+def context(item):
+    seq=item['sequence'];sid,action=seq.split('/');oid=item['object']
+    return seq,sid,oid,seq.replace('/','_'),DOC/oid,TMP/seq.replace('/','_')
+
+def prepare(item):
+    seq,sid,oid,seqid,doc,tmp=context(item);doc.mkdir(parents=True,exist_ok=True);tmp.mkdir(parents=True,exist_ok=True)
+    target=tmp/'reference.npz'
+    if target.exists() and (doc/'reference.json').exists():return
+    with zipfile.ZipFile(BASE/'raw/data/meta.zip') as z:
+        pref=f'meta/object_vtemplates/{oid}/';params=json.loads(z.read(pref+'object_params.json'));parts=np.asarray(json.loads(z.read(pref+'parts.json')),np.int64)
+        vertices=np.array([[float(x) for x in r.split()[1:4]] for r in z.read(pref+'mesh.obj').decode().splitlines() if r.startswith('v ')])/1000
+        misc=json.loads(z.read('meta/misc.json'))[sid]
+    with zipfile.ZipFile(BASE/'raw/data/raw_seqs.zip') as z:
+        objects=native_npy_bytes(z.read(f'raw_seqs/{seq}.object.npy'));hand=native_npy_bytes(z.read(f'raw_seqs/{seq}.mano.npy'));ego=native_npy_bytes(z.read(f'raw_seqs/{seq}.egocam.dist.npy'))
+    with zipfile.ZipFile(BASE/'raw/data/images_zips'/f'{seq}.zip') as z:
+        names=[n for n in z.namelist() if n.endswith('.jpg')];cams=sorted({n.split('/')[0] for n in names});assert cams==list(map(str,range(9)))
+        image_counts={c:sum(n.startswith(c+'/') for n in names) for c in cams}
+    assert len(parts)==len(vertices) and set(np.unique(parts))=={0,1} and objects.shape[1]==7
+    world=[];keypoints=[]
+    for start in range(0,len(objects),24):
+        v,k=official_object(objects[start:start+24],vertices,parts,params,oid);world.append(v);keypoints.append(k)
+    world=np.concatenate(world);keypoints=np.concatenate(keypoints)
+    # Deliberately wrong part IDs, angle sign, and units must diverge from upstream.
+    i=int(np.argmax(objects[:,0]));o=objects[i];R=Rotation.from_rotvec(o[1:4]);H=Rotation.from_rotvec([0,0,-o[0]])
+    negative={}
+    for name,selector,angle in [('swapped_parts',parts==1,-o[0]),('reversed_joint_angle',parts==0,o[0])]:
+        v=vertices.copy();v[selector]=Rotation.from_rotvec([0,0,angle]).apply(v[selector]);v=R.apply(v)+o[4:7]/1000
+        negative[name]=float(np.linalg.norm(v-world[i],axis=-1).max());assert negative[name]>1e-4,(oid,name)
+    negative['millimetres_as_metres']=float(np.linalg.norm(world[i]*1000-world[i],axis=-1).max());assert negative['millimetres_as_metres']>1
+    hands=[]
+    for side in ['left','right']:
+        model=Mano(side);h=hand[side];pose=np.c_[h['rot'],h['pose']];pose[:,3:]+=model.data['hands_mean'];v=[]
+        for start in range(0,len(pose),64):
+            q=pose[start:start+64];v.append(model.geometry(q,np.broadcast_to(h['shape'],(len(q),10)),h['trans'][start:start+64])[0])
+        hands.append(np.concatenate(v).astype(np.float32))
+    np.savez(target,object_world=world,keypoints_world=keypoints,hand_left=hands[0],hand_right=hands[1],vertices=vertices,parts=parts,
+        object_params=objects,world2cam=np.asarray(misc['world2cam']),Kexo=np.asarray(misc['intris_mat']),Rego=ego['R_k_cam_np'],Tego=ego['T_k_cam_np'],Kego=ego['intrinsics'],Dego=ego['dist8'])
+    save(doc/'reference.json',{'object':oid,'sequence':seq,'annotation_frames':len(objects),'image_frames_per_camera':image_counts,'vertices':len(vertices),'ioi_offset':misc['ioi_offset'],
+        'angle_min_max_rad':[float(objects[:,0].min()),float(objects[:,0].max())],'negative_control_max_error_m':negative,
+        'reference_method':'Pinned official ObjectTensors.forward executed with raw state; SciPy quaternion primitives, all vertices and annotation frames; raw world MANO reconstructed separately.',
+        'official_code':'https://github.com/zc-alexfan/arctic/blob/49f3eb4b7f01663172576357bb3d95695d7a7689/common/object_tensors.py',
+        'official_code_sha256':digest(OFFICIAL/'common/object_tensors.py'),'scope_limit':'Comparison proves faithful conversion, not source-to-image accuracy. Human acceptance remains pending.'})
+    print('REFERENCE_READY',oid,len(objects),len(vertices),flush=True)
+
+def check_geometry(item,p):
+    seq,sid,oid,seqid,doc,tmp=context(item);m,frames,structure=validate_view(p);cam=m['camera_id'];ref=load(doc/'reference.json')
+    with np.load(tmp/'reference.npz') as loaded:a={k:loaded[k] for k in loaded.files}
+    ix=np.array([int(f['source_frame_id'])-ref['ioi_offset'] for f in frames]);known=(ix>=0)&(ix<len(a['object_world']))
+    assert [f['source_annotation_index'] for f in frames]==[int(j) if ok else None for j,ok in zip(ix,known)]
+    E=arr(p/'camera/extrinsic.pt');K=arr(p/'camera/intrinsic.pt');op=p/'annotation/object'/oid;meta=load(op/'metainfo.json')
+    vertices=np.array([[float(x) for x in row.split()[1:4]] for row in (p/meta['mesh_ref']).read_text().splitlines() if row.startswith('v ')])
+    assert np.max(np.abs(vertices-a['vertices']))<1e-10
+    part=arr(p/meta['vertex_part_ids_ref']);assert np.array_equal(part,1-a['parts'])
+    expectedE=np.broadcast_to(np.eye(4),(len(frames),4,4)).copy()
+    if cam=='0':
+        expectedE[:]=np.nan;expectedE[known]=np.eye(4);expectedE[known,:3,:3]=a['Rego'][ix[known]];expectedE[known,:3,3]=a['Tego'][ix[known]].reshape(-1,3)
+        expectedK=a['Kego'];assert np.max(np.abs(arr(p/'camera/distortion.pt')-a['Dego']))<1e-5
+    else:expectedE[:]=a['world2cam'][int(cam)-1];expectedK=a['Kexo'][int(cam)-1]
+    assert np.max(np.abs(E[known]-expectedE[known]))<2e-6 and np.max(np.abs(K-expectedK))<.002
+    pose=arr(op/'pose.pt');top=arr(op/'parts/top/pose.pt');joint=arr(op/'joint_state.pt');valid=arr(op/'pose_valid.pt');kp=arr(op/'keypoints/xyz.pt')
+    errors={'object_vertices_m':0.,'object_keypoints_m':0.,'hand_vertices_m':0.};angleerror=float(np.max(np.abs(joint[valid,0]-a['object_params'][ix[valid],0])))
+    assert angleerror<1e-6
+    articulation=load(op/'articulation.json');assert articulation['joints'][0]['axis']==[0,0,-1]
+    for i in np.flatnonzero(valid):
+        v=vertices@pose[i,:3,:3].T+pose[i,:3,3];sel=part==1;v[sel]=vertices[sel]@top[i,:3,:3].T+top[i,:3,3]
+        golden=a['object_world'][ix[i]]@expectedE[i,:3,:3].T+expectedE[i,:3,3]
+        golden_k=a['keypoints_world'][ix[i]]@expectedE[i,:3,:3].T+expectedE[i,:3,3]
+        errors['object_vertices_m']=max(errors['object_vertices_m'],float(np.linalg.norm(v-golden,axis=-1).max()))
+        errors['object_keypoints_m']=max(errors['object_keypoints_m'],float(np.linalg.norm(kp[i]-golden_k,axis=-1).max()))
+    motion=arr(p/'annotation/mano/motion.pt');hv=arr(p/'annotation/mano/motion_valid.pt')
+    for side,name in enumerate(['left','right']):
+        model=Mano(name);indices=np.flatnonzero(hv[:,side])
+        for start in range(0,len(indices),64):
+            ii=indices[start:start+64];v=motion[ii,side];actual=model.geometry(v[:,:48],v[:,48:58],v[:,58:])[0]
+            golden=a['hand_'+name][ix[ii]]@expectedE[ii,:3,:3].transpose(0,2,1)+expectedE[ii,None,:3,3]
+            errors['hand_vertices_m']=max(errors['hand_vertices_m'],float(np.linalg.norm(actual-golden,axis=-1).max()))
+    assert max(errors.values())<2e-5,(oid,cam,errors)
+    # Separate top/bottom/textured meshes have their own topology; validate their
+    # native canonical coordinates independently, never reuse mesh.obj part IDs.
+    asset_checks=[]
+    with zipfile.ZipFile(BASE/'raw/data/meta.zip') as z:
+        for name in ['mesh.obj','mesh_tex.obj','top.obj','bottom.obj']:
+            source=z.read(f'meta/object_vtemplates/{oid}/{name}').decode().splitlines();dest=(op/'assets'/name).read_text().splitlines()
+            sv=np.array([[float(x) for x in r.split()[1:4]] for r in source if r.startswith('v ')])/1000
+            dv=np.array([[float(x) for x in r.split()[1:4]] for r in dest if r.startswith('v ')])
+            assert sv.shape==dv.shape and np.max(np.abs(sv-dv))<1e-10
+            assert [r for r in source if r.startswith(('f ','vt ','vn '))]==[r for r in dest if r.startswith(('f ','vt ','vn '))]
+            asset_checks.append({'mesh':name,'vertices':len(sv),'canonical_scale_topology_UV_normals':'passed'})
+    result={'camera_id':cam,'sequence_id':seqid,'passed':True,'max_errors':errors,'tolerance_m':2e-5,'checked_object_frames':int(valid.sum()),
+        'checked_vertices_per_frame':len(vertices),'mano_frames_per_side':hv.sum(0).tolist(),'asset_checks':asset_checks,'joint_angle_error_rad':angleerror}
+    save(doc/'checks'/f'{cam}.json',result)
+    return result
+
+def snapshots_for(item,p):
+    oid=item['object'];q=arr(p/f'annotation/object/{oid}/joint_state.pt')[:,0];T=len(q);finite=np.flatnonzero(np.isfinite(q));later=finite[finite>=min(12,T//10)]
+    ix=[0,1,T-1,*np.linspace(2,T-2,13,dtype=int).tolist(),int(later[np.argmin(q[later])]),int(later[np.argmax(q[later])])]
+    return sorted(set(i for i in ix if 0<=i<T))
+
+def worker(item,cam):
+    seq,sid,oid,seqid,doc,tmp=context(item);p=OUT/'data'/seqid/cam;resultfile=doc/'checks'/f'{cam}-complete.json'
+    if resultfile.exists():
+        result=load(resultfile)
+        if (doc/result['video']['video_ref']).exists() and digest(doc/result['video']['video_ref'])==result['video']['video_sha256']:
+            print('RESUME_CAMERA',oid,cam,flush=True);return
+    if shutil.disk_usage(BASE).free<400_000_000_000:raise RuntimeError('E free space below 400 GB')
+    arctic(seq,[cam],initialize=False)
+    numeric=check_geometry(item,p);indices=snapshots_for(item,p)
+    video=render_view(p,'ARCTIC',None,doc=doc,tmp=tmp/'render',snapshot_indices=indices)
+    save(resultfile,{'object':oid,'sequence':seq,'camera_id':cam,'numeric':numeric,'video':video,'visual_sample_indices':indices,'human_review':'pending'})
+    print('CAMERA_DONE',oid,cam,flush=True)
+
+def contact_sheets(item,results):
+    seq,sid,oid,seqid,doc,tmp=context(item);files=[]
+    for r in results:
+        cam=r['camera_id'];p=OUT/'data'/seqid/cam;q=arr(p/f'annotation/object/{oid}/joint_state.pt')[:,0]
+        picks=r['visual_sample_indices'];canvas=Image.new('RGB',(4*480,((len(picks)+3)//4)*285),(20,20,20));draw=ImageDraw.Draw(canvas)
+        for cell,i in enumerate(picks):
+                row,col=divmod(cell,4)
+                im=Image.open(doc/'video_snapshots'/cam/f'{i:06d}.jpg').convert('RGB');a=np.array(im)
+                # Identify colored rendered geometry to crop both original and
+                # overlay by the same region. This is a review crop, no data edit.
+                mid=a[40:552,512:1024];diff=np.abs(mid.astype(int)-a[40:552,:512].astype(int)).max(2)>35
+                yy,xx=np.where(diff)
+                if len(xx):
+                    x0=max(0,int(xx.min())-25);x1=min(512,int(xx.max())+26);y0=max(0,int(yy.min())-25);y1=min(512,int(yy.max())+26)
+                else:x0,y0,x1,y1=0,0,512,512
+                for side,xoff in enumerate([0,512]):
+                    crop=im.crop((xoff+x0,40+y0,xoff+x1,40+y1));crop.thumbnail((236,246))
+                    canvas.paste(crop,(col*480+side*240+(236-crop.width)//2,row*285+30))
+                draw.text((col*480+6,row*285+5),f'{oid} cam {cam} idx {i} q={q[i]:.2f}: RGB | projection',fill='white')
+        file=doc/f'contact-camera-{cam}.jpg';canvas.save(file,quality=94);files.append(file.name)
+    return files
+
+def object_page(item):
+    seq,sid,oid,seqid,doc,tmp=context(item);cameras=load(SELECTION)['cameras'];results=[load(doc/'checks'/f'{c}-complete.json') for c in cameras]
+    contacts=contact_sheets(item,results)
+    asset=asset_preview(OUT/'data'/seqid/cameras[0]/'annotation/object'/oid,doc/'assets'/f'{oid}.mp4')
+    worst=max(max(r['numeric']['max_errors'].values()) for r in results)
+    ref=load(doc/'reference.json');dark={r['camera_id']:r['video']['exposure_diagnostics']['suspect_dark_frame_indices'] for r in results}
+    report={'object':oid,'sequence':seq,'camera_count':len(cameras),'camera_frames':sum(r['video']['frames'] for r in results),'full_frame_numeric_checks':'passed','maximum_geometry_error_m':worst,
+        'reference':ref,'views':results,'contact_sheets':contacts,'asset_preview':asset,'agent_visual_review':{'status':'pending'},'human_review':'pending',
+        'dark_candidates':dark,'valid_policy_note':'Existing laptop reviewed exposure exclusion is retained. New sequences preserve source availability; dark candidates and suspicious alignment are reported, not silently blacklisted.'}
+    save(doc/'report.json',report);write_object_html(report)
+    print('OBJECT_READY',oid,report['camera_frames'],worst,flush=True)
+    return report
+
+STYLE='body{font:16px/1.6 system-ui;max-width:1550px;margin:24px auto;padding:0 20px;background:#f2f5f9;color:#162137}section,header{background:white;padding:20px;margin:18px 0;border-radius:10px}video,img{max-width:100%}table{border-collapse:collapse;width:100%}td,th{border:1px solid #ccd3de;padding:8px;text-align:left}.warn{color:#9b4414}a{color:#1265a7}'
+
+def write_object_html(r):
+    doc=DOC/r['object'];status=r['agent_visual_review'];cards=''
+    review_text=html.escape(status.get('summary',status['status']))
+    if status.get('notes'):review_text+='<ul>'+''.join('<li>'+html.escape(n)+'</li>' for n in status['notes'])+'</ul>'
+    for v in r['views']:
+        vv=v['video'];cards+=f'<section><h2>Camera {v["camera_id"]} · {vv["frames"]}帧</h2><video controls preload="metadata" src="{vv["video_ref"]}"></video><p>几何数值检查通过；暗帧候选索引：{r["dark_candidates"][v["camera_id"]]}。有效MANO左右：{vv["mano_valid_counts"]}。</p></section>'
+    page=f'<!doctype html><html lang="zh-CN"><meta charset="utf-8"><title>ARCTIC {r["object"]} 完整序列验收</title><style>{STYLE}</style><header><a href="../index.html">全部物体</a><h1>{r["object"]} · {r["sequence"]}</h1><p>固定相机6、{r["camera_frames"]}个相机帧，全部已渲染。最大几何数值差{r["maximum_geometry_error_m"]:.3g}米。<a href="report.json">完整结果</a> · <a href="reference.json">官方对照及反例测试</a></p><p class="warn">代理视觉检查：{html.escape(json.dumps(status,ensure_ascii=False))}；用户人工确认待完成，不自动全量或上传。</p><p>左：shared原图；中：源参数诊断投影；右：shared当前valid监督。顶盖紫、底座绿，左手青、右手橙。透视线框显示遮挡面，不是分割。播放30fps仅用于预览，源采集FPS未知。</p><p>暗帧指标仅作候选；除已有laptop曝光规则外，未添加按帧号的人工屏蔽。该序列通过不代表该物体全部序列通过。</p></header>'+cards+'<section><h2>资产及纹理</h2><video controls src="'+r['asset_preview']['video_ref']+'"></video></section><section><h2>视觉抽查对照（不代替完整视频）</h2>'+''.join(f'<a href="{f}"><img src="{f}"></a>' for f in r['contact_sheets'])+'</section><script>document.querySelectorAll("video").forEach(v=>v.addEventListener("play",()=>document.querySelectorAll("video").forEach(o=>{if(o!==v)o.pause()})));</script></html>'
+    page=page.replace(html.escape(json.dumps(status,ensure_ascii=False)),review_text)
+    page=page.replace('<p class="warn">代理视觉检查：','<div class="warn">代理视觉检查：').replace('用户人工确认待完成，不自动全量或上传。</p>','用户人工确认待完成，不自动全量或上传。</div>')
+    (doc/'index.html').write_text(page,encoding='utf-8')
+
+def overview(selection):
+    rows=[];reports=[]
+    for item in selection['objects']:
+        doc=DOC/item['object'];p=doc/'report.json'
+        if p.exists():
+            r=load(p);reports.append(r);state='数值通过；'+('视觉已抽查（见问题记录）' if r['agent_visual_review']['status']=='reviewed_with_findings' else '视觉待检查');count=r['camera_frames'];link=f'<a href="{item["object"]}/index.html">{item["object"]}</a>'
+        else:state='制作中';count='—';link=item['object']
+        rows.append(f'<tr><td>{link}</td><td>{item["sequence"]}</td><td>{count}</td><td>{state}</td></tr>')
+    DOC.mkdir(parents=True,exist_ok=True)
+    player=''
+    if reports:
+        options=''.join(f'<option value="{r["object"]}">{r["object"]} · {r["camera_frames"]}帧</option>' for r in reports)
+        player='<section><h2>固定相机6，切换物体</h2><label>物体 <select id="object-picker">'+options+'</select></label> · <a id="object-detail">该物体检查记录与原图对照</a><p id="object-result"></p><video id="object-video" controls preload="metadata" style="width:100%"></video></section>'
+        data={r['object']:{'video':r['object']+'/'+r['views'][0]['video']['video_ref'],'summary':r['agent_visual_review'].get('summary','数值通过；视觉检查待完成')} for r in reports}
+        player+='<script>const objects='+json.dumps(data,ensure_ascii=False)+';const picker=document.getElementById("object-picker"),video=document.getElementById("object-video");function choose(){const k=picker.value;video.pause();video.src=objects[k].video;document.getElementById("object-detail").href=k+"/index.html";document.getElementById("object-result").textContent=objects[k].summary;video.load();}picker.addEventListener("change",choose);choose();</script>'
+    (DOC/'index.html').write_text(f'<!doctype html><html lang="zh-CN"><meta charset="utf-8"><title>ARCTIC 11种物体完整序列验收</title><style>{STYLE}</style><header><h1>ARCTIC · 11种物体逐项验收</h1><p>新增组统一采用s06的固定相机6，11种物体各一段完整序列；相机内外参保持一致。<a href="../index.html">原s09 laptop九路相机预览</a>保留。按明显开合幅度选择较短完整序列，不裁切源帧。</p><p>已完成{len(reports)}/11种物体。<a href="selection.json">选择依据</a> · <a href="REPORT.md">验收结论</a> · <a href="report.json">总报告</a></p><p class="warn">自动数值通过≠图像对齐无误≠用户人工通过。未开始全量处理或上传。</p></header>'+player+'<section><table><tr><th>物体/完整视频</th><th>序列</th><th>相机帧数</th><th>检查状态</th></tr>'+''.join(rows)+'</table></section></html>',encoding='utf-8')
+    save(DOC/'selection.json',selection);save(DOC/'report.json',{'scope':'11 whole sequences, fixed subject s06 and fixed static camera 6; original laptop 9-camera group retained separately','completed_objects':len(reports),'expected_objects':11,'objects':[{'object':r['object'],'sequence':r['sequence'],'camera_frames':r['camera_frames'],'maximum_geometry_error_m':r['maximum_geometry_error_m'],'agent_visual_review':r['agent_visual_review']} for r in reports],'human_review':'pending','full_preprocessing_authorized':False,'upload_authorized':False})
+
+def run():
+    selection=load(SELECTION);TMP.mkdir(parents=True,exist_ok=True);DOC.mkdir(parents=True,exist_ok=True)
+    state=load(WORK/'tmp/dataset-work-state.json');state['active']['ARCTIC'].update(stage='pilot_all_objects_rendering',object_review_scope=selection['objects']);save(WORK/'tmp/dataset-work-state.json',state)
+    setup('ARCTIC','eleven_object_review');overview(selection)
+    def launch(item):
+        seq,sid,oid,seqid,doc,tmp=context(item)
+        for cam in selection['cameras']:
+            log=tmp/f'worker-{cam}.log'
+            with log.open('ab') as f:
+                proc=subprocess.run([sys.executable,'-u',str(Path(__file__).resolve()),'worker','--object',oid,'--camera',cam],stdout=f,stderr=subprocess.STDOUT)
+            if proc.returncode:raise RuntimeError(f'{oid} camera {cam} failed, see {log}')
+        return item
+    with concurrent.futures.ThreadPoolExecutor(max_workers=2) as pool:
+        futures=[]
+        for item in selection['objects']:
+            prepare(item)
+            if (context(item)[4]/'report.json').exists():print('RESUME_OBJECT',item['object'],flush=True);continue
+            futures.append(pool.submit(launch,item))
+        for future in concurrent.futures.as_completed(futures):
+            item=future.result();object_page(item);overview(selection)
+            save(TMP/'status.json',{'object':item['object'],'stage':'converting_rendering','completed_views_total':len(list(DOC.glob('*/checks/*-complete.json')))})
+    for name in ['arctic_object_review.py','convert_fresh_sequence_pilots.py','render_full_sequence_review.py','diagnose_arctic_review.py','fresh_sequence_common.py','review_fresh_sequence_pilots.py']:
+        shutil.copyfile(WORK/'scripts'/name,OUT/'script'/name)
+    save(TMP/'status.json',{'stage':'all_objects_rendered_pending_visual_inspection','completed_views_total':len(selection['objects'])*len(selection['cameras'])})
+    print('ALL_OBJECTS_RENDERED',flush=True)
+
+if __name__=='__main__':
+    parser=argparse.ArgumentParser();parser.add_argument('mode',choices=['run','worker','overview']);parser.add_argument('--object');parser.add_argument('--camera');args=parser.parse_args()
+    if args.mode=='worker':worker(next(x for x in load(SELECTION)['objects'] if x['object']==args.object),args.camera)
+    elif args.mode=='overview':overview(load(SELECTION))
+    else:run()
diff --git a/ARCTIC/shared-v1.0/script/audit_source_fields.py b/ARCTIC/shared-v1.0/script/audit_source_fields.py
new file mode 100644
index 0000000000000000000000000000000000000000..9dd63f10811edb78e5140e940368c3ccf6909e23
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/audit_source_fields.py
@@ -0,0 +1,55 @@
+"""Read small existing native annotations; do not convert or download data."""
+from pilot_build import *
+
+def shape_tree(x, depth=0):
+ if isinstance(x,(np.ndarray,torch.Tensor)):
+  if isinstance(x,np.ndarray) and x.dtype==object and x.shape==():return shape_tree(x.item(),depth)
+  return {'shape':list(x.shape),'dtype':str(x.dtype)}
+ if isinstance(x,dict):
+  if len(x)>30:return {'count':len(x),'first_key':str(next(iter(x))),'first_value':shape_tree(next(iter(x.values())),depth+1)}
+  return {str(k):shape_tree(v,depth+1) for k,v in x.items()}
+ if isinstance(x,(list,tuple)):
+  try:
+   a=np.asarray(x)
+   if a.dtype.kind in 'biuf':return {'shape':list(a.shape),'dtype':str(a.dtype)}
+  except (ValueError,TypeError):pass
+  return {'count':len(x),'first':shape_tree(x[0],depth+1) if x else None}
+ return x if isinstance(x,(str,int,float,bool)) or x is None else type(x).__name__
+
+def native_npy(p):
+ with p.open('rb') as f:
+  version=np.lib.format.read_magic(f)
+  if version==(1,0):_,_,dtype=np.lib.format.read_array_header_1_0(f)
+  else:_,_,dtype=np.lib.format.read_array_header_2_0(f)
+  if dtype.hasobject:return Restricted(f,encoding='latin1').load()
+ return np.load(p,allow_pickle=False)
+
+def main():
+ rows=[]
+ def add(ds,p,x,detail=None):rows.append({'dataset':ds,'path':str(p.relative_to(WORK)).replace('\\','/'),'fields':shape_tree(x),'detail':detail})
+ p=next((WORK/'tmp/pilot-dex').rglob('labels_000001.npz'))
+ with np.load(p,allow_pickle=False) as z:add('Dex-YCB',p,{k:z[k] for k in z.files})
+ p=next((ROOT/'VISOR/raw/data').rglob('P01_03.json'));x=load(p);frames=x['video_annotations'];keys=sorted({k for f in frames for a in f['annotations'] for k in a})
+ add('VISOR',p,frames[0],{'top_level_keys':list(x),'annotation_keys_all_local_frames':keys,'frames':len(frames),'image_keys':sorted({k for f in frames for k in f['image']})})
+ for folder in ['hand_pose','hand_pose_mano','cam_pose']:
+  p=ROOT/'H2O/raw/data/local_subset/subject4_ego/h1/0/cam4'/folder/'000000.txt';add('H2O',p,np.loadtxt(p))
+ p=next((ROOT/'HOI4D/raw/data').rglob('*.pickle'));add('HOI4D',p,read_pickle(p))
+ for name in ['training_K.json','training_mano.json','training_scale.json']:
+  p=next((ROOT/'FreiHAND/raw/data').rglob(name));x=load(p);add('FreiHAND',p,x[0],{'rows':len(x)})
+ p=next((ROOT/'RHD/raw/data').rglob('anno_evaluation.pickle'));x=read_pickle(p);a=x[0];add('RHD',p,a,{'records':len(x),'visibility_values':np.unique(a['uv_vis'][:,2]).tolist(),'invisible_points_first_sample':int((a['uv_vis'][:,2]==0).sum())})
+ p=next((ROOT/'ARCTIC/raw/data').rglob('*.mano.npy'));add('ARCTIC',p,native_npy(p))
+ p=next((ROOT/'ARCTIC/raw/data').rglob('*.egocam.dist.npy'));x=native_npy(p).item();add('ARCTIC',p,x,{'dist8':np.asarray(x['dist8']).tolist()})
+ p=next((ROOT/'OakInk2/raw/data').rglob('*.pkl'));x=read_pickle(p);add('OakInk2',p,x,{'frame_ids_first':x['frame_id_list'][:5],'mocap_ids_first':x['mocap_frame_id_list'][:5],'object_ids':x['obj_list']})
+ p=next((ROOT/'ReInterHand/raw/data').rglob('*_left.json'));add('ReInterHand',p,load(p))
+ p=OLD/'HOT3D/P0001_10a27bf7/mano_hand_pose_trajectory.jsonl';add('HOT3D',p,json.loads(p.open().readline()))
+ p=next((ROOT/'HOT3D/raw/data').glob('*ground_truth.zip'))
+ with zipfile.ZipFile(p) as z:
+  headers={}
+  for name in z.namelist():
+   if name.endswith('.csv'):
+    with z.open(name) as f:headers[name]=f.readline().decode().strip()
+  add('HOT3D',p,headers)
+ save(WORK/'docs/source-field-audit-evidence.json',{'purpose':'native field shape probe; not a full conversion or whole-dataset validation','records':rows})
+ print(json.dumps({'datasets':sorted({r['dataset'] for r in rows}),'records':len(rows),'output':'docs/source-field-audit-evidence.json'},indent=2))
+
+if __name__=='__main__':main()
diff --git a/ARCTIC/shared-v1.0/script/convert_fresh_sequence_pilots.py b/ARCTIC/shared-v1.0/script/convert_fresh_sequence_pilots.py
new file mode 100644
index 0000000000000000000000000000000000000000..46cfc35d9d8e82d707ba2dcb7624d897ceb7b095
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/convert_fresh_sequence_pilots.py
@@ -0,0 +1,343 @@
+"""Convert one complete, freshly downloaded sequence per dataset, all cameras.
+
+Run: python scripts/convert_fresh_sequence_pilots.py ARCTIC|HOT3D
+Requires the local CPU runtime and licensed MANO models described in README.md.
+No full-dataset conversion or upload is performed by this entry point.
+"""
+from fresh_sequence_common import *
+from pilot_hot_normalize import csv_transform, transform
+from accelerated_media import zip_members,file_members
+import concurrent.futures
+
+ARCTIC_SEQ='s09/laptop_use_03'
+HOT3D_SEQ='P0015_179e1b84'
+
+def loadzipcsv(z,name):return list(csv.DictReader(io.StringIO(z.read(name).decode())))
+
+def expand(rows,indices,shape):
+    a=np.full(shape,np.nan);good=(indices>=0)&(indices<len(rows));a[good]=np.asarray(rows)[indices[good]]
+    return a,good
+
+def arctic_unlabelled(seq,cameras=None):
+    """Official held-out s03 RGB: static calibration/assets, no hidden GT guesses."""
+    base=DATA_ROOT/'ARCTIC';raw=base/'raw/data';sid,action=seq.split('/');oid=action.split('_')[0];seqid=seq.replace('/','_')
+    with zipfile.ZipFile(raw/'splits_json.zip') as z:splitdef=json.loads(z.read('splits_json/protocol_p1.json'))
+    assert seq in splitdef['test']
+    source={'archive':f'raw/data/images_zips/{seq}.zip','member_pattern':'{camera_id}/{source_frame_id:05d}.jpg'}
+    with zipfile.ZipFile(raw/'meta.zip') as za,zipfile.ZipFile(raw/f'images_zips/{seq}.zip') as zi:
+        misc=json.loads(za.read('meta/misc.json'))[sid];prefix=f'meta/object_vtemplates/{oid}/'
+        native_parts=np.asarray(json.loads(za.read(prefix+'parts.json')),np.int64)
+        names=sorted(n for n in zi.namelist() if re.fullmatch(r'[0-8]/[0-9]+\.jpg',n))
+        for cam in sorted({n.split('/')[0] for n in names},key=int):
+            if cameras is not None and cam not in cameras:continue
+            selected=[n for n in names if n.split('/')[0]==cam];ids=[int(Path(n).stem) for n in selected]
+            size=list(Image.open(io.BytesIO(zi.read(selected[0]))).size);assert size==misc['image_size'][int(cam)]
+            v=FreshView('ARCTIC',seqid,cam,ids,size,dict(source,camera_id=cam))
+            if (v.p/'metainfo.json').exists() and load(v.p/'metainfo.json').get('pilot_complete'):
+                print('RESUME verified view',seq,cam,flush=True);continue
+            t=v.T;v.m.update(split='test',split_scheme='ARCTIC_protocol_p1',split_source_locator={'archive':'raw/data/splits_json.zip','member':'splits_json/protocol_p1.json','sequence':seq},view_type='ego' if cam=='0' else 'exo',source_hand_object_gt_available=False)
+            v.m['sampling'].update(source_fps=None,source_fps_reason='Capture FPS not asserted by supplied source; original image numbers retained')
+            if cam!='0':
+                src={'archive':'raw/data/meta.zip','member':'meta/misc.json','record':sid}
+                v.m['camera'].update(model='pinhole',schema_id='camera/pinhole/v1')
+                K=np.broadcast_to(misc['intris_mat'][int(cam)-1],(t,3,3));E=np.broadcast_to(misc['world2cam'][int(cam)-1],(t,4,4))
+                v.array('intrinsic','camera/intrinsic.pt','camera/pinhole/v1',K,np.ones(t,bool),['T','row','column'],source=src)
+                v.array('extrinsic','camera/extrinsic.pt','transform/se3/v1',E,np.ones(t,bool),['T','row','column'],source=src,unit='m',convention='T_camera_from_world');assert_se3(E)
+            for row,name in zip(v.frames,selected):row.update(sync_id=f'{seq}:{row["source_frame_id"]}',source_annotation_index=None);row['source_locator'].update(member=name)
+            rel=f'annotation/object/{oid}';op=v.p/rel;assets=op/'assets';object_textured_assets(za,prefix,assets)
+            parts=1-native_parts;tensor(op/'vertex_part_ids.pt',parts)
+            articulation={'hierarchy_semantics':'mechanical_articulation','parts':[{'part_id':'bottom','parent_id':None,'has_model':True,'mesh_ref':rel+'/assets/bottom.obj'},{'part_id':'top','parent_id':'bottom','has_model':True,'mesh_ref':rel+'/assets/top.obj'}],
+                'joints':[{'joint_id':'hinge','parent_part':'bottom','child_part':'top','type':'revolute','unit':'rad','axis':[0,0,-1],'axis_frame':'parent_part','zero_transform':np.eye(4).tolist(),'positive_direction':'right_hand_rule_about_axis','limits':None}],'joint_order':['hinge']}
+            save(op/'articulation.json',articulation);sourceasset={'archive':'raw/data/meta.zip','member_prefix':prefix}
+            v.component('object_'+oid+'_articulation',rel+'/articulation.json','object/articulated/v1',[],np.array(True),source=sourceasset,static=True)
+            v.mesh(oid,rel+'/assets/mesh.obj',sourceasset)
+            original=np.asarray([[float(x) for x in l.split()[1:4]] for l in za.read(prefix+'mesh.obj').decode().splitlines() if l.startswith('v ')])/1000
+            actual=np.asarray([[float(x) for x in l.split()[1:4]] for l in (assets/'mesh.obj').read_text().splitlines() if l.startswith('v ')])
+            assert np.array_equal(original,actual) and len(parts)==len(actual)
+            save(op/'metainfo.json',{'object_id':oid,'identity_scope':'sequence','scope_id':seqid,'association_basis':'official_sequence_object_name; no per-frame object pose released','has_model':True,'mesh_ref':rel+'/assets/mesh.obj','articulation_ref':rel+'/articulation.json','vertex_part_ids_ref':rel+'/vertex_part_ids.pt','topology_id':'sha256:'+digest(assets/'mesh.obj'),'part_mapping':{'0':'bottom','1':'top'},'part_id_conversion':{'formula':'shared_id = 1 - raw_parts_json_value','source_locator':dict(sourceasset,member=prefix+'parts.json')},'unit':'m','source_locator':sourceasset})
+            v.checks.append({'test':'source_static_mesh_vertices_metres_and_part_ids','passed':True,'vertices':len(actual)})
+            v.m['omitted_components']={'mano':'Official test ground truth withheld: no sequence MANO file in raw_seqs.zip','object_pose':'Official test ground truth withheld: no sequence object trajectory, joint states or camera-frame keypoints','keypoints':'No released sequence hand keypoint supervision','visibility':'Not provided in selected official release','depth':'No source depth images','segment':'No source pixel masks','contact':'No native contact labels in downloaded scope'}
+            if cam=='0':v.m['omitted_components'].update(intrinsic='No test sequence egocam.dist.npy; do not reuse another subject calibration',extrinsic='No released egocentric trajectory for test sequence')
+            print('MEDIA ARCTIC',cam,t,flush=True);submit_media(v,lambda archive=raw/f'images_zips/{seq}.zip',items=tuple(selected):zip_members(archive,items))
+
+def arctic(sequence=None,cameras=None,initialize=True):
+    dataset='ARCTIC';seq=sequence or ARCTIC_SEQ;sid,action=seq.split('/');oid=action.split('_')[0];seqid=seq.replace('/','_')
+    if initialize:setup(dataset,seqid)
+    base=DATA_ROOT/dataset;raw=base/'raw/data'
+    with zipfile.ZipFile(raw/'raw_seqs.zip') as z:
+        if f'raw_seqs/{seq}.mano.npy' not in z.namelist():
+            assert all(f'raw_seqs/{seq}.{x}.npy' not in z.namelist() for x in ['object','egocam.dist'])
+            return arctic_unlabelled(seq,cameras)
+        hand=native_npy_bytes(z.read(f'raw_seqs/{seq}.mano.npy'))
+        ego=native_npy_bytes(z.read(f'raw_seqs/{seq}.egocam.dist.npy'))
+        objects=native_npy_bytes(z.read(f'raw_seqs/{seq}.object.npy'))
+    with zipfile.ZipFile(raw/'splits_json.zip') as z:splitdef=json.loads(z.read('splits_json/protocol_p1.json'))
+    split=next((k for k,v in splitdef.items() if seq in v),'unknown')
+    if split=='val':split='val'
+    za=zipfile.ZipFile(raw/'meta.zip');misc=json.loads(za.read('meta/misc.json'))[sid]
+    params=json.loads(za.read(f'meta/object_vtemplates/{oid}/object_params.json'))
+    native_parts=np.asarray(json.loads(za.read(f'meta/object_vtemplates/{oid}/parts.json')),np.int64)
+    models=[Mano('left'),Mano('right')];N=len(objects)
+    world=np.broadcast_to(np.eye(4),(N,4,4)).copy();world[:,:3,:3]=Rotation.from_rotvec(objects[:,1:4]).as_matrix();world[:,:3,3]=objects[:,4:7]/1000
+    hinge=np.broadcast_to(np.eye(4),(N,4,4)).copy();hinge[:,:3,:3]=Rotation.from_rotvec(np.c_[np.zeros((N,2)),-objects[:,0]]).as_matrix()
+    kp=np.concatenate([params['keypoints_top'],params['keypoints_bottom']])/1000;n_top=len(params['keypoints_top'])
+    source={'archive':f'raw/data/images_zips/{seq}.zip','member_pattern':'{camera_id}/{source_frame_id:05d}.jpg'}
+    zimage=zipfile.ZipFile(raw/f'images_zips/{seq}.zip')
+    names=sorted(n for n in zimage.namelist() if re.fullmatch(r'[0-8]/[0-9]+\.jpg',n))
+    extra=[n for n in zimage.namelist() if n.endswith('.jpg') and not re.fullmatch(r'[0-8]/[0-9]+\.jpg',n)]
+    assert all(n.startswith('__MACOSX/') or Path(n).name.startswith('._') for n in extra),extra[:5]
+    for cam in sorted({n.split('/')[0] for n in names},key=int):
+        if cameras is not None and cam not in cameras:continue
+        selected=[n for n in names if n.split('/')[0]==cam];ids=np.asarray([int(Path(n).stem) for n in selected]);ix=ids-misc['ioi_offset']
+        size=list(Image.open(io.BytesIO(zimage.read(selected[0]))).size)
+        assert size==misc['image_size'][int(cam)]
+        v=FreshView(dataset,seqid,cam,ids,size,dict(source,camera_id=cam))
+        if (v.p/'metainfo.json').exists():
+            old=load(v.p/'metainfo.json')
+            if old.get('pilot_complete') and old.get('num_frames')==len(ids):print('RESUME verified view',seq,cam,flush=True);continue
+        t=v.T;good=(ix>=0)&(ix<N)
+        v.m.update(split=split,split_scheme='ARCTIC_protocol_p1',split_source_locator={'archive':'raw/data/splits_json.zip','member':'splits_json/protocol_p1.json','sequence':seq},view_type='ego' if cam=='0' else 'exo')
+        v.m['sampling'].update(source_fps=None,source_fps_reason='Actual capture FPS not asserted by the selected raw files; frame indices and official synchronization offset retained',selection='Every released image of this selected camera in the complete official sequence ZIP; boundary images without annotations retained')
+        srcanno={'archive':'raw/data/raw_seqs.zip','member':f'raw_seqs/{seq}.mano.npy','index_ref':'camera/frames.jsonl:source_annotation_index'}
+        srccam={'archive':'raw/data/meta.zip','member':'meta/misc.json','record':sid}
+        E=np.broadcast_to(np.eye(4),(t,4,4)).copy()
+        if cam=='0':
+            K=np.broadcast_to(ego['intrinsics'],(t,3,3));D=np.broadcast_to(ego['dist8'],(t,8))
+            E[:]=np.nan;E[good]=np.eye(4);E[good,:3,:3]=ego['R_k_cam_np'][ix[good]];E[good,:3,3]=ego['T_k_cam_np'][ix[good]].reshape(-1,3)
+            src={'archive':'raw/data/raw_seqs.zip','member':f'raw_seqs/{seq}.egocam.dist.npy'}
+            v.m['camera'].update(model='rational8',schema_id='camera/rational8/v1')
+            v.array('intrinsic','camera/intrinsic.pt','camera/rational8/v1',K,np.ones(t,bool),['T','row','column'],source=src,distortion_ref='camera/distortion.pt')
+            tensor(v.p/'camera/distortion.pt',np.asarray(D,np.float32));validE=good
+        else:
+            K=np.broadcast_to(misc['intris_mat'][int(cam)-1],(t,3,3));D=None;E[:]=misc['world2cam'][int(cam)-1];validE=np.ones(t,bool);src=srccam
+            v.m['camera'].update(model='pinhole',schema_id='camera/pinhole/v1')
+            v.array('intrinsic','camera/intrinsic.pt','camera/pinhole/v1',K,np.ones(t,bool),['T','row','column'],source=srccam)
+        v.array('extrinsic','camera/extrinsic.pt','transform/se3/v1',E,validE,['T','row','column'],source=src,unit='m',convention='T_camera_from_world')
+        assert_se3(E[validE]);motion=np.full((t,2,61),np.nan)
+        for side,model in enumerate(models):
+            h=hand[['left','right'][side]];pose=np.c_[h['rot'],h['pose']].copy();pose[:,3:]+=model.data['hands_mean']
+            result,err=camera_mano(model,pose[ix[good]],np.broadcast_to(h['shape'],(good.sum(),10)),h['trans'][ix[good]],E[good])
+            motion[good,side]=result
+            # Independent source-model mean configuration, separate from camera conversion algebra.
+            from types import SimpleNamespace
+            import smplx
+            native=smplx.MANO(model_path=str(WORK/'tmp/pilot-models'),data_struct=SimpleNamespace(**model.data),is_rhand=side==1,use_pca=False,flat_hand_mean=False)
+            checkidx=np.arange(0,N,max(1,N//12));b=torch.tensor(np.broadcast_to(h['shape'],(len(checkidx),10)).copy(),dtype=torch.float32)
+            with torch.no_grad():expected=native(global_orient=torch.tensor(h['rot'][checkidx]),hand_pose=torch.tensor(h['pose'][checkidx]),betas=b,transl=torch.tensor(h['trans'][checkidx])).vertices.numpy()
+            actual=model.geometry(pose[checkidx],b.numpy(),h['trans'][checkidx])[0]
+            delta=float(np.max(np.linalg.norm(expected-actual,axis=-1)));assert delta<2e-6
+            v.checks.append({'test':'MANO_mean_and_camera_transform','side':side,'all_aligned_frames':int(good.sum()),'camera_max_m':err,'native_mean_max_m':delta,'passed':True})
+        v.array('mano','annotation/mano/motion.pt','mano/axis_angle61/v1',motion,np.broadcast_to(good[:,None],(t,2)),['T','hand','parameter'],source=srcanno,unit='m',angle_unit='rad',coordinate_frame='camera',model_profile={'version':'MANO-v1.2','flat_hand_mean':True,'left_shapedirs_fix':False},method='Official nonflat mean expanded; shape-dependent root corrected during world-to-camera conversion; image index minus official ioi_offset')
+        for i,(row,name,j) in enumerate(zip(v.frames,selected,ix)):
+            row.update(source_annotation_index=int(j) if good[i] else None,sync_id=f'{seq}:{ids[i]}')
+            row['source_locator'].update(member=name)
+            row['annotation_matches']={'mano':{'source_frame_id':str(j) if good[i] else None,'source_timestamp_ns':None,'clock_domain':None,'match_method':'index_map','delta_ns':None,'mapping':'image_number - misc.ioi_offset','source_record_present':bool(good[i])}}
+        rel=f'annotation/object/{oid}';op=v.p/rel;assets=op/'assets';object_textured_assets(za,f'meta/object_vtemplates/{oid}/',assets)
+        # Parse exact OBJ vertex ordering before loading a render-oriented mesh.
+        vertices=np.array([[float(x) for x in r.split()[1:4]] for r in (assets/'mesh.obj').read_text().splitlines() if r.startswith('v ')])
+        # Official ObjectTensors adds 1 to raw parts, then articulates ==1: raw 0 is TOP.
+        shared_parts=1-native_parts
+        assert len(vertices)==len(native_parts);tensor(op/'vertex_part_ids.pt',shared_parts)
+        P=np.full((t,4,4),np.nan);top=P.copy();P[good]=E[good]@world[ix[good]];top[good]=P[good]@hinge[ix[good]]
+        srcobj={'archive':'raw/data/raw_seqs.zip','member':f'raw_seqs/{seq}.object.npy','index_ref':'camera/frames.jsonl:source_annotation_index'}
+        for name,data in [('root',P),('bottom',P),('top',top)]:
+            path=rel+('/pose.pt' if name=='root' else f'/parts/{name}/pose.pt')
+            v.array(f'object_{oid}_{name}_pose',path,'object/rigid/v1',data,good,['T','row','column'],source=srcobj,unit='m',convention='T_camera_from_object' if name=='root' else 'T_camera_from_part')
+            assert_se3(data[good])
+        joint,_=expand(objects[:,:1],ix,(t,1));v.array(f'object_{oid}_joint',rel+'/joint_state.pt','object/joint_state/v1',joint,good[:,None],['T','joint'],source=srcobj,joint_order=['hinge'],unit='rad')
+        articulation={'hierarchy_semantics':'mechanical_articulation','parts':[{'part_id':'bottom','parent_id':None,'has_model':True,'mesh_ref':rel+'/assets/bottom.obj'},{'part_id':'top','parent_id':'bottom','has_model':True,'mesh_ref':rel+'/assets/top.obj'}],
+            'joints':[{'joint_id':'hinge','parent_part':'bottom','child_part':'top','type':'revolute','unit':'rad','axis':[0,0,-1],'axis_frame':'parent_part','zero_transform':np.eye(4).tolist(),'positive_direction':'right_hand_rule_about_axis','limits':None}],'joint_order':['hinge']}
+        save(op/'articulation.json',articulation)
+        v.component('object_'+oid+'_articulation',rel+'/articulation.json','object/articulated/v1',[],np.array(True),source=srcobj,static=True)
+        sourceasset={'archive':'raw/data/meta.zip','member_prefix':f'meta/object_vtemplates/{oid}/'}
+        v.mesh(oid,rel+'/assets/mesh.obj',sourceasset)
+        topology=digest(assets/'mesh.obj')
+        save(op/'metainfo.json',{'object_id':oid,'identity_scope':'sequence','scope_id':seqid,'association_basis':'official_sequence_object_identity_shared_across_cameras','has_model':True,'mesh_ref':rel+'/assets/mesh.obj','articulation_ref':rel+'/articulation.json','vertex_part_ids_ref':rel+'/vertex_part_ids.pt','topology_id':'sha256:'+topology,'part_mapping':{'0':'bottom','1':'top'},'part_id_conversion':{'formula':'shared_id = 1 - raw_parts_json_value','source_locator':dict(sourceasset,member=f'meta/object_vtemplates/{oid}/parts.json')},'unit':'m','source_locator':srcobj})
+        xyz=np.full((t,len(kp),3),np.nan);errs=[]
+        for i in np.flatnonzero(good):
+            xyz[i]=kp@P[i,:3,:3].T+P[i,:3,3];xyz[i,:n_top]=kp[:n_top]@top[i,:3,:3].T+top[i,:3,3]
+            ref=vertices.copy();j=ix[i];ref[native_parts==0]=Rotation.from_rotvec([0,0,-float(objects[j,0])]).apply(ref[native_parts==0]);ref=Rotation.from_rotvec(objects[j,1:4]).apply(ref)+objects[j,4:7]/1000;ref=ref@E[i,:3,:3].T+E[i,:3,3]
+            out=vertices@P[i,:3,:3].T+P[i,:3,3];out[shared_parts==1]=vertices[shared_parts==1]@top[i,:3,:3].T+top[i,:3,3];errs.append(float(np.linalg.norm(out-ref,axis=1).max()))
+        assert max(errs)<2e-5
+        v.array('object_'+oid+'_keypoints',rel+'/keypoints/xyz.pt','object_keypoints/camera/v1',xyz,np.broadcast_to(good[:,None],(t,len(kp))),['T','joint','coordinate'],source=sourceasset,unit='m',coordinate_frame='camera',keypoint_schema='arctic_object_keypoints/v1',keypoint_names=[f'top_{i}' for i in range(n_top)]+[f'bottom_{i}' for i in range(len(kp)-n_top)],kind='derived',method='Official template keypoints driven by released articulation and camera transforms')
+        v.checks.append({'test':'all_frame_articulated_geometry_independent_scipy_rotation','passed':True,'max_m':max(errs),'angle_min_max_rad':[float(objects[:,0].min()),float(objects[:,0].max())]})
+        v.m['omitted_components']={'depth':'No depth images in this official raw release','segment':'No pixel segmentation in this official raw release','contact':'No source contact labels in selected official raw packages','keypoints':'No separate measured hand keypoints in raw; MANO available. Render-only joints are not published as additional labels','visibility':'Not supplied in selected raw annotations; fitting_err is outside current shared scope'}
+        print('MEDIA',dataset,cam,t,flush=True);submit_media(v,lambda archive=raw/f'images_zips/{seq}.zip',items=tuple(selected):zip_members(archive,items))
+    zimage.close();za.close()
+
+def vrs_records(path,cam):
+    raw=command([VRS,'print-json',path,'+',cam,'+','data']).decode(errors='replace')
+    rows=[]
+    for line in raw.splitlines():
+        if line.startswith('{"data_layout"'):rows.append({x['name']:x['value'] for x in json.loads(line)['data_layout']})
+    return rows
+
+def nearest(times,query,limit):
+    times=np.asarray(times,np.int64);q=np.asarray(query,np.int64)
+    if not len(times):return np.zeros(len(q),np.int64),np.zeros(len(q),bool),np.zeros(len(q),np.int64)
+    j=np.searchsorted(times,q);a=np.clip(j-1,0,len(times)-1);b=np.clip(j,0,len(times)-1)
+    ix=np.where(np.abs(times[a]-q)<=np.abs(times[b]-q),a,b);delta=times[ix]-q
+    return ix,np.abs(delta)<=limit,delta
+
+def verify_hot3d_frame_order_and_metadata(sequence=None):
+    """Independent extraction-index and VRS capture-rate check, also on resume."""
+    seq=sequence or HOT3D_SEQ
+    base=DATA_ROOT/'HOT3D';src=base/'raw/data/Aria'/seq;recording=next(src.glob('*.vrs'))
+    extract=base/'tmp'/RUN/seq/'images'
+    for cam in ['214-1','1201-1','1201-2']:
+        p=base/RELEASE/'data'/seq/cam;m=load(p/'metainfo.json');rows=vrs_records(recording,cam)
+        paths=sorted(extract.glob(cam+'-*.jpg'))
+        indices=[int(x.name.split('-')[-2]) for x in paths];assert indices==list(range(1,len(rows)+1))
+        file_times=np.array([float(x.stem.split('-')[-1]) for x in paths]);times=np.array([r['capture_timestamp_ns'] for r in rows])/1e9
+        assert np.max(np.abs(file_times-times))<.001,'extracted images do not match camera record ordering'
+        raw=command([VRS,'print-json',recording,'+',cam,'+','configuration']).decode(errors='replace')
+        cfg=[{x['name']:x.get('value') for x in json.loads(line)['data_layout']} for line in raw.splitlines() if line.startswith('{"data_layout"')][0]
+        rate=cfg['nominal_rate'];assert abs(np.median(np.diff(times))-1/rate)<.001
+        m['sampling']['source_fps']=rate
+        m['sampling']['source_fps_basis']='VRS camera configuration nominal_rate; actual intervals checked separately'
+        m['sampling']['duration_seconds']=float(times[-1]-times[0])
+        frames=[json.loads(line) for line in (p/'camera/frames.jsonl').read_text().splitlines()]
+        for frame in frames:
+            for match in frame['annotation_matches'].values():
+                if 'available' in match:match['source_record_present']=match.pop('available')
+        lines(p/'camera/frames.jsonl',frames)
+        info=load(base/'raw/data/assets/object_models/models_info.json')
+        reverse={str(v['original_id']):str(k) for k,v in info.items()}
+        for op in (p/'annotation/object').glob('*'):
+            om=load(op/'metainfo.json');oid=om['object_id'];idx=int(reverse[oid])
+            glb=base/'raw/data/assets/object_models'/f'obj_{idx:06d}.glb';assert glb.is_file()
+            # Retain the official material-bearing companion byte-for-byte. OBJ remains the declared standard mesh.
+            target=op/'assets'/glb.name;shutil.copyfile(glb,target)
+            assert digest(glb)==digest(target)
+            om.update(category_name=info[str(idx)]['name'],original_glb_ref=target.relative_to(p).as_posix(),original_glb_source_locator={'file':glb.relative_to(base).as_posix(),'semantics':'native companion; standard geometry is the referenced metre OBJ'})
+            save(op/'metainfo.json',om)
+        save(p/'metainfo.json',m)
+        checkpath=base/'tmp'/RUN/'checks'/f'{seq}__{cam}.json';checks=load(checkpath)
+        checks['checks']=[c for c in checks['checks'] if c['test']!='VRS_extracted_filename_order_and_nominal_rate']
+        checks['checks'].append({'test':'VRS_extracted_filename_order_and_nominal_rate','frames':len(rows),'nominal_fps':rate,'filename_rounding_max_ms':float(np.max(np.abs(file_times-times))*1000),'passed':True});save(checkpath,checks)
+
+def hot3d(sequence=None,initialize=True):
+    dataset='HOT3D';seq=sequence or HOT3D_SEQ
+    if initialize:setup(dataset,seq)
+    base=DATA_ROOT/dataset;raw=base/'raw/data';src=raw/'Aria'/seq;tmp=base/'tmp'/RUN
+    recording=next(src.glob('*.vrs'));extract=tmp/seq/'images'
+    if not (tmp/seq/'extracted.json').exists() or not extract.exists() or not any(extract.glob('*.jpg')):
+        extract.mkdir(parents=True,exist_ok=True)
+        log=tmp/seq/'vrs-extraction.log'
+        with log.open('wb') as out:
+            run=subprocess.run([str(VRS),'extract-images',str(recording),'--to',str(extract),'+','214-1','+','1201-1','+','1201-2'],stdout=out,stderr=subprocess.STDOUT)
+        assert run.returncode==0,'VRS extraction failed; see tmp log'
+        save(tmp/seq/'extracted.json',{'source':recording.name,'streams':['214-1','1201-1','1201-2'],'complete':True})
+    gt=zipfile.ZipFile(next(src.glob('*ground_truth.zip')));handszip=zipfile.ZipFile(next(src.glob('*hand_data.zip')))
+    meta=json.loads(gt.read('metadata.json'));cams=json.loads(gt.read('camera_models.json'));mapping=loadzipcsv(gt,'timecode_devicetime_mapping.csv');mapping.sort(key=lambda r:int(r['devicetime_ns']))
+    assert meta['headset']=='Aria' and meta['recording_name']==seq
+    dev=np.array([int(r['devicetime_ns']) for r in mapping],np.int64);tc=np.array([int(r['timecode_ns']) for r in mapping],np.int64)
+    heads={int(r['timestamp[ns]']):csv_transform(r) for r in loadzipcsv(gt,'headset_trajectory.csv')};head_times=sorted(heads)
+    hands={r['timestamp_ns']:r['hand_poses'] for r in map(json.loads,handszip.read('mano_hand_pose_trajectory.jsonl').splitlines())};hand_times=sorted(hands)
+    objects=loadzipcsv(gt,'dynamic_objects.csv');vis_h=loadzipcsv(gt,'box2d_hands.csv');vis_o=loadzipcsv(gt,'box2d_objects.csv')
+    models=[Mano('left'),Mano('right')];fixed=np.abs(models[0].data['shapedirs'][:,0]-models[1].data['shapedirs'][:,0]).sum()<1
+    if fixed:
+        models[0].data['shapedirs'][:,0]*=-1
+        with torch.no_grad():models[0].layer.shapedirs[:,0]*=-1
+    assetszip=zipfile.ZipFile(raw/'assets/hot3d_models.zip');assetinfo=json.loads(assetszip.read('models/models_info.json'));reverse={str(v['original_id']):str(k) for k,v in assetinfo.items()}
+    for camera in cams:
+        cam=camera['stream_id'];rows=vrs_records(recording,cam);paths=sorted(extract.glob(cam+'-*.jpg'))
+        assert len(rows)==len(paths)>1,(cam,len(rows),len(paths))
+        v=FreshView(dataset,seq,cam,[r['frame_number'] for r in rows],[camera['imageWidth'],camera['imageHeight']],{'file':recording.relative_to(base).as_posix(),'stream_id':cam})
+        if (v.p/'metainfo.json').exists() and load(v.p/'metainfo.json').get('pilot_complete'):print('RESUME verified view',cam,flush=True);continue
+        t=v.T;capture=np.array([int(r['capture_timestamp_ns']) for r in rows],np.int64)
+        ix,good,delta=nearest(dev,capture,1000);assert good.all(),'Missing official device-time/timecode mapping'
+        times=tc[ix];sourcegt={'archive':next(src.glob('*ground_truth.zip')).relative_to(base).as_posix()}
+        test_ids={'P0004','P0005','P0006','P0008','P0016','P0020'}
+        split='test' if meta['participant_id'] in test_ids else 'train'
+        assert meta['have_hand_object_pose_gt']==(split=='train')
+        v.m.update(view_type='ego',split=split,split_scheme='HOT3D_official_participant_split',split_source_locator={'url':'https://github.com/facebookresearch/hot3d/blob/146b34afef8c1a32adeef7e981c070109f225c87/README.md#testtrain-split','participant_id':meta['participant_id'],'archive':sourcegt['archive'],'member':'metadata.json'},source_hand_object_gt_available=meta['have_hand_object_pose_gt'])
+        v.m['sampling'].update(source_fps=30,source_fps_basis='camera VRS nominal_rate; confirmed against capture timestamps',selection='All data records in each of the three camera streams of the complete official VRS',capture_interval_median_ns=int(np.median(np.diff(capture))))
+        v.m['camera'].update(model='fisheye624',schema_id='camera/fisheye624/v1',native_channel_semantics='RGB' if cam=='214-1' else 'monochrome; unified RGB decoder repeats intensity in three channels',max_solid_angle=camera['maxSolidAngle'])
+        v.array('intrinsic','camera/intrinsic.pt','camera/fisheye624/v1',np.broadcast_to(camera['projectionParams'],(t,15)),np.ones(t,bool),['T','parameter'],source=dict(sourcegt,member='camera_models.json',stream_id=cam))
+        dc=camera['T_Device_Camera'];Tdc=transform(dc['quaternion_wxyz'],dc['translation_xyz'])
+        hi,hvalid,hdelta=nearest(head_times,times,1_000_000);E=np.full((t,4,4),np.nan)
+        for i in np.flatnonzero(hvalid):E[i]=np.linalg.inv(heads[head_times[hi[i]]]@Tdc)
+        assert hvalid.all(),'Headset GT missing at image times'
+        assert_se3(E);v.array('extrinsic','camera/extrinsic.pt','transform/se3/v1',E,hvalid,['T','row','column'],source=dict(sourcegt,member='headset_trajectory.csv',calibration_member='camera_models.json'),unit='m',convention='T_camera_from_world',method='inverse(T_world_from_device @ T_device_from_camera); official timecode mapping then <=1ms nearest source pose')
+        mi,mvalid,mdelta=nearest(hand_times,times,1_000_000);motion=np.full((t,2,61),np.nan);valid=np.zeros((t,2),bool)
+        for side,model in enumerate(models):
+            chosen=[i for i in range(t) if mvalid[i] and str(side) in hands[hand_times[mi[i]]]]
+            if not chosen:continue
+            native=[hands[hand_times[mi[i]]][str(side)] for i in chosen];pose=[];beta=[];trans=[]
+            for h in native:
+                wrist=h['wrist_xform'];pose.append(np.r_[Rotation.from_quat(np.roll(wrist['q_wxyz'],-1)).as_rotvec(),np.asarray(h['pose'])@model.data['hands_components'][:15]+model.data['hands_mean']]);beta.append(h['betas']);trans.append(wrist['t_xyz'])
+            motion[chosen,side],err=camera_mano(model,pose,beta,trans,E[chosen]);valid[chosen,side]=True
+            # Official source model uses PCA15 and nonflat mean, unlike output AA48.
+            import smplx
+            from types import SimpleNamespace
+            native_model=smplx.MANO(model_path=str(WORK/'tmp/pilot-models'),data_struct=SimpleNamespace(**model.data),is_rhand=side==1,use_pca=True,num_pca_comps=15,flat_hand_mean=False)
+            ixcheck=np.arange(0,len(chosen),max(1,len(chosen)//16));arr=np.asarray(pose);bet=np.asarray(beta);tr=np.asarray(trans)
+            with torch.no_grad():expected=native_model(global_orient=torch.tensor(arr[ixcheck,:3],dtype=torch.float32),hand_pose=torch.tensor(np.array([native[i]['pose'] for i in ixcheck]),dtype=torch.float32),betas=torch.tensor(bet[ixcheck],dtype=torch.float32),transl=torch.tensor(tr[ixcheck],dtype=torch.float32)).vertices.numpy()
+            actual=model.geometry(arr[ixcheck],bet[ixcheck],tr[ixcheck])[0];pcaerr=float(np.max(np.linalg.norm(expected-actual,axis=-1)));assert pcaerr<2e-6
+            v.checks.append({'test':'official_PCA15_vs_AA48_and_camera_geometry','side':side,'camera_max_m':err,'pca_max_m':pcaerr,'valid_frames':len(chosen),'passed':True})
+        if hand_times:
+            v.array('mano','annotation/mano/motion.pt','mano/axis_angle61/v1',motion,valid,['T','hand','parameter'],source={'archive':next(src.glob('*hand_data.zip')).relative_to(base).as_posix(),'member':'mano_hand_pose_trajectory.jsonl','time_index_ref':'camera/frames.jsonl:annotation_matches.mano'},unit='m',angle_unit='rad',coordinate_frame='camera',model_profile={'version':'MANO-v1.2','flat_hand_mean':True,'left_shapedirs_fix':bool(fixed)},method='Official PCA15 expanded with model mean and handedness fix; transform all matched MANO parameters into current camera')
+        def vis_values(records,identity_field,identities):
+            out=np.full((t,len(identities)),np.nan)
+            for c,identity in enumerate(identities):
+                values={int(r['timestamp[ns]']):float(r['visibility_ratio[%]']) for r in records if r['stream_id']==cam and str(r[identity_field])==str(identity) and r['visibility_ratio[%]']!=''}
+                if not values:continue
+                ts=sorted(values);ii,vv,dd=nearest(ts,times,1_000_000)
+                for i in np.flatnonzero(vv):out[i,c]=values[ts[ii[i]]]
+            return out
+        if hand_times and vis_h:
+            visibility=vis_values(vis_h,'hand_index',[0,1]);v.attribute('mano','annotation/mano/visibility.pt',visibility,['T','hand'],dict(sourcegt,member='box2d_hands.csv',stream_id=cam,match_method='nearest',max_delta_ns=1000000),'whole_hand')
+        for i,row in enumerate(v.frames):
+            row.update(timestamp_ns=int(capture[i]),clock_domain='device_time',timecode_ns=int(times[i]),timecode_clock_domain='timecode',mapping_delta_ns=int(delta[i]))
+            row['source_locator'].update(record_index=i,capture_timestamp_ns=int(capture[i]))
+            row['annotation_matches']={}
+            for key,ts,idxs,flags,dd in [('extrinsic',head_times,hi,hvalid,hdelta),('mano',hand_times,mi,mvalid,mdelta)]:
+                row['annotation_matches'][key]={'source_frame_id':str(ts[idxs[i]]) if flags[i] else None,'source_timestamp_ns':int(ts[idxs[i]]) if flags[i] else None,'clock_domain':'timecode','match_method':'exact' if dd[i]==0 else 'nearest','max_delta_ns':1000000,'delta_ns':int(dd[i]) if flags[i] else None,'image_time_ref':'timecode_ns','source_record_present':bool(flags[i])}
+        for oid in sorted(set(meta['object_uids'])|{r['object_uid'] for r in objects}):
+            rel=f'annotation/object/{oid}';op=v.p/rel;assets=op/'assets';assets.mkdir(parents=True,exist_ok=True)
+            sourceposes={int(r['timestamp[ns]']):csv_transform(r) for r in objects if r['object_uid']==oid};ts=sorted(sourceposes);oi,ok,od=nearest(ts,times,1_000_000)
+            P=np.full((t,4,4),np.nan)
+            for i in np.flatnonzero(ok):P[i]=E[i]@sourceposes[ts[oi[i]]]
+            key=f'object_{oid}_pose'
+            if sourceposes:
+                assert_se3(P[ok]);v.array(key,rel+'/pose.pt','object/rigid/v1',P,ok,['T','row','column'],source=dict(sourcegt,member='dynamic_objects.csv',object_uid=oid),unit='m',convention='T_camera_from_object')
+            for i,row in enumerate(v.frames):row['annotation_matches'][key]={'source_frame_id':str(ts[oi[i]]) if ok[i] else None,'source_timestamp_ns':int(ts[oi[i]]) if ok[i] else None,'clock_domain':'timecode','match_method':'exact' if od[i]==0 else 'nearest','max_delta_ns':1000000,'delta_ns':int(od[i]) if ok[i] else None,'image_time_ref':'timecode_ns','source_record_present':bool(ok[i])}
+            if sourceposes and vis_o:
+                ov=vis_values(vis_o,'object_uid',[oid])[:,0];v.attribute(key,rel+'/visibility.pt',ov,['T'],dict(sourcegt,member='box2d_objects.csv',stream_id=cam,object_uid=oid,match_method='nearest',max_delta_ns=1000000),'whole_object')
+            idx=int(reverse[oid]);prefix=f'models/obj_{idx:06d}'
+            for member in assetszip.namelist():
+                if member.startswith(prefix+'.'):(assets/Path(member).name).write_bytes(assetszip.read(member))
+            ply=assets/f'obj_{idx:06d}.ply';mesh=trimesh.load(ply,process=False,force='mesh');mesh.vertices=np.asarray(mesh.vertices)*.001
+            expected_extent=np.array([assetinfo[str(idx)]['size_'+x] for x in 'xyz']);assert np.allclose(mesh.extents,expected_extent,atol=1e-5)
+            text,textures=trimesh.exchange.obj.export_obj(mesh,return_texture=True,digits=12);meshpath=assets/f'obj_{idx:06d}.obj';meshpath.write_text(text)
+            for filename,data in textures.items():(assets/filename).write_bytes(data)
+            restored=trimesh.load(meshpath,process=False,force='mesh');assert np.allclose(restored.triangles,mesh.triangles,atol=1e-7)
+            assert np.array_equal(np.asarray(restored.visual.material.image),np.asarray(mesh.visual.material.image))
+            # Keep the official GLB as a companion asset so richer original material properties remain available.
+            glb=raw/'assets'/f'{oid}.glb'
+            if glb.exists():shutil.copyfile(glb,assets/glb.name)
+            # The millimetre PLY was a temporary converter input, not a second shared geometry convention.
+            assert ply.resolve().is_relative_to(assets.resolve());ply.unlink()
+            sourceasset={'archive':'raw/data/assets/hot3d_models.zip','member':prefix+'.ply','original_id':oid,'bop_id':idx}
+            v.mesh(oid,meshpath.relative_to(v.p).as_posix(),sourceasset)
+            save(op/'metainfo.json',{'object_id':oid,'source_object_id':oid,'identity_scope':'sequence','scope_id':seq,'association_basis':'official_object_uid_in_shared_world_coordinates','has_model':True,'mesh_ref':meshpath.relative_to(v.p).as_posix(),'unit':'m','source_locator':dict(sourcegt,member='dynamic_objects.csv' if sourceposes else 'metadata.json',object_uid=oid),'original_glb_ref':(assets/glb.name).relative_to(v.p).as_posix() if glb.exists() else None})
+            v.checks.append({'test':'object_mesh_scale_triangle_and_texture_roundtrip','object_id':oid,'vertices':len(mesh.vertices),'faces':len(mesh.faces),'extent_m':mesh.extents.tolist(),'passed':True})
+        v.checks.append({'test':'all_camera_frame_device_time_mapping','frames':t,'max_delta_ns':int(np.abs(delta).max()),'passed':True})
+        v.m['omitted_components']={'depth':'No depth stream in selected official Aria raw recording','segment':'Ground-truth masks/*.csv are frame-selection predicates, not pixel masks; no source pixel segmentation provided here','contact':'No native contact labels in downloaded scope','keypoints':'No independent MANO21 keypoint array in this raw release; selected MANO parameters retained','alternative_hand_model':'UmeTrack representation retained in raw; this shared contract selects MANO','alternative_camera_estimates':'Raw MPS online calibration and SLAM trajectories retained; shared uses the official GT camera_models and headset trajectory'}
+        if not hand_times:v.m['omitted_components']['mano']='Official MANO trajectory is empty; no pose labels released for this sequence'
+        if not objects:v.m['omitted_components']['object_pose']='Official object trajectory is empty; static source models retained without fabricated pose arrays'
+        print('MEDIA',dataset,cam,t,flush=True);submit_media(v,lambda files=tuple(paths):file_members(files))
+    assetszip.close();gt.close();handszip.close()
+
+if __name__=='__main__':
+    p=argparse.ArgumentParser();p.add_argument('dataset',choices=['ARCTIC','HOT3D']);a=p.parse_args()
+    torch.set_num_threads(4)
+    {'ARCTIC':arctic,'HOT3D':hot3d}[a.dataset]()
+    if a.dataset=='HOT3D':
+        verify_hot3d_frame_order_and_metadata()
+        from apply_hot3d_qa import apply_qa
+        apply_qa()
+    elif (DATA_ROOT/'ARCTIC'/RELEASE/'doc/validation/exposure.json').exists():
+        from repair_arctic_pilot import repair
+        repair()
diff --git a/ARCTIC/shared-v1.0/script/diagnose_arctic_review.py b/ARCTIC/shared-v1.0/script/diagnose_arctic_review.py
new file mode 100644
index 0000000000000000000000000000000000000000..a6a76873a6890c7632983d1d811d329057068fda
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/diagnose_arctic_review.py
@@ -0,0 +1,125 @@
+"""Inspect source exposure and execute upstream articulation for the selected pilot.
+
+Read-only for raw and shared data. Writes portable evidence under doc/validation.
+"""
+from fresh_sequence_common import *
+from review_fresh_sequence_pilots import arr, project_for_view
+import ast
+
+SEQ='s09_laptop_use_03'
+BASE=DATA_ROOT/'ARCTIC'
+ROOT=BASE/RELEASE
+DOC=ROOT/'doc/validation'
+OFFICIAL=WORK/'tmp/fresh-sequence-pilot/official/ARCTIC'
+
+def exposure():
+    result={}
+    with zipfile.ZipFile(BASE/'raw/data/images_zips/s09/laptop_use_03.zip') as z:
+        for cam in map(str,range(9)):
+            names=sorted(n for n in z.namelist() if n.startswith(cam+'/') and n.endswith('.jpg'))
+            rows=[];canvas=Image.new('RGB',(10*160,180),(20,20,20))
+            from PIL import ImageDraw
+            draw=ImageDraw.Draw(canvas)
+            for i,n in enumerate(names):
+                im=Image.open(io.BytesIO(z.read(n)));im.draft('RGB',(160,160));im.thumbnail((160,150));gray=np.asarray(im.convert('L'))
+                rows.append({'frame_index':i,'source_frame_id':int(Path(n).stem),'median_luma':float(np.median(gray)),
+                    'p95_luma':float(np.percentile(gray,95)),'mean_luma':float(gray.mean()),'fraction_gt_32':float((gray>32).mean())})
+                if i<10:canvas.paste(im,(i*160,25));draw.text((i*160+3,5),f'idx {i}, src {i+1}',fill='white')
+            path=DOC/'diagnosis'/f'exposure-{cam}.jpg';path.parent.mkdir(parents=True,exist_ok=True);canvas.save(path,quality=94)
+            result[cam]={'method':'Official raw JPEG, PIL draft preview <=160x150, 8bit luma; diagnostics, not a general exposure classifier','frames':rows,'preview_ref':path.relative_to(DOC).as_posix()}
+            print('EXPOSURE',cam,'first10_p95',[r['p95_luma'] for r in rows[:10]],'p95_lt_35_indices',[r['frame_index'] for r in rows if r['p95_luma']<35],flush=True)
+    save(DOC/'exposure.json',result)
+    return result
+
+class Xdict(dict):
+    def overwrite(self,k,v):self[k]=v
+
+def official_object(objects,vertices,raw_parts,params,object_id='laptop'):
+    # Execute the pinned official class, bypassing its multi-object file loader.
+    # Only quaternion primitives use a SciPy implementation; no semantic rewriting.
+    tree=ast.parse((OFFICIAL/'common/object_tensors.py').read_text())
+    klass=next(n for n in tree.body if isinstance(n,ast.ClassDef) and n.name=='ObjectTensors')
+    def aa_quat(a):
+        a=a.numpy();q=Rotation.from_rotvec(a.reshape(-1,3)).as_quat().reshape(a.shape[:-1]+(4,))
+        return torch.tensor(q[...,[3,0,1,2]],dtype=torch.float32)
+    def qapply(q,v):
+        q=q.numpy();v=v.numpy();shape=np.broadcast_shapes(q.shape[:-1],v.shape[:-1]);qs=np.broadcast_to(q,shape+(4,)).reshape(-1,4)
+        out=Rotation.from_quat(qs[:,[1,2,3,0]]).apply(np.broadcast_to(v,shape+(3,)).reshape(-1,3)).reshape(shape+(3,))
+        return torch.tensor(out,dtype=torch.float32)
+    ns={'nn':torch.nn,'torch':torch,'np':np,'xdict':Xdict,'axis_angle_to_quaternion':aa_quat,'quaternion_apply':qapply}
+    exec(compile(ast.Module(body=[klass],type_ignores=[]),'pinned_ARCTIC_ObjectTensors','exec'),ns)
+    layer=ns['ObjectTensors'].__new__(ns['ObjectTensors']);torch.nn.Module.__init__(layer)
+    n=len(vertices);ft=lambda a:torch.tensor(np.asarray(a),dtype=torch.float32)
+    d={'names':[object_id],'diameter':ft([1]),'f':torch.zeros((1,1,3),dtype=torch.long),'f_len':torch.tensor([1]),'v_len':torch.tensor([n]),
+       'v':ft(vertices[None]),'mask':ft(np.ones((1,n))),'v_sub':ft(vertices[None]),
+       'parts_ids':torch.tensor(raw_parts[None]+1),'parts_sub_ids':torch.tensor(raw_parts[None]+1),'z_axis':ft([[0,0,-1]])}
+    for out,key in [('bbox_top','bbox_top'),('bbox_bottom','bbox_bottom'),('kp_top','keypoints_top'),('kp_bottom','keypoints_bottom')]:d[out]=ft([np.array(params[key])/1000])
+    layer.obj_tensors=d
+    out=layer.forward(ft(objects[:,:1]),ft(objects[:,1:4]),ft(objects[:,4:7]/1000),[object_id]*len(objects))
+    return out['v'].numpy(),out['kp3d'].numpy()
+
+def geometry():
+    with zipfile.ZipFile(BASE/'raw/data/meta.zip') as z:
+        pref='meta/object_vtemplates/laptop/'
+        vertices=np.array([[float(x) for x in r.split()[1:4]] for r in z.read(pref+'mesh.obj').decode().splitlines() if r.startswith('v ')])/1000
+        parts=np.array(json.loads(z.read(pref+'parts.json')),int);params=json.loads(z.read(pref+'object_params.json'))
+        misc=json.loads(z.read('meta/misc.json'))['s09']
+    with zipfile.ZipFile(BASE/'raw/data/raw_seqs.zip') as z:
+        objects=native_npy_bytes(z.read('raw_seqs/s09/laptop_use_03.object.npy'))
+        hand=native_npy_bytes(z.read('raw_seqs/s09/laptop_use_03.mano.npy'))
+    # Every annotation frame, a uniform vertex subset; plus every vertex at source 271.
+    vi=np.arange(0,len(vertices),max(1,len(vertices)//500));native,kp=official_object(objects,vertices[vi],parts[vi],params)
+    hands=[]
+    for side in ['left','right']:
+        model=Mano(side);h=hand[side];pose=np.c_[h['rot'],h['pose']];pose[:,3:]+=model.data['hands_mean']
+        xyz=model.geometry(pose,np.broadcast_to(h['shape'],(len(pose),10)),h['trans'])[0];hands.append(xyz)
+    tests=[]
+    for meta in sorted((ROOT/'data'/SEQ).glob('*/metainfo.json')):
+        p=meta.parent;m=load(meta);cam=m['camera_id'];source=BASE/'tmp/full-video-review-v2/diagnostic'/cam
+        if not (source/'snapshot.json').exists():source=p
+        E=arr(source/'camera/extrinsic.pt');P=arr(source/'annotation/object/laptop/pose.pt');top=arr(source/'annotation/object/laptop/parts/top/pose.pt')
+        actual=vertices[vi][None]@P[1:,:3,:3].transpose(0,2,1)+P[1:,None,:3,3]
+        sel=parts[vi]==0;actual[:,sel]=vertices[vi][sel][None]@top[1:,:3,:3].transpose(0,2,1)+top[1:,None,:3,3]
+        expected=native@E[1:,:3,:3].transpose(0,2,1)+E[1:,None,:3,3]
+        error=float(np.max(np.linalg.norm(actual-expected,axis=-1)));assert error<2e-5
+        motion=arr(source/'annotation/mano/motion.pt');handerrs=[]
+        for side in range(2):
+            model=Mano(['left','right'][side]);a=motion[1:,side];shared=model.geometry(a[:,:48],a[:,48:58],a[:,58:])[0]
+            expected_hand=hands[side]@E[1:,:3,:3].transpose(0,2,1)+E[1:,None,:3,3]
+            err=float(np.max(np.linalg.norm(shared-expected_hand,axis=-1)));assert err<2e-5;handerrs.append(err)
+        tests.append({'camera_id':cam,'annotation_frames':540,'vertices_per_frame':len(vi),'official_object_max_m':error,'source_to_shared_MANO_max_m':handerrs})
+        print('GEOMETRY',tests[-1],flush=True)
+    report={'issue':'Raw parts.json 0=top,1=bottom; previous converter wrongly treated 1 as top. Official code adds 1 and articulates parts_ids==1.',
+        'correction':'Shared convention remains 0=bottom,1=top; shared_id=1-raw_id. No time shift, pose change or image edit.',
+        'upstream_code':'https://github.com/zc-alexfan/arctic/blob/49f3eb4b7f01663172576357bb3d95695d7a7689/common/object_tensors.py',
+        'method':'Executed pinned upstream ObjectTensors class with actual raw tensor state and SciPy quaternion primitives, compared all 540 frames and 9 cameras. MANO verified independently from source world parameters.',
+        'tests':tests,'limitation':'Numerical agreement with source geometry does not establish source-to-image accuracy. Residual hand alignment remains for full-video human review.'}
+    save(DOC/'diagnosis.json',report)
+    return report
+
+def comparison():
+    from render_full_sequence_review import model_vertices, fast_edges, dots
+    p=ROOT/'data'/SEQ/'6';m=load(p/'metainfo.json');i=270
+    source=BASE/'tmp/full-video-review-v2/diagnostic/6'
+    if not (source/'snapshot.json').exists():source=p
+    with zipfile.ZipFile(BASE/'raw/data/images_zips/s09/laptop_use_03.zip') as z:rgb=np.array(Image.open(io.BytesIO(z.read('6/00271.jpg'))).convert('RGB').resize((500,700)))
+    with zipfile.ZipFile(BASE/'raw/data/meta.zip') as z:parts=np.array(json.loads(z.read('meta/object_vtemplates/laptop/parts.json')))
+    vertices=np.array([[float(x) for x in r.split()[1:4]] for r in (p/'annotation/object/laptop/assets/mesh.obj').read_text().splitlines() if r.startswith('v ')])
+    pose=arr(p/'annotation/object/laptop/pose.pt')[i];top=arr(p/'annotation/object/laptop/parts/top/pose.pt')[i]
+    K=arr(p/'camera/intrinsic.pt')[i];scale=np.array([.25,.25]);motion=arr(source/'annotation/mano/motion.pt')[i:i+1];valid=arr(source/'annotation/mano/motion_valid.pt')[i:i+1]
+    panels=[rgb.copy(),rgb.copy(),rgb.copy()]
+    for col,sel in [(1,parts==1),(2,parts==0)]:
+        xyz=vertices@pose[:3,:3].T+pose[:3,3];xyz[sel]=vertices[sel]@top[:3,:3].T+top[:3,3]
+        uv=project_for_view(xyz,m,(K,None),scale);dots(panels[col],uv[::4],(100,245,90),1)
+    for side,color in [(0,(0,225,240)),(1,(255,160,20))]:
+        v,j,e=model_vertices(motion,valid,side,False);uv=project_for_view(v[0],m,(K,None),scale)
+        for im in panels[1:]:fast_edges(im,uv,e,color)
+    canvas=np.full((750,1500,3),22,np.uint8)
+    for col,(im,title) in enumerate(zip(panels,['RAW source 271','OLD wrong object parts','FIXED official object parts'])):
+        canvas[45:745,col*500:(col+1)*500]=im;cv2.putText(canvas,title,(col*500+10,26),cv2.FONT_HERSHEY_SIMPLEX,.62,(250,250,250),1)
+    Image.fromarray(canvas).save(DOC/'diagnosis/camera6-270-before-after.jpg',quality=95)
+
+if __name__=='__main__':
+    import argparse
+    ap=argparse.ArgumentParser();ap.add_argument('mode',choices=['exposure','geometry','comparison']);args=ap.parse_args()
+    {'exposure':exposure,'geometry':geometry,'comparison':comparison}[args.mode]()
diff --git a/ARCTIC/shared-v1.0/script/diagnose_hot3d_alignment.py b/ARCTIC/shared-v1.0/script/diagnose_hot3d_alignment.py
new file mode 100644
index 0000000000000000000000000000000000000000..900891a64a5423524cebe337443877682f357cdb
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/diagnose_hot3d_alignment.py
@@ -0,0 +1,47 @@
+"""Independent native pose/bbox checks for the HOT3D opening-frame complaint."""
+from fresh_sequence_common import *
+from review_fresh_sequence_pilots import arr
+
+def pose(row):
+    T=np.eye(4);T[:3,:3]=Rotation.from_quat([float(row['q_wo_'+k]) for k in 'xyzw']).as_matrix();T[:3,3]=[float(row['t_wo_'+k+'[m]']) for k in 'xyz'];return T
+
+def rows(z,name):return list(csv.DictReader(io.StringIO(z.read(name).decode())))
+
+def diagnose():
+    base=DATA_ROOT/'HOT3D';root=base/RELEASE;seq='P0015_179e1b84';src=base/'raw/data/Aria'/seq;cache=base/'tmp/full-video-review-v2/diagnostic';doc=root/'doc/validation'
+    with zipfile.ZipFile(next(src.glob('*ground_truth.zip'))) as z:
+        cams=json.loads(z.read('camera_models.json'));heads={int(r['timestamp[ns]']):pose(r) for r in rows(z,'headset_trajectory.csv')}
+        objects={(int(r['timestamp[ns]']),r['object_uid']):pose(r) for r in rows(z,'dynamic_objects.csv')};boxes=rows(z,'box2d_objects.csv');masks=rows(z,'masks/mask_qa_pass.csv')
+    diagnostics=[];plate='248247003992116'
+    for c in cams:
+        cam=c['stream_id'];p=root/'data'/seq/cam;fr=[json.loads(x) for x in (p/'camera/frames.jsonl').read_text().splitlines()];meta=load(p/'metainfo.json')
+        dc=c['T_Device_Camera'];Tdc=np.eye(4);Tdc[:3,:3]=Rotation.from_quat(np.roll(dc['quaternion_wxyz'],-1)).as_matrix();Tdc[:3,3]=dc['translation_xyz']
+        nativeE=np.stack([np.linalg.inv(heads[f['timecode_ns']]@Tdc) for f in fr]);storedE=arr(p/'camera/extrinsic.pt');eerr=float(np.max(np.abs(nativeE-storedE)));assert eerr<2e-6
+        rawhead=np.stack([heads[f['timecode_ns']] for f in fr]);firstchange=next(i for i in range(len(fr)) if not np.array_equal(rawhead[i],rawhead[0]))
+        op=p/'annotation/object'/plate;om=load(op/'metainfo.json');mesh=trimesh.load(p/om['original_glb_ref'],process=False,force='mesh');v=np.asarray(mesh.vertices)
+        storedP=arr(cache/cam/'annotation/object'/plate/'pose.pt');qa={int(r['timestamp[ns]']):r['mask']=='True' for r in masks if r['stream_id']==cam}
+        lookup={int(r['timestamp[ns]']):r for r in boxes if r['stream_id']==cam and r['object_uid']==plate}
+        tests=[]
+        for i in [0,1,26,27,179,180,253,507,760]:
+            ts=fr[i]['timecode_ns'];P=nativeE[i]@objects[(ts,plate)];error=float(np.max(np.abs(P-storedP[i])));assert error<2e-6
+            xyz=v@P[:3,:3].T+P[:3,3];uv=project_points(xyz,'fisheye624',c['projectionParams']);good=np.isfinite(uv).all(1)
+            box=lookup.get(ts);diff=None;expected=None
+            if box and box['x_min[pixel]'] and good.all():
+                expected=np.array([float(box[k]) for k in ['x_min[pixel]','y_min[pixel]','x_max[pixel]','y_max[pixel]']]);bb=np.r_[uv.min(0),uv.max(0)];diff=float(np.max(np.abs(expected-bb)))
+            tests.append({'frame_index':i,'source_frame_id':fr[i]['source_frame_id'],'official_qa_pass':qa[ts],'native_vs_cached_pose_max_abs':error,'source_plate_bbox_vs_projection_max_pixel':diff})
+            if i in [0,180]:
+                files=sorted((base/'tmp'/RUN/seq/'images').glob(cam+'-*.jpg'));image=np.asarray(Image.open(files[i]).convert('RGB')).copy()
+                if expected is not None:cv2.rectangle(image,tuple(expected[:2].astype(int)),tuple(expected[2:].astype(int)),(255,235,0),2)
+                inds=np.arange(0,len(uv),max(1,len(uv)//5000));pix=uv[inds];ok=np.isfinite(pix).all(1)&(pix[:,0]>=0)&(pix[:,0]<image.shape[1])&(pix[:,1]>=0)&(pix[:,1]<image.shape[0]);pix=np.rint(pix[ok]).astype(int)
+                image[np.clip(pix[:,1],0,image.shape[0]-1),np.clip(pix[:,0],0,image.shape[1]-1)]=[50,255,80]
+                cv2.putText(image,f'idx {i}: native QA={qa[ts]} | yellow=official box, green=native pose',(8,24),cv2.FONT_HERSHEY_SIMPLEX,.45,(255,255,255),1)
+                out=doc/'diagnosis';out.mkdir(parents=True,exist_ok=True);Image.fromarray(image).save(out/f'{cam}-{i:04d}.png')
+        diagnostics.append({'camera_id':cam,'native_vs_shared_extrinsic_max_abs':eerr,'initial_identical_headset_pose_frames':firstchange,'tests':tests})
+    report={'conclusion':'Native official hand/object poses already fail official manual QA at the complained frames. Shared media order and native timestamp matching are unchanged; independent native pose reconstruction reproduces the diagnostic geometry. Do not time-shift labels to hide rejected source data.',
+        'unproven':'Exact upstream reason for each QA rejection is not provided; initial repeated headset poses are an observation, not a claim of the unique cause.',
+        'official_qa_semantics':'Manual QA found no issue with poses of visible hands/objects when true; pose_available does not assess accuracy.',
+        'source':'https://github.com/facebookresearch/hot3d/blob/146b34afef8c1a32adeef7e981c070109f225c87/README.md#using-mask-files','cameras':diagnostics}
+    save(doc/'diagnosis.json',report)
+    for d in diagnostics:print(d['camera_id'],'head frozen frames',d['initial_identical_headset_pose_frames'],'native box errors',[x['source_plate_bbox_vs_projection_max_pixel'] for x in d['tests']],flush=True)
+
+if __name__=='__main__':diagnose()
diff --git a/ARCTIC/shared-v1.0/script/document_fresh_sequence_pilots.py b/ARCTIC/shared-v1.0/script/document_fresh_sequence_pilots.py
new file mode 100644
index 0000000000000000000000000000000000000000..0ee7772d868a4ee508d515dadceed8f5ad84620e
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/document_fresh_sequence_pilots.py
@@ -0,0 +1,90 @@
+"""Write source-to-contract mappings for the two fresh sequence pilots."""
+from fresh_sequence_common import *
+
+COMMON='''
+## 公共约定与运行
+
+这是完整短序列试点，不是全量发布。每个视图使用 [CONTRACT.md](CONTRACT.md) 与 [schema.json](schema.json)。输入是本轮全新下载的 E 盘 raw，不读取 F 盘旧数据或旧试点产物。原始 ZIP/VRS 不改写，临时解包与日志只在数据集 tmp。
+
+统一下标 `frame_index=0..T-1`；原帧号与已知时间在 `camera/frames.jsonl`。数值张量是 CPU float32，长度为米，角度为弧度；布尔 valid 表示是否有对应可用标注，缺失的浮点位置填 NaN。valid 不表示可见性。手槽固定左、右。
+
+图像由源 JPEG 直接装入 MJPEG/MKV，每包字节不变。容器使用 1 fps、`timebase_role=index_only`，不能按播放器的时长推断真实时间。FFmpeg 固定关闭自动旋转、输出 rgb24，源/目标逐帧 SHA256 相等。实际版本与像素格式写入各视图 metainfo。预览图缩小以方便审核，不是训练输入。
+
+校验范围和数值结果见 [validation/REPORT.md](validation/REPORT.md) 与 [validation/report.json](validation/report.json)，原始输入 SHA256 见 [pilot-source-manifest.json](pilot-source-manifest.json)。源官方摘要和原包 CRC 在 raw/doc 的下载验收记录中。人工确认仍待完成，不以数学一致性替代投影核验。
+
+运行脚本来自本项目 scripts；已将所需本地模块快照放入本发布 script。当前运行环境为 Python 3、PyTorch 2.5.1 CPU、NumPy、smplx、SciPy、Pillow、OpenCV、trimesh、imageio-ffmpeg；HOT3D 另需官方 VRS 工具。MANO v1.2 模型由使用者依授权在本地提供，不包含在发布目录。
+
+脚本默认 `E:/hand-datasets` 数据根目录，项目 `tmp/pilot-runtime` 运行环境和 `tmp/pilot-models` 模型目录；迁移时需按机器调整 `fresh_sequence_common.py` 的 DATA_ROOT/VRS 和本地模型路径。这份试点脚本不是无需配置的独立安装包。
+
+```text
+python scripts/convert_fresh_sequence_pilots.py ARCTIC
+python scripts/convert_fresh_sequence_pilots.py HOT3D
+python scripts/document_fresh_sequence_pilots.py
+python scripts/diagnose_arctic_review.py exposure
+python scripts/diagnose_arctic_review.py geometry
+python scripts/repair_arctic_pilot.py
+python scripts/render_full_sequence_review.py ARCTIC
+python scripts/render_full_sequence_review.py HOT3D
+```
+'''
+
+ARCTIC='''# ARCTIC：raw → shared 字段映射
+
+本序列原包的 top.obj / bottom.obj 引用 material_0.mtl，但包内实际为 material.mtl；已核对 newmtl/usemtl 均为 material_0。共享层仅修正 mtllib 文件路径，不改材质、UV或纹理，raw 原包保持不变。
+
+本次选择 `s09/laptop_use_03`，原图 ZIP 的全部九路相机各 541 张。输出序列名 `s09_laptop_use_03`；相机 0 为 ego，其余 1–8 为 exo。选择官方 P1，split=train。
+
+2026-09-20复核：此前顶盖/底座编码解释反了，现已按官方ObjectTensors重新映射。9路相机的index 0/1是原始欠曝图，各动态几何组件valid=false；camera 6的index 270手部对齐问题仅记录，按用户要求撤销单帧屏蔽，恢复原MANO数值及valid=true。不针对单个可疑投影硬编码排除。曝光筛选不构成其他帧质量保证。整套试点未验收，详见[问题与证据](validation/ARCTIC_FINDINGS.md)。
+
+| raw 中实际数据 | 共享层 | 转换与保留范围 |
+|---|---|---|
+| images_zips/s09/laptop_use_03.zip，0–8/00001.jpg…00541.jpg | observation/rgb.mkv | 所有原图，无裁剪、缩放、旋转或重新压缩；全部 4869 帧 |
+| meta.zip：misc.json 的 ioi_offset=2 | camera/frames.jsonl | 动捕索引=原图编号−2；原图 1 没有标注，2…541 对应标注 0…539；同步标识含源序列和图号 |
+| raw_seqs.zip：*.egocam.dist.npy 的 intrinsics、dist8、R_k_cam_np、T_k_cam_np | camera/intrinsic.pt、distortion.pt、extrinsic.pt | 相机 0 rational8，畸变顺序 k1,k2,p1,p2,k3,k4,k5,k6；外参 camera_from_world；首图外参缺失，其余 540 帧有效 |
+| meta.zip：misc.json 的 intris_mat、world2cam | 八个外部相机的 camera/ | 官方 processing.forward_project2d 以针孔模型投影；静态参数展开为 T 份，全部 541 帧有标定 |
+| raw_seqs.zip：*.mano.npy，左右 rot、pose、shape、trans | annotation/mano/motion.pt [541,2,61] | 非平手均值加回，48 轴角+10形状+3平移；世界系转相机系，按形状相关腕根修正 translation；首帧两手 valid=false |
+| raw_seqs.zip：*.object.npy [540,7] | object/laptop/pose.pt、parts/*/pose.pt、joint_state.pt | 角度 rad；全局旋转轴角转 SE(3)；物体平移 mm→m；顶盖相对底座绕 -Z 旋转，首帧无效 |
+| meta.zip：object_vtemplates/laptop/mesh.obj、top.obj、bottom.obj、MTL/JPG | object/laptop/assets/ | OBJ 顶点 mm→m；顶点顺序/拓扑/UV、材质和纹理引用保留；每个视图一份 |
+| 同目录 parts.json | object/laptop/vertex_part_ids.pt 和 metainfo.json | 源0顶盖/1底座，转换为 int64：shared_id=1-source_id，统一0底座/1顶盖；绑定 mesh_ref 和 topology_id |
+| object_params.json：keypoints_top、keypoints_bottom | object/laptop/keypoints/xyz.pt [541,32,3] | 原模板各16点的固定顺序，按已发布姿态驱动到相机系；标记 derived，不冒充实测手关键点 |
+| splits_json.zip：protocol_p1.json | 视图 metainfo 的 split 与来源 | 保留官方划分；不随意拆帧 |
+
+未作为 shared 模态发布：fitting_err、身体数据、动捕 marker、替代采样产物。选定原始包没有独立深度图、像素分割、接触标签或实测手关键点数组；MANO 派生网格只在内存中做校验，不保存 hand_mesh。当前 raw 未提供可直接核实的逐帧时间，timestamp_ns 与 source_fps 保留 null，不用容器 1 fps 伪造真实采集时间。
+
+依据固定官方提交 [`49f3eb4`](https://github.com/zc-alexfan/arctic/tree/49f3eb4b7f01663172576357bb3d95695d7a7689)：`src/datasets/arctic_dataset.py` 的 ioi_offset 映射、`common/body_models.py` 的非平手均值、`common/object_tensors.py` 的 -Z 关节轴与毫米单位、`src/arctic/processing.py` 和 `common/transforms.py` 的九相机投影。
+'''
+
+HOT='''# HOT3D：raw → shared 字段映射
+
+源标注边界：原始 MANO JSONL 共3042条相机时刻记录，2418条含手参数，624条 hand_poses 为空。各视图1014帧中806帧含双手参数、208帧无参数；最终 valid 还要与该相机官方 mask_qa_pass 取交集，不把806称为有效监督数。官方QA是人工检查可见手/物体姿态是否有问题，无法定位到某一实体；这里保守排除整相机帧的手物监督。图像保留，源QA记录和逐帧计数见 validation/source_qa.json。
+
+本次选择完整 Aria 序列 `P0015_179e1b84`。三路相机各 1014 帧，源 nominal_rate=30 Hz，首尾采集时间跨度约33.77秒；没有 Quest。
+
+| raw 中实际数据 | 共享层 | 转换与保留范围 |
+|---|---|---|
+| *_main_recording.vrs：214-1、1201-1、1201-2 | 各视图 observation/rgb.mkv | 原 JPEG 包；RGB 1408×1408，两灰度640×480。统一读取uint8 RGB，灰度重复到三通道；保留相机类型说明 |
+| VRS data 的 frame_number、capture_timestamp_ns | camera/frames.jsonl | 原设备时间与帧号；检查提取文件的记录序号和时间与 VRS 一致 |
+| ground_truth.zip：timecode_devicetime_mapping.csv | frames 的 timecode_ns、mapping_delta_ns | 官方 device→timecode 映射，允许最多1000ns定位差并记录实差；标注在 timecode 域内匹配 |
+| ground_truth.zip：camera_models.json | camera/intrinsic.pt [1014,15] | 官方静态 FISHEYE624 模型，保留全部15参数及 maxSolidAngle；不改成针孔或人为去畸变 |
+| ground_truth.zip：headset_trajectory.csv 和 T_Device_Camera | camera/extrinsic.pt [1014,4,4] | inverse(world_from_device @ device_from_camera)，平移米；原标注与图像 timecode 最近邻，最大允许1ms，并记录实际偏差 |
+| hand_data.zip：mano_hand_pose_trajectory.jsonl | annotation/mano/motion.pt [1014,2,61] | 源 PCA15 按 MANO v1.2 官方主成分展开+均值；保留10维shape，源q_wxyz转轴角；修正官方左手shapedirs问题；转当前相机系 |
+| ground_truth.zip：box2d_hands.csv 的 visibility_ratio[%] | annotation/mano/visibility.pt [1014,2] | 实际源值已在0–1，按官方加载器直接保留，不除100；独立有效性还需通过姿态QA；0可见比例不是缺失；不引入quality |
+| ground_truth.zip：dynamic_objects.csv，6个object_uid | object/<uid>/pose.pt [1014,4,4] | 源 world_from_object 转 camera_from_object；米，固定实例UID在序列各相机间保持对应 |
+| ground_truth.zip：box2d_objects.csv 的 visibility_ratio[%] | 各物体 visibility.pt [1014] | 比例与独立有效性，归所属物体组件属性 |
+| assets/hot3d_models.zip 的 BOP PLY+PNG+models_info.json | object/<uid>/assets/*.obj、MTL、纹理 | BOP id 通过 original_id 对应原UID；PLY顶点mm→m。检查尺寸、三角形及纹理像素往返一致 |
+| assets/object_models/obj_*.glb | 各物体 assets 中原生材质伴随资产 | 按 BOP id 对齐，原文件字节不变，保留可能超出 OBJ/MTL 的材质信息；标准 mesh_ref 始终指向米制 OBJ |
+
+本序列使用6个物体：spatula_red、potato_masher、can_parmesan、plate_bamboo、carton_oj、bottle_ranch。部件为刚体，未伪造机械关节。图像/手/物体的标注匹配与最大误差在各帧 annotation_matches 明确保存。整手/物体 visibility 的源时间匹配阈值另在属性 source_locator 声明。
+
+缺省与选择：无深度相机流；`masks/*.csv` 是帧条件，不是像素分割，QA已合入手/物体valid及依赖姿态的visibility有效性。没有额外生成接触、手关键点或独立hand_mesh。UmeTrack参数、MPS在线相机与SLAM轨迹保留在raw；本版shared选择MANO和ground_truth内相机参数。依据固定提交官方README的参与者划分，P0015属于train，metainfo记录来源。
+
+转换语义依据 HOT3D 官方固定提交 [`146b34a`](https://github.com/facebookresearch/hot3d/tree/146b34afef8c1a32adeef7e981c070109f225c87) 的 `mano_layer.py`、`loader_hand_poses.py`、`AriaDataProvider.py`。物体包依据官方 BOP HOT3D仓库固定提交 [`30fe967`](https://huggingface.co/datasets/bop-benchmark/hot3d/tree/30fe9674782f32e1e5edba98476b6ff4300132c5)。Fisheye624预览按Project Aria官方FisheyeRadTanThinPrism公式实现；内参原值不修改。
+'''
+
+if __name__=='__main__':
+    for dataset,text in [('ARCTIC',ARCTIC),('HOT3D',HOT)]:
+        root=DATA_ROOT/dataset/RELEASE
+        (root/'doc/RAW_TO_SHARED.md').write_text(text+COMMON,encoding='utf-8')
+        shutil.copyfile(Path(__file__),root/'script'/Path(__file__).name)
+        (root/'script/README.md').write_text('运行条件与配置见 ../doc/RAW_TO_SHARED.md。\n模型权重不分发；日志与提取缓存不分发。\n',encoding='utf-8')
+        print('DOCUMENTED',dataset)
diff --git a/ARCTIC/shared-v1.0/script/finalize_arctic_objects_review.py b/ARCTIC/shared-v1.0/script/finalize_arctic_objects_review.py
new file mode 100644
index 0000000000000000000000000000000000000000..f4a135498c7c811431030ee490cded96c92e5cbf
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/finalize_arctic_objects_review.py
@@ -0,0 +1,144 @@
+"""Assemble evidence after the bounded eleven-object review; never authorizes production."""
+from arctic_object_review import *
+from html.parser import HTMLParser
+from urllib.parse import unquote, urlsplit
+
+NOTES=WORK/'tmp/arctic-object-visual-review.json'
+
+def asset_contacts():
+    selection=load(SELECTION);outputs=[]
+    for item in selection['objects']:
+        _,_,oid,seqid,doc,tmp=context(item)
+        video=doc/'assets'/f'{oid}.mp4'
+        if not video.exists():continue
+        asset=asset_preview(OUT/'data'/seqid/'6'/'annotation/object'/oid,video)
+        r=load(doc/'report.json');r['asset_preview']=asset;save(doc/'report.json',r)
+        target=tmp/'asset-angles.jpg'
+        command([FF,'-v','error','-y','-threads','2','-i',video,'-vf',
+            'select=eq(n\\,12)+eq(n\\,36)+eq(n\\,60),scale=480:-1,tile=3x1',
+            '-frames:v','1','-threads','2',target])
+        outputs.append((oid,target))
+    for group,start in enumerate(range(0,len(outputs),6),1):
+        chosen=outputs[start:start+6];images=[Image.open(p).convert('RGB') for _,p in chosen]
+        height=max(im.height for im in images)+32
+        sheet=Image.new('RGB',(1440,height*len(images)),(20,20,20));draw=ImageDraw.Draw(sheet)
+        for i,((oid,_),im) in enumerate(zip(chosen,images)):
+            draw.text((10,i*height+6),f'{oid}: turntable frames 12 / 36 / 60',fill='white')
+            sheet.paste(im,(0,i*height+32))
+        target=DOC/f'assets-contact-{group}.jpg';sheet.save(target,quality=94);print(target,flush=True)
+
+class Links(HTMLParser):
+    def __init__(self):super().__init__();self.links=[]
+    def handle_starttag(self,tag,attrs):
+        for k,v in attrs:
+            if k in ('href','src','poster') and v:self.links.append(v)
+
+def finalize():
+    selection=load(SELECTION);notes=json.loads(NOTES.read_text(encoding='utf-8-sig'))
+    assert set(notes)=={x['object'] for x in selection['objects']}
+    assert load(TMP/'status.json')['stage'] in ('all_objects_rendered_pending_visual_inspection','review_complete_pending_user')
+    reports=[];cameras=[];extra=[];old=load(DOC.parent/'report.json')
+    for item in selection['objects']:
+        seq,sid,oid,seqid,doc,tmp=context(item);r=load(doc/'report.json');p=OUT/'data'/seqid/'6'
+        assert r['reference']['ioi_offset']==1 and r['camera_frames']==r['reference']['annotation_frames']+1
+        assert notes[oid]['status']=='reviewed_with_findings' and notes[oid].get('asset_visual_review')
+        r['agent_visual_review']=notes[oid]
+        r['agent_visual_review']['inspected_frame_indices']=r['views'][0]['visual_sample_indices']
+        r['agent_visual_review']['entire_video_watched']=False
+        r['agent_visual_review']['sample_method']='beginning/end, uniform temporal samples, joint-angle extremes; exact indices recorded'
+        K=arr(p/'camera/intrinsic.pt');E=arr(p/'camera/extrinsic.pt')
+        assert np.array_equal(K,np.broadcast_to(K[0],K.shape)) and np.array_equal(E,np.broadcast_to(E[0],E.shape))
+        cameras.append((K[0],E[0]))
+        m=load(p/'metainfo.json');assert not m['components']['mano'].get('valid_policy',{}).get('local_excluded_frame_indices')
+        m['sampling']['selection']='Every released image of this selected camera in the complete official sequence ZIP; boundary images without annotations retained'
+        save(p/'metainfo.json',m)
+        view=r['views'][0];v=doc/view['video']['video_ref'];assert digest(v)==view['video']['video_sha256']
+        decoded=framehash(['-i',str(v)]);assert len(decoded)==r['camera_frames']
+        asset=doc/r['asset_preview']['video_ref'];assert len(framehash(['-i',str(asset)]))==72
+        texture_checks=[]
+        with zipfile.ZipFile(BASE/'raw/data/meta.zip') as z:
+            for f in (p/f'annotation/object/{oid}/assets').rglob('*'):
+                if f.suffix.lower() in ('.mtl','.jpg','.jpeg','.png'):
+                    source=z.read(f'meta/object_vtemplates/{oid}/'+f.name)
+                    assert f.read_bytes()==source
+                    texture_checks.append({'file':f.name,'sha256':digest(f),'raw_bytes_equal':True})
+        r['delivery_checks']={'preview_fully_decoded_frames':len(decoded),'preview_sha256':digest(v),
+            'asset_preview_frames':72,'asset_video_sha256':digest(asset),'texture_and_material_files':texture_checks}
+        save(doc/'report.json',r);write_object_html(r);reports.append(r)
+        extra.append({'object':oid,'preview_frames':len(decoded),'asset_frames':72,'texture_files_checked':len(texture_checks)})
+        print('FINAL_CHECK',oid,len(decoded),flush=True)
+    assert all(np.array_equal(k,cameras[0][0]) and np.array_equal(e,cameras[0][1]) for k,e in cameras)
+    oldviews=old['full_video_views'];assert len(oldviews)==9
+    for v in oldviews:assert digest(DOC.parent/v['video_ref'])==v['video_sha256']
+    # The same canonical laptop asset is shared by both review groups; fix its
+    # original report's texture preview too, without touching its nine frame videos.
+    laptop=next(r for r in reports if r['object']=='laptop')
+    shutil.copyfile(DOC/'laptop'/laptop['asset_preview']['video_ref'],DOC.parent/'assets/laptop.mp4')
+    old['asset_previews']=[laptop['asset_preview']]
+    overview(selection)
+    total=sum(r['camera_frames'] for r in reports);n=sum(len(r['views'][0]['visual_sample_indices']) for r in reports)
+    maximum=max(r['maximum_geometry_error_m'] for r in reports)
+    rows='\n'.join(f'| [{r["object"]}]({r["object"]}/index.html) | {r["sequence"]} | {r["camera_frames"]} | {r["maximum_geometry_error_m"]:.3g} | {r["agent_visual_review"]["summary"]} |' for r in reports)
+    text=f'''# ARCTIC 11种物体验收结果
+
+[固定相机切换物体](index.html) · [原laptop九视角](../index.html) · [方法](README.md)
+
+新增组：s06、静态相机6，11个完整序列，共{total}帧；内外参逐帧恒定且在11个物体间完全一致。原s09/laptop_use_03九视角视频摘要未变。
+
+**转换数值检查通过；源图像和标注质量仍有已知问题，用户确认待完成。** 不据此启动全量处理或上传。
+
+| 物体 | 完整序列 | 帧数 | 最大几何差/m | 视觉抽查结果 |
+|---|---|---:|---:|---|
+{rows}
+
+## 检查边界
+
+- 数值：全部有标注帧、全部物体顶点/关键点和双手MANO几何，对照raw及固定官方实现；最大差{maximum:.6g}m，容差2e-5m。交换部件、反转关节角、误用毫米的反例在11个物体上均被检出。
+- 媒体：全部原JPEG包和统一解码RGB逐帧一致；完整预览视频再次全部解码，无缺帧/解码错误。MP4仅用于doc展示，不替代原码流MKV。
+- 视觉：实际查看{n}个原图/投影对照时刻（每段的起止、均匀样本及开合极值），检查资产旋转预览的3个方位。完整视频已全部渲染；代理未逐帧人工观看全部视频。
+- 模态：MANO、物体姿态/铰接/32个模板关键点、内外参、同步索引、RGB、资产纹理、valid均有检查/展示。未提供的深度、像素分割、接触、源visibility没有伪造。
+- 开头暗帧与局部手指贴合误差如实记录；本轮没有修改按帧号的排除mask。末帧无源动捕标注是源范围边界，RGB保留、几何valid=false。
+- 数值忠实不证明原标注准确；本次检查11种物体各一段，不能推导其它所有序列质量通过。
+
+[资产预览抽查1](assets-contact-1.jpg) · [资产预览抽查2](assets-contact-2.jpg)
+'''
+    (DOC/'REPORT.md').write_text(text,encoding='utf-8')
+    summary=load(DOC/'report.json');summary.update(camera_parameter_equality='exact_across_all_11_views_and_all_frames',
+        total_frames=total,visually_inspected_frame_count=n,maximum_geometry_error_m=maximum,
+        original_nine_camera_videos_unchanged=True,delivery_checks=extra,known_issues=['Source dark opening frames','Residual hand-to-image fit differences','Source annotation absent on final image'],
+        interpretation='All-frame faithful conversion passed; sampled visual review has findings; source quality and human acceptance not asserted')
+    save(DOC/'report.json',summary)
+    # Add navigation to the original page without rebuilding or replacing its nine videos.
+    home=DOC.parent/'index.html';body=home.read_text(encoding='utf-8');marker='id="eleven-object-review"'
+    if marker not in body:
+        body=body.replace('</header>','<p id="eleven-object-review"><strong>新增：</strong><a href="objects/index.html">固定相机6，切换11种物体的完整序列预览</a> · <a href="objects/REPORT.md">逐物体验收结果</a></p></header>',1)
+        home.write_text(body,encoding='utf-8')
+    mapping=OUT/'doc/RAW_TO_SHARED.md';body=mapping.read_text(encoding='utf-8')
+    if '## 新增固定相机物体组' not in body:
+        body+='\n## 新增固定相机物体组\n\n在原s09/laptop_use_03九视角之外，新增s06相机6的11物体完整序列。字段规则相同，具体选择、来源、帧数和校验见[逐物体报告](validation/objects/REPORT.md)及[来源摘要](validation/objects/source-manifest.json)。本组ioi_offset=1，图像编号减1对应标注索引；末图超出标注范围，保留RGB且几何valid=false。本组前两帧暗图只记录，没有套用旧s09曝光排除。仍非全量验收。\n'
+        mapping.write_text(body,encoding='utf-8')
+    for name in ['arctic_object_review.py','finalize_arctic_objects_review.py','convert_fresh_sequence_pilots.py','render_full_sequence_review.py','diagnose_arctic_review.py','fresh_sequence_common.py','review_fresh_sequence_pilots.py']:
+        shutil.copyfile(WORK/'scripts'/name,OUT/'script'/name)
+    manifest=[{'path':p.relative_to(OUT).as_posix(),'sha256':digest(p)} for p in sorted((OUT/'script').glob('*.py'))]
+    save(DOC/'script-manifest.json',manifest)
+    old.setdefault('previous_script_manifest_before_object_review',old['script_manifest']);old['script_manifest']=manifest
+    old['additional_fixed_camera_group']={'report_ref':'objects/report.json','sequence_count':11,'camera_id':'6','subject_id':'s06','human_review':'pending'}
+    save(DOC.parent/'report.json',old)
+    missing=[]
+    for page in [home,*DOC.rglob('*.html')]:
+        parser=Links();parser.feed(page.read_text(encoding='utf-8'))
+        for url in parser.links:
+            parts=urlsplit(url)
+            if parts.scheme or not parts.path:continue
+            if not (page.parent/unquote(parts.path)).exists():missing.append([str(page),url])
+    assert not missing,missing
+    artifacts=[{'path':p.relative_to(DOC).as_posix(),'size':p.stat().st_size,'sha256':digest(p)} for p in sorted(DOC.rglob('*')) if p.is_file() and p.name!='artifact-manifest.json']
+    save(DOC/'artifact-manifest.json',{'files':artifacts,'offline_html_links':'passed','log_files_in_report':False})
+    status={'stage':'review_complete_pending_user','completed_views_total':11,'human_review':'pending','report':str(DOC/'index.html'),'full_preprocessing_authorized':False,'upload_authorized':False}
+    save(TMP/'status.json',status);save(DOC.parent/'status.json',status)
+    state=load(WORK/'tmp/dataset-work-state.json');state['active']['ARCTIC'].update(stage='pilot_two_group_review_awaiting_human_review',object_review_report=str(DOC/'index.html'),object_review_completed=11);save(WORK/'tmp/dataset-work-state.json',state)
+    print('FINALIZED',total,n,maximum,flush=True)
+
+if __name__=='__main__':
+    parser=argparse.ArgumentParser();parser.add_argument('mode',choices=['assets','finalize']);args=parser.parse_args()
+    asset_contacts() if args.mode=='assets' else finalize()
diff --git a/ARCTIC/shared-v1.0/script/finalize_fresh_sequence_review.py b/ARCTIC/shared-v1.0/script/finalize_fresh_sequence_review.py
new file mode 100644
index 0000000000000000000000000000000000000000..06b629a9af67ecea87e9692de2b5f631995a4458
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/finalize_fresh_sequence_review.py
@@ -0,0 +1,48 @@
+"""Check full-video artifacts; never infer visual acceptance from numeric success."""
+from fresh_sequence_common import *
+from review_fresh_sequence_pilots import validate_view, arr
+from html.parser import HTMLParser
+
+class Links(HTMLParser):
+    def __init__(self):super().__init__();self.refs=[];self.videos=[]
+    def handle_starttag(self,tag,attrs):
+        a=dict(attrs)
+        for k in ['href','src']:
+            if a.get(k):self.refs.append(a[k])
+        if tag=='video':self.videos.append(a.get('src'))
+
+def finalize():
+    for ds,n in [('ARCTIC',9),('HOT3D',3)]:
+        root=DATA_ROOT/ds/RELEASE;doc=root/'doc/validation';r=load(doc/'report.json')
+        assert r['human_review']=='pending' and len(r['full_video_views'])==n
+        parser=Links();parser.feed((doc/'index.html').read_text(encoding='utf-8'))
+        assert len(parser.videos)==n+len(r['asset_previews'])
+        for ref in parser.refs:
+            if not ref.startswith(('http:','https:','#')):assert (doc/ref).is_file(),ref
+        for view in r['full_video_views']:
+            file=doc/view['video_ref'];assert digest(file)==view['video_sha256']
+            assert len(framehash(['-i',str(file)],copy=True))==view['frames']==view['rendered_frames']
+        for p in (root/'data').glob('*/*/metainfo.json'):validate_view(p.parent)
+        if ds=='ARCTIC':
+            with zipfile.ZipFile(DATA_ROOT/ds/'raw/data/meta.zip') as z:raw=np.asarray(json.loads(z.read('meta/object_vtemplates/laptop/parts.json')),np.int64)
+            for mp in (root/'data/s09_laptop_use_03').glob('*/metainfo.json'):
+                p=mp.parent;m=load(mp);assert np.array_equal(arr(p/'annotation/object/laptop/vertex_part_ids.pt'),1-raw)
+                for c in m['components'].values():
+                    if c.get('valid_policy',{}).get('local_excluded_frame_indices'):
+                        indices=c['valid_policy']['local_excluded_frame_indices'];assert not arr(p/c['valid_ref'])[indices].any()
+                assert m['acceptance']['training_release_approved'] is False
+            r['stage']='blocked_pending_visual_review';r['agent_visual_inspection']={'status':'known_residual_hand_alignment_issue','scope':'user screenshots, camera6 frame270 before/after, source exposure contact sheets and selected video snapshots','remaining':'full-camera sequence human acceptance; no claim of all-frame visual pass'}
+        else:
+            r['agent_visual_inspection']={'status':'source_QA_failures_excluded_new_full_video_review_pending','scope':'official QA, independent native pose/bbox checks; diagnostic snapshots','remaining':'human review of full sequence'}
+        for p in (root/'script').glob('*.py'):compile(p.read_text(encoding='utf-8-sig'),str(p),'exec')
+        assert not list(root.rglob('*.log')) and not list((root/'data').rglob('hand_mesh*'))
+        r['offline_artifact_check']={'relative_links':'passed','full_camera_videos':n,'frames':sum(v['frames'] for v in r['full_video_views']),'video_hash_and_frame_counts':'passed','tensor_validity_and_refs':'passed','script_compile':'passed','visual_acceptance':'not_granted'}
+        save(doc/'report.json',r);save(doc/'status.json',{'stage':r['stage'],'human_review':'pending','full_preprocessing_authorized':False,'upload_authorized':False})
+        shutil.copyfile(Path(__file__),root/'script'/Path(__file__).name)
+        print(ds,'ARTIFACT_CHECK_PASSED','human acceptance remains pending',flush=True)
+    p=WORK/'tmp/dataset-work-state.json';state=load(p)
+    for ds in ['ARCTIC','HOT3D']:
+        state['active'][ds].update(stage='blocked_pending_visual_review' if ds=='ARCTIC' else 'pilot_full_video_awaiting_human_review',human_sample_review='pending',full_preprocessing_authorized=False,upload_authorized=False)
+    save(p,state)
+
+if __name__=='__main__':finalize()
diff --git a/ARCTIC/shared-v1.0/script/fresh_sequence_common.py b/ARCTIC/shared-v1.0/script/fresh_sequence_common.py
new file mode 100644
index 0000000000000000000000000000000000000000..082d5389619a845fc254c899399971bd4ce59a1b
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/fresh_sequence_common.py
@@ -0,0 +1,212 @@
+"""Helpers for the fresh ARCTIC/HOT3D whole-sequence v1.0 pilots.
+
+Original archives are read only. Temporary extraction is outside raw.
+Licensed MANO weights stay in the local model directory and are never published.
+"""
+from sequence_pilot_world import camera_mano, Mano
+from pilot_build import *
+from scipy.spatial.transform import Rotation
+import hashlib
+import re
+import time
+import trimesh
+from concurrent.futures import ThreadPoolExecutor,wait,FIRST_COMPLETED
+from contextlib import contextmanager
+
+DATA_ROOT = Path('E:/hand-datasets')
+RELEASE = 'shared-v1.0'
+RUN = 'fresh-sequence-pilot-20260920'
+VRS = WORK/'scripts/tools/vrs_env/Library/bin/vrs.exe'
+MEDIA_POOL=None
+MEDIA_PENDING=[]
+MEDIA_WORKERS=1
+MEDIA_POLICY='legacy_full_pixels'
+OUTPUT_OVERRIDE=None
+TMP_OVERRIDE=None
+
+@contextmanager
+def media_pipeline(workers=1,policy='jpeg-packets-all-rgb-sample16/v1'):
+    global MEDIA_POOL,MEDIA_PENDING,MEDIA_WORKERS,MEDIA_POLICY
+    assert MEDIA_POOL is None and not MEDIA_PENDING
+    MEDIA_WORKERS=workers;MEDIA_POLICY=policy
+    MEDIA_POOL=ThreadPoolExecutor(max_workers=workers)
+    try:
+        yield
+        for future in MEDIA_PENDING:future.result()
+    finally:
+        MEDIA_POOL.shutdown(wait=True,cancel_futures=True)
+        MEDIA_POOL=None;MEDIA_PENDING=[];MEDIA_POLICY='legacy_full_pixels'
+
+def submit_media(view,members_factory):
+    global MEDIA_PENDING
+    def execute():
+        jpeg_video(view,members_factory());view.m['pilot_complete']=True;view.finish()
+    if MEDIA_POOL is None:execute();return
+    # Bound pending views and source/video spools, not only running FFmpeg tasks.
+    while len(MEDIA_PENDING)>=MEDIA_WORKERS:
+        finished,_=wait(MEDIA_PENDING,return_when=FIRST_COMPLETED)
+        for future in finished:future.result()
+        MEDIA_PENDING=[f for f in MEDIA_PENDING if f not in finished]
+    MEDIA_PENDING.append(MEDIA_POOL.submit(execute))
+
+def digest(p):
+    h=hashlib.sha256()
+    with Path(p).open('rb') as f:
+        for b in iter(lambda:f.read(4*1024*1024),b''): h.update(b)
+    return h.hexdigest()
+
+def native_npy_bytes(data):
+    f=io.BytesIO(data);version=np.lib.format.read_magic(f)
+    shape,order,dtype=np.lib.format._read_array_header(f,version)
+    if dtype.hasobject:return Restricted(f,encoding='latin1').load().item()
+    return np.frombuffer(f.read(),dtype=dtype).reshape(shape,order='F' if order else 'C').copy()
+
+def command(args):
+    r=subprocess.run([str(x) for x in args],capture_output=True)
+    if r.returncode:raise RuntimeError(r.stderr.decode(errors='replace')[-3000:])
+    return r.stdout
+
+def framehash(args,copy=False):
+    output=command([FF,'-v','error','-threads','2','-noautorotate',*args,'-map','0:v:0',
+        *(['-c:v','copy'] if copy else ['-pix_fmt','rgb24','-threads','2']),
+        '-f','framehash','-hash','sha256','-fps_mode','passthrough','pipe:1']).decode()
+    return [(int(r.split(',')[-2]),r.split(',')[-1].strip()) for r in output.splitlines() if r and not r.startswith('#')]
+
+def jpeg_video(view, members):
+    """members is a fresh iterator of (locator, JPEG bytes). Verify every packet/frame."""
+    if MEDIA_POLICY!='legacy_full_pixels':
+        from accelerated_media import write_video
+        return write_video(view,members)
+    temporary=view.tmp/f'{view.m["sequence_id"]}__{view.m["camera_id"]}.mjpeg'
+    temporary.parent.mkdir(parents=True,exist_ok=True)
+    hashes=[]
+    with temporary.open('wb') as out:
+        for locator,b in members:
+            assert b[:2]==b'\xff\xd8',locator
+            hashes.append((len(b),hashlib.sha256(b).hexdigest()));out.write(b)
+    assert len(hashes)==view.T
+    dst=view.p/'observation/rgb.mkv';dst.parent.mkdir(parents=True,exist_ok=True)
+    source=['-f','image2pipe','-framerate','1','-c:v','mjpeg','-i',str(temporary)]
+    command([FF,'-v','error','-y',*source,'-map','0:v:0','-c:v','copy',dst])
+    packet=framehash(['-i',str(dst)],True)
+    assert packet==hashes,'source JPEG packets changed in MKV'
+    before=framehash(source);after=framehash(['-i',str(dst)])
+    assert before==after and len(before)==view.T,'RGB decoding differs'
+    w,h=view.m['camera']['image_size'];assert all(n==w*h*3 for n,_ in after)
+    info=subprocess.run([FF,'-hide_banner','-i',str(dst)],capture_output=True).stderr.decode(errors='replace')
+    line=next(s.strip() for s in info.splitlines() if 'Video: mjpeg' in s)
+    pixel=re.search(r'Video: mjpeg[^,]*,\s*(\w+)',line).group(1)
+    version=command([FF,'-version']).decode().splitlines()[0]
+    view.component('rgb','observation/rgb.mkv','rgb/uint8/v1',['T','height','width','channel'],np.ones(view.T,bool),
+        source=view.source,method='Original JPEG packets copied from raw into MKV; no recompression or rotation',
+        storage={'profile':'mkv/mjpeg_copy/v1','codec':'mjpeg','pixel_format':pixel,'encoded_size':[w,h],
+          'source_lossy':True,'decoder':{'name':'ffmpeg','version':version,'output_pixel_format':'rgb24','autorotate':False,'source_color_declaration':'unknown','actual_stream_description':line},
+          'encoder_parameters':{'mode':'stream_copy','version':version}},storage_fps=1,timebase_role='index_only')
+    view.checks.append({'test':'all_source_JPEG_packets_and_RGB_pixels_exact','passed':True,'frames':view.T,
+        'packet_manifest_sha256':hashlib.sha256(json.dumps(hashes).encode()).hexdigest(),
+        'pixel_manifest_sha256':hashlib.sha256(json.dumps(after).encode()).hexdigest(),'output_bytes':dst.stat().st_size})
+    save(view.tmp/'media_checks'/f'{view.m["sequence_id"]}__{view.m["camera_id"]}.json',{'packets':hashes,'pixels':after})
+    # This precise file was created by this call within the current run's temporary directory.
+    assert temporary.resolve().is_relative_to(view.tmp.resolve());temporary.unlink()
+
+class FreshView:
+    def __init__(self,dataset,seq,cam,ids,size,source):
+        self.root=(OUTPUT_OVERRIDE/dataset/RELEASE if OUTPUT_OVERRIDE else DATA_ROOT/dataset/RELEASE);self.p=self.root/'data'/seq/cam
+        self.tmp=(TMP_OVERRIDE/dataset/RUN if TMP_OVERRIDE else DATA_ROOT/dataset/'tmp'/RUN);self.p.mkdir(parents=True,exist_ok=True)
+        self.T=len(ids);self.source=source;self.checks=[]
+        self.m={'schema_version':'hand-warehouse/1.0','storage_revision':'lossless-video-2026-09-18',
+            'dataset_id':dataset,'release_id':RELEASE,'sequence_id':seq,'camera_id':cam,
+            'sample_type':'video','num_frames':self.T,'hand_slots':['left','right'],
+            'sampling':{'type':'dense','source_fps':None,'frames_ref':'camera/frames.jsonl','selection':'all released images in selected complete source sequence'},
+            'camera':{'model':'unknown','schema_id':None,'image_size':size},'split':'unknown','components':{},'pilot':True}
+        self.frames=[{'frame_index':i,'source_frame_id':str(fid),'timestamp_ns':None,'clock_domain':None,
+            'source_locator':dict(source,source_frame_id=str(fid))} for i,fid in enumerate(ids)]
+    def component(self,name,path,schema,axes,valid,source=None,method='source values retained',kind='format_conversion',**kw):
+        valid=np.asarray(valid,bool);vp=Path(path).with_name(Path(path).stem+'_valid.pt').as_posix();tensor(self.p/vp,valid)
+        self.m['components'][name]={'available':True,'schema_id':schema,'data_ref':path,'axes':axes,
+            'valid_ref':vp,'valid_axes':axes[:valid.ndim],'provenance':{'kind':kind,'method':method},
+            'source_locator':source or self.source,**kw}
+    def array(self,name,path,schema,data,valid,axes,**kw):
+        a=np.asarray(data,np.float32).copy();valid=np.asarray(valid,bool)
+        assert a.shape[0]==self.T and a.shape[:valid.ndim]==valid.shape
+        assert np.isfinite(a[valid]).all(),name
+        a[~valid]=np.nan;tensor(self.p/path,a)
+        self.component(name,path,schema,axes,valid,**kw)
+        restored=torch.load(self.p/path,map_location='cpu',weights_only=True).numpy()
+        assert np.array_equal(a,restored,equal_nan=True)
+        self.checks.append({'test':'tensor_exact_roundtrip','file':path,'shape':list(a.shape),'valid_count':int(valid.sum()),'passed':True})
+    def attribute(self,parent,path,data,axes,source,meaning):
+        data=np.asarray(data,np.float32);valid=np.isfinite(data)
+        assert ((data[valid]>=0)&(data[valid]<=1)).all()
+        vp=Path(path).with_name(Path(path).stem+'_valid.pt').as_posix();tensor(self.p/path,data);tensor(self.p/vp,valid)
+        self.m['components'][parent].setdefault('attributes',{})['visibility']={'data_ref':path,'valid_ref':vp,
+            'axes':axes,'valid_axes':axes,'dtype':'float32','visibility_type':'ratio','meaning':meaning,
+            'provenance':{'kind':'source','method':'native HOT3D visibility ratio retained in [0,1]; misleading [%] header is not a percentage scale'},'source_locator':source}
+    def mesh(self,oid,path,source):
+        self.component('object_'+oid+'_mesh',path,'mesh/obj/v1',['vertex','coordinate'],np.array(True),source=source,
+            method='Object vertices in metres; topology, material and texture dependencies retained',static=True,unit='m',coordinate_frame='object')
+    def finish(self):
+        lines(self.p/'camera/frames.jsonl',self.frames);save(self.p/'metainfo.json',self.m)
+        save(self.tmp/'checks'/f'{self.m["sequence_id"]}__{self.m["camera_id"]}.json',{'checks':self.checks,'frames':self.T,'input':self.source})
+        print('CONVERTED',self.m['dataset_id'],self.m['sequence_id'],self.m['camera_id'],self.T,flush=True)
+
+def assert_se3(a):
+    a=np.asarray(a);assert a.shape[-2:]==(4,4)
+    # A component may exist while QA leaves no usable poses in this view.
+    if a.size==0:return
+    assert np.isfinite(a).all() and np.allclose(a[...,3,:],[0,0,0,1],atol=1e-6)
+    r=a[...,:3,:3];assert np.max(np.abs(r@np.swapaxes(r,-1,-2)-np.eye(3)))<2e-5
+    assert np.max(np.abs(np.linalg.det(r)-1))<2e-5
+
+def setup(dataset,seq):
+    base=DATA_ROOT/dataset;out=base/RELEASE
+    for folder in ['doc','script','doc/validation']: (out/folder).mkdir(parents=True,exist_ok=True)
+    for f in ['CONTRACT.md','schema.json']:shutil.copyfile(WORK/'distribution/hand-warehouse-1.0'/f,out/'doc'/f)
+    save(out/'doc/validation/status.json',{'stage':'building_sequence_pilot','human_review':'pending','sequence_id':seq,'full_preprocessing_authorized':False,'upload_authorized':False})
+    sources=['fresh_sequence_common.py','convert_fresh_sequence_pilots.py','pilot_build.py','pilot_fetch.py','pilot_mano.py',
+        'sequence_pilot_common.py','sequence_pilot_world.py','pilot_hot_normalize.py','audit_source_fields.py']
+    for name in sources:
+        if (WORK/'scripts'/name).exists():shutil.copyfile(WORK/'scripts'/name,out/'script'/name)
+    (base/'tmp'/RUN).mkdir(parents=True,exist_ok=True)
+
+def object_textured_assets(z,prefix,target):
+    target.mkdir(parents=True,exist_ok=True)
+    for name in z.namelist():
+        if not name.startswith(prefix) or name.endswith('/'):continue
+        rel=Path(name[len(prefix):]);dest=target/rel
+        assert dest.resolve().is_relative_to(target.resolve());dest.parent.mkdir(parents=True,exist_ok=True)
+        if dest.suffix.lower() in ('.obj','.mtl','.jpg','.jpeg','.png'):
+            b=z.read(name)
+            if dest.suffix=='.obj':
+                rows=b.decode().splitlines()
+                for i,row in enumerate(rows):
+                    if row.startswith('v '):
+                        v=row.split();v[1:4]=[format(float(x)/1000,'.17g') for x in v[1:4]];rows[i]=' '.join(v)
+                b=('\n'.join(rows)+'\n').encode()
+            dest.write_bytes(b)
+    repair_arctic_material_refs(target)
+
+def repair_arctic_material_refs(target):
+    """Repair a source filename mismatch only when material identifiers agree."""
+    material=target/'material.mtl'
+    if not material.exists():return
+    ids={x.split(maxsplit=1)[1] for x in material.read_text().splitlines() if x.startswith('newmtl ')}
+    for obj in target.glob('*.obj'):
+        text=obj.read_text();rows=text.splitlines();used={x.split(maxsplit=1)[1] for x in rows if x.startswith('usemtl ')}
+        if 'mtllib material_0.mtl' in rows and not (target/'material_0.mtl').exists():
+            assert used and used.issubset(ids),(obj,used,ids)
+            obj.write_text(text.replace('mtllib material_0.mtl','mtllib material.mtl'))
+
+def project_points(xyz,camera,params):
+    xyz=np.asarray(xyz,float);uv=np.full((len(xyz),2),np.nan);good=np.isfinite(xyz).all(1)&(xyz[:,2]>1e-6)
+    if camera=='fisheye624':
+        p=np.asarray(params);xy=xyz[good,:2]/xyz[good,2,None];r=np.linalg.norm(xy,axis=1);th=np.arctan(r)
+        radial=th*(1+sum(p[3+i]*th**(2*i+2) for i in range(6)))
+        xy=xy*np.divide(radial,r,out=np.ones_like(r),where=r>1e-12)[:,None]
+        x,y=xy.T;rr=x*x+y*y
+        xd=x+(2*x*x+rr)*p[9]+2*x*y*p[10]+p[11]*rr+p[12]*rr*rr
+        yd=y+(2*y*y+rr)*p[10]+2*x*y*p[9]+p[13]*rr+p[14]*rr*rr
+        uv[good]=p[0]*np.stack([xd,yd],1)+p[1:3]
+    else:
+        K,D=params;uv[good]=cv2.projectPoints(xyz[good],np.zeros(3),np.zeros(3),K,D)[0].reshape(-1,2)
+    return uv
diff --git a/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/apply_hot3d_qa.py b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/apply_hot3d_qa.py
new file mode 100644
index 0000000000000000000000000000000000000000..942ced8b3d5b27195bd14b959a20d07e92c5ed9b
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/apply_hot3d_qa.py
@@ -0,0 +1,65 @@
+"""Apply native HOT3D manual QA to supervision validity, preserving raw diagnostics."""
+from fresh_sequence_common import *
+
+VIDEO_RUN='full-video-review-v2'
+SEQ='P0015_179e1b84'
+
+def csvrows(z,name):return list(csv.DictReader(io.StringIO(z.read(name).decode())))
+
+def apply_qa(sequence=None,publish=True):
+    seq=sequence or SEQ
+    base=DATA_ROOT/'HOT3D';root=base/RELEASE;src=base/'raw/data/Aria'/seq
+    gtfile=next(src.glob('*ground_truth.zip'));qa_report={}
+    with zipfile.ZipFile(gtfile) as z:
+        masks=csvrows(z,'masks/mask_qa_pass.csv');hands=csvrows(z,'box2d_hands.csv');objects=csvrows(z,'box2d_objects.csv')
+    for meta in sorted((root/'data'/seq).glob('*/metainfo.json')):
+        p=meta.parent;m=load(meta);cam=m['camera_id'];T=m['num_frames'];frames=[json.loads(x) for x in (p/'camera/frames.jsonl').read_text().splitlines()]
+        cache=base/'tmp'/VIDEO_RUN/'diagnostic'/(cam if seq==SEQ else seq+'/'+cam)
+        if (cache/'snapshot.json').exists():assert Path(load(cache/'snapshot.json')['source_view']).resolve()==p.resolve()
+        if not (cache/'snapshot.json').exists():
+            cache.mkdir(parents=True,exist_ok=True)
+            for file in p.rglob('*.pt'):
+                target=cache/file.relative_to(p);target.parent.mkdir(parents=True,exist_ok=True);shutil.copyfile(file,target)
+            save(cache/'metainfo.json',m);save(cache/'snapshot.json',{'purpose':'pre-QA native parameter diagnostics only; not shared usable labels','source_view':str(p),'frames':T})
+        lookup={int(r['timestamp[ns]']):r['mask']=='True' for r in masks if r['stream_id']==cam}
+        qa=np.array([lookup[f['timecode_ns']] for f in frames],bool)
+        source={'archive':gtfile.relative_to(base).as_posix(),'member':'masks/mask_qa_pass.csv','stream_id':cam,'time_index_ref':'camera/frames.jsonl:timecode_ns'}
+        policy={'semantics':'usable_annotation','formula':'source_annotation_present AND source_numerically_valid AND official_manual_qa_pass',
+            'qa_scope':'all visible hand/object poses in this camera frame; conservatively exclude all hand/object supervision if false',
+            'source_locator':source,'no_qa_source_rule':'use the component source-validity rules; do not invent QA'}
+        for key,c in m['components'].items():
+            # Camera calibration and RGB are not labels covered by this manual hand/object QA flag.
+            if key=='mano' or c['schema_id'] in ['object/rigid/v1','object/joint_state/v1','object_keypoints/camera/v1']:
+                values=torch.load(cache/c['data_ref'],weights_only=True).numpy();present=torch.load(cache/c['valid_ref'],weights_only=True).numpy()
+                valid=present & qa.reshape((T,)+(1,)*(present.ndim-1));values=values.copy();values[~valid]=np.nan
+                tensor(p/c['data_ref'],values);tensor(p/c['valid_ref'],valid);c['valid_policy']=policy
+            for attrname,a in c.get('attributes',{}).items():
+                if attrname!='visibility':continue
+                ishand=key=='mano';oid=c['source_locator'].get('object_uid');ids=[0,1] if ishand else [oid]
+                rows=hands if ishand else objects;identity='hand_index' if ishand else 'object_uid'
+                native={(int(r['timestamp[ns]']),str(r[identity])):float(r['visibility_ratio[%]']) for r in rows if r['stream_id']==cam and r['visibility_ratio[%]']!=''}
+                values=np.array([[native.get((f['timecode_ns'],str(k)),np.nan) for k in ids] for f in frames],np.float32)
+                if not ishand:values=values[:,0]
+                present=np.isfinite(values);assert ((values[present]>=0)&(values[present]<=1)).all()
+                # Correct the diagnostic copy as well: these values are fractions, despite the source header.
+                tensor(cache/a['data_ref'],values);tensor(cache/a['valid_ref'],present)
+                valid=present & qa.reshape((T,)+(1,)*(present.ndim-1));filtered=values.copy();filtered[~valid]=np.nan
+                tensor(p/a['data_ref'],filtered);tensor(p/a['valid_ref'],valid)
+                a['provenance']={'kind':'source','method':'native [0,1] ratio retained; official loader does not divide misleading [%] header by 100'}
+                a['valid_policy']={'semantics':'usable_visibility_annotation','formula':'source_ratio_present AND official_manual_qa_pass','source_locator':source}
+                a['source_locator']['match_method']='exact'
+        m['validity_revision']='usable-annotation-2026-09-20'
+        m['split']='train';m['split_scheme']='HOT3D_official_participant_split';m.pop('split_reason',None)
+        m['split_source_locator']={'url':'https://github.com/facebookresearch/hot3d/blob/146b34afef8c1a32adeef7e981c070109f225c87/README.md#testtrain-split','participant_id':seq.split('_')[0]}
+        assert seq.split('_')[0] not in {'P0004','P0005','P0006','P0008','P0016','P0020'}
+        save(meta,m)
+        qa_report[cam]={'frames':T,'qa_pass':qa.tolist(),'passed_count':int(qa.sum()),'source_locator':source,
+            'interpretation':'manual QA of visible hand/object poses; not RGB validity or individual error localization'}
+        print('QA_APPLIED',cam,'pass',int(qa.sum()),'fail',int((~qa).sum()),flush=True)
+    save(root/'doc/validation/source_qa.json' if publish else base/'tmp'/RUN/'qa'/f'{seq}.json',qa_report)
+    if publish:
+        for f in ['apply_hot3d_qa.py','fresh_sequence_common.py','convert_fresh_sequence_pilots.py']:
+            shutil.copyfile(WORK/'scripts'/f,root/'script'/f)
+    return qa_report
+
+if __name__=='__main__':apply_qa()
diff --git a/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/audit_source_fields.py b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/audit_source_fields.py
new file mode 100644
index 0000000000000000000000000000000000000000..9dd63f10811edb78e5140e940368c3ccf6909e23
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/audit_source_fields.py
@@ -0,0 +1,55 @@
+"""Read small existing native annotations; do not convert or download data."""
+from pilot_build import *
+
+def shape_tree(x, depth=0):
+ if isinstance(x,(np.ndarray,torch.Tensor)):
+  if isinstance(x,np.ndarray) and x.dtype==object and x.shape==():return shape_tree(x.item(),depth)
+  return {'shape':list(x.shape),'dtype':str(x.dtype)}
+ if isinstance(x,dict):
+  if len(x)>30:return {'count':len(x),'first_key':str(next(iter(x))),'first_value':shape_tree(next(iter(x.values())),depth+1)}
+  return {str(k):shape_tree(v,depth+1) for k,v in x.items()}
+ if isinstance(x,(list,tuple)):
+  try:
+   a=np.asarray(x)
+   if a.dtype.kind in 'biuf':return {'shape':list(a.shape),'dtype':str(a.dtype)}
+  except (ValueError,TypeError):pass
+  return {'count':len(x),'first':shape_tree(x[0],depth+1) if x else None}
+ return x if isinstance(x,(str,int,float,bool)) or x is None else type(x).__name__
+
+def native_npy(p):
+ with p.open('rb') as f:
+  version=np.lib.format.read_magic(f)
+  if version==(1,0):_,_,dtype=np.lib.format.read_array_header_1_0(f)
+  else:_,_,dtype=np.lib.format.read_array_header_2_0(f)
+  if dtype.hasobject:return Restricted(f,encoding='latin1').load()
+ return np.load(p,allow_pickle=False)
+
+def main():
+ rows=[]
+ def add(ds,p,x,detail=None):rows.append({'dataset':ds,'path':str(p.relative_to(WORK)).replace('\\','/'),'fields':shape_tree(x),'detail':detail})
+ p=next((WORK/'tmp/pilot-dex').rglob('labels_000001.npz'))
+ with np.load(p,allow_pickle=False) as z:add('Dex-YCB',p,{k:z[k] for k in z.files})
+ p=next((ROOT/'VISOR/raw/data').rglob('P01_03.json'));x=load(p);frames=x['video_annotations'];keys=sorted({k for f in frames for a in f['annotations'] for k in a})
+ add('VISOR',p,frames[0],{'top_level_keys':list(x),'annotation_keys_all_local_frames':keys,'frames':len(frames),'image_keys':sorted({k for f in frames for k in f['image']})})
+ for folder in ['hand_pose','hand_pose_mano','cam_pose']:
+  p=ROOT/'H2O/raw/data/local_subset/subject4_ego/h1/0/cam4'/folder/'000000.txt';add('H2O',p,np.loadtxt(p))
+ p=next((ROOT/'HOI4D/raw/data').rglob('*.pickle'));add('HOI4D',p,read_pickle(p))
+ for name in ['training_K.json','training_mano.json','training_scale.json']:
+  p=next((ROOT/'FreiHAND/raw/data').rglob(name));x=load(p);add('FreiHAND',p,x[0],{'rows':len(x)})
+ p=next((ROOT/'RHD/raw/data').rglob('anno_evaluation.pickle'));x=read_pickle(p);a=x[0];add('RHD',p,a,{'records':len(x),'visibility_values':np.unique(a['uv_vis'][:,2]).tolist(),'invisible_points_first_sample':int((a['uv_vis'][:,2]==0).sum())})
+ p=next((ROOT/'ARCTIC/raw/data').rglob('*.mano.npy'));add('ARCTIC',p,native_npy(p))
+ p=next((ROOT/'ARCTIC/raw/data').rglob('*.egocam.dist.npy'));x=native_npy(p).item();add('ARCTIC',p,x,{'dist8':np.asarray(x['dist8']).tolist()})
+ p=next((ROOT/'OakInk2/raw/data').rglob('*.pkl'));x=read_pickle(p);add('OakInk2',p,x,{'frame_ids_first':x['frame_id_list'][:5],'mocap_ids_first':x['mocap_frame_id_list'][:5],'object_ids':x['obj_list']})
+ p=next((ROOT/'ReInterHand/raw/data').rglob('*_left.json'));add('ReInterHand',p,load(p))
+ p=OLD/'HOT3D/P0001_10a27bf7/mano_hand_pose_trajectory.jsonl';add('HOT3D',p,json.loads(p.open().readline()))
+ p=next((ROOT/'HOT3D/raw/data').glob('*ground_truth.zip'))
+ with zipfile.ZipFile(p) as z:
+  headers={}
+  for name in z.namelist():
+   if name.endswith('.csv'):
+    with z.open(name) as f:headers[name]=f.readline().decode().strip()
+  add('HOT3D',p,headers)
+ save(WORK/'docs/source-field-audit-evidence.json',{'purpose':'native field shape probe; not a full conversion or whole-dataset validation','records':rows})
+ print(json.dumps({'datasets':sorted({r['dataset'] for r in rows}),'records':len(rows),'output':'docs/source-field-audit-evidence.json'},indent=2))
+
+if __name__=='__main__':main()
diff --git a/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/convert_fresh_sequence_pilots.py b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/convert_fresh_sequence_pilots.py
new file mode 100644
index 0000000000000000000000000000000000000000..ba9baf7b2f8e28ff41bd022f53bb4def194f3224
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/convert_fresh_sequence_pilots.py
@@ -0,0 +1,344 @@
+"""Convert one complete, freshly downloaded sequence per dataset, all cameras.
+
+Run: python scripts/convert_fresh_sequence_pilots.py ARCTIC|HOT3D
+Requires the local CPU runtime and licensed MANO models described in README.md.
+No full-dataset conversion or upload is performed by this entry point.
+"""
+from fresh_sequence_common import *
+from pilot_hot_normalize import csv_transform, transform
+import concurrent.futures
+
+ARCTIC_SEQ='s09/laptop_use_03'
+HOT3D_SEQ='P0015_179e1b84'
+
+def loadzipcsv(z,name):return list(csv.DictReader(io.StringIO(z.read(name).decode())))
+
+def expand(rows,indices,shape):
+    a=np.full(shape,np.nan);good=(indices>=0)&(indices<len(rows));a[good]=np.asarray(rows)[indices[good]]
+    return a,good
+
+def arctic_unlabelled(seq,cameras=None):
+    """Official held-out s03 RGB: static calibration/assets, no hidden GT guesses."""
+    base=DATA_ROOT/'ARCTIC';raw=base/'raw/data';sid,action=seq.split('/');oid=action.split('_')[0];seqid=seq.replace('/','_')
+    with zipfile.ZipFile(raw/'splits_json.zip') as z:splitdef=json.loads(z.read('splits_json/protocol_p1.json'))
+    assert seq in splitdef['test']
+    source={'archive':f'raw/data/images_zips/{seq}.zip','member_pattern':'{camera_id}/{source_frame_id:05d}.jpg'}
+    with zipfile.ZipFile(raw/'meta.zip') as za,zipfile.ZipFile(raw/f'images_zips/{seq}.zip') as zi:
+        misc=json.loads(za.read('meta/misc.json'))[sid];prefix=f'meta/object_vtemplates/{oid}/'
+        native_parts=np.asarray(json.loads(za.read(prefix+'parts.json')),np.int64)
+        names=sorted(n for n in zi.namelist() if re.fullmatch(r'[0-8]/[0-9]+\.jpg',n))
+        for cam in sorted({n.split('/')[0] for n in names},key=int):
+            if cameras is not None and cam not in cameras:continue
+            selected=[n for n in names if n.split('/')[0]==cam];ids=[int(Path(n).stem) for n in selected]
+            size=list(Image.open(io.BytesIO(zi.read(selected[0]))).size);assert size==misc['image_size'][int(cam)]
+            v=FreshView('ARCTIC',seqid,cam,ids,size,dict(source,camera_id=cam))
+            if (v.p/'metainfo.json').exists() and load(v.p/'metainfo.json').get('pilot_complete'):
+                print('RESUME verified view',seq,cam,flush=True);continue
+            t=v.T;v.m.update(split='test',split_scheme='ARCTIC_protocol_p1',split_source_locator={'archive':'raw/data/splits_json.zip','member':'splits_json/protocol_p1.json','sequence':seq},view_type='ego' if cam=='0' else 'exo',source_hand_object_gt_available=False)
+            v.m['sampling'].update(source_fps=None,source_fps_reason='Capture FPS not asserted by supplied source; original image numbers retained')
+            if cam!='0':
+                src={'archive':'raw/data/meta.zip','member':'meta/misc.json','record':sid}
+                v.m['camera'].update(model='pinhole',schema_id='camera/pinhole/v1')
+                K=np.broadcast_to(misc['intris_mat'][int(cam)-1],(t,3,3));E=np.broadcast_to(misc['world2cam'][int(cam)-1],(t,4,4))
+                v.array('intrinsic','camera/intrinsic.pt','camera/pinhole/v1',K,np.ones(t,bool),['T','row','column'],source=src)
+                v.array('extrinsic','camera/extrinsic.pt','transform/se3/v1',E,np.ones(t,bool),['T','row','column'],source=src,unit='m',convention='T_camera_from_world');assert_se3(E)
+            for row,name in zip(v.frames,selected):row.update(sync_id=f'{seq}:{row["source_frame_id"]}',source_annotation_index=None);row['source_locator'].update(member=name)
+            rel=f'annotation/object/{oid}';op=v.p/rel;assets=op/'assets';object_textured_assets(za,prefix,assets)
+            parts=1-native_parts;tensor(op/'vertex_part_ids.pt',parts)
+            articulation={'hierarchy_semantics':'mechanical_articulation','parts':[{'part_id':'bottom','parent_id':None,'has_model':True,'mesh_ref':rel+'/assets/bottom.obj'},{'part_id':'top','parent_id':'bottom','has_model':True,'mesh_ref':rel+'/assets/top.obj'}],
+                'joints':[{'joint_id':'hinge','parent_part':'bottom','child_part':'top','type':'revolute','unit':'rad','axis':[0,0,-1],'axis_frame':'parent_part','zero_transform':np.eye(4).tolist(),'positive_direction':'right_hand_rule_about_axis','limits':None}],'joint_order':['hinge']}
+            save(op/'articulation.json',articulation);sourceasset={'archive':'raw/data/meta.zip','member_prefix':prefix}
+            v.component('object_'+oid+'_articulation',rel+'/articulation.json','object/articulated/v1',[],np.array(True),source=sourceasset,static=True)
+            v.mesh(oid,rel+'/assets/mesh.obj',sourceasset)
+            original=np.asarray([[float(x) for x in l.split()[1:4]] for l in za.read(prefix+'mesh.obj').decode().splitlines() if l.startswith('v ')])/1000
+            actual=np.asarray([[float(x) for x in l.split()[1:4]] for l in (assets/'mesh.obj').read_text().splitlines() if l.startswith('v ')])
+            assert np.array_equal(original,actual) and len(parts)==len(actual)
+            save(op/'metainfo.json',{'object_id':oid,'identity_scope':'sequence','scope_id':seqid,'association_basis':'official_sequence_object_name; no per-frame object pose released','has_model':True,'mesh_ref':rel+'/assets/mesh.obj','articulation_ref':rel+'/articulation.json','vertex_part_ids_ref':rel+'/vertex_part_ids.pt','topology_id':'sha256:'+digest(assets/'mesh.obj'),'part_mapping':{'0':'bottom','1':'top'},'part_id_conversion':{'formula':'shared_id = 1 - raw_parts_json_value','source_locator':dict(sourceasset,member=prefix+'parts.json')},'unit':'m','source_locator':sourceasset})
+            v.checks.append({'test':'source_static_mesh_vertices_metres_and_part_ids','passed':True,'vertices':len(actual)})
+            v.m['omitted_components']={'mano':'Official test ground truth withheld: no sequence MANO file in raw_seqs.zip','object_pose':'Official test ground truth withheld: no sequence object trajectory, joint states or camera-frame keypoints','keypoints':'No released sequence hand keypoint supervision','visibility':'Not provided in selected official release','depth':'No source depth images','segment':'No source pixel masks','contact':'No native contact labels in downloaded scope'}
+            if cam=='0':v.m['omitted_components'].update(intrinsic='No test sequence egocam.dist.npy; do not reuse another subject calibration',extrinsic='No released egocentric trajectory for test sequence')
+            print('MEDIA ARCTIC',cam,t,flush=True);jpeg_video(v,((n,zi.read(n)) for n in selected));v.m['pilot_complete']=True;v.finish()
+
+def arctic(sequence=None,cameras=None,initialize=True):
+    dataset='ARCTIC';seq=sequence or ARCTIC_SEQ;sid,action=seq.split('/');oid=action.split('_')[0];seqid=seq.replace('/','_')
+    if initialize:setup(dataset,seqid)
+    base=DATA_ROOT/dataset;raw=base/'raw/data'
+    with zipfile.ZipFile(raw/'raw_seqs.zip') as z:
+        if f'raw_seqs/{seq}.mano.npy' not in z.namelist():
+            assert all(f'raw_seqs/{seq}.{x}.npy' not in z.namelist() for x in ['object','egocam.dist'])
+            return arctic_unlabelled(seq,cameras)
+        hand=native_npy_bytes(z.read(f'raw_seqs/{seq}.mano.npy'))
+        ego=native_npy_bytes(z.read(f'raw_seqs/{seq}.egocam.dist.npy'))
+        objects=native_npy_bytes(z.read(f'raw_seqs/{seq}.object.npy'))
+    with zipfile.ZipFile(raw/'splits_json.zip') as z:splitdef=json.loads(z.read('splits_json/protocol_p1.json'))
+    split=next((k for k,v in splitdef.items() if seq in v),'unknown')
+    if split=='val':split='val'
+    za=zipfile.ZipFile(raw/'meta.zip');misc=json.loads(za.read('meta/misc.json'))[sid]
+    params=json.loads(za.read(f'meta/object_vtemplates/{oid}/object_params.json'))
+    native_parts=np.asarray(json.loads(za.read(f'meta/object_vtemplates/{oid}/parts.json')),np.int64)
+    models=[Mano('left'),Mano('right')];N=len(objects)
+    world=np.broadcast_to(np.eye(4),(N,4,4)).copy();world[:,:3,:3]=Rotation.from_rotvec(objects[:,1:4]).as_matrix();world[:,:3,3]=objects[:,4:7]/1000
+    hinge=np.broadcast_to(np.eye(4),(N,4,4)).copy();hinge[:,:3,:3]=Rotation.from_rotvec(np.c_[np.zeros((N,2)),-objects[:,0]]).as_matrix()
+    kp=np.concatenate([params['keypoints_top'],params['keypoints_bottom']])/1000;n_top=len(params['keypoints_top'])
+    source={'archive':f'raw/data/images_zips/{seq}.zip','member_pattern':'{camera_id}/{source_frame_id:05d}.jpg'}
+    zimage=zipfile.ZipFile(raw/f'images_zips/{seq}.zip')
+    names=sorted(n for n in zimage.namelist() if re.fullmatch(r'[0-8]/[0-9]+\.jpg',n))
+    extra=[n for n in zimage.namelist() if n.endswith('.jpg') and not re.fullmatch(r'[0-8]/[0-9]+\.jpg',n)]
+    assert all(n.startswith('__MACOSX/') or Path(n).name.startswith('._') for n in extra),extra[:5]
+    for cam in sorted({n.split('/')[0] for n in names},key=int):
+        if cameras is not None and cam not in cameras:continue
+        selected=[n for n in names if n.split('/')[0]==cam];ids=np.asarray([int(Path(n).stem) for n in selected]);ix=ids-misc['ioi_offset']
+        size=list(Image.open(io.BytesIO(zimage.read(selected[0]))).size)
+        assert size==misc['image_size'][int(cam)]
+        v=FreshView(dataset,seqid,cam,ids,size,dict(source,camera_id=cam))
+        if (v.p/'metainfo.json').exists():
+            old=load(v.p/'metainfo.json')
+            if old.get('pilot_complete') and old.get('num_frames')==len(ids):print('RESUME verified view',seq,cam,flush=True);continue
+        t=v.T;good=(ix>=0)&(ix<N)
+        v.m.update(split=split,split_scheme='ARCTIC_protocol_p1',split_source_locator={'archive':'raw/data/splits_json.zip','member':'splits_json/protocol_p1.json','sequence':seq},view_type='ego' if cam=='0' else 'exo')
+        v.m['sampling'].update(source_fps=None,source_fps_reason='Actual capture FPS not asserted by the selected raw files; frame indices and official synchronization offset retained',selection='Every released image of this selected camera in the complete official sequence ZIP; boundary images without annotations retained')
+        srcanno={'archive':'raw/data/raw_seqs.zip','member':f'raw_seqs/{seq}.mano.npy','index_ref':'camera/frames.jsonl:source_annotation_index'}
+        srccam={'archive':'raw/data/meta.zip','member':'meta/misc.json','record':sid}
+        E=np.broadcast_to(np.eye(4),(t,4,4)).copy()
+        if cam=='0':
+            K=np.broadcast_to(ego['intrinsics'],(t,3,3));D=np.broadcast_to(ego['dist8'],(t,8))
+            E[:]=np.nan;E[good]=np.eye(4);E[good,:3,:3]=ego['R_k_cam_np'][ix[good]];E[good,:3,3]=ego['T_k_cam_np'][ix[good]].reshape(-1,3)
+            src={'archive':'raw/data/raw_seqs.zip','member':f'raw_seqs/{seq}.egocam.dist.npy'}
+            v.m['camera'].update(model='rational8',schema_id='camera/rational8/v1')
+            v.array('intrinsic','camera/intrinsic.pt','camera/rational8/v1',K,np.ones(t,bool),['T','row','column'],source=src,distortion_ref='camera/distortion.pt')
+            tensor(v.p/'camera/distortion.pt',np.asarray(D,np.float32));validE=good
+        else:
+            K=np.broadcast_to(misc['intris_mat'][int(cam)-1],(t,3,3));D=None;E[:]=misc['world2cam'][int(cam)-1];validE=np.ones(t,bool);src=srccam
+            v.m['camera'].update(model='pinhole',schema_id='camera/pinhole/v1')
+            v.array('intrinsic','camera/intrinsic.pt','camera/pinhole/v1',K,np.ones(t,bool),['T','row','column'],source=srccam)
+        v.array('extrinsic','camera/extrinsic.pt','transform/se3/v1',E,validE,['T','row','column'],source=src,unit='m',convention='T_camera_from_world')
+        assert_se3(E[validE]);motion=np.full((t,2,61),np.nan)
+        for side,model in enumerate(models):
+            h=hand[['left','right'][side]];pose=np.c_[h['rot'],h['pose']].copy();pose[:,3:]+=model.data['hands_mean']
+            result,err=camera_mano(model,pose[ix[good]],np.broadcast_to(h['shape'],(good.sum(),10)),h['trans'][ix[good]],E[good])
+            motion[good,side]=result
+            # Independent source-model mean configuration, separate from camera conversion algebra.
+            from types import SimpleNamespace
+            import smplx
+            native=smplx.MANO(model_path=str(WORK/'tmp/pilot-models'),data_struct=SimpleNamespace(**model.data),is_rhand=side==1,use_pca=False,flat_hand_mean=False)
+            checkidx=np.arange(0,N,max(1,N//12));b=torch.tensor(np.broadcast_to(h['shape'],(len(checkidx),10)).copy(),dtype=torch.float32)
+            with torch.no_grad():expected=native(global_orient=torch.tensor(h['rot'][checkidx]),hand_pose=torch.tensor(h['pose'][checkidx]),betas=b,transl=torch.tensor(h['trans'][checkidx])).vertices.numpy()
+            actual=model.geometry(pose[checkidx],b.numpy(),h['trans'][checkidx])[0]
+            delta=float(np.max(np.linalg.norm(expected-actual,axis=-1)));assert delta<2e-6
+            v.checks.append({'test':'MANO_mean_and_camera_transform','side':side,'all_aligned_frames':int(good.sum()),'camera_max_m':err,'native_mean_max_m':delta,'passed':True})
+        v.array('mano','annotation/mano/motion.pt','mano/axis_angle61/v1',motion,np.broadcast_to(good[:,None],(t,2)),['T','hand','parameter'],source=srcanno,unit='m',angle_unit='rad',coordinate_frame='camera',model_profile={'version':'MANO-v1.2','flat_hand_mean':True,'left_shapedirs_fix':False},method='Official nonflat mean expanded; shape-dependent root corrected during world-to-camera conversion; image index minus official ioi_offset')
+        for i,(row,name,j) in enumerate(zip(v.frames,selected,ix)):
+            row.update(source_annotation_index=int(j) if good[i] else None,sync_id=f'{seq}:{ids[i]}')
+            row['source_locator'].update(member=name)
+            row['annotation_matches']={'mano':{'source_frame_id':str(j) if good[i] else None,'source_timestamp_ns':None,'clock_domain':None,'match_method':'index_map','delta_ns':None,'mapping':'image_number - misc.ioi_offset','source_record_present':bool(good[i])}}
+        rel=f'annotation/object/{oid}';op=v.p/rel;assets=op/'assets';object_textured_assets(za,f'meta/object_vtemplates/{oid}/',assets)
+        # Parse exact OBJ vertex ordering before loading a render-oriented mesh.
+        vertices=np.array([[float(x) for x in r.split()[1:4]] for r in (assets/'mesh.obj').read_text().splitlines() if r.startswith('v ')])
+        # Official ObjectTensors adds 1 to raw parts, then articulates ==1: raw 0 is TOP.
+        shared_parts=1-native_parts
+        assert len(vertices)==len(native_parts);tensor(op/'vertex_part_ids.pt',shared_parts)
+        P=np.full((t,4,4),np.nan);top=P.copy();P[good]=E[good]@world[ix[good]];top[good]=P[good]@hinge[ix[good]]
+        srcobj={'archive':'raw/data/raw_seqs.zip','member':f'raw_seqs/{seq}.object.npy','index_ref':'camera/frames.jsonl:source_annotation_index'}
+        for name,data in [('root',P),('bottom',P),('top',top)]:
+            path=rel+('/pose.pt' if name=='root' else f'/parts/{name}/pose.pt')
+            v.array(f'object_{oid}_{name}_pose',path,'object/rigid/v1',data,good,['T','row','column'],source=srcobj,unit='m',convention='T_camera_from_object' if name=='root' else 'T_camera_from_part')
+            assert_se3(data[good])
+        joint,_=expand(objects[:,:1],ix,(t,1));v.array(f'object_{oid}_joint',rel+'/joint_state.pt','object/joint_state/v1',joint,good[:,None],['T','joint'],source=srcobj,joint_order=['hinge'],unit='rad')
+        articulation={'hierarchy_semantics':'mechanical_articulation','parts':[{'part_id':'bottom','parent_id':None,'has_model':True,'mesh_ref':rel+'/assets/bottom.obj'},{'part_id':'top','parent_id':'bottom','has_model':True,'mesh_ref':rel+'/assets/top.obj'}],
+            'joints':[{'joint_id':'hinge','parent_part':'bottom','child_part':'top','type':'revolute','unit':'rad','axis':[0,0,-1],'axis_frame':'parent_part','zero_transform':np.eye(4).tolist(),'positive_direction':'right_hand_rule_about_axis','limits':None}],'joint_order':['hinge']}
+        save(op/'articulation.json',articulation)
+        v.component('object_'+oid+'_articulation',rel+'/articulation.json','object/articulated/v1',[],np.array(True),source=srcobj,static=True)
+        sourceasset={'archive':'raw/data/meta.zip','member_prefix':f'meta/object_vtemplates/{oid}/'}
+        v.mesh(oid,rel+'/assets/mesh.obj',sourceasset)
+        topology=digest(assets/'mesh.obj')
+        save(op/'metainfo.json',{'object_id':oid,'identity_scope':'sequence','scope_id':seqid,'association_basis':'official_sequence_object_identity_shared_across_cameras','has_model':True,'mesh_ref':rel+'/assets/mesh.obj','articulation_ref':rel+'/articulation.json','vertex_part_ids_ref':rel+'/vertex_part_ids.pt','topology_id':'sha256:'+topology,'part_mapping':{'0':'bottom','1':'top'},'part_id_conversion':{'formula':'shared_id = 1 - raw_parts_json_value','source_locator':dict(sourceasset,member=f'meta/object_vtemplates/{oid}/parts.json')},'unit':'m','source_locator':srcobj})
+        xyz=np.full((t,len(kp),3),np.nan);errs=[]
+        for i in np.flatnonzero(good):
+            xyz[i]=kp@P[i,:3,:3].T+P[i,:3,3];xyz[i,:n_top]=kp[:n_top]@top[i,:3,:3].T+top[i,:3,3]
+            ref=vertices.copy();j=ix[i];ref[native_parts==0]=Rotation.from_rotvec([0,0,-float(objects[j,0])]).apply(ref[native_parts==0]);ref=Rotation.from_rotvec(objects[j,1:4]).apply(ref)+objects[j,4:7]/1000;ref=ref@E[i,:3,:3].T+E[i,:3,3]
+            out=vertices@P[i,:3,:3].T+P[i,:3,3];out[shared_parts==1]=vertices[shared_parts==1]@top[i,:3,:3].T+top[i,:3,3];errs.append(float(np.linalg.norm(out-ref,axis=1).max()))
+        assert max(errs)<2e-5
+        v.array('object_'+oid+'_keypoints',rel+'/keypoints/xyz.pt','object_keypoints/camera/v1',xyz,np.broadcast_to(good[:,None],(t,len(kp))),['T','joint','coordinate'],source=sourceasset,unit='m',coordinate_frame='camera',keypoint_schema='arctic_object_keypoints/v1',keypoint_names=[f'top_{i}' for i in range(n_top)]+[f'bottom_{i}' for i in range(len(kp)-n_top)],kind='derived',method='Official template keypoints driven by released articulation and camera transforms')
+        v.checks.append({'test':'all_frame_articulated_geometry_independent_scipy_rotation','passed':True,'max_m':max(errs),'angle_min_max_rad':[float(objects[:,0].min()),float(objects[:,0].max())]})
+        v.m['omitted_components']={'depth':'No depth images in this official raw release','segment':'No pixel segmentation in this official raw release','contact':'No source contact labels in selected official raw packages','keypoints':'No separate measured hand keypoints in raw; MANO available. Render-only joints are not published as additional labels','visibility':'Not supplied in selected raw annotations; fitting_err is outside current shared scope'}
+        print('MEDIA',dataset,cam,t,flush=True);jpeg_video(v,((n,zimage.read(n)) for n in selected))
+        v.m['pilot_complete']=True;v.finish()
+    zimage.close();za.close()
+
+def vrs_records(path,cam):
+    raw=command([VRS,'print-json',path,'+',cam,'+','data']).decode(errors='replace')
+    rows=[]
+    for line in raw.splitlines():
+        if line.startswith('{"data_layout"'):rows.append({x['name']:x['value'] for x in json.loads(line)['data_layout']})
+    return rows
+
+def nearest(times,query,limit):
+    times=np.asarray(times,np.int64);q=np.asarray(query,np.int64)
+    if not len(times):return np.zeros(len(q),np.int64),np.zeros(len(q),bool),np.zeros(len(q),np.int64)
+    j=np.searchsorted(times,q);a=np.clip(j-1,0,len(times)-1);b=np.clip(j,0,len(times)-1)
+    ix=np.where(np.abs(times[a]-q)<=np.abs(times[b]-q),a,b);delta=times[ix]-q
+    return ix,np.abs(delta)<=limit,delta
+
+def verify_hot3d_frame_order_and_metadata(sequence=None):
+    """Independent extraction-index and VRS capture-rate check, also on resume."""
+    seq=sequence or HOT3D_SEQ
+    base=DATA_ROOT/'HOT3D';src=base/'raw/data/Aria'/seq;recording=next(src.glob('*.vrs'))
+    extract=base/'tmp'/RUN/seq/'images'
+    for cam in ['214-1','1201-1','1201-2']:
+        p=base/RELEASE/'data'/seq/cam;m=load(p/'metainfo.json');rows=vrs_records(recording,cam)
+        paths=sorted(extract.glob(cam+'-*.jpg'))
+        indices=[int(x.name.split('-')[-2]) for x in paths];assert indices==list(range(1,len(rows)+1))
+        file_times=np.array([float(x.stem.split('-')[-1]) for x in paths]);times=np.array([r['capture_timestamp_ns'] for r in rows])/1e9
+        assert np.max(np.abs(file_times-times))<.001,'extracted images do not match camera record ordering'
+        raw=command([VRS,'print-json',recording,'+',cam,'+','configuration']).decode(errors='replace')
+        cfg=[{x['name']:x.get('value') for x in json.loads(line)['data_layout']} for line in raw.splitlines() if line.startswith('{"data_layout"')][0]
+        rate=cfg['nominal_rate'];assert abs(np.median(np.diff(times))-1/rate)<.001
+        m['sampling']['source_fps']=rate
+        m['sampling']['source_fps_basis']='VRS camera configuration nominal_rate; actual intervals checked separately'
+        m['sampling']['duration_seconds']=float(times[-1]-times[0])
+        frames=[json.loads(line) for line in (p/'camera/frames.jsonl').read_text().splitlines()]
+        for frame in frames:
+            for match in frame['annotation_matches'].values():
+                if 'available' in match:match['source_record_present']=match.pop('available')
+        lines(p/'camera/frames.jsonl',frames)
+        info=load(base/'raw/data/assets/object_models/models_info.json')
+        reverse={str(v['original_id']):str(k) for k,v in info.items()}
+        for op in (p/'annotation/object').glob('*'):
+            om=load(op/'metainfo.json');oid=om['object_id'];idx=int(reverse[oid])
+            glb=base/'raw/data/assets/object_models'/f'obj_{idx:06d}.glb';assert glb.is_file()
+            # Retain the official material-bearing companion byte-for-byte. OBJ remains the declared standard mesh.
+            target=op/'assets'/glb.name;shutil.copyfile(glb,target)
+            assert digest(glb)==digest(target)
+            om.update(category_name=info[str(idx)]['name'],original_glb_ref=target.relative_to(p).as_posix(),original_glb_source_locator={'file':glb.relative_to(base).as_posix(),'semantics':'native companion; standard geometry is the referenced metre OBJ'})
+            save(op/'metainfo.json',om)
+        save(p/'metainfo.json',m)
+        checkpath=base/'tmp'/RUN/'checks'/f'{seq}__{cam}.json';checks=load(checkpath)
+        checks['checks']=[c for c in checks['checks'] if c['test']!='VRS_extracted_filename_order_and_nominal_rate']
+        checks['checks'].append({'test':'VRS_extracted_filename_order_and_nominal_rate','frames':len(rows),'nominal_fps':rate,'filename_rounding_max_ms':float(np.max(np.abs(file_times-times))*1000),'passed':True});save(checkpath,checks)
+
+def hot3d(sequence=None,initialize=True):
+    dataset='HOT3D';seq=sequence or HOT3D_SEQ
+    if initialize:setup(dataset,seq)
+    base=DATA_ROOT/dataset;raw=base/'raw/data';src=raw/'Aria'/seq;tmp=base/'tmp'/RUN
+    recording=next(src.glob('*.vrs'));extract=tmp/seq/'images'
+    if not (tmp/seq/'extracted.json').exists() or not extract.exists() or not any(extract.glob('*.jpg')):
+        extract.mkdir(parents=True,exist_ok=True)
+        log=tmp/seq/'vrs-extraction.log'
+        with log.open('wb') as out:
+            run=subprocess.run([str(VRS),'extract-images',str(recording),'--to',str(extract),'+','214-1','+','1201-1','+','1201-2'],stdout=out,stderr=subprocess.STDOUT)
+        assert run.returncode==0,'VRS extraction failed; see tmp log'
+        save(tmp/seq/'extracted.json',{'source':recording.name,'streams':['214-1','1201-1','1201-2'],'complete':True})
+    gt=zipfile.ZipFile(next(src.glob('*ground_truth.zip')));handszip=zipfile.ZipFile(next(src.glob('*hand_data.zip')))
+    meta=json.loads(gt.read('metadata.json'));cams=json.loads(gt.read('camera_models.json'));mapping=loadzipcsv(gt,'timecode_devicetime_mapping.csv');mapping.sort(key=lambda r:int(r['devicetime_ns']))
+    assert meta['headset']=='Aria' and meta['recording_name']==seq
+    dev=np.array([int(r['devicetime_ns']) for r in mapping],np.int64);tc=np.array([int(r['timecode_ns']) for r in mapping],np.int64)
+    heads={int(r['timestamp[ns]']):csv_transform(r) for r in loadzipcsv(gt,'headset_trajectory.csv')};head_times=sorted(heads)
+    hands={r['timestamp_ns']:r['hand_poses'] for r in map(json.loads,handszip.read('mano_hand_pose_trajectory.jsonl').splitlines())};hand_times=sorted(hands)
+    objects=loadzipcsv(gt,'dynamic_objects.csv');vis_h=loadzipcsv(gt,'box2d_hands.csv');vis_o=loadzipcsv(gt,'box2d_objects.csv')
+    models=[Mano('left'),Mano('right')];fixed=np.abs(models[0].data['shapedirs'][:,0]-models[1].data['shapedirs'][:,0]).sum()<1
+    if fixed:
+        models[0].data['shapedirs'][:,0]*=-1
+        with torch.no_grad():models[0].layer.shapedirs[:,0]*=-1
+    assetszip=zipfile.ZipFile(raw/'assets/hot3d_models.zip');assetinfo=json.loads(assetszip.read('models/models_info.json'));reverse={str(v['original_id']):str(k) for k,v in assetinfo.items()}
+    for camera in cams:
+        cam=camera['stream_id'];rows=vrs_records(recording,cam);paths=sorted(extract.glob(cam+'-*.jpg'))
+        assert len(rows)==len(paths)>1,(cam,len(rows),len(paths))
+        v=FreshView(dataset,seq,cam,[r['frame_number'] for r in rows],[camera['imageWidth'],camera['imageHeight']],{'file':recording.relative_to(base).as_posix(),'stream_id':cam})
+        if (v.p/'metainfo.json').exists() and load(v.p/'metainfo.json').get('pilot_complete'):print('RESUME verified view',cam,flush=True);continue
+        t=v.T;capture=np.array([int(r['capture_timestamp_ns']) for r in rows],np.int64)
+        ix,good,delta=nearest(dev,capture,1000);assert good.all(),'Missing official device-time/timecode mapping'
+        times=tc[ix];sourcegt={'archive':next(src.glob('*ground_truth.zip')).relative_to(base).as_posix()}
+        test_ids={'P0004','P0005','P0006','P0008','P0016','P0020'}
+        split='test' if meta['participant_id'] in test_ids else 'train'
+        assert meta['have_hand_object_pose_gt']==(split=='train')
+        v.m.update(view_type='ego',split=split,split_scheme='HOT3D_official_participant_split',split_source_locator={'url':'https://github.com/facebookresearch/hot3d/blob/146b34afef8c1a32adeef7e981c070109f225c87/README.md#testtrain-split','participant_id':meta['participant_id'],'archive':sourcegt['archive'],'member':'metadata.json'},source_hand_object_gt_available=meta['have_hand_object_pose_gt'])
+        v.m['sampling'].update(source_fps=30,source_fps_basis='camera VRS nominal_rate; confirmed against capture timestamps',selection='All data records in each of the three camera streams of the complete official VRS',capture_interval_median_ns=int(np.median(np.diff(capture))))
+        v.m['camera'].update(model='fisheye624',schema_id='camera/fisheye624/v1',native_channel_semantics='RGB' if cam=='214-1' else 'monochrome; unified RGB decoder repeats intensity in three channels',max_solid_angle=camera['maxSolidAngle'])
+        v.array('intrinsic','camera/intrinsic.pt','camera/fisheye624/v1',np.broadcast_to(camera['projectionParams'],(t,15)),np.ones(t,bool),['T','parameter'],source=dict(sourcegt,member='camera_models.json',stream_id=cam))
+        dc=camera['T_Device_Camera'];Tdc=transform(dc['quaternion_wxyz'],dc['translation_xyz'])
+        hi,hvalid,hdelta=nearest(head_times,times,1_000_000);E=np.full((t,4,4),np.nan)
+        for i in np.flatnonzero(hvalid):E[i]=np.linalg.inv(heads[head_times[hi[i]]]@Tdc)
+        assert hvalid.all(),'Headset GT missing at image times'
+        assert_se3(E);v.array('extrinsic','camera/extrinsic.pt','transform/se3/v1',E,hvalid,['T','row','column'],source=dict(sourcegt,member='headset_trajectory.csv',calibration_member='camera_models.json'),unit='m',convention='T_camera_from_world',method='inverse(T_world_from_device @ T_device_from_camera); official timecode mapping then <=1ms nearest source pose')
+        mi,mvalid,mdelta=nearest(hand_times,times,1_000_000);motion=np.full((t,2,61),np.nan);valid=np.zeros((t,2),bool)
+        for side,model in enumerate(models):
+            chosen=[i for i in range(t) if mvalid[i] and str(side) in hands[hand_times[mi[i]]]]
+            if not chosen:continue
+            native=[hands[hand_times[mi[i]]][str(side)] for i in chosen];pose=[];beta=[];trans=[]
+            for h in native:
+                wrist=h['wrist_xform'];pose.append(np.r_[Rotation.from_quat(np.roll(wrist['q_wxyz'],-1)).as_rotvec(),np.asarray(h['pose'])@model.data['hands_components'][:15]+model.data['hands_mean']]);beta.append(h['betas']);trans.append(wrist['t_xyz'])
+            motion[chosen,side],err=camera_mano(model,pose,beta,trans,E[chosen]);valid[chosen,side]=True
+            # Official source model uses PCA15 and nonflat mean, unlike output AA48.
+            import smplx
+            from types import SimpleNamespace
+            native_model=smplx.MANO(model_path=str(WORK/'tmp/pilot-models'),data_struct=SimpleNamespace(**model.data),is_rhand=side==1,use_pca=True,num_pca_comps=15,flat_hand_mean=False)
+            ixcheck=np.arange(0,len(chosen),max(1,len(chosen)//16));arr=np.asarray(pose);bet=np.asarray(beta);tr=np.asarray(trans)
+            with torch.no_grad():expected=native_model(global_orient=torch.tensor(arr[ixcheck,:3],dtype=torch.float32),hand_pose=torch.tensor(np.array([native[i]['pose'] for i in ixcheck]),dtype=torch.float32),betas=torch.tensor(bet[ixcheck],dtype=torch.float32),transl=torch.tensor(tr[ixcheck],dtype=torch.float32)).vertices.numpy()
+            actual=model.geometry(arr[ixcheck],bet[ixcheck],tr[ixcheck])[0];pcaerr=float(np.max(np.linalg.norm(expected-actual,axis=-1)));assert pcaerr<2e-6
+            v.checks.append({'test':'official_PCA15_vs_AA48_and_camera_geometry','side':side,'camera_max_m':err,'pca_max_m':pcaerr,'valid_frames':len(chosen),'passed':True})
+        if hand_times:
+            v.array('mano','annotation/mano/motion.pt','mano/axis_angle61/v1',motion,valid,['T','hand','parameter'],source={'archive':next(src.glob('*hand_data.zip')).relative_to(base).as_posix(),'member':'mano_hand_pose_trajectory.jsonl','time_index_ref':'camera/frames.jsonl:annotation_matches.mano'},unit='m',angle_unit='rad',coordinate_frame='camera',model_profile={'version':'MANO-v1.2','flat_hand_mean':True,'left_shapedirs_fix':bool(fixed)},method='Official PCA15 expanded with model mean and handedness fix; transform all matched MANO parameters into current camera')
+        def vis_values(records,identity_field,identities):
+            out=np.full((t,len(identities)),np.nan)
+            for c,identity in enumerate(identities):
+                values={int(r['timestamp[ns]']):float(r['visibility_ratio[%]']) for r in records if r['stream_id']==cam and str(r[identity_field])==str(identity) and r['visibility_ratio[%]']!=''}
+                if not values:continue
+                ts=sorted(values);ii,vv,dd=nearest(ts,times,1_000_000)
+                for i in np.flatnonzero(vv):out[i,c]=values[ts[ii[i]]]
+            return out
+        if hand_times and vis_h:
+            visibility=vis_values(vis_h,'hand_index',[0,1]);v.attribute('mano','annotation/mano/visibility.pt',visibility,['T','hand'],dict(sourcegt,member='box2d_hands.csv',stream_id=cam,match_method='nearest',max_delta_ns=1000000),'whole_hand')
+        for i,row in enumerate(v.frames):
+            row.update(timestamp_ns=int(capture[i]),clock_domain='device_time',timecode_ns=int(times[i]),timecode_clock_domain='timecode',mapping_delta_ns=int(delta[i]))
+            row['source_locator'].update(record_index=i,capture_timestamp_ns=int(capture[i]))
+            row['annotation_matches']={}
+            for key,ts,idxs,flags,dd in [('extrinsic',head_times,hi,hvalid,hdelta),('mano',hand_times,mi,mvalid,mdelta)]:
+                row['annotation_matches'][key]={'source_frame_id':str(ts[idxs[i]]) if flags[i] else None,'source_timestamp_ns':int(ts[idxs[i]]) if flags[i] else None,'clock_domain':'timecode','match_method':'exact' if dd[i]==0 else 'nearest','max_delta_ns':1000000,'delta_ns':int(dd[i]) if flags[i] else None,'image_time_ref':'timecode_ns','source_record_present':bool(flags[i])}
+        for oid in sorted(set(meta['object_uids'])|{r['object_uid'] for r in objects}):
+            rel=f'annotation/object/{oid}';op=v.p/rel;assets=op/'assets';assets.mkdir(parents=True,exist_ok=True)
+            sourceposes={int(r['timestamp[ns]']):csv_transform(r) for r in objects if r['object_uid']==oid};ts=sorted(sourceposes);oi,ok,od=nearest(ts,times,1_000_000)
+            P=np.full((t,4,4),np.nan)
+            for i in np.flatnonzero(ok):P[i]=E[i]@sourceposes[ts[oi[i]]]
+            key=f'object_{oid}_pose'
+            if sourceposes:
+                assert_se3(P[ok]);v.array(key,rel+'/pose.pt','object/rigid/v1',P,ok,['T','row','column'],source=dict(sourcegt,member='dynamic_objects.csv',object_uid=oid),unit='m',convention='T_camera_from_object')
+            for i,row in enumerate(v.frames):row['annotation_matches'][key]={'source_frame_id':str(ts[oi[i]]) if ok[i] else None,'source_timestamp_ns':int(ts[oi[i]]) if ok[i] else None,'clock_domain':'timecode','match_method':'exact' if od[i]==0 else 'nearest','max_delta_ns':1000000,'delta_ns':int(od[i]) if ok[i] else None,'image_time_ref':'timecode_ns','source_record_present':bool(ok[i])}
+            if sourceposes and vis_o:
+                ov=vis_values(vis_o,'object_uid',[oid])[:,0];v.attribute(key,rel+'/visibility.pt',ov,['T'],dict(sourcegt,member='box2d_objects.csv',stream_id=cam,object_uid=oid,match_method='nearest',max_delta_ns=1000000),'whole_object')
+            idx=int(reverse[oid]);prefix=f'models/obj_{idx:06d}'
+            for member in assetszip.namelist():
+                if member.startswith(prefix+'.'):(assets/Path(member).name).write_bytes(assetszip.read(member))
+            ply=assets/f'obj_{idx:06d}.ply';mesh=trimesh.load(ply,process=False,force='mesh');mesh.vertices=np.asarray(mesh.vertices)*.001
+            expected_extent=np.array([assetinfo[str(idx)]['size_'+x] for x in 'xyz']);assert np.allclose(mesh.extents,expected_extent,atol=1e-5)
+            text,textures=trimesh.exchange.obj.export_obj(mesh,return_texture=True,digits=12);meshpath=assets/f'obj_{idx:06d}.obj';meshpath.write_text(text)
+            for filename,data in textures.items():(assets/filename).write_bytes(data)
+            restored=trimesh.load(meshpath,process=False,force='mesh');assert np.allclose(restored.triangles,mesh.triangles,atol=1e-7)
+            assert np.array_equal(np.asarray(restored.visual.material.image),np.asarray(mesh.visual.material.image))
+            # Keep the official GLB as a companion asset so richer original material properties remain available.
+            glb=raw/'assets'/f'{oid}.glb'
+            if glb.exists():shutil.copyfile(glb,assets/glb.name)
+            # The millimetre PLY was a temporary converter input, not a second shared geometry convention.
+            assert ply.resolve().is_relative_to(assets.resolve());ply.unlink()
+            sourceasset={'archive':'raw/data/assets/hot3d_models.zip','member':prefix+'.ply','original_id':oid,'bop_id':idx}
+            v.mesh(oid,meshpath.relative_to(v.p).as_posix(),sourceasset)
+            save(op/'metainfo.json',{'object_id':oid,'source_object_id':oid,'identity_scope':'sequence','scope_id':seq,'association_basis':'official_object_uid_in_shared_world_coordinates','has_model':True,'mesh_ref':meshpath.relative_to(v.p).as_posix(),'unit':'m','source_locator':dict(sourcegt,member='dynamic_objects.csv' if sourceposes else 'metadata.json',object_uid=oid),'original_glb_ref':(assets/glb.name).relative_to(v.p).as_posix() if glb.exists() else None})
+            v.checks.append({'test':'object_mesh_scale_triangle_and_texture_roundtrip','object_id':oid,'vertices':len(mesh.vertices),'faces':len(mesh.faces),'extent_m':mesh.extents.tolist(),'passed':True})
+        v.checks.append({'test':'all_camera_frame_device_time_mapping','frames':t,'max_delta_ns':int(np.abs(delta).max()),'passed':True})
+        v.m['omitted_components']={'depth':'No depth stream in selected official Aria raw recording','segment':'Ground-truth masks/*.csv are frame-selection predicates, not pixel masks; no source pixel segmentation provided here','contact':'No native contact labels in downloaded scope','keypoints':'No independent MANO21 keypoint array in this raw release; selected MANO parameters retained','alternative_hand_model':'UmeTrack representation retained in raw; this shared contract selects MANO','alternative_camera_estimates':'Raw MPS online calibration and SLAM trajectories retained; shared uses the official GT camera_models and headset trajectory'}
+        if not hand_times:v.m['omitted_components']['mano']='Official MANO trajectory is empty; no pose labels released for this sequence'
+        if not objects:v.m['omitted_components']['object_pose']='Official object trajectory is empty; static source models retained without fabricated pose arrays'
+        print('MEDIA',dataset,cam,t,flush=True);jpeg_video(v,((p.name,p.read_bytes()) for p in paths))
+        v.m['pilot_complete']=True;v.finish()
+    assetszip.close();gt.close();handszip.close()
+
+if __name__=='__main__':
+    p=argparse.ArgumentParser();p.add_argument('dataset',choices=['ARCTIC','HOT3D']);a=p.parse_args()
+    torch.set_num_threads(4)
+    {'ARCTIC':arctic,'HOT3D':hot3d}[a.dataset]()
+    if a.dataset=='HOT3D':
+        verify_hot3d_frame_order_and_metadata()
+        from apply_hot3d_qa import apply_qa
+        apply_qa()
+    elif (DATA_ROOT/'ARCTIC'/RELEASE/'doc/validation/exposure.json').exists():
+        from repair_arctic_pilot import repair
+        repair()
diff --git a/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/diagnose_arctic_review.py b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/diagnose_arctic_review.py
new file mode 100644
index 0000000000000000000000000000000000000000..a6a76873a6890c7632983d1d811d329057068fda
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/diagnose_arctic_review.py
@@ -0,0 +1,125 @@
+"""Inspect source exposure and execute upstream articulation for the selected pilot.
+
+Read-only for raw and shared data. Writes portable evidence under doc/validation.
+"""
+from fresh_sequence_common import *
+from review_fresh_sequence_pilots import arr, project_for_view
+import ast
+
+SEQ='s09_laptop_use_03'
+BASE=DATA_ROOT/'ARCTIC'
+ROOT=BASE/RELEASE
+DOC=ROOT/'doc/validation'
+OFFICIAL=WORK/'tmp/fresh-sequence-pilot/official/ARCTIC'
+
+def exposure():
+    result={}
+    with zipfile.ZipFile(BASE/'raw/data/images_zips/s09/laptop_use_03.zip') as z:
+        for cam in map(str,range(9)):
+            names=sorted(n for n in z.namelist() if n.startswith(cam+'/') and n.endswith('.jpg'))
+            rows=[];canvas=Image.new('RGB',(10*160,180),(20,20,20))
+            from PIL import ImageDraw
+            draw=ImageDraw.Draw(canvas)
+            for i,n in enumerate(names):
+                im=Image.open(io.BytesIO(z.read(n)));im.draft('RGB',(160,160));im.thumbnail((160,150));gray=np.asarray(im.convert('L'))
+                rows.append({'frame_index':i,'source_frame_id':int(Path(n).stem),'median_luma':float(np.median(gray)),
+                    'p95_luma':float(np.percentile(gray,95)),'mean_luma':float(gray.mean()),'fraction_gt_32':float((gray>32).mean())})
+                if i<10:canvas.paste(im,(i*160,25));draw.text((i*160+3,5),f'idx {i}, src {i+1}',fill='white')
+            path=DOC/'diagnosis'/f'exposure-{cam}.jpg';path.parent.mkdir(parents=True,exist_ok=True);canvas.save(path,quality=94)
+            result[cam]={'method':'Official raw JPEG, PIL draft preview <=160x150, 8bit luma; diagnostics, not a general exposure classifier','frames':rows,'preview_ref':path.relative_to(DOC).as_posix()}
+            print('EXPOSURE',cam,'first10_p95',[r['p95_luma'] for r in rows[:10]],'p95_lt_35_indices',[r['frame_index'] for r in rows if r['p95_luma']<35],flush=True)
+    save(DOC/'exposure.json',result)
+    return result
+
+class Xdict(dict):
+    def overwrite(self,k,v):self[k]=v
+
+def official_object(objects,vertices,raw_parts,params,object_id='laptop'):
+    # Execute the pinned official class, bypassing its multi-object file loader.
+    # Only quaternion primitives use a SciPy implementation; no semantic rewriting.
+    tree=ast.parse((OFFICIAL/'common/object_tensors.py').read_text())
+    klass=next(n for n in tree.body if isinstance(n,ast.ClassDef) and n.name=='ObjectTensors')
+    def aa_quat(a):
+        a=a.numpy();q=Rotation.from_rotvec(a.reshape(-1,3)).as_quat().reshape(a.shape[:-1]+(4,))
+        return torch.tensor(q[...,[3,0,1,2]],dtype=torch.float32)
+    def qapply(q,v):
+        q=q.numpy();v=v.numpy();shape=np.broadcast_shapes(q.shape[:-1],v.shape[:-1]);qs=np.broadcast_to(q,shape+(4,)).reshape(-1,4)
+        out=Rotation.from_quat(qs[:,[1,2,3,0]]).apply(np.broadcast_to(v,shape+(3,)).reshape(-1,3)).reshape(shape+(3,))
+        return torch.tensor(out,dtype=torch.float32)
+    ns={'nn':torch.nn,'torch':torch,'np':np,'xdict':Xdict,'axis_angle_to_quaternion':aa_quat,'quaternion_apply':qapply}
+    exec(compile(ast.Module(body=[klass],type_ignores=[]),'pinned_ARCTIC_ObjectTensors','exec'),ns)
+    layer=ns['ObjectTensors'].__new__(ns['ObjectTensors']);torch.nn.Module.__init__(layer)
+    n=len(vertices);ft=lambda a:torch.tensor(np.asarray(a),dtype=torch.float32)
+    d={'names':[object_id],'diameter':ft([1]),'f':torch.zeros((1,1,3),dtype=torch.long),'f_len':torch.tensor([1]),'v_len':torch.tensor([n]),
+       'v':ft(vertices[None]),'mask':ft(np.ones((1,n))),'v_sub':ft(vertices[None]),
+       'parts_ids':torch.tensor(raw_parts[None]+1),'parts_sub_ids':torch.tensor(raw_parts[None]+1),'z_axis':ft([[0,0,-1]])}
+    for out,key in [('bbox_top','bbox_top'),('bbox_bottom','bbox_bottom'),('kp_top','keypoints_top'),('kp_bottom','keypoints_bottom')]:d[out]=ft([np.array(params[key])/1000])
+    layer.obj_tensors=d
+    out=layer.forward(ft(objects[:,:1]),ft(objects[:,1:4]),ft(objects[:,4:7]/1000),[object_id]*len(objects))
+    return out['v'].numpy(),out['kp3d'].numpy()
+
+def geometry():
+    with zipfile.ZipFile(BASE/'raw/data/meta.zip') as z:
+        pref='meta/object_vtemplates/laptop/'
+        vertices=np.array([[float(x) for x in r.split()[1:4]] for r in z.read(pref+'mesh.obj').decode().splitlines() if r.startswith('v ')])/1000
+        parts=np.array(json.loads(z.read(pref+'parts.json')),int);params=json.loads(z.read(pref+'object_params.json'))
+        misc=json.loads(z.read('meta/misc.json'))['s09']
+    with zipfile.ZipFile(BASE/'raw/data/raw_seqs.zip') as z:
+        objects=native_npy_bytes(z.read('raw_seqs/s09/laptop_use_03.object.npy'))
+        hand=native_npy_bytes(z.read('raw_seqs/s09/laptop_use_03.mano.npy'))
+    # Every annotation frame, a uniform vertex subset; plus every vertex at source 271.
+    vi=np.arange(0,len(vertices),max(1,len(vertices)//500));native,kp=official_object(objects,vertices[vi],parts[vi],params)
+    hands=[]
+    for side in ['left','right']:
+        model=Mano(side);h=hand[side];pose=np.c_[h['rot'],h['pose']];pose[:,3:]+=model.data['hands_mean']
+        xyz=model.geometry(pose,np.broadcast_to(h['shape'],(len(pose),10)),h['trans'])[0];hands.append(xyz)
+    tests=[]
+    for meta in sorted((ROOT/'data'/SEQ).glob('*/metainfo.json')):
+        p=meta.parent;m=load(meta);cam=m['camera_id'];source=BASE/'tmp/full-video-review-v2/diagnostic'/cam
+        if not (source/'snapshot.json').exists():source=p
+        E=arr(source/'camera/extrinsic.pt');P=arr(source/'annotation/object/laptop/pose.pt');top=arr(source/'annotation/object/laptop/parts/top/pose.pt')
+        actual=vertices[vi][None]@P[1:,:3,:3].transpose(0,2,1)+P[1:,None,:3,3]
+        sel=parts[vi]==0;actual[:,sel]=vertices[vi][sel][None]@top[1:,:3,:3].transpose(0,2,1)+top[1:,None,:3,3]
+        expected=native@E[1:,:3,:3].transpose(0,2,1)+E[1:,None,:3,3]
+        error=float(np.max(np.linalg.norm(actual-expected,axis=-1)));assert error<2e-5
+        motion=arr(source/'annotation/mano/motion.pt');handerrs=[]
+        for side in range(2):
+            model=Mano(['left','right'][side]);a=motion[1:,side];shared=model.geometry(a[:,:48],a[:,48:58],a[:,58:])[0]
+            expected_hand=hands[side]@E[1:,:3,:3].transpose(0,2,1)+E[1:,None,:3,3]
+            err=float(np.max(np.linalg.norm(shared-expected_hand,axis=-1)));assert err<2e-5;handerrs.append(err)
+        tests.append({'camera_id':cam,'annotation_frames':540,'vertices_per_frame':len(vi),'official_object_max_m':error,'source_to_shared_MANO_max_m':handerrs})
+        print('GEOMETRY',tests[-1],flush=True)
+    report={'issue':'Raw parts.json 0=top,1=bottom; previous converter wrongly treated 1 as top. Official code adds 1 and articulates parts_ids==1.',
+        'correction':'Shared convention remains 0=bottom,1=top; shared_id=1-raw_id. No time shift, pose change or image edit.',
+        'upstream_code':'https://github.com/zc-alexfan/arctic/blob/49f3eb4b7f01663172576357bb3d95695d7a7689/common/object_tensors.py',
+        'method':'Executed pinned upstream ObjectTensors class with actual raw tensor state and SciPy quaternion primitives, compared all 540 frames and 9 cameras. MANO verified independently from source world parameters.',
+        'tests':tests,'limitation':'Numerical agreement with source geometry does not establish source-to-image accuracy. Residual hand alignment remains for full-video human review.'}
+    save(DOC/'diagnosis.json',report)
+    return report
+
+def comparison():
+    from render_full_sequence_review import model_vertices, fast_edges, dots
+    p=ROOT/'data'/SEQ/'6';m=load(p/'metainfo.json');i=270
+    source=BASE/'tmp/full-video-review-v2/diagnostic/6'
+    if not (source/'snapshot.json').exists():source=p
+    with zipfile.ZipFile(BASE/'raw/data/images_zips/s09/laptop_use_03.zip') as z:rgb=np.array(Image.open(io.BytesIO(z.read('6/00271.jpg'))).convert('RGB').resize((500,700)))
+    with zipfile.ZipFile(BASE/'raw/data/meta.zip') as z:parts=np.array(json.loads(z.read('meta/object_vtemplates/laptop/parts.json')))
+    vertices=np.array([[float(x) for x in r.split()[1:4]] for r in (p/'annotation/object/laptop/assets/mesh.obj').read_text().splitlines() if r.startswith('v ')])
+    pose=arr(p/'annotation/object/laptop/pose.pt')[i];top=arr(p/'annotation/object/laptop/parts/top/pose.pt')[i]
+    K=arr(p/'camera/intrinsic.pt')[i];scale=np.array([.25,.25]);motion=arr(source/'annotation/mano/motion.pt')[i:i+1];valid=arr(source/'annotation/mano/motion_valid.pt')[i:i+1]
+    panels=[rgb.copy(),rgb.copy(),rgb.copy()]
+    for col,sel in [(1,parts==1),(2,parts==0)]:
+        xyz=vertices@pose[:3,:3].T+pose[:3,3];xyz[sel]=vertices[sel]@top[:3,:3].T+top[:3,3]
+        uv=project_for_view(xyz,m,(K,None),scale);dots(panels[col],uv[::4],(100,245,90),1)
+    for side,color in [(0,(0,225,240)),(1,(255,160,20))]:
+        v,j,e=model_vertices(motion,valid,side,False);uv=project_for_view(v[0],m,(K,None),scale)
+        for im in panels[1:]:fast_edges(im,uv,e,color)
+    canvas=np.full((750,1500,3),22,np.uint8)
+    for col,(im,title) in enumerate(zip(panels,['RAW source 271','OLD wrong object parts','FIXED official object parts'])):
+        canvas[45:745,col*500:(col+1)*500]=im;cv2.putText(canvas,title,(col*500+10,26),cv2.FONT_HERSHEY_SIMPLEX,.62,(250,250,250),1)
+    Image.fromarray(canvas).save(DOC/'diagnosis/camera6-270-before-after.jpg',quality=95)
+
+if __name__=='__main__':
+    import argparse
+    ap=argparse.ArgumentParser();ap.add_argument('mode',choices=['exposure','geometry','comparison']);args=ap.parse_args()
+    {'exposure':exposure,'geometry':geometry,'comparison':comparison}[args.mode]()
diff --git a/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/fresh_sequence_common.py b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/fresh_sequence_common.py
new file mode 100644
index 0000000000000000000000000000000000000000..b7a53b980e09a07459fc831b1abe5be0e7d1ec5d
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/fresh_sequence_common.py
@@ -0,0 +1,176 @@
+"""Helpers for the fresh ARCTIC/HOT3D whole-sequence v1.0 pilots.
+
+Original archives are read only. Temporary extraction is outside raw.
+Licensed MANO weights stay in the local model directory and are never published.
+"""
+from sequence_pilot_world import camera_mano, Mano
+from pilot_build import *
+from scipy.spatial.transform import Rotation
+import hashlib
+import re
+import time
+import trimesh
+
+DATA_ROOT = Path('E:/hand-datasets')
+RELEASE = 'shared-v1.0'
+RUN = 'fresh-sequence-pilot-20260920'
+VRS = WORK/'scripts/tools/vrs_env/Library/bin/vrs.exe'
+
+def digest(p):
+    h=hashlib.sha256()
+    with Path(p).open('rb') as f:
+        for b in iter(lambda:f.read(4*1024*1024),b''): h.update(b)
+    return h.hexdigest()
+
+def native_npy_bytes(data):
+    f=io.BytesIO(data);version=np.lib.format.read_magic(f)
+    shape,order,dtype=np.lib.format._read_array_header(f,version)
+    if dtype.hasobject:return Restricted(f,encoding='latin1').load().item()
+    return np.frombuffer(f.read(),dtype=dtype).reshape(shape,order='F' if order else 'C').copy()
+
+def command(args):
+    r=subprocess.run([str(x) for x in args],capture_output=True)
+    if r.returncode:raise RuntimeError(r.stderr.decode(errors='replace')[-3000:])
+    return r.stdout
+
+def framehash(args,copy=False):
+    output=command([FF,'-v','error','-threads','2','-noautorotate',*args,'-map','0:v:0',
+        *(['-c:v','copy'] if copy else ['-pix_fmt','rgb24','-threads','2']),
+        '-f','framehash','-hash','sha256','-fps_mode','passthrough','pipe:1']).decode()
+    return [(int(r.split(',')[-2]),r.split(',')[-1].strip()) for r in output.splitlines() if r and not r.startswith('#')]
+
+def jpeg_video(view, members):
+    """members is a fresh iterator of (locator, JPEG bytes). Verify every packet/frame."""
+    temporary=view.tmp/f'{view.m["sequence_id"]}__{view.m["camera_id"]}.mjpeg'
+    temporary.parent.mkdir(parents=True,exist_ok=True)
+    hashes=[]
+    with temporary.open('wb') as out:
+        for locator,b in members:
+            assert b[:2]==b'\xff\xd8',locator
+            hashes.append((len(b),hashlib.sha256(b).hexdigest()));out.write(b)
+    assert len(hashes)==view.T
+    dst=view.p/'observation/rgb.mkv';dst.parent.mkdir(parents=True,exist_ok=True)
+    source=['-f','image2pipe','-framerate','1','-c:v','mjpeg','-i',str(temporary)]
+    command([FF,'-v','error','-y',*source,'-map','0:v:0','-c:v','copy',dst])
+    packet=framehash(['-i',str(dst)],True)
+    assert packet==hashes,'source JPEG packets changed in MKV'
+    before=framehash(source);after=framehash(['-i',str(dst)])
+    assert before==after and len(before)==view.T,'RGB decoding differs'
+    w,h=view.m['camera']['image_size'];assert all(n==w*h*3 for n,_ in after)
+    info=subprocess.run([FF,'-hide_banner','-i',str(dst)],capture_output=True).stderr.decode(errors='replace')
+    line=next(s.strip() for s in info.splitlines() if 'Video: mjpeg' in s)
+    pixel=re.search(r'Video: mjpeg[^,]*,\s*(\w+)',line).group(1)
+    version=command([FF,'-version']).decode().splitlines()[0]
+    view.component('rgb','observation/rgb.mkv','rgb/uint8/v1',['T','height','width','channel'],np.ones(view.T,bool),
+        source=view.source,method='Original JPEG packets copied from raw into MKV; no recompression or rotation',
+        storage={'profile':'mkv/mjpeg_copy/v1','codec':'mjpeg','pixel_format':pixel,'encoded_size':[w,h],
+          'source_lossy':True,'decoder':{'name':'ffmpeg','version':version,'output_pixel_format':'rgb24','autorotate':False,'source_color_declaration':'unknown','actual_stream_description':line},
+          'encoder_parameters':{'mode':'stream_copy','version':version}},storage_fps=1,timebase_role='index_only')
+    view.checks.append({'test':'all_source_JPEG_packets_and_RGB_pixels_exact','passed':True,'frames':view.T,
+        'packet_manifest_sha256':hashlib.sha256(json.dumps(hashes).encode()).hexdigest(),
+        'pixel_manifest_sha256':hashlib.sha256(json.dumps(after).encode()).hexdigest(),'output_bytes':dst.stat().st_size})
+    save(view.tmp/'media_checks'/f'{view.m["sequence_id"]}__{view.m["camera_id"]}.json',{'packets':hashes,'pixels':after})
+    # This precise file was created by this call within the current run's temporary directory.
+    assert temporary.resolve().is_relative_to(view.tmp.resolve());temporary.unlink()
+
+class FreshView:
+    def __init__(self,dataset,seq,cam,ids,size,source):
+        self.root=DATA_ROOT/dataset/RELEASE;self.p=self.root/'data'/seq/cam
+        self.tmp=DATA_ROOT/dataset/'tmp'/RUN;self.p.mkdir(parents=True,exist_ok=True)
+        self.T=len(ids);self.source=source;self.checks=[]
+        self.m={'schema_version':'hand-warehouse/1.0','storage_revision':'lossless-video-2026-09-18',
+            'dataset_id':dataset,'release_id':RELEASE,'sequence_id':seq,'camera_id':cam,
+            'sample_type':'video','num_frames':self.T,'hand_slots':['left','right'],
+            'sampling':{'type':'dense','source_fps':None,'frames_ref':'camera/frames.jsonl','selection':'all released images in selected complete source sequence'},
+            'camera':{'model':'unknown','schema_id':None,'image_size':size},'split':'unknown','components':{},'pilot':True}
+        self.frames=[{'frame_index':i,'source_frame_id':str(fid),'timestamp_ns':None,'clock_domain':None,
+            'source_locator':dict(source,source_frame_id=str(fid))} for i,fid in enumerate(ids)]
+    def component(self,name,path,schema,axes,valid,source=None,method='source values retained',kind='format_conversion',**kw):
+        valid=np.asarray(valid,bool);vp=Path(path).with_name(Path(path).stem+'_valid.pt').as_posix();tensor(self.p/vp,valid)
+        self.m['components'][name]={'available':True,'schema_id':schema,'data_ref':path,'axes':axes,
+            'valid_ref':vp,'valid_axes':axes[:valid.ndim],'provenance':{'kind':kind,'method':method},
+            'source_locator':source or self.source,**kw}
+    def array(self,name,path,schema,data,valid,axes,**kw):
+        a=np.asarray(data,np.float32).copy();valid=np.asarray(valid,bool)
+        assert a.shape[0]==self.T and a.shape[:valid.ndim]==valid.shape
+        assert np.isfinite(a[valid]).all(),name
+        a[~valid]=np.nan;tensor(self.p/path,a)
+        self.component(name,path,schema,axes,valid,**kw)
+        restored=torch.load(self.p/path,map_location='cpu',weights_only=True).numpy()
+        assert np.array_equal(a,restored,equal_nan=True)
+        self.checks.append({'test':'tensor_exact_roundtrip','file':path,'shape':list(a.shape),'valid_count':int(valid.sum()),'passed':True})
+    def attribute(self,parent,path,data,axes,source,meaning):
+        data=np.asarray(data,np.float32);valid=np.isfinite(data)
+        assert ((data[valid]>=0)&(data[valid]<=1)).all()
+        vp=Path(path).with_name(Path(path).stem+'_valid.pt').as_posix();tensor(self.p/path,data);tensor(self.p/vp,valid)
+        self.m['components'][parent].setdefault('attributes',{})['visibility']={'data_ref':path,'valid_ref':vp,
+            'axes':axes,'valid_axes':axes,'dtype':'float32','visibility_type':'ratio','meaning':meaning,
+            'provenance':{'kind':'source','method':'native HOT3D visibility ratio retained in [0,1]; misleading [%] header is not a percentage scale'},'source_locator':source}
+    def mesh(self,oid,path,source):
+        self.component('object_'+oid+'_mesh',path,'mesh/obj/v1',['vertex','coordinate'],np.array(True),source=source,
+            method='Object vertices in metres; topology, material and texture dependencies retained',static=True,unit='m',coordinate_frame='object')
+    def finish(self):
+        lines(self.p/'camera/frames.jsonl',self.frames);save(self.p/'metainfo.json',self.m)
+        save(self.tmp/'checks'/f'{self.m["sequence_id"]}__{self.m["camera_id"]}.json',{'checks':self.checks,'frames':self.T,'input':self.source})
+        print('CONVERTED',self.m['dataset_id'],self.m['sequence_id'],self.m['camera_id'],self.T,flush=True)
+
+def assert_se3(a):
+    a=np.asarray(a);assert a.shape[-2:]==(4,4)
+    # A component may exist while QA leaves no usable poses in this view.
+    if a.size==0:return
+    assert np.isfinite(a).all() and np.allclose(a[...,3,:],[0,0,0,1],atol=1e-6)
+    r=a[...,:3,:3];assert np.max(np.abs(r@np.swapaxes(r,-1,-2)-np.eye(3)))<2e-5
+    assert np.max(np.abs(np.linalg.det(r)-1))<2e-5
+
+def setup(dataset,seq):
+    base=DATA_ROOT/dataset;out=base/RELEASE
+    for folder in ['doc','script','doc/validation']: (out/folder).mkdir(parents=True,exist_ok=True)
+    for f in ['CONTRACT.md','schema.json']:shutil.copyfile(WORK/'distribution/hand-warehouse-1.0'/f,out/'doc'/f)
+    save(out/'doc/validation/status.json',{'stage':'building_sequence_pilot','human_review':'pending','sequence_id':seq,'full_preprocessing_authorized':False,'upload_authorized':False})
+    sources=['fresh_sequence_common.py','convert_fresh_sequence_pilots.py','pilot_build.py','pilot_fetch.py','pilot_mano.py',
+        'sequence_pilot_common.py','sequence_pilot_world.py','pilot_hot_normalize.py','audit_source_fields.py']
+    for name in sources:
+        if (WORK/'scripts'/name).exists():shutil.copyfile(WORK/'scripts'/name,out/'script'/name)
+    (base/'tmp'/RUN).mkdir(parents=True,exist_ok=True)
+
+def object_textured_assets(z,prefix,target):
+    target.mkdir(parents=True,exist_ok=True)
+    for name in z.namelist():
+        if not name.startswith(prefix) or name.endswith('/'):continue
+        rel=Path(name[len(prefix):]);dest=target/rel
+        assert dest.resolve().is_relative_to(target.resolve());dest.parent.mkdir(parents=True,exist_ok=True)
+        if dest.suffix.lower() in ('.obj','.mtl','.jpg','.jpeg','.png'):
+            b=z.read(name)
+            if dest.suffix=='.obj':
+                rows=b.decode().splitlines()
+                for i,row in enumerate(rows):
+                    if row.startswith('v '):
+                        v=row.split();v[1:4]=[format(float(x)/1000,'.17g') for x in v[1:4]];rows[i]=' '.join(v)
+                b=('\n'.join(rows)+'\n').encode()
+            dest.write_bytes(b)
+    repair_arctic_material_refs(target)
+
+def repair_arctic_material_refs(target):
+    """Repair a source filename mismatch only when material identifiers agree."""
+    material=target/'material.mtl'
+    if not material.exists():return
+    ids={x.split(maxsplit=1)[1] for x in material.read_text().splitlines() if x.startswith('newmtl ')}
+    for obj in target.glob('*.obj'):
+        text=obj.read_text();rows=text.splitlines();used={x.split(maxsplit=1)[1] for x in rows if x.startswith('usemtl ')}
+        if 'mtllib material_0.mtl' in rows and not (target/'material_0.mtl').exists():
+            assert used and used.issubset(ids),(obj,used,ids)
+            obj.write_text(text.replace('mtllib material_0.mtl','mtllib material.mtl'))
+
+def project_points(xyz,camera,params):
+    xyz=np.asarray(xyz,float);uv=np.full((len(xyz),2),np.nan);good=np.isfinite(xyz).all(1)&(xyz[:,2]>1e-6)
+    if camera=='fisheye624':
+        p=np.asarray(params);xy=xyz[good,:2]/xyz[good,2,None];r=np.linalg.norm(xy,axis=1);th=np.arctan(r)
+        radial=th*(1+sum(p[3+i]*th**(2*i+2) for i in range(6)))
+        xy=xy*np.divide(radial,r,out=np.ones_like(r),where=r>1e-12)[:,None]
+        x,y=xy.T;rr=x*x+y*y
+        xd=x+(2*x*x+rr)*p[9]+2*x*y*p[10]+p[11]*rr+p[12]*rr*rr
+        yd=y+(2*y*y+rr)*p[10]+2*x*y*p[9]+p[13]*rr+p[14]*rr*rr
+        uv[good]=p[0]*np.stack([xd,yd],1)+p[1:3]
+    else:
+        K,D=params;uv[good]=cv2.projectPoints(xyz[good],np.zeros(3),np.zeros(3),K,D)[0].reshape(-1,2)
+    return uv
diff --git a/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/pilot_build.py b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/pilot_build.py
new file mode 100644
index 0000000000000000000000000000000000000000..2881d0878bd6d3456c2bf84edbf151cb3851a3ac
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/pilot_build.py
@@ -0,0 +1,209 @@
+"""Small, source-traceable dataset pilot. Originals are copied, never edited.
+
+Run with Python + numpy + torch + Pillow + opencv and imageio-ffmpeg.
+Optional runtime installed at tmp/pilot-runtime. Outputs: hand-datasets/.
+"""
+from pathlib import Path
+import argparse, codecs, collections, csv, io, json, pickle, shutil, subprocess, sys, zipfile
+sys.path.insert(0,str(Path(__file__).resolve().parents[1]/'tmp/pilot-runtime'))
+import numpy as np
+import torch
+from PIL import Image, ImageDraw
+import cv2
+from pilot_fetch import ROOT, sha
+
+WORK=ROOT.parent
+OLD=WORK/'datasets'
+sys.path.insert(0,str(WORK/'tmp/pilot-runtime'))
+import imageio_ffmpeg
+FF=imageio_ffmpeg.get_ffmpeg_exe()
+N=16
+
+def save(p,value):
+ p.parent.mkdir(parents=True,exist_ok=True)
+ p.write_text(json.dumps(value,ensure_ascii=False,indent=2,allow_nan=False),encoding='utf-8')
+def lines(p,rows):
+ p.parent.mkdir(parents=True,exist_ok=True)
+ p.write_text(''.join(json.dumps(x,ensure_ascii=False,allow_nan=False)+'\n' for x in rows),encoding='utf-8')
+def tensor(p,a):
+ p.parent.mkdir(parents=True,exist_ok=True);torch.save(torch.as_tensor(np.asarray(a).copy()),p)
+def load(p):return json.loads(p.read_text(encoding='utf-8'))
+
+class Restricted(pickle.Unpickler):
+ def find_class(self,module,name):
+  allowed={('numpy.core.multiarray','_reconstruct'):np.core.multiarray._reconstruct,
+   ('numpy._core.multiarray','_reconstruct'):np.core.multiarray._reconstruct,
+   ('numpy','ndarray'):np.ndarray,('numpy','dtype'):np.dtype,
+   ('numpy.core.numeric','_frombuffer'):np.core.numeric._frombuffer,
+   ('numpy._core.numeric','_frombuffer'):np.core.numeric._frombuffer,
+   ('_codecs','encode'):codecs.encode,
+   ('collections','OrderedDict'):collections.OrderedDict,
+   ('torch._utils','_rebuild_tensor_v2'):torch._utils._rebuild_tensor_v2,
+   ('torch.storage','_load_from_bytes'):lambda b:torch.load(io.BytesIO(b),map_location='cpu',weights_only=True)}
+  if (module,name) not in allowed:raise pickle.UnpicklingError(f'Unsupported class: {module}.{name}')
+  return allowed[(module,name)]
+def read_pickle(p):
+ with p.open('rb') as f:return Restricted(f,encoding='latin1').load()
+
+class Build:
+ def __init__(self,name):
+  self.name=name;self.base=ROOT/name;self.out=self.base/'pilot-v0';self.refs=[];self.views=[];self.checks=[]
+  for p in ['raw/doc','raw/script','raw/data','pilot-v0/doc','pilot-v0/script','pilot-v0/data']:(self.base/p).mkdir(parents=True,exist_ok=True)
+ def copy(self,src):
+  rel=src.relative_to(OLD/self.name);dst=self.base/'raw/data/local_subset'/rel
+  dst.parent.mkdir(parents=True,exist_ok=True)
+  if not dst.exists():shutil.copyfile(src,dst)
+  source_hash=sha(src);assert sha(dst)==source_hash
+  self.refs.append({'source':str(src),'stored':str(dst.relative_to(self.base)), 'sha256':source_hash,'bytes':src.stat().st_size,'kind':'historical_extracted_file_copy','official_archive_verified':False})
+  return dst
+ def view(self,seq,cam,ids,**meta):
+  p=self.out/'data'/seq/cam;p.mkdir(parents=True,exist_ok=True)
+  m={'schema_version':'pilot-v0','dataset':self.name,'sequence_id':seq,'camera_id':cam,'num_frames':len(ids),'fps':None,'split':None,'view_type':None,'time_semantics':'unknown','missing_modalities':{},**meta}
+  self.views.append((p,m))
+  lines(p/'frames.jsonl',[{'frame_index':i,'source_frame_id':str(s),'timestamp_ns':None,'clock_domain':None} for i,s in enumerate(ids)])
+  return p,m
+ def video(self,p,arrays):
+  a=np.stack(arrays);h,w=a.shape[1:3]
+  pixel='rgb24' if a.ndim==4 else ('gray16le' if a.dtype==np.uint16 else 'gray')
+  if pixel=='gray16le':a=a.astype('<u2')
+  a=np.ascontiguousarray(a);p.parent.mkdir(parents=True,exist_ok=True)
+  command=[FF,'-hide_banner','-loglevel','error','-y','-f','rawvideo','-pixel_format',pixel,'-video_size',f'{w}x{h}','-framerate','1','-i','pipe:0','-an','-c:v','ffv1','-level','3','-g','1',str(p)]
+  subprocess.run(command,input=a.tobytes(),check=True,capture_output=True)
+  decoded=subprocess.run([FF,'-v','error','-i',str(p),'-f','rawvideo','-pix_fmt',pixel,'pipe:1'],check=True,capture_output=True).stdout
+  assert decoded==a.tobytes(),f'Lossless roundtrip failed: {p}'
+  self.checks.append({'file':str(p.relative_to(self.out)),'test':'decode_all_frames_pixel_exact','passed':True,'frames':len(a),'dtype':str(a.dtype),'shape':list(a.shape)})
+  return {'codec':'FFV1','pixel_format':pixel,'container_fps':1,'container_time_is_capture_time':False}
+ def finish(self,notes):
+  for p,m in self.views:save(p/'metainfo.json',m)
+  save(self.base/'raw/doc/local_subset_manifest.json',self.refs)
+  report={'dataset':self.name,'views':len(self.views),'observations':sum(m['num_frames'] for _,m in self.views),'checks':self.checks,'notes':notes,'status':'partial' if notes else 'passed'}
+  save(self.out/'doc/report.json',report)
+  for script in ['pilot_build.py','pilot_fetch.py']:
+   shutil.copyfile(WORK/'scripts'/script,self.out/'script'/script)
+  (self.out/'script/README.md').write_text('执行入口：工作区 scripts/pilot_build.py --dataset '+self.name+'。此目录为脚本快照；脚本当前依赖工作区的 datasets 与 tmp/pilot-runtime，尚不是独立发布包。\n',encoding='utf-8')
+  files=[{'path':str(p.relative_to(self.out)), 'bytes':p.stat().st_size,'sha256':sha(p)} for p in sorted((self.out/'data').rglob('*')) if p.is_file()]
+  save(self.out/'doc/output_manifest.json',files)
+  print(json.dumps(report,ensure_ascii=False),flush=True)
+  return report
+
+def h2o():
+ b=Build('H2O');src=OLD/'H2O/subject4_ego/h1/0/cam4';ids=list(range(N))
+ p,m=b.view('subject4_h1_0','cam4',ids,split='test',view_type='ego',time_semantics='ordered_source_frames; acquisition FPS not yet verified')
+ for folder in ['rgb','depth','hand_pose','hand_pose_mano','cam_pose']:
+  for i in ids:b.copy(src/folder/f'{i:06d}.{ "png" if folder in ["rgb","depth"] else "txt"}')
+ intr=np.loadtxt(b.copy(src/'cam_intrinsics.txt'));fx,fy,cx,cy,w,h=intr
+ tensor(p/'camera/intrinsic.pt',np.array([[fx,0,cx],[0,fy,cy],[0,0,1]],np.float32))
+ ext=np.stack([np.loadtxt(src/'cam_pose'/f'{i:06d}.txt').reshape(4,4) for i in ids])
+ tensor(p/'camera/extrinsic.pt',np.linalg.inv(ext).astype(np.float32))
+ m['camera']={'model':'pinhole','width':int(w),'height':int(h),'extrinsic':'T_camera_from_world','source_extrinsic':'T_world_from_camera','unit':'m','distortion':'not supplied in selected intrinsics'}
+ a=np.stack([np.loadtxt(src/'hand_pose_mano'/f'{i:06d}.txt').reshape(2,62) for i in ids])
+ valid=a[:,:,0]==1;motion=np.concatenate([a[:,:,4:52],a[:,:,52:62],a[:,:,1:4]],axis=-1).astype(np.float32)
+ motion[~valid]=np.nan;tensor(p/'annotation/mano/motion.pt',motion);tensor(p/'annotation/mano/valid_flag.pt',valid)
+ m['mano']={'layout':['pose_axis_angle:48','betas:10','translation:3'],'shape':[N,2,61],'slots':['left','right'],'coordinate_frame':'camera','translation_unit':'m','flat_hand_mean':True,'model_version':'not supplied; MANO model not bundled','valid_flag':'source annotation flag, not visibility','geometry_validation':'not_tested; model unavailable'}
+ joints=np.stack([np.loadtxt(src/'hand_pose'/f'{i:06d}.txt').reshape(2,64) for i in ids]);tensor(p/'annotation/keypoints/xyz.pt',joints[:,:,1:].reshape(N,2,21,3).astype(np.float32));tensor(p/'annotation/keypoints/valid_flag.pt',joints[:,:,0]==1)
+ rgb=[np.asarray(Image.open(src/'rgb'/f'{i:06d}.png').convert('RGB')) for i in ids]
+ depth=[np.asarray(Image.open(src/'depth'/f'{i:06d}.png'),dtype=np.uint16) for i in ids]
+ m['rgb']=b.video(p/'observation/rgb.mkv',rgb);m['depth']={**b.video(p/'observation/depth.mkv',depth),'unit':'mm','scale_to_meters':0.001,'invalid_value':0,'rgb_registration':'not_tested'}
+ m['missing_modalities']={'object':'not present in selected historical subset','segmentation':'no native mask in selected subset'}
+ # Independent check: 3D keypoint projections must be finite where the source flags valid.
+ xyz=joints[:,:,1:].reshape(N,2,21,3);projection=xyz@np.array([[fx,0,cx],[0,fy,cy],[0,0,1]]).T
+ uv=projection[...,:2]/projection[...,2:];assert np.isfinite(uv[joints[:,:,0]==1]).all()
+ image=Image.fromarray(rgb[0]);draw=ImageDraw.Draw(image)
+ for side in range(2):
+  for u,v in uv[0,side]:draw.ellipse((u-3,v-3,u+3,v+3),fill=('red' if side==0 else 'cyan'))
+ image.save(b.out/'doc/keypoints_overlay.png')
+ return b.finish(['MANO geometry not rendered; model asset unavailable. RGB-depth registration not tested. No native segmentation or object input in subset.'])
+
+def visor():
+ b=Build('VISOR');raw=b.base/'raw/data/GroundTruth-SparseAnnotations';jp=raw/'annotations/train/P01_03.json';zp=raw/'rgb_frames/train/P01/P01_03.zip'
+ source=load(jp);annotations=source['video_annotations'][:N]
+ ids=[x['image']['name'] for x in annotations];p,m=b.view('P01_03','ego',ids,split='train',view_type='ego',time_semantics='sparse_annotations; use source frame IDs; MKV timestamps are storage indices')
+ rgbs=[];masks=[];records=[]
+ with zipfile.ZipFile(zp) as z:
+  assert z.testzip() is None
+  for i,frame in enumerate(annotations):
+   rgb=np.asarray(Image.open(io.BytesIO(z.read(frame['image']['name']))).convert('RGB'));rgbs.append(rgb)
+   mask=np.zeros((1080,1920),np.uint16);entities=[]
+   for value,entity in enumerate(frame['annotations'],1):
+    # Same polygon rasterization primitive and integer conversion as official VISOR-VIS.
+    polygons=[np.asarray(poly,dtype=np.int32) for poly in entity['segments'] if len(poly)>=3]
+    if polygons:cv2.fillPoly(mask,polygons,int(value))
+    entities.append({'mask_value':value,'instance_id':entity['id'],'track_id':None,'source':{k:v for k,v in entity.items() if k!='segments'}})
+   if mask.shape!=rgb.shape[:2]:mask=cv2.resize(mask,(rgb.shape[1],rgb.shape[0]),interpolation=cv2.INTER_NEAREST)
+   masks.append(mask);records.append({'frame_index':i,'source_frame_id':frame['image']['name'],'subsequence':frame['image']['subsequence'],'annotation_valid':True,'instances':entities})
+   assert set(np.unique(mask)) <= {0,*[x['mask_value'] for x in entities]}
+ m['rgb']=b.video(p/'observation/rgb.mkv',rgbs);m['segmentation']={**b.video(p/'annotation/segmentation/mask.mkv',masks),'provenance':'native_sparse_polygon_rasterized','background_value':0,'background_semantics':'unlabeled/background; source is not globally exhaustive','ignore_value':None,'polygon_canvas':[1920,1080],'overlap_policy':'source annotation order; last polygon wins','track_id':'unknown; no heuristic tracking'}
+ lines(p/'annotation/segmentation/instances.jsonl',records)
+ m['missing_modalities']={'mano':'not supplied','depth':'not supplied','camera':'not supplied','object_pose_mesh':'not supplied'}
+ im=rgbs[2].copy();mask=masks[2];colors=np.array([[0,0,0],[240,70,70],[70,240,90],[70,100,240],[230,190,60],[150,80,200]],np.uint8)
+ used=mask>0;im[used]=(im[used]*0.55+colors[mask[used]%len(colors)]*0.45).astype(np.uint8)
+ Image.fromarray(im).save(b.out/'doc/segmentation_overlay.png')
+ b.checks.append({'test':'official_zip_crc_all_members','passed':True,'members':len(zipfile.ZipFile(zp).namelist())})
+ b.checks.append({'test':'all_mask_values_resolve_to_frame_instances','passed':True})
+ return b.finish([])
+
+def rhd():
+ b=Build('RHD');src=OLD/'RHD/RHD_published_v2/evaluation';ann=read_pickle(b.copy(src/'anno_evaluation.pickle'))
+ for i in range(N):
+  p,m=b.view(f'evaluation_{i:05d}','camera',[i],split='evaluation',time_semantics='independent_image')
+  rgb=np.asarray(Image.open(b.copy(src/'color'/f'{i:05d}.png')).convert('RGB'))
+  depth=np.asarray(Image.open(b.copy(src/'depth'/f'{i:05d}.png')).convert('RGB'),dtype=np.uint16)
+  code=depth[:,:,0]*256+depth[:,:,1]
+  m['rgb']=b.video(p/'observation/rgb.mkv',[rgb]);m['depth']={**b.video(p/'observation/depth.mkv',[code]),'encoding':'source R*256+G','metric_unit':None,'metric_conversion_status':'not_tested; preserved integer code only'}
+  tensor(p/'camera/intrinsic.pt',ann[i]['K']);tensor(p/'annotation/keypoints/xyz.pt',ann[i]['xyz'].reshape(1,2,21,3));tensor(p/'annotation/keypoints/uv.pt',ann[i]['uv_vis'][:,:2].reshape(1,2,21,2));tensor(p/'annotation/keypoints/visible.pt',(ann[i]['uv_vis'][:,2]==1).reshape(1,2,21))
+  m['keypoints']={'unit':'m','layout':'source 42-joint order reshaped as 2x21','hand_order':['left','right'],'joint_order':'official RHD: wrist then thumb/index/middle/ring/pinky, fingers tip-to-palm','definition_reference':'raw/doc/dataset_README.txt','visibility':'per joint; invisible does not mean unannotated'}
+  m['missing_modalities']={'mano':'not supplied by source','segmentation':'omitted by historical downloader','extrinsic':'not supplied in selected annotations'}
+ return b.finish(['Depth metric conversion pending; segmentation absent from historical input.'])
+
+def freihand():
+ b=Build('FreiHAND');src=OLD/'FreiHAND';K=load(b.copy(src/'training_K.json'));mano=load(b.copy(src/'training_mano.json'));scale=load(b.copy(src/'training_scale.json'))
+ for base in range(4):
+  for variant in range(4):
+   i=base+variant*32560;p,m=b.view(f'training_{base:08d}',f'variant_{variant}',[i],split='training',time_semantics='independent_image',base_sample_id=str(base),variant_id=variant)
+   m['camera_group_semantics']='image_variant_not_physical_camera';m['physical_camera_id']=None
+   rgb=np.asarray(Image.open(b.copy(src/'training/rgb'/f'{i:08d}.jpg')).convert('RGB'));m['rgb']=b.video(p/'observation/rgb.mkv',[rgb])
+   tensor(p/'camera/intrinsic.pt',np.asarray(K[base],np.float32));tensor(p/'annotation/mano/source_theta.pt',np.asarray(mano[base],np.float32));tensor(p/'annotation/mano/source_scale.pt',np.asarray(scale[base],np.float32))
+   m['mano']={'source_theta_layout':['pose48','shape10','uv_root2','scale1'],'motion_status':'not_generated','reason':'To recover MANO translation, official code subtracts posed joint 9 from backprojected root. Requires correct MANO model; last three source values are not translation.'}
+   m['missing_modalities']={'depth':'not supplied','segmentation':'not present in historical input','extrinsic':'not supplied','normalized_mano':'model dependency unresolved'}
+ return b.finish(['Preserved source MANO theta; no misleading motion.pt produced. Correct model needed for normalized translation.'])
+
+def hoi4d():
+ b=Build('HOI4D');rel=Path('ZY20210800001/H1/C20/N18/S286/s01/T1');src=OLD/'HOI4D/Hand_pose/handpose_left_hand'/rel
+ files=sorted(src.glob('*.pickle'),key=lambda p:int(p.stem))[:N];ids=[x.stem for x in files]
+ p,m=b.view('H1_C20_N18_S286_s01_T1','ZY20210800001',ids,view_type='ego',time_semantics='annotation_frame_ids_only')
+ a=np.full((len(ids),2,61),np.nan,np.float32);v=np.zeros((len(ids),2),bool);uv=np.full((len(ids),2,21,2),np.nan,np.float32)
+ for i,frame in enumerate(ids):
+  for side,folder in enumerate(['handpose_left_hand','handpose_right_hand']):
+   f=OLD/'HOI4D/Hand_pose'/folder/rel/(frame+'.pickle')
+   if not f.exists():continue
+   x=read_pickle(b.copy(f));a[i,side]=np.concatenate([x['poseCoeff'].reshape(-1),x['beta'].reshape(-1),x['trans'].reshape(-1)]);v[i,side]=True;uv[i,side]=x['kps2D']
+ tensor(p/'annotation/mano/motion.pt',a);tensor(p/'annotation/mano/valid_flag.pt',v);tensor(p/'annotation/keypoints/uv.pt',uv)
+ tensor(p/'camera/intrinsic.pt',np.load(b.copy(OLD/'HOI4D/camera_params/ZY20210800001/intrin.npy'),allow_pickle=False))
+ m['mano']={'layout':['pose_axis_angle48','betas10','translation3'],'slots':['left','right'],'translation_unit':'m','coordinate_frame':'camera','flat_hand_mean':True,'valid_flag':'source file present and parsed; not visibility','geometry_validation':'not_tested'}
+ m['missing_modalities']={'rgb':'not downloaded','depth':'not downloaded','segmentation':'not downloaded','extrinsic':'not available in selected subset','object':'not available in selected subset'}
+ return b.finish(['Annotation-only mini; no matching RGB/depth/segmentation downloaded. MANO geometry not rendered.'])
+
+def reinterhand():
+ b=Build('ReInterHand');src=next(p for p in sorted((OLD/'ReInterHand').glob('m--*')) if (p/'mano_fits/params').exists());files=sorted((src/'mano_fits/params').glob('*_left.json'))[:N];ids=[x.stem.rsplit('_',1)[0] for x in files]
+ # Capture-level poses have no established camera transform in current local subset.
+ p,m=b.view(src.name,'source_unresolved',ids,time_semantics='capture_annotations_only; not a camera observation')
+ rows=[]
+ for frame in ids:
+  row={'source_frame_id':frame,'hands':{}}
+  for side in ['left','right']:
+   f=src/'mano_fits/params'/f'{frame}_{side}.json'
+   if f.exists():row['hands'][side]=load(b.copy(f))
+  rows.append(row)
+ lines(p/'annotation/mano/source_parameters.jsonl',rows)
+ m['missing_modalities']={'rgb':'not extracted from incomplete archive set','camera':'not available','normalized_mano':'source-to-camera mapping unresolved'}
+ m['not_a_real_camera']=True
+ return b.finish(['Source-only annotation sample. source_unresolved is an explicit staging group, not an invented camera.'])
+
+def main():
+ parser=argparse.ArgumentParser();parser.add_argument('--dataset',nargs='*',default=['VISOR','H2O','RHD','FreiHAND','HOI4D','ReInterHand']);args=parser.parse_args()
+ funcs={'VISOR':visor,'H2O':h2o,'RHD':rhd,'FreiHAND':freihand,'HOI4D':hoi4d,'ReInterHand':reinterhand}
+ for name in args.dataset:
+  try:funcs[name]()
+  except Exception as e:
+   save(ROOT/name/'pilot-v0/doc/build_error.json',{'error':str(e)});raise
+if __name__=='__main__':main()
diff --git a/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/pilot_fetch.py b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/pilot_fetch.py
new file mode 100644
index 0000000000000000000000000000000000000000..7671f83ac7a97438b95528fe9c13c00768eb7861
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/pilot_fetch.py
@@ -0,0 +1,63 @@
+"""Download bounded official pilot inputs; keep archives intact. No upstream code execution."""
+from pathlib import Path
+import concurrent.futures, hashlib, json, requests
+
+ROOT = Path(__file__).resolve().parents[1] / 'hand-datasets'
+REFS = {
+ 'Dex-YCB': ('NVlabs/dex-ycb-toolkit', ['README.md','dex_ycb_toolkit/dex_ycb.py','dex_ycb_toolkit/sequence_loader.py']),
+ 'VISOR': ('epic-kitchens/VISOR-VIS', ['README.md','vis.py']),
+ 'H2O': ('taeinkwon/h2odataset', ['README.md','download_script.py']),
+ 'HOI4D': ('leolyliu/HOI4D-Instructions', ['README.md','utils/parse_2Dmask_img.py','utils/color_map.py']),
+ 'ARCTIC': ('zc-alexfan/arctic', ['docs/data/README.md','docs/data/data_doc.md']),
+ 'FreiHAND': ('lmb-freiburg/freihand', ['README.md','utils/model.py','utils/fh_utils.py','view_samples.py']),
+ 'RHD': ('lmb-freiburg/hand3d', ['README.md','create_binary_db.py']),
+ 'HOT3D': ('facebookresearch/hot3d', ['README.md','hot3d/data_loaders/loader_hand_poses.py']),
+ 'OakInk2': ('oakink/OakInk2', ['README.md']),
+ 'ReInterHand': ('facebookresearch/InterWild', ['tool/ReInterHand/download/download_checksum_framelist.py','tool/ReInterHand/download/download_ego_per_frame.py','tool/ReInterHand/download/download_mano_fits.py']),
+}
+BASE='https://data.bris.ac.uk/datasets/2v6cgv1x04ol22qp9rm9x2j6a7/'
+
+def sha(p):
+ h=hashlib.sha256()
+ with p.open('rb') as f:
+  for b in iter(lambda:f.read(1024*1024),b''):h.update(b)
+ return h.hexdigest()
+
+def fetch(dataset, url, relative, limit=30_000_000):
+ p=ROOT/dataset/'raw'/relative;p.parent.mkdir(parents=True,exist_ok=True)
+ receipt=p.with_name(p.name+'.receipt.json')
+ if p.exists() and receipt.exists():
+  info=json.loads(receipt.read_text()); assert sha(p)==info['sha256'];return info
+ part=p.with_name(p.name+'.part')
+ with requests.get(url,stream=True,headers={'Accept-Encoding':'identity'},timeout=(15,40)) as r:
+  r.raise_for_status(); expected=int(r.headers.get('Content-Length',0))
+  if expected>limit:raise ValueError(f'File exceeds pilot limit: {expected}')
+  size=0
+  with part.open('wb') as f:
+   for chunk in r.iter_content(1024*1024):
+    size+=len(chunk)
+    if size>limit:raise ValueError('Stream exceeds pilot limit')
+    f.write(chunk)
+  if expected and not r.headers.get('Content-Encoding') and size!=expected:raise ValueError('Incomplete transfer')
+ part.replace(p)
+ info={'url':url,'path':str(p.relative_to(ROOT/dataset)), 'bytes':size,'sha256':sha(p),'official_checksum':None,'verification':'HTTP success; content length where supplied; locally computed SHA256'}
+ receipt.write_text(json.dumps(info,indent=2),encoding='utf-8');return info
+
+def reference(item):
+ name,(repo,paths)=item;results=[]
+ for section in ['raw/doc','raw/script','raw/data','pilot-v0/doc','pilot-v0/script','pilot-v0/data']:(ROOT/name/section).mkdir(parents=True,exist_ok=True)
+ for path in paths:
+  try:results.append(fetch(name,f'https://raw.githubusercontent.com/{repo}/HEAD/{path}',('script/' if path.endswith('.py') else 'doc/')+path))
+  except Exception as e:results.append({'path':path,'error':str(e)})
+ return name,results
+
+def main():
+ results=dict(concurrent.futures.ThreadPoolExecutor(6).map(reference,REFS.items()))
+ for path in ['GroundTruth-SparseAnnotations/annotations/train/P01_03.json','GroundTruth-SparseAnnotations/rgb_frames/train/P01/P01_03.zip','EPIC_100_noun_classes_v2.csv','frame_mapping.json']:
+  try:results['VISOR'].append(fetch('VISOR',BASE+path,'data/'+path))
+  except Exception as e:results['VISOR'].append({'path':path,'error':str(e)})
+ for name,rows in results.items():
+  (ROOT/name/'raw/doc/download_manifest.json').write_text(json.dumps(rows,indent=2),encoding='utf-8')
+  print(name,[(r.get('path'),r.get('bytes',r.get('error'))) for r in rows],flush=True)
+
+if __name__=='__main__':main()
diff --git a/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/pilot_hot_normalize.py b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/pilot_hot_normalize.py
new file mode 100644
index 0000000000000000000000000000000000000000..3b224e4d405082be63eab19b8461328423626a66
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/pilot_hot_normalize.py
@@ -0,0 +1,39 @@
+from pilot_mano import *
+
+def transform(q,t):
+ q=np.asarray(q,float);q=q/np.linalg.norm(q);w,x,y,z=q
+ R=np.array([[1-2*(y*y+z*z),2*(x*y-z*w),2*(x*z+y*w)],[2*(x*y+z*w),1-2*(x*x+z*z),2*(y*z-x*w)],[2*(x*z-y*w),2*(y*z+x*w),1-2*(x*x+y*y)]])
+ T=np.eye(4);T[:3,:3]=R;T[:3,3]=t;return T
+def csv_transform(row):return transform([float(row['q_wo_'+a]) for a in ['w','x','y','z']],[float(row[f't_wo_{a}[m]']) for a in ['x','y','z']])
+def main():
+ base=ROOT/'HOT3D/pilot-v0';p=next((base/'data').rglob('metainfo.json')).parent;m=load(p/'metainfo.json');frames=[json.loads(l) for l in (p/'frames.jsonl').read_text().splitlines()];source=OLD/'HOT3D/P0001_10a27bf7'
+ camera=next(c for c in load(source/'camera_models.json') if c['stream_id']=='214-1');dc=camera['T_Device_Camera'];Tdc=transform(dc['quaternion_wxyz'],dc['translation_xyz'])
+ device={int(r['timestamp[ns]']):csv_transform(r) for r in csv.DictReader((source/'headset_trajectory.csv').open())}
+ Tcws=np.stack([np.linalg.inv(device[f['timecode_ns']]@Tdc) for f in frames]);tensor(p/'camera/extrinsic.pt',Tcws.astype(np.float32));tensor(p/'camera/intrinsic.pt',np.array(camera['projectionParams'],np.float32))
+ m['camera']={'projection_model':'FISHEYE624','intrinsic_layout':'15 native projectionParams, not a 3x3 K','width':camera['imageWidth'],'height':camera['imageHeight'],'extrinsic':'T_camera_from_world','unit':'m','calibration_source':'static camera_models.json + exact timecode headset trajectory; online calibration not substituted'}
+ models=[Mano('left'),Mano('right')]
+ # Match the explicitly documented fix in HOT3D official mano_layer.py.
+ fixed=np.abs(models[0].data['shapedirs'][:,0]-models[1].data['shapedirs'][:,0]).sum()<1
+ if fixed:
+  models[0].data['shapedirs'][:,0]*=-1
+  with torch.no_grad():models[0].layer.shapedirs[:,0]*=-1
+ rows=[json.loads(l) for l in (p/'annotation/mano/source_parameters.jsonl').read_text().splitlines()];motion=np.full((len(rows),2,61),np.nan,np.float32);valid=np.zeros((len(rows),2),bool);errors=[]
+ for i,row in enumerate(rows):
+  for side,hand in row['source_annotation']['hand_poses'].items():
+   side=int(side);model=models[side];beta=np.array(hand['betas']);wrist=hand['wrist_xform'];Tw=transform(wrist['q_wxyz'],wrist['t_xyz']);pose=np.r_[cv2.Rodrigues(Tw[:3,:3])[0].ravel(),np.array(hand['pose'])@model.data['hands_components'][:15]+model.data['hands_mean']]
+   world_vertices,_,_=model.geometry(pose,beta,Tw[:3,3]);root=model.data['J_regressor'][0]@(model.data['v_template']+np.einsum('vci,i->vc',model.data['shapedirs'],beta))
+   Tc=Tcws[i];camera_pose=pose.copy();camera_pose[:3]=cv2.Rodrigues(Tc[:3,:3]@Tw[:3,:3])[0].ravel();camera_t=Tc[:3,:3]@(root+Tw[:3,3])+Tc[:3,3]-root
+   camera_vertices,_,_=model.geometry(camera_pose,beta,camera_t);transformed=world_vertices@Tc[:3,:3].T+Tc[:3,3];error=float(np.linalg.norm(camera_vertices-transformed,axis=-1).max());errors.append(error);assert error<1e-5
+   motion[i,side]=np.r_[camera_pose,beta,camera_t];valid[i,side]=True
+ tensor(p/'annotation/mano/motion.pt',motion);tensor(p/'annotation/mano/valid_flag.pt',valid)
+ m['mano']={'layout':['pose_axis_angle48','betas10','translation3'],'slots':['left','right'],'coordinate_frame':'camera','translation_unit':'m','flat_hand_mean':True,'model_version':'MANO v1.2','left_shapedirs_fix':bool(fixed),'conversion':'official 15-PCA expansion; world-to-camera rotation and shape-dependent root correction','valid_flag':'source annotation present, not visibility'}
+ objects=[json.loads(l) for l in (p/'annotation/object/source_world_poses.jsonl').read_text().splitlines()];object_ids=sorted({r['object_uid'] for r in objects})
+ for oid in object_ids:
+  lookup={int(r['timestamp[ns]']):r for r in objects if r['object_uid']==oid};poses=np.full((len(frames),4,4),np.nan,np.float32);flags=np.zeros(len(frames),bool)
+  for i,frame in enumerate(frames):
+   if frame['timecode_ns'] in lookup:poses[i]=Tcws[i]@csv_transform(lookup[frame['timecode_ns']]);flags[i]=True
+  tensor(p/'annotation/object'/oid/'pose.pt',poses);tensor(p/'annotation/object'/oid/'valid_flag.pt',flags)
+ m['object']={'pose':'T_camera_from_object','unit':'m','mesh_status':'not downloaded','object_ids':object_ids};save(p/'metainfo.json',m)
+ save(base/'doc/mano_validation.json',{'test':'camera-space vertices equal transformed world-space vertices','max_vertex_error_m':max(errors),'passed':True,'limitation':'algebraic transform check; no independent fisheye reprojection evaluation','left_shapedirs_fix':bool(fixed)})
+ print('HOT3D normalized',motion.shape,'max transform error',max(errors))
+if __name__=='__main__':main()
diff --git a/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/pilot_mano.py b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/pilot_mano.py
new file mode 100644
index 0000000000000000000000000000000000000000..d4596f225a22b61f642a2ef0193588172a5b581e
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/pilot_mano.py
@@ -0,0 +1,87 @@
+"""Local licensed MANO model dependency; never copies models into shared outputs."""
+from pilot_build import *
+from types import SimpleNamespace
+import smplx
+
+class StoredState:
+ def __setstate__(self,state):self.__dict__.update(state)
+class ModelReader(Restricted):
+ def find_class(self,module,name):
+  if (module,name) in [('chumpy.ch','Ch'),('chumpy.reordering','Select'),('scipy.sparse.csc','csc_matrix')]:return StoredState
+  if (module,name)==('__builtin__','set'):return set
+  return super().find_class(module,name)
+
+def model_data(side):
+ p=WORK/'tmp/pilot-models'/f'MANO_{side.upper()}.pkl'
+ with p.open('rb') as f:d=ModelReader(f,encoding='latin1').load()
+ # Materialize only the two known serialized structures; execute no Chumpy code.
+ s=d['shapedirs']
+ if isinstance(s,StoredState):d['shapedirs']=s.a.x.reshape(-1)[s.idxs].reshape(s.preferred_shape)
+ sparse=d['J_regressor'];dense=np.zeros(sparse._shape)
+ for col in range(sparse._shape[1]):
+  a,z=sparse.indptr[col:col+2];dense[sparse.indices[a:z],col]=sparse.data[a:z]
+ d['J_regressor']=dense
+ assert d['shapedirs'].shape==(778,3,10) and dense.shape==(16,778)
+ return d
+
+class Mano:
+ def __init__(self,side):
+  self.side=side
+  self.data=model_data(side)
+  self.layer=smplx.MANO(model_path=str(WORK/'tmp/pilot-models'),data_struct=SimpleNamespace(**self.data),is_rhand=side=='right',use_pca=False,flat_hand_mean=True)
+ def geometry(self,pose,betas,translation=None,keypoint_convention='freihand'):
+  pose=torch.as_tensor(np.asarray(pose).copy(),dtype=torch.float32).reshape(-1,48);betas=torch.as_tensor(np.asarray(betas).copy(),dtype=torch.float32).reshape(-1,10)
+  with torch.no_grad():o=self.layer(global_orient=pose[:,:3],hand_pose=pose[:,3:],betas=betas)
+  v=o.vertices.numpy();j=o.joints.numpy()[:,:16]
+  if translation is not None:
+   t=np.asarray(translation).reshape(-1,1,3);v=v+t;j=j+t
+  # Standard wrist + thumb,index,middle,ring,pinky proximal-to-distal order.
+  coords=[]
+  for k in [0,13,14,15,1,2,3,4,5,6,10,11,12,7,8,9]:coords.append(j[:,k])
+  out=np.stack([j[:,0],j[:,13],j[:,14],j[:,15],v[:,744],j[:,1],j[:,2],j[:,3],v[:,320],j[:,4],j[:,5],j[:,6],v[:,443],j[:,10],j[:,11],j[:,12],v[:,555],j[:,7],j[:,8],j[:,9],v[:,672]],axis=1)
+  if keypoint_convention=='manopth':out[:,[4,8,12,16,20]]=v[:,[745,317,444 if self.side=='right' else 445,556,673]]
+  return v,out,j
+
+def freihand_normalize():
+ base=ROOT/'FreiHAND/pilot-v0';model=Mano('right');reports=[]
+ for meta_file in sorted((base/'data').rglob('metainfo.json')):
+  p=meta_file.parent;m=load(meta_file);theta=torch.load(p/'annotation/mano/source_theta.pt',weights_only=True).numpy().reshape(1,61)
+  K=torch.load(p/'camera/intrinsic.pt',weights_only=True).numpy();focal=(K[0,0]+K[1,1])/2
+  root=np.r_[(theta[0,58:60]-K[:2,2]),focal]/theta[0,60]
+  pose=theta[:,:48].copy();pose[:,3:]+=model.data['hands_mean']
+  _,_,j=model.geometry(pose,theta[:,48:58]);translation=root-j[0,4]
+  motion=np.full((1,2,61),np.nan,np.float32);motion[0,1]=np.r_[pose[0],theta[0,48:58],translation]
+  tensor(p/'annotation/mano/motion.pt',motion);tensor(p/'annotation/mano/valid_flag.pt',[[False,True]])
+  _,keypoints,_=model.geometry(pose,theta[:,48:58],translation)
+  assert np.allclose(keypoints[0,9],root,atol=1e-6)
+  m['mano']={'layout':['pose_axis_angle48','betas10','translation3'],'slots':['left','right'],'coordinate_frame':'camera','translation_unit':'m','flat_hand_mean':True,'model_version':'MANO v1.2','conversion':'source non-flat mean added to finger pose; official orthographic root recovery minus posed MANO joint 4 (21-joint index9)','valid_flag':'right-hand source annotation available; left not supplied'}
+  m['missing_modalities'].pop('normalized_mano',None);save(meta_file,m)
+  reports.append({'path':str(p.relative_to(base)),'root_recovery_error_m':float(np.linalg.norm(keypoints[0,9]-root))})
+ save(base/'doc/mano_validation.json',{'status':'passed_algebraic_root_check','model_sha256':sha(WORK/'tmp/pilot-models/MANO_RIGHT.pkl'),'samples':reports,'limit':'No independent 3D ground truth currently in input; algebraic check does not prove fit accuracy.'})
+ print('FreiHAND normalized',len(reports))
+
+def validate_mano(name):
+ base=ROOT/name/'pilot-v0';models=[Mano('left'),Mano('right')];summaries=[]
+ for mp in (base/'data').rglob('metainfo.json'):
+  p=mp.parent;motionp=p/'annotation/mano/motion.pt'
+  if not motionp.exists():continue
+  a=torch.load(motionp,weights_only=True).numpy();valid=torch.load(p/'annotation/mano/valid_flag.pt',weights_only=True).numpy();K=torch.load(p/'camera/intrinsic.pt',weights_only=True).numpy();pred=np.full((len(a),2,21,3),np.nan,np.float32)
+  for side,model in enumerate(models):
+   ix=np.flatnonzero(valid[:,side])
+   if len(ix):_,pred[ix,side],_=model.geometry(a[ix,side,:48],a[ix,side,48:58],a[ix,side,58:],keypoint_convention='manopth')
+  tensor(p/'annotation/mano/derived_keypoints.pt',pred)
+  uv=pred@K.T;uv=uv[...,:2]/uv[...,2:]
+  row={'sequence':str(p.relative_to(base)),'keypoint_convention':'official manopth, including handedness-specific fingertip vertices','finite_valid_geometry':bool(np.isfinite(pred[valid]).all())}
+  target=p/('annotation/keypoints/xyz.pt' if name=='H2O' else 'annotation/keypoints/uv.pt')
+  if target.exists():
+   gt=torch.load(target,weights_only=True).numpy();err=np.linalg.norm((pred if name=='H2O' else uv)-gt,axis=-1);good=np.isfinite(err)&valid[:,:,None]
+   row['mean_error']=float(err[good].mean());row['max_error']=float(err[good].max());row['unit']='m' if name=='H2O' else 'pixels'
+  summaries.append(row)
+  meta=load(p/'metainfo.json');meta['mano']['geometry_validation']='measured against source keypoints; see doc/mano_validation.json';meta['mano']['model_version']='MANO v1.2'
+  meta['mano']['derived_keypoints_convention']='official manopth 21 joints, including side-specific fingertip vertices'
+  save(p/'metainfo.json',meta)
+ save(base/'doc/mano_validation.json',{'status':'measured_not_auto_accepted','samples':summaries,'model_sha256':{s:sha(WORK/'tmp/pilot-models'/f'MANO_{s.upper()}.pkl') for s in ['left','right']}})
+ print(name,summaries)
+
+if __name__=='__main__':
+ freihand_normalize();validate_mano('H2O');validate_mano('HOI4D')
diff --git a/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/preprocess_fresh_full.py b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/preprocess_fresh_full.py
new file mode 100644
index 0000000000000000000000000000000000000000..0965c395d6db42cce5db4a5ff75e952c178cb576
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/preprocess_fresh_full.py
@@ -0,0 +1,301 @@
+"""Authorized ARCTIC/HOT3D production queue. One heavy converter, immutable raw.
+
+Checkpoints and logs live under dataset/tmp; distributable validation lives in doc.
+No upload or raw deletion. All output is independently checked before completion.
+"""
+from fresh_sequence_common import *
+from convert_fresh_sequence_pilots import arctic,hot3d,verify_hot3d_frame_order_and_metadata
+from review_fresh_sequence_pilots import validate_view,arr
+from apply_hot3d_qa import apply_qa
+from diagnose_arctic_review import official_object
+from datetime import datetime
+import os,ctypes,msvcrt,traceback,html,random
+
+JOB='full-preprocess-v1.0'
+CONTROL=WORK/'tmp'/JOB
+RESERVE=400_000_000_000
+SCRIPT=Path(__file__).resolve()
+TEST_IDS={'P0004','P0005','P0006','P0008','P0016','P0020'}
+torch.set_num_threads(4)
+
+def now():return datetime.now().astimezone().isoformat()
+
+def atomic(p,value):
+    p.parent.mkdir(parents=True,exist_ok=True);q=p.with_name(p.name+'.writing')
+    q.write_text(json.dumps(value,ensure_ascii=False,indent=2,allow_nan=False),encoding='utf-8');os.replace(q,p)
+
+def temporary(ds):return DATA_ROOT/ds/'tmp'/JOB
+
+PRODUCTION_SCRIPTS=['preprocess_fresh_full.py','convert_fresh_sequence_pilots.py','apply_hot3d_qa.py','review_fresh_sequence_pilots.py','render_full_sequence_review.py','fresh_sequence_common.py','diagnose_arctic_review.py','pilot_build.py','pilot_fetch.py','pilot_mano.py','sequence_pilot_common.py','sequence_pilot_world.py','pilot_hot_normalize.py','audit_source_fields.py']
+
+def code_manifest():
+    return [{'path':'script/'+f,'sha256':digest(WORK/'scripts'/f)} for f in PRODUCTION_SCRIPTS]
+
+def snapshot_scripts(ds):
+    root=DATA_ROOT/ds/RELEASE
+    for f in PRODUCTION_SCRIPTS:shutil.copyfile(WORK/'scripts'/f,root/'script'/f)
+    atomic(root/'doc/validation/full/script-manifest.json',{'created_at':now(),'scope':'Current production converter and validation scripts; earlier pilot manifests describe their historical run','files':code_manifest()})
+
+def inputs(ds):
+    base=DATA_ROOT/ds
+    name='archive_checksums.json' if ds=='ARCTIC' else 'file_checksums.json'
+    return load(base/'raw/doc'/name)
+
+def check_input(ds,rel,records):
+    p=DATA_ROOT/ds/'raw/data'/rel;r=records[rel];st=p.stat()
+    assert st.st_size==r['size'],f'Raw size changed: {p}'
+    assert r.get('official_sha256_match',r.get('official_digest_match',False))
+    if st.st_mtime_ns!=r['mtime_ns']:
+        assert digest(p)==r['sha256'],f'Raw digest changed: {p}'
+        if p.suffix=='.zip':
+            with zipfile.ZipFile(p) as z:assert z.testzip() is None
+    return {'file':'raw/data/'+rel,'size':st.st_size,'sha256':r['sha256'],'download_checksum_and_crc_record_reused':st.st_mtime_ns==r['mtime_ns']}
+
+def plan():
+    CONTROL.mkdir(parents=True,exist_ok=True);out={'created_at':now(),'order':['ARCTIC','HOT3D'],'datasets':{},'reserve_bytes':RESERVE}
+    for ds in out['order']:
+        base=DATA_ROOT/ds;records=inputs(ds)
+        for rel in records:check_input(ds,rel,records)
+        rows=[];auxiliary=[]
+        if ds=='ARCTIC':
+            with zipfile.ZipFile(base/'raw/data/splits_json.zip') as z:splits=json.loads(z.read('splits_json/protocol_p1.json'))
+            for p in sorted((base/'raw/data/images_zips').rglob('*.zip')):
+                seq=p.relative_to(base/'raw/data/images_zips').with_suffix('').as_posix()
+                if '/' not in seq:
+                    assert seq=='backgrounds',seq
+                    auxiliary.append({'file':'raw/data/images_zips/backgrounds.zip','handling':'retain original archive in raw; background reference photographs are not hand-object capture sequences'})
+                    continue
+                with zipfile.ZipFile(p) as z:
+                    im=[i for i in z.infolist() if re.fullmatch(r'[0-8]/[0-9]+\.jpg',i.filename)]
+                    extra=[i.filename for i in z.infolist() if i.filename.endswith('.jpg') and not re.fullmatch(r'[0-8]/[0-9]+\.jpg',i.filename)]
+                    assert all(n.startswith('__MACOSX/') or Path(n).name.startswith('._') for n in extra),(p,extra[:5])
+                    cams=sorted({i.filename.split('/')[0] for i in im},key=int)
+                    counts={c:sum(i.filename.startswith(c+'/') for i in im) for c in cams}
+                    b=sum(i.file_size for i in im)
+                assert cams,(seq,counts)
+                rows.append({'sequence':seq,'sequence_id':seq.replace('/','_'),'object':seq.split('/')[1].split('_')[0],
+                    'cameras':cams,'frames_per_camera':counts,'jpeg_bytes':b,'split':next((k for k,v in splits.items() if seq in v),'unknown'),
+                    'inputs':['raw_seqs.zip','meta.zip','splits_json.zip','images_zips/'+seq+'.zip']})
+            budget=sum(r['jpeg_bytes'] for r in rows)+160_000_000_000
+        else:
+            for p in sorted((base/'raw/data/Aria').iterdir()):
+                if not p.is_dir():continue
+                gt=next(p.glob('*ground_truth.zip'));hz=next(p.glob('*hand_data.zip'));vrs=next(p.glob('*.vrs'))
+                with zipfile.ZipFile(gt) as z:
+                    m=json.loads(z.read('metadata.json'));cams=json.loads(z.read('camera_models.json'))
+                    assert m['headset']=='Aria'
+                    hasgt=m['have_hand_object_pose_gt'];split='test' if m['participant_id'] in TEST_IDS else 'train';assert hasgt==(split=='train')
+                    if hasgt:assert 'masks/mask_qa_pass.csv' in z.namelist()
+                with zipfile.ZipFile(hz) as z:mano_bytes=z.getinfo('mano_hand_pose_trajectory.jsonl').file_size
+                assert hasgt or mano_bytes==0
+                rows.append({'sequence':p.name,'sequence_id':p.name,'cameras':[c['stream_id'] for c in cams],'split':split,
+                    'source_gt_available':hasgt,'vrs_bytes':vrs.stat().st_size,'inputs':[q.relative_to(base/'raw/data').as_posix() for q in [gt,hz,vrs]]})
+            # Includes persistent extracted JPEGs, current media spool, meshes, tensors and reports.
+            budget=sum(r['vrs_bytes'] for r in rows)*3+150_000_000_000
+        assert len(rows)==(339 if ds=='ARCTIC' else 198)
+        out['datasets'][ds]={'sequences':rows,'count':len(rows),'views':sum(len(r['cameras']) for r in rows),'additional_budget_bytes':budget,'raw_files_verified':len(records),'auxiliary_raw_only':auxiliary}
+        print('PLAN',ds,len(rows),out['datasets'][ds]['views'],budget,flush=True)
+    out['free_bytes_at_plan']=shutil.disk_usage(DATA_ROOT).free
+    out['additional_peak_budget_bytes']=sum(v['additional_budget_bytes'] for v in out['datasets'].values())
+    assert out['free_bytes_at_plan']-out['additional_peak_budget_bytes']>=RESERVE,out
+    atomic(CONTROL/'plan.json',out)
+    return out
+
+def approve():
+    p=load(CONTROL/'plan.json');confirmation={'confirmed_at':now(),'user_statement':'HOT3D和ARCTIC我都确认了，可以开始做预处理了。',
+        'scope':'Reviewed pilots accepted; full preprocessing authorized for ARCTIC and HOT3D Aria only',
+        'full_preprocessing_authorized':True,'full_output_review':'pending','upload_authorized':False,'raw_retention':True}
+    atomic(CONTROL/'authorization.json',confirmation)
+    state=load(WORK/'tmp/dataset-work-state.json')
+    for ds in p['order']:
+        root=DATA_ROOT/ds/RELEASE;doc=root/'doc/validation'
+        atomic(doc/'pilot-human-confirmation.json',confirmation)
+        for rp in [doc/'report.json',*sorted((doc/'objects').glob('*/report.json')),*([doc/'objects/report.json'] if (doc/'objects/report.json').exists() else [])]:
+            r=load(rp);r.update(human_review='approved',full_preprocessing_authorized=True,upload_authorized=False,pilot_confirmation=confirmation)
+            atomic(rp,r)
+        for page in [doc/'index.html',*sorted((doc/'objects').rglob('index.html'))]:
+            if not page.exists():continue
+            s=page.read_text(encoding='utf-8')
+            s=s.replace('人工确认待完成','试点人工确认已通过；全量产物待单独验收').replace('用户人工确认待完成','用户试点确认已通过').replace('用户确认前不全量处理','用户已确认试点，允许全量预处理')
+            if 'id="production-authorization"' not in s:
+                s=s.replace('<header>','<header><p id="production-authorization" style="color:#176b3a"><strong>用户已确认本页试点，授权全量预处理。</strong> 全量完成后另行随机抽查；尚未授权上传。</p>',1)
+            page.write_text(s,encoding='utf-8')
+        if (doc/'objects/artifact-manifest.json').exists():
+            mp=doc/'objects/artifact-manifest.json';a=load(mp)
+            for f in a['files']:
+                file=mp.parent/f['path'];f.update(size=file.stat().st_size,sha256=digest(file))
+            atomic(mp,a)
+        for meta in (root/'data').glob('*/*/metainfo.json'):
+            m=load(meta);m['pilot_human_review']='approved';m['full_preprocessing_authorized']=True
+            if 'acceptance' in m:m['acceptance'].update(status='pilot_user_approved',training_release_approved=False,reason='Pilot confirmed; full outputs and release still require separate review')
+            atomic(meta,m)
+        state['active'][ds].update(stage='full_preprocessing_queued',human_sample_review='approved',full_preprocessing_authorized=True,upload_authorized=False,
+            production_plan=str(CONTROL/'plan.json'),full_sequence_count=p['datasets'][ds]['count'],full_view_count=p['datasets'][ds]['views'])
+        atomic(doc/'status.json',{'stage':'full_preprocessing_queued','pilot_review':'approved','full_output_review':'pending','upload_authorized':False})
+        docprod=doc/'full';docprod.mkdir(parents=True,exist_ok=True)
+        atomic(docprod/'plan.json',p['datasets'][ds])
+        (docprod/'README.md').write_text('# 全量预处理\n\n试点已获用户确认。逐序列转换、逐帧码流/像素和几何校验，完成后另建随机抽查报告。原始包保留，不自动上传。\n\n每个序列的正式自动检查在本目录sequences/；运行日志和进度只在数据集tmp。断点以输入官方摘要及全部输出文件摘要核验。\n\nHOT3D测试集未发布手物姿态真值：保留RGB、相机和物体资产，姿态及可见性缺省。ARCTIC仅保留已确认s09/laptop的曝光排除，其它序列不新增帧号屏蔽。\n',encoding='utf-8')
+        for f in ['preprocess_fresh_full.py','convert_fresh_sequence_pilots.py','apply_hot3d_qa.py','review_fresh_sequence_pilots.py','render_full_sequence_review.py','fresh_sequence_common.py','diagnose_arctic_review.py']:
+            shutil.copyfile(WORK/'scripts'/f,root/'script'/f)
+    atomic(WORK/'tmp/dataset-work-state.json',state)
+
+def official_arctic_check(row,views):
+    base=DATA_ROOT/'ARCTIC';seq=row['sequence'];oid=row['object'];sid=seq.split('/')[0]
+    if row['split']=='test':
+        with zipfile.ZipFile(base/'raw/data/raw_seqs.zip') as z:
+            assert all(f'raw_seqs/{seq}.{x}.npy' not in z.namelist() for x in ['mano','object','egocam.dist'])
+        for p in views:
+            m=load(p/'metainfo.json');assert 'mano' not in m['components'] and not list((p/'annotation/object').glob('*/pose.pt'))
+            if p.name=='0':assert 'intrinsic' not in m['components'] and 'extrinsic' not in m['components'] and m['camera']['schema_id'] is None
+        return {'passed':True,'test':'official_held_out_test_ground_truth_omitted; static_mesh_metres_and_part_ids_checked_per_view','source_gt_available':False}
+    with zipfile.ZipFile(base/'raw/data/meta.zip') as z:
+        pre=f'meta/object_vtemplates/{oid}/';verts=np.array([[float(x) for x in s.split()[1:4]] for s in z.read(pre+'mesh.obj').decode().splitlines() if s.startswith('v ')])/1000
+        parts=np.asarray(json.loads(z.read(pre+'parts.json')),np.int64);params=json.loads(z.read(pre+'object_params.json'));offset=json.loads(z.read('meta/misc.json'))[sid]['ioi_offset']
+    with zipfile.ZipFile(base/'raw/data/raw_seqs.zip') as z:objects=native_npy_bytes(z.read(f'raw_seqs/{seq}.object.npy'))
+    world=[];kp=[]
+    for start in range(0,len(objects),32):
+        v,k=official_object(objects[start:start+32],verts,parts,params,oid);world.append(v);kp.append(k)
+    world=np.concatenate(world);kp=np.concatenate(kp);worst=0.;checked=0
+    for p in views:
+        frames=[json.loads(l) for l in (p/'camera/frames.jsonl').read_text().splitlines()];ix=np.array([int(f['source_frame_id'])-offset for f in frames]);op=p/'annotation/object'/oid
+        E=arr(p/'camera/extrinsic.pt');P=arr(op/'pose.pt');top=arr(op/'parts/top/pose.pt');valid=arr(op/'pose_valid.pt');ids=arr(op/'vertex_part_ids.pt');points=arr(op/'keypoints/xyz.pt')
+        assert np.array_equal(ids,1-parts)
+        for i in np.flatnonzero(valid):
+            actual=verts@P[i,:3,:3].T+P[i,:3,3];sel=ids==1;actual[sel]=verts[sel]@top[i,:3,:3].T+top[i,:3,3]
+            expected=world[ix[i]]@E[i,:3,:3].T+E[i,:3,3];expected_kp=kp[ix[i]]@E[i,:3,:3].T+E[i,:3,3]
+            worst=max(worst,float(np.linalg.norm(actual-expected,axis=-1).max()),float(np.linalg.norm(points[i]-expected_kp,axis=-1).max()));checked+=1
+    assert worst<2e-5,(seq,worst)
+    return {'test':'all_valid_frames_all_vertices_and_keypoints_vs_pinned_official_ObjectTensors','passed':True,'camera_frames':checked,'vertices_per_frame':len(verts),'max_error_m':worst}
+
+def worker(ds,seq):
+    plan=load(CONTROL/'plan.json');row=next(r for r in plan['datasets'][ds]['sequences'] if r['sequence']==seq)
+    assert load(CONTROL/'authorization.json')['full_preprocessing_authorized']
+    base=DATA_ROOT/ds;root=base/RELEASE;tmp=temporary(ds);result=tmp/'completed'/f'{row["sequence_id"]}.json';records=inputs(ds)
+    source_files=list(row['inputs'])
+    if ds=='HOT3D':source_files+=sorted(rel for rel in records if rel.startswith('assets/'))
+    source=[check_input(ds,rel,records) for rel in source_files]
+    if result.exists() and load(result).get('scripts'):
+        assert load(result)['scripts']==code_manifest(),'Converter code changed since checkpoint; explicit revalidation is required'
+    if result.exists() and load(result).get('scripts')==code_manifest():
+        done=load(result)
+        assert done['source']==source
+        for f in done['files']:
+            p=root/f['path'];assert p.stat().st_size==f['size'] and digest(p)==f['sha256'],p
+        print('VERIFIED_RESUME',ds,seq,flush=True);return
+    assert shutil.disk_usage(base).free>RESERVE+15_000_000_000,'Disk reserve reached'
+    reused={c for c in row['cameras'] if (root/'data'/row['sequence_id']/c/'metainfo.json').exists() and load(root/'data'/row['sequence_id']/c/'metainfo.json').get('pilot_complete')}
+    if ds=='ARCTIC':arctic(seq,initialize=False)
+    else:
+        hot3d(seq,initialize=False);verify_hot3d_frame_order_and_metadata(seq)
+        if row['source_gt_available']:apply_qa(seq,publish=False)
+    views=[root/'data'/row['sequence_id']/cam for cam in row['cameras']]
+    results=[]
+    for p in views:
+        m,frames,r=validate_view(p);m.update(pilot=False,production_complete=True,pilot_human_review='approved',full_output_human_review='pending',upload_authorized=False)
+        if p.name in reused:
+            media=next(x for x in r['conversion_checks'] if x['test']=='all_source_JPEG_packets_and_RGB_pixels_exact')
+            for copy,key in [(True,'packet_manifest_sha256'),(False,'pixel_manifest_sha256')]:
+                actual=framehash(['-i',str(p/'observation/rgb.mkv')],copy=copy)
+                assert len(actual)==m['num_frames'] and hashlib.sha256(json.dumps(actual).encode()).hexdigest()==media[key],(p,'reused media changed',key)
+            r['reused_media_revalidated']=True
+        if ds=='HOT3D' and not row['source_gt_available']:
+            assert 'mano' not in m['components'] and not list((p/'annotation/object').glob('*/pose.pt'))
+        atomic(p/'metainfo.json',m);results.append(r)
+    independent=official_arctic_check(row,views) if ds=='ARCTIC' else {'passed':True,'test':'all_frame_camera_MANO_geometry_PCA_and_source_QA_per_view' if row['source_gt_available'] else 'all_frame_media_camera_geometry_and_omitted_pose_components','source_gt_available':row['source_gt_available']}
+    files=[]
+    for p in sorted((root/'data'/row['sequence_id']).rglob('*')):
+        if p.is_file():files.append({'path':p.relative_to(root).as_posix(),'size':p.stat().st_size,'sha256':digest(p)})
+    done={'dataset':ds,'sequence':seq,'sequence_id':row['sequence_id'],'completed_at':now(),'source':source,'scripts':code_manifest(),'views':results,'independent_geometry':independent,
+        'files':files,'bytes':sum(f['size'] for f in files),'frames':sum(v['frames'] for v in results),'full_output_human_review':'pending','upload_authorized':False}
+    atomic(root/'doc/validation/full/sequences'/f'{row["sequence_id"]}.json',done);atomic(result,done)
+    print('SEQUENCE_COMPLETE',ds,seq,done['frames'],done['bytes'],flush=True)
+
+def random_review(ds,rows):
+    from render_full_sequence_review import render_view,asset_preview
+    doc=DATA_ROOT/ds/RELEASE/'doc/validation/full_random';doc.mkdir(parents=True,exist_ok=True);rng=random.Random(20260920);selected=[]
+    if ds=='ARCTIC':
+        for oid in sorted({r['object'] for r in rows}):
+            group=[r for r in rows if r['object']==oid and r['split']!='test'];r=rng.choice(group);selected.append((r,rng.choice(r['cameras']),'labelled_object:'+oid,len(group)))
+        if not any(c=='0' for _,c,_,_ in selected):
+            group=[r for r in rows if '0' in r['cameras'] and r['split']!='test'];r=rng.choice(group);selected.append((r,'0','labelled_ego_camera_coverage',len(group)))
+        for split in ['train','val','test']:
+            if not any(r['split']==split for r,_,_,_ in selected):
+                group=[r for r in rows if r['split']==split];r=rng.choice(group);selected.append((r,'0' if split=='test' else rng.choice(r['cameras']),'split_coverage:'+split,len(group)))
+    else:
+        for split in ['train','test']:
+            group=[r for r in rows if r['split']==split];r=rng.choice(group)
+            for cam in r['cameras']:selected.append((r,cam,'split:'+split,len(group)))
+    sampling={'seed':20260920,'method':'ARCTIC: equal allocation by object among labelled train/val, uniform sequence then available camera; add labelled ego and missing train/val/test strata. Held-out test uses ego to show missing calibration/pose. HOT3D: equal allocation train/test, uniform whole sequence per split, every camera. Not frame-weighted; whole selected sequence is rendered.',
+        'human_review':'pending','selections':[{'sequence_id':r['sequence_id'],'camera_id':c,'stratum':s,'stratum_sequences':n,'frame_indices':'all'} for r,c,s,n in selected]}
+    atomic(doc/'sampling.json',sampling);cards=[];results=[];assets={}
+    for row,cam,_,_ in selected:
+        st={'stage':'generating_full_random_review','dataset':ds,'sequence':row['sequence'],'camera_id':cam,'pid':os.getpid(),'time':now(),'completed_previews':len(results),'total_previews':len(selected)}
+        atomic(temporary(ds)/'status.json',st);atomic(CONTROL/'status.json',st)
+        p=DATA_ROOT/ds/RELEASE/'data'/row['sequence_id']/cam;sub=doc/row['sequence_id'];tmp=temporary(ds)/'random-review'/row['sequence_id'];qa=None
+        qfile=DATA_ROOT/ds/'tmp'/RUN/'qa'/f'{row["sequence_id"]}.json'
+        if ds=='HOT3D' and row['source_gt_available']:
+            if not qfile.exists():apply_qa(row['sequence'],publish=False)
+            qa=load(qfile)
+        result=render_view(p,ds,qa,doc=sub,tmp=tmp);results.append(result)
+        for op in (p/'annotation/object').glob('*'):
+            out=sub/'assets'/f'{op.name}.mp4'
+            if not out.exists():asset_preview(op,out)
+            assets[out.relative_to(doc).as_posix()]=row['sequence_id']+' / '+op.name
+        cards.append(f'<section><h2>{row["sequence_id"]} · camera {cam} · {result["frames"]}帧</h2><video controls preload="metadata" src="{row["sequence_id"]}/{result["video_ref"]}"></video></section>')
+    for ref,label in assets.items():cards.append(f'<section><h2>物体资产 · {label}</h2><video controls preload="metadata" src="{ref}"></video></section>')
+    atomic(doc/'report.json',{'dataset':ds,'sampling':sampling,'views':results,'asset_previews':assets,'human_review':'pending','upload_authorized':False})
+    (doc/'index.html').write_text('<!doctype html><meta charset="utf-8"><title>'+ds+' 全量随机抽查</title><style>body{font:16px/1.7 system-ui;max-width:1536px;margin:25px auto;background:#f3f5f9}section,header{background:white;padding:20px;margin:20px}video{width:100%}</style><header><h1>'+ds+' 全量随机抽查</h1><p>从全部完成产物读取，选中序列全部帧已渲染；用户人工确认待完成，不自动上传。</p><a href="sampling.json">随机种子与抽样范围/权重</a> · <a href="report.json">逐模态检查</a></header>'+''.join(cards),encoding='utf-8')
+    (doc/'REPORT.md').write_text('# 全量随机抽查\n\n[离线完整序列](index.html) · [抽样定义](sampling.json) · [模态检查](report.json)\n\n人工确认待完成。源缺失模态缺省，不以数值通过替代视觉质量确认。\n',encoding='utf-8')
+
+def run():
+    CONTROL.mkdir(parents=True,exist_ok=True);guard=(CONTROL/'process.lock').open('a+b');guard.seek(0)
+    try:msvcrt.locking(guard.fileno(),msvcrt.LK_NBLCK,1)
+    except OSError:raise RuntimeError('Another production queue owns the lock')
+    assert load(CONTROL/'authorization.json')['full_preprocessing_authorized'];p=load(CONTROL/'plan.json')
+    power=ctypes.windll.kernel32.SetThreadExecutionState(0x80000001)
+    try:
+        for ds in p['order']:
+            snapshot_scripts(ds)
+            rows=p['datasets'][ds]['sequences'];tmp=temporary(ds);tmp.mkdir(parents=True,exist_ok=True)
+            # Smaller full sequences first; sequence boundaries remain unchanged.
+            rows=sorted(rows,key=lambda r:r.get('jpeg_bytes',r.get('vrs_bytes',0)))
+            for n,row in enumerate(rows):
+                st={'stage':'converting','dataset':ds,'sequence':row['sequence'],'completed_sequences':n,'total_sequences':len(rows),'pid':os.getpid(),'time':now()}
+                atomic(tmp/'status.json',st);atomic(CONTROL/'status.json',st)
+                registry=load(WORK/'tmp/dataset-work-state.json');registry['active'][ds].update(stage='full_preprocessing',production_status=str(tmp/'status.json'),production_process_id=os.getpid());atomic(WORK/'tmp/dataset-work-state.json',registry)
+                assert shutil.disk_usage(DATA_ROOT).free>RESERVE+15_000_000_000,'Disk reserve reached'
+                log=tmp/'logs'/f'{row["sequence_id"]}.log';log.parent.mkdir(parents=True,exist_ok=True)
+                with log.open('ab') as out:
+                    child=subprocess.Popen([sys.executable,'-u',str(SCRIPT),'worker','--dataset',ds,'--sequence',row['sequence']],stdout=out,stderr=subprocess.STDOUT)
+                    while child.poll() is None:
+                        st.update(worker_pid=child.pid,time=now(),log=str(log));atomic(tmp/'status.json',st);atomic(CONTROL/'status.json',st)
+                        if shutil.disk_usage(DATA_ROOT).free<RESERVE:child.terminate();child.wait();raise RuntimeError('Disk reserve reached; active sequence stopped, raw retained')
+                        time.sleep(20)
+                    if child.returncode:raise RuntimeError(f'Worker failed: {ds} {row["sequence"]}; see {log}')
+                print('QUEUE_PROGRESS',ds,n+1,len(rows),row['sequence'],flush=True)
+            st={'stage':'generating_full_random_review','dataset':ds,'pid':os.getpid(),'time':now(),'completed_sequences':len(rows),'total_sequences':len(rows)}
+            atomic(tmp/'status.json',st);atomic(CONTROL/'status.json',st)
+            state=load(WORK/'tmp/dataset-work-state.json');state['active'][ds].update(stage='generating_full_random_review');atomic(WORK/'tmp/dataset-work-state.json',state)
+            random_review(ds,rows)
+            state=load(WORK/'tmp/dataset-work-state.json');state['active'][ds].update(stage='full_random_awaiting_human_review',full_output_human_review='pending',upload_authorized=False);atomic(WORK/'tmp/dataset-work-state.json',state)
+            atomic(tmp/'status.json',{'stage':'full_random_awaiting_human_review','time':now(),'completed_sequences':len(rows),'total_sequences':len(rows),'report':str(DATA_ROOT/ds/RELEASE/'doc/validation/full_random/index.html')})
+        atomic(CONTROL/'status.json',{'stage':'all_preprocessing_complete_awaiting_full_random_review','time':now(),'upload_authorized':False})
+    except BaseException as e:
+        error={'stage':'failed_needs_attention','time':now(),'error':str(e),'dataset':locals().get('ds'),'sequence':locals().get('row',{}).get('sequence')}
+        atomic(CONTROL/'status.json',error)
+        if 'ds' in locals():atomic(temporary(ds)/'status.json',error)
+        raise
+    finally:ctypes.windll.kernel32.SetThreadExecutionState(0x80000000);guard.close()
+
+if __name__=='__main__':
+    ap=argparse.ArgumentParser();ap.add_argument('mode',choices=['plan','approve','worker','run']);ap.add_argument('--dataset',choices=['ARCTIC','HOT3D']);ap.add_argument('--sequence');a=ap.parse_args()
+    if a.mode=='worker':
+        CONTROL.mkdir(parents=True,exist_ok=True)
+        with (CONTROL/'worker.lock').open('a+b') as lock:
+            lock.seek(0)
+            try:msvcrt.locking(lock.fileno(),msvcrt.LK_NBLCK,1)
+            except OSError:raise RuntimeError('A production worker is already active; refusing overlapping conversion')
+            worker(a.dataset,a.sequence)
+    elif a.mode=='plan':plan()
+    elif a.mode=='approve':approve()
+    else:run()
diff --git a/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/render_full_sequence_review.py b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/render_full_sequence_review.py
new file mode 100644
index 0000000000000000000000000000000000000000..a29dc18687e4ebe26f4db71a514ae4c34220e0e6
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/render_full_sequence_review.py
@@ -0,0 +1,256 @@
+"""Full-frame, all-camera diagnostic videos read from actual shared tensors/media.
+
+Preview MP4 is explicitly display-only; warehouse MKV packets remain untouched.
+"""
+from fresh_sequence_common import *
+from review_fresh_sequence_pilots import arr, validate_view, project_for_view
+from datetime import datetime
+import html
+
+VIDEO_RUN='full-video-review-v2'
+CELL=512
+WIDTH=3*CELL
+HEIGHT=800
+COLORS=[(110,255,100),(255,80,210),(240,220,60),(90,180,255),(190,160,255),(255,120,140)]
+
+def text(im,txt,xy,color=(235,235,235),size=.5):cv2.putText(im,txt,xy,cv2.FONT_HERSHEY_SIMPLEX,size,color,1,cv2.LINE_AA)
+
+def fast_edges(im,uv,edges,color):
+    xy=uv[edges];good=np.isfinite(xy).all((1,2)) & (np.max(np.nan_to_num(np.abs(xy),nan=CELL*20),axis=(1,2))<CELL*12)
+    xy=xy[good];xy=xy[np.linalg.norm(xy[:,0]-xy[:,1],axis=1)<CELL*.7]
+    if len(xy):cv2.polylines(im,np.rint(xy).astype(np.int32),False,color,1,cv2.LINE_AA)
+
+def dots(im,uv,color,radius=2):
+    good=np.isfinite(uv).all(1)&(uv[:,0]>=0)&(uv[:,0]<im.shape[1])&(uv[:,1]>=0)&(uv[:,1]<im.shape[0])
+    xy=np.rint(uv[good]).astype(int);xy[:,0]=np.clip(xy[:,0],0,im.shape[1]-1);xy[:,1]=np.clip(xy[:,1],0,im.shape[0]-1)
+    for d in range(max(1,radius)):
+        im[np.clip(xy[:,1]+d,0,im.shape[0]-1),xy[:,0]]=color
+        im[xy[:,1],np.clip(xy[:,0]+d,0,im.shape[1]-1)]=color
+
+def model_vertices(motion,valid,side,fix):
+    model=Mano(['left','right'][side])
+    if side==0 and fix:
+        model.data['shapedirs'][:,0]*=-1
+        with torch.no_grad():model.layer.shapedirs[:,0]*=-1
+    verts=np.full((len(motion),778,3),np.nan,np.float32);joints=np.full((len(motion),16,3),np.nan,np.float32)
+    ids=np.flatnonzero(valid[:,side])
+    for start in range(0,len(ids),64):
+        ix=ids[start:start+64];a=motion[ix,side];verts[ix],_,joints[ix]=model.geometry(a[:,:48],a[:,48:58],a[:,58:])
+    faces=np.asarray(model.layer.faces);edges=np.unique(np.sort(np.concatenate([faces[:,[0,1]],faces[:,[1,2]],faces[:,[2,0]]]),axis=1),axis=0)
+    return verts,joints,edges
+
+def encoder(path,fps,log,size=(WIDTH,HEIGHT)):
+    path.parent.mkdir(parents=True,exist_ok=True)
+    return subprocess.Popen([str(FF),'-v','error','-y','-f','rawvideo','-pix_fmt','rgb24','-s',f'{size[0]}x{size[1]}','-r',str(fps),'-i','pipe:0',
+        '-an','-c:v','libx264','-preset','veryfast','-crf','19','-pix_fmt','yuv420p','-threads','2','-bf','0','-g','60','-movflags','+faststart',str(path)],stdin=subprocess.PIPE,stderr=log)
+
+def read_exact(pipe,n):
+    chunks=[];count=0
+    while count<n:
+        b=pipe.read(n-count)
+        if not b:break
+        chunks.append(b);count+=len(b)
+    return b''.join(chunks)
+
+def trajectory_plot(E,width=265,height=145):
+    good=np.isfinite(E).all((1,2));xyz=np.full((len(E),3),np.nan)
+    xyz[good]=-(np.swapaxes(E[good,:3,:3],1,2)@E[good,:3,3,None])[:,:,0]
+    base=np.full((height,width,3),24,np.uint8);xy=xyz[:,[0,2]];valid=np.isfinite(xy).all(1)
+    if not valid.any():return base,np.zeros((len(E),2),int),xyz
+    lo=np.nanmin(xy,0);hi=np.nanmax(xy,0);span=np.maximum(hi-lo,.05);scale=min((width-24)/span[0],(height-35)/span[1]);pix=(xy-lo)*scale+12
+    pix[:,1]=height-12-pix[:,1];pixels=np.nan_to_num(pix).astype(int)
+    for i in range(1,len(E)):
+        if valid[i-1] and valid[i]:cv2.line(base,tuple(pixels[i-1]),tuple(pixels[i]),(110,110,110),1)
+    text(base,'camera world trajectory X/Z (m)',(7,15),size=.38)
+    return base,pixels,xyz
+
+def object_specs(p,source):
+    result=[]
+    for k,op in enumerate(sorted((p/'annotation/object').glob('*'))):
+        if not (op/'pose.pt').exists():continue
+        om=load(op/'metainfo.json');rel=op.relative_to(p);mesh=trimesh.load(p/om['mesh_ref'],process=False,force='mesh')
+        rawpose=arr(source/rel/'pose.pt');rawvalid=arr(source/rel/'pose_valid.pt');valid=arr(op/'pose_valid.pt')
+        spec={'id':om['object_id'],'name':om.get('category_name',om['object_id']),'color':COLORS[k%len(COLORS)],'P':rawpose,'rawvalid':rawvalid,'valid':valid,'op':op}
+        if 'vertex_part_ids_ref' in om:
+            xyz=np.array([[float(x) for x in r.split()[1:4]] for r in (p/om['mesh_ref']).read_text().splitlines() if r.startswith('v ')])
+            part=arr(p/om['vertex_part_ids_ref']);idx=np.arange(0,len(xyz),max(1,len(xyz)//4500))
+            spec.update(vertices=xyz[idx],parts=part[idx],top=arr(source/rel/'parts/top/pose.pt'),edges=None,
+                keypoints=arr(source/rel/'keypoints/xyz.pt'),joint=arr(source/rel/'joint_state.pt')[:,0])
+        else:
+            edges=np.asarray(mesh.edges_unique);edges=edges[::max(1,len(edges)//1100)]
+            unique,inv=np.unique(edges,return_inverse=True);spec.update(vertices=np.asarray(mesh.vertices)[unique],edges=inv.reshape(-1,2),parts=None)
+        spec['visibility']=arr(op/'visibility.pt') if (op/'visibility.pt').exists() else None
+        spec['raw_visibility']=arr(source/rel/'visibility.pt') if (source/rel/'visibility.pt').exists() else None
+        result.append(spec)
+    return result
+
+def render_view(p,dataset,qa_data,doc=None,tmp=None,source=None,snapshot_indices=None):
+    m,frames,structural=validate_view(p);cam=m['camera_id'];T=m['num_frames'];root=DATA_ROOT/dataset/RELEASE
+    doc=doc or root/'doc/validation';tmp=tmp or DATA_ROOT/dataset/'tmp'/VIDEO_RUN;tmp.mkdir(parents=True,exist_ok=True)
+    if source is None:
+        source=p
+        for candidate in [DATA_ROOT/dataset/'tmp'/VIDEO_RUN/'diagnostic'/m['sequence_id']/cam,DATA_ROOT/dataset/'tmp'/VIDEO_RUN/'diagnostic'/cam]:
+            if (candidate/'snapshot.json').exists() and Path(load(candidate/'snapshot.json')['source_view']).resolve()==p.resolve():source=candidate;break
+    qa=np.array(qa_data[cam]['qa_pass'],bool) if dataset=='HOT3D' and qa_data and qa_data[cam].get('qa_pass') is not None else None
+    if 'mano' in m['components']:
+        motion=arr(source/'annotation/mano/motion.pt');present=arr(source/'annotation/mano/motion_valid.pt');valid=arr(p/'annotation/mano/motion_valid.pt')
+    else:motion=np.full((T,2,61),np.nan);present=np.zeros((T,2),bool);valid=present.copy()
+    visibility=arr(p/'annotation/mano/visibility.pt') if (p/'annotation/mano/visibility.pt').exists() else None
+    rawvis=arr(source/'annotation/mano/visibility.pt') if (source/'annotation/mano/visibility.pt').exists() else None
+    hands=[model_vertices(motion,present,s,m['components'].get('mano',{}).get('model_profile',{}).get('left_shapedirs_fix',False)) for s in range(2)]
+    objs=object_specs(p,source);K=arr(p/'camera/intrinsic.pt') if (p/'camera/intrinsic.pt').exists() else np.full((T,3,3),np.nan);D=arr(p/'camera/distortion.pt') if (p/'camera/distortion.pt').exists() else None;E=arr(p/'camera/extrinsic.pt') if (p/'camera/extrinsic.pt').exists() else np.full((T,4,4),np.nan)
+    W,H=m['camera']['image_size'];scale0=min(CELL/W,CELL/H);w=int(W*scale0)//2*2;h=int(H*scale0)//2*2;scale=np.array([w/W,h/H])
+    fps=m['sampling']['source_fps'] or 30;path=doc/'videos'/f'{cam}-full.mp4';logpath=tmp/f'render-{cam}.log';log=logpath.open('wb')
+    dec=subprocess.Popen([str(FF),'-v','error','-threads','2','-noautorotate','-i',str(p/'observation/rgb.mkv'),'-vf',f'scale={w}:{h}','-pix_fmt','rgb24','-fps_mode','passthrough','-f','rawvideo','pipe:1'],stdout=subprocess.PIPE,stderr=log)
+    enc=encoder(path,fps,log);traj,tp,world=trajectory_plot(E);rendered=0;exposure=[]
+    snapshots=sorted(set(snapshot_indices or [0,1,26,27,179,180,T//4,T//2,3*T//4,T-1]))
+    timeline=np.zeros((12,WIDTH,3),np.uint8)
+    for x in range(WIDTH):timeline[:,x]=(40,135,65) if qa is not None and qa[min(T-1,x*T//WIDTH)] else ((170,42,50) if qa is not None else (75,80,95))
+    for i in range(T):
+        b=read_exact(dec.stdout,w*h*3);assert len(b)==w*h*3,(cam,i,'decoder ended early')
+        original=np.frombuffer(b,np.uint8).reshape(h,w,3).copy();raw=original.copy();usable=original.copy()
+        gray=cv2.cvtColor(cv2.resize(original,(96,96)),cv2.COLOR_RGB2GRAY)
+        exposure.append({'frame_index':i,'p95_luma':float(np.percentile(gray,95)),'median_luma':float(np.median(gray))})
+        params=K[i] if m['camera']['model']=='fisheye624' else (K[i],D[i] if D is not None else None)
+        for side,(verts,joints,edges) in enumerate(hands):
+            if not present[i,side]:continue
+            uv=project_for_view(verts[i],m,params,scale);color=(0,235,255) if side==0 else (255,135,20)
+            fast_edges(raw,uv,edges,color)
+            if valid[i,side]:fast_edges(usable,uv,edges,color)
+            # White MANO joints are a diagnostic derivation, not a new keypoint component.
+            dots(raw,project_for_view(joints[i],m,params,scale),(255,255,255),2)
+            if valid[i,side]:dots(usable,project_for_view(joints[i],m,params,scale),(255,255,255),2)
+        for obj in objs:
+            if not obj['rawvalid'][i]:continue
+            P=obj['P'][i];xyz=obj['vertices']@P[:3,:3].T+P[:3,3]
+            if obj['parts'] is not None:
+                top=obj['top'][i];sel=obj['parts']==1;xyz[sel]=obj['vertices'][sel]@top[:3,:3].T+top[:3,3]
+            uv=project_for_view(xyz,m,params,scale)
+            for image,enabled in [(raw,True),(usable,obj['valid'][i])]:
+                if not enabled:continue
+                if obj['edges'] is not None:fast_edges(image,uv,obj['edges'],obj['color'])
+                else:
+                    dots(image,uv[obj['parts']==0],(100,240,90));dots(image,uv[obj['parts']==1],(230,80,230))
+                    dots(image,project_for_view(obj['keypoints'][i],m,params,scale),(255,255,255),3)
+                # Local object axes in metres make pose orientation visible.
+                axes=np.vstack([np.zeros(3),np.eye(3)*.045])@P[:3,:3].T+P[:3,3];auv=project_for_view(axes,m,params,scale)
+                for j,c in enumerate([(255,60,60),(60,255,60),(60,120,255)]):fast_edges(image,auv,np.array([[0,j+1]]),c)
+        canvas=np.full((HEIGHT,WIDTH,3),20,np.uint8)
+        for col,(im,label) in enumerate([(original,'SOURCE RGB'),(raw,'NATIVE LABEL DIAGNOSTIC'),(usable,'SHARED USABLE SUPERVISION')]):
+            canvas[40+(CELL-h)//2:40+(CELL-h)//2+h,col*CELL+(CELL-w)//2:col*CELL+(CELL-w)//2+w]=im
+            text(canvas,label,(col*CELL+12,25),size=.58)
+        status='NO OFFICIAL QA PROVIDED' if qa is None else ('OFFICIAL QA PASS' if qa[i] else 'OFFICIAL QA FAIL - NOT SUPERVISION')
+        if dataset=='ARCTIC' and i in m['components'].get('mano',{}).get('valid_policy',{}).get('local_excluded_frame_indices',[]):status='LOCAL REVIEW REJECT - NOT HAND SUPERVISION'
+        color=(235,195,65) if qa is None else ((80,240,110) if qa[i] else (255,90,90))
+        text(canvas,f'{dataset} {cam}  index={i}/{T-1}  source={frames[i]["source_frame_id"]}   {status}',(12,578),color,.6)
+        text(canvas,f'MANO source L/R={int(present[i,0])}/{int(present[i,1])}  shared valid L/R={int(valid[i,0])}/{int(valid[i,1])}  (m/rad)',(12,603),size=.49)
+        def vals(a):return 'n/a' if a is None else ','.join('unknown' if not np.isfinite(x) else f'{100*x:.1f}%' for x in np.atleast_1d(a))
+        text(canvas,f'visibility native L/R: {vals(rawvis[i] if rawvis is not None else None)}  usable: {vals(visibility[i] if visibility is not None else None)}',(12,626),size=.48)
+        text(canvas,f'camera={m["camera"]["model"]}; intrinsic '+np.array2string(K[i].ravel()[:5],precision=2),(12,649),size=.45)
+        text(canvas,'camera world position (m): '+np.array2string(world[i],precision=3),(12,672),size=.45)
+        clock=frames[i]['timestamp_ns'];text(canvas,f'timestamp_ns={clock}  clock={frames[i]["clock_domain"]}',(12,695),size=.44)
+        text(canvas,f'preview FPS={fps}; '+('capture FPS unknown, playback only' if m['sampling']['source_fps'] is None else 'source capture time is stored separately'),(12,718),size=.44)
+        text(canvas,'cyan=left; orange=right; white=derived MANO joints / source object points',(12,741),size=.41)
+        text(canvas,'Projection shows hidden surfaces; it is NOT native segmentation.',(12,762),(180,180,180),.43)
+        for j,obj in enumerate(objs):
+            a=obj['raw_visibility'];v=obj['visibility'];s=f'{obj["name"]}: valid={int(obj["valid"][i])}'
+            if a is not None:s+=f' vis={vals(v[i])} (raw {vals(a[i])})'
+            if 'joint' in obj:s+=f' hinge={obj["joint"][i]:.3f}rad  keypoints=32'
+            text(canvas,s,(660,603+j*24),obj['color'],.44)
+        tr=traj.copy()
+        if np.isfinite(world[i]).all():cv2.circle(tr,tuple(tp[i]),4,(255,220,70),-1)
+        canvas[610:755,WIDTH-277:WIDTH-12]=tr
+        canvas[-12:]=timeline;cursor=min(WIDTH-1,int(i*WIDTH/T));canvas[-12:,cursor:cursor+2]=255
+        enc.stdin.write(canvas.tobytes());rendered+=1
+        if i in snapshots:
+            snap=doc/'video_snapshots'/cam;snap.mkdir(parents=True,exist_ok=True);Image.fromarray(canvas).save(snap/f'{i:06d}.jpg',quality=92)
+        if i%200==0:print('VIDEO_FRAME',dataset,cam,i,T,flush=True)
+    assert not dec.stdout.read(1),'extra source frames';dec.stdout.close();assert dec.wait()==0
+    enc.stdin.close();assert enc.wait()==0;log.close();assert rendered==T
+    packets=framehash(['-i',str(path)],copy=True);assert len(packets)==T
+    result={'sequence_id':m['sequence_id'],'camera_id':cam,'frames':T,'rendered_frames':rendered,'video_ref':path.relative_to(doc).as_posix(),'video_sha256':digest(path),'preview_fps':fps,
+        'preview_is_training_data':False,'preview_encoding':'H264 CRF19, scaled, display only; shared original MKV unchanged',
+        'source_labels_from':'raw-derived diagnostic cache' if source!=p else 'shared unfiltered source-compatible labels',
+        'all_source_and_shared_valid_masks_used':True,'structure_checks':structural,'mano_valid_counts':valid.sum(0).tolist(),
+        'official_qa_pass_frames':int(qa.sum()) if qa is not None else None,'official_qa_fail_frames':int((~qa).sum()) if qa is not None else None,
+        'exposure_diagnostics':{'method':'scaled shared decoded RGB to 96x96 gray; p95<35 flags a candidate only, does not modify valid','suspect_dark_frame_indices':[x['frame_index'] for x in exposure if x['p95_luma']<35]}}
+    save(tmp/f'{cam}-exposure.json',exposure)
+    save(tmp/f'{cam}-video-result.json',result);print('VIDEO_COMPLETE',dataset,cam,T,flush=True)
+    return result
+
+def asset_preview(op,out):
+    """Textured point-splat turntable plus the source texture; diagnostic, not photoreal rendering."""
+    p=op.parents[2];m=load(op/'metainfo.json');meshpath=p/m['mesh_ref']
+    if (op/'assets/mesh_tex.obj').exists():meshpath=op/'assets/mesh_tex.obj'
+    mesh=trimesh.load(meshpath,process=False,force='mesh');xyz=np.asarray(mesh.vertices);idx=np.arange(0,len(xyz),max(1,len(xyz)//35000));xyz=xyz[idx]
+    texture=getattr(mesh.visual.material,'image',None) if hasattr(mesh.visual,'material') else None
+    texture_source=None
+    if meshpath.name=='mesh_tex.obj' and op.relative_to(DATA_ROOT).parts[0]=='ARCTIC':
+        # The loader can silently return a 2x2 default material. Resolve the
+        # official single-texture MTL explicitly instead of accepting that placeholder.
+        mtl=[r.split(maxsplit=1)[1] for r in meshpath.read_text().splitlines() if r.startswith('mtllib ')]
+        diffuse=[]
+        for name in mtl:
+            material=meshpath.parent/name
+            diffuse.extend(material.parent/r.split(maxsplit=1)[1] for r in material.read_text().splitlines() if r.startswith('map_Kd '))
+        diffuse=list(set(diffuse));assert len(diffuse)==1,(meshpath,diffuse)
+        assert diffuse[0].resolve().is_relative_to(meshpath.parent.resolve())
+        texture=Image.open(diffuse[0]).convert('RGB');assert min(texture.size)>2
+        texture_source={'file':diffuse[0].relative_to(p).as_posix(),'sha256':digest(diffuse[0]),'size':list(texture.size),'resolution_method':'explicit official MTL map_Kd; reject loader placeholder'}
+    if texture is not None and getattr(mesh.visual,'uv',None) is not None:
+        tex=np.asarray(texture.convert('RGB'));uv=np.asarray(mesh.visual.uv)[idx]
+        rgb=tex[np.clip(np.rint((1-uv[:,1])*(len(tex)-1)).astype(int),0,len(tex)-1),np.clip(np.rint(uv[:,0]*(tex.shape[1]-1)).astype(int),0,tex.shape[1]-1)]
+    else:rgb=np.broadcast_to([150,210,245],xyz.shape).astype(np.uint8)
+    extent=np.ptp(xyz,axis=0);xyz=xyz-(xyz.min(0)+xyz.max(0))/2;scale=300/max(np.linalg.norm(extent),.001)
+    logpath=DATA_ROOT/op.relative_to(DATA_ROOT).parts[0]/'tmp'/VIDEO_RUN/'assets.log'
+    logpath.parent.mkdir(parents=True,exist_ok=True)
+    out.parent.mkdir(parents=True,exist_ok=True)
+    with logpath.open('ab') as log:
+        enc=encoder(out,12,log,size=(800,480))
+        for i in range(72):
+            image=np.full((480,800,3),24,np.uint8);R=Rotation.from_euler('xy',[.35,i*2*np.pi/72]).as_matrix();v=xyz@R.T
+            order=np.argsort(v[:,2]);xy=np.rint(v[:,:2]*scale+[220,220]).astype(int)
+            good=(xy[:,0]>=0)&(xy[:,0]<430)&(xy[:,1]>=35)&(xy[:,1]<400);order=order[good[order]]
+            for dx,dy in [(0,0),(1,0),(0,1)]:image[np.clip(xy[order,1]+dy,0,479),np.clip(xy[order,0]+dx,0,429)]=rgb[order]
+            if texture is not None:
+                thumb=texture.convert('RGB');thumb.thumbnail((350,345));image[40:40+thumb.height,440:440+thumb.width]=np.asarray(thumb)
+            text(image,m.get('category_name',m['object_id']),(12,25),size=.65)
+            text(image,'TEXTURED 3D POINT PREVIEW',(12,420),size=.55);text(image,'SOURCE MATERIAL TEXTURE',(445,420),size=.5)
+            text(image,'OBJ extent (m): '+np.array2string(extent,precision=4),(12,450),size=.5);enc.stdin.write(image.tobytes())
+        enc.stdin.close();assert enc.wait()==0
+    assert len(framehash(['-i',str(out)],copy=True))==72
+    return {'object_id':m['object_id'],'video_ref':'assets/'+out.name,'frames':72,'texture_present':texture is not None,'texture_source':texture_source,'extent_m':extent.tolist(),'method':'diagnostic textured point turntable, not photoreal material rendering'}
+
+def publish_review(dataset,results,assets):
+    root=DATA_ROOT/dataset/RELEASE;doc=root/'doc/validation'
+    scope=[('RGB','全部帧原图与逐帧码流/像素校验'),('相机内参/畸变','投影使用声明模型和全部参数；HUD显示相机模型'),('相机外参','全帧姿态、米制相机位置与世界XZ轨迹'),('MANO','逐帧网格线框、派生关节点、左右手valid'),('物体姿态/几何','逐帧轮廓、局部XYZ轴、物体ID'),('材质/纹理','每个物体的原纹理与完整旋转预览')]
+    scope+=([('部件/关节','底座/顶盖分别着色，逐帧关节角'),('物体关键点','源32个物体点逐帧叠加'),('visibility','原数据未提供，明确缺省')] if dataset=='ARCTIC' else [('visibility','原值0–1已修正，原生/可用比例同时显示'),('官方QA','完整红绿时间带；QA失败帧保留诊断、停用监督'),('部件/关节、独立物体关键点','本序列无该源模态，不虚构')])
+    scope+=[('深度/像素分割/接触','本次raw范围未提供，不以投影或帧筛选CSV冒充')]
+    table=''.join(f'<tr><td>{a}</td><td>{b}</td></tr>' for a,b in scope)
+    cards=''.join(f'<section><h2>Camera {r["camera_id"]} · {r["frames"]} 帧 · 有效MANO左右 {r["mano_valid_counts"]}</h2><video controls preload="metadata" src="{r["video_ref"]}"></video><p>全部{r["rendered_frames"]}帧已渲染，无码帧跳过。官方QA通过/未通过：{r["official_qa_pass_frames"]}/{r["official_qa_fail_frames"]}（null表示来源无此QA）。</p></section>' for r in results)
+    gallery=''.join(f'<section><h3>物体 {a["object_id"]}</h3><video controls loop preload="metadata" src="{a["video_ref"]}"></video><p>纹理点云旋转预览及原始材质图；此预览不证明完整PBR渲染等价。</p></section>' for a in assets)
+    known=('已修正转换器颠倒顶盖/底座编码的错误：原parts.json的0是顶盖、1是底座，共享层重新映射为0底座/1顶盖。9路相机前两帧均为源欠曝图，已排除对应几何监督；camera 6 index 270的手部偏移作为已知问题记录，按用户要求撤销单帧屏蔽，恢复原标注和valid。整套试点仍待验收。中栏显示源参数诊断，右栏按当前valid绘制。详见 <a href="ARCTIC_FINDINGS.md">问题与修复证据</a>、<a href="diagnosis/camera6-270-before-after.jpg">修复前后对照</a>。预览30fps仅用于播放，不代表真实采集帧率；没有源准确性QA。' if dataset=='ARCTIC' else '原始开头帧存在明显错位，官方QA为False；源头戴位姿开头27帧保持不变，帧号/时间/包顺序均未人为移动。中栏保留原始诊断投影，右栏只绘制当前valid监督。已修正visibility误除100。QA按相机整帧保守筛选，不能断言具体哪个实体错。')
+    page=f'''<!doctype html><html lang="zh-CN"><meta charset="utf-8"><meta name="viewport" content="width=device-width, initial-scale=1"><title>{dataset} 完整视频审核 v2</title><style>body{{font:16px/1.7 system-ui;background:#edf1f7;color:#172136;max-width:1550px;margin:25px auto;padding:0 20px}}header,section{{background:white;padding:22px;border-radius:12px;margin-bottom:20px}}video{{width:100%;background:#151515}}h1{{margin:0}}.warn{{color:#a71d2b;font-weight:bold}}table{{border-collapse:collapse;width:100%}}td,th{{border:1px solid #ccd4df;padding:8px;text-align:left}}.assets{{display:grid;grid-template-columns:repeat(auto-fit,minmax(500px,1fr));gap:18px}}a{{color:#1567a3}}</style><header><h1>{dataset} · 完整序列审核 v2</h1><p class="warn">人工确认待完成。旧“抽帧投影通过”结论撤回，本页替代旧报告；尚未全量或发布。</p><p>视频三栏：原图 ｜ 原始标注诊断 ｜ 共享层有效监督。所有相机、所有帧都已渲染；青=左手，橙=右手，其他色=物体/部件，白点=MANO派生关节点或源物体关键点。投影不做遮挡剔除，也不是分割真值。</p><p>{known}</p><p>标注 valid=实际标注存在且数值合法，并通过适用官方QA及已明确登记的本地排除规则；可见性不代替valid，剩余帧不因此自动认定人工通过。诊断中被排除的参数不进入有效监督，raw保留。MP4经过缩放与展示编码，仅为doc预览，训练输入仍是无新增损失的shared MKV。</p><p><a href="REPORT.md">正式说明</a> · <a href="report.json">逐相机检查结果</a> · <a href="../RAW_TO_SHARED.md">字段映射</a> · <a href="../CONTRACT.md">更新后的valid定义</a></p><h2>逐模态检查覆盖</h2><table><tr><th>模态</th><th>检查/可视化方式</th></tr>{table}</table></header>{cards}<h2>物体资产检查</h2><div class="assets">{gallery}</div><script>document.querySelectorAll('video').forEach(v=>v.addEventListener('play',()=>document.querySelectorAll('video').forEach(o=>{{if(o!==v)o.pause()}})));</script></html>'''
+    (doc/'index.html').write_text(page,encoding='utf-8')
+    old=load(doc/'report.json');old.update(stage='full_video_review_awaiting_human',human_review='pending',review_revision='full-video-v2',
+        old_sample_projection_conclusion='withdrawn',full_video_views=results,asset_previews=assets,modality_coverage=[{'modality':a,'check':b} for a,b in scope],known_issues=known,
+        agent_visual_inspection={'status':'pending_new_full_video_review','not_equivalent_to_user_approval':True},created_at=datetime.now().astimezone().isoformat())
+    old.pop('views',None);old.pop('offline_artifact_check',None);save(doc/'report.json',old)
+    save(doc/'status.json',{'stage':'full_video_review_awaiting_human','human_review':'pending','full_preprocessing_authorized':False,'upload_authorized':False})
+    rows='\n'.join(f'| {r["camera_id"]} | {r["frames"]} | {r["rendered_frames"]} | {r["mano_valid_counts"]} | {r["official_qa_pass_frames"]} |' for r in results)
+    (doc/'REPORT.md').write_text(f'# {dataset} 完整序列审核 v2\n\n[离线完整视频](index.html) · [机器结果](report.json)\n\n人工确认待完成；旧抽帧投影通过结论撤回。{known}\n\n|相机|源帧数|渲染帧数|左右手有效标注帧数|官方QA通过帧数|\n|---|---:|---:|---|---:|\n{rows}\n\n## 逐模态覆盖\n\n'+''.join(f'- {a}：{b}。\n' for a,b in scope)+'\n完整视频表示每帧已渲染，并不等于每帧人工审核通过。被官方QA排除的源标注仅在诊断栏显示；不挪帧、不隐藏原始异常。所有有效性筛选、可见性与图像编码验收分开记录。源标注质量问题无法通过格式转换修好；通过valid屏蔽，并保留raw溯源。\n',encoding='utf-8')
+    for f in ['CONTRACT.md','schema.json']:shutil.copyfile(WORK/'distribution/hand-warehouse-1.0'/f,root/'doc'/f)
+    for f in ['render_full_sequence_review.py','apply_hot3d_qa.py','fresh_sequence_common.py','review_fresh_sequence_pilots.py','convert_fresh_sequence_pilots.py']:shutil.copyfile(WORK/'scripts'/f,root/'script'/f)
+    print('FULL_VIDEO_REVIEW_READY',dataset,flush=True)
+
+if __name__=='__main__':
+    ap=argparse.ArgumentParser();ap.add_argument('dataset',choices=['HOT3D','ARCTIC']);a=ap.parse_args();torch.set_num_threads(3)
+    root=DATA_ROOT/a.dataset/RELEASE;qa=load(root/'doc/validation/source_qa.json') if a.dataset=='HOT3D' else None
+    # This entry point owns the original single-sequence report. Other review
+    # sequences have their own output directories in arctic_object_review.py.
+    sequence={'ARCTIC':'s09_laptop_use_03','HOT3D':'P0015_179e1b84'}[a.dataset]
+    views=sorted(p.parent for p in (root/'data'/sequence).glob('*/metainfo.json'));results=[]
+    for p in views:results.append(render_view(p,a.dataset,qa))
+    assets=[]
+    for op in sorted((views[0]/'annotation/object').iterdir()):assets.append(asset_preview(op,root/'doc/validation/assets'/f'{op.name}.mp4'))
+    publish_review(a.dataset,results,assets)
diff --git a/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/review_fresh_sequence_pilots.py b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/review_fresh_sequence_pilots.py
new file mode 100644
index 0000000000000000000000000000000000000000..4f812ae3ab2052ca4440157fbc07271787b2e143
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/review_fresh_sequence_pilots.py
@@ -0,0 +1,221 @@
+"""Validate actual shared outputs and create portable, offline pilot reviews.
+
+Run after convert_fresh_sequence_pilots.py. Never authorizes full processing.
+"""
+from fresh_sequence_common import *
+import html
+from datetime import datetime
+
+def arr(p):
+    value=torch.load(p,map_location='cpu',weights_only=True)
+    assert isinstance(value,torch.Tensor) and value.device.type=='cpu'
+    return value.numpy()
+
+def safe_ref(root,ref):
+    path=(root/ref).resolve()
+    assert path.is_relative_to(root.resolve()) and path.is_file(),ref
+    return path
+
+def validate_view(p):
+    m=load(p/'metainfo.json');T=m['num_frames'];registry=load(WORK/'distribution/hand-warehouse-1.0/schema.json')
+    assert m['schema_version']==registry['schema_version'] and m['pilot_complete']
+    frames=[json.loads(line) for line in (p/'camera/frames.jsonl').read_text().splitlines()]
+    assert len(frames)==T and [f['frame_index'] for f in frames]==list(range(T))
+    assert len({f['source_frame_id'] for f in frames})==T
+    for frame in frames:
+        for match in frame.get('annotation_matches',{}).values():
+            if 'available' in match:match['source_record_present']=match.pop('available')
+    lines(p/'camera/frames.jsonl',frames)
+    summary=[]
+    for key,c in m['components'].items():
+        assert c['schema_id'] in registry['components'],(key,c['schema_id'])
+        assert c['available'] and c['source_locator'] and c['provenance']['kind'] in ['source','derived','format_conversion']
+        data=safe_ref(p,c['data_ref']);valid=arr(safe_ref(p,c['valid_ref']))
+        assert valid.dtype==bool and valid.ndim==len(c['valid_axes'])
+        if data.suffix=='.pt':
+            a=arr(data);assert a.dtype==np.float32 and a.shape[0]==T
+            assert a.shape[:valid.ndim]==valid.shape and len(c['axes'])==a.ndim
+            assert np.isfinite(a[valid]).all() and np.isnan(a[~valid]).all()
+            if c['schema_id'] in ['object/rigid/v1','transform/se3/v1']:assert_se3(a[valid])
+            if c['schema_id']=='mano/axis_angle61/v1':assert a.shape==(T,2,61)
+        elif data.suffix=='.mkv':
+            assert c['storage']['profile'] in registry['storage_profiles']
+            assert valid.shape==(T,) and valid.all()
+        for attr in c.get('attributes',{}).values():
+            a=arr(safe_ref(p,attr['data_ref']));b=arr(safe_ref(p,attr['valid_ref']))
+            assert b.dtype==bool and a.shape==b.shape
+            assert np.isfinite(a[b]).all() and ((a[b]>=0)&(a[b]<=1)).all() and np.isnan(a[~b]).all()
+        if 'distortion_ref' in c:
+            d=arr(safe_ref(p,c['distortion_ref']));assert d.shape==(T,8) and np.isfinite(d).all()
+        summary.append({'component':key,'schema':c['schema_id'],'valid':int(valid.sum()),'total':int(valid.size)})
+    for op in sorted((p/'annotation/object').glob('*')):
+        om=load(op/'metainfo.json');mesh=safe_ref(p,om['mesh_ref'])
+        if 'vertex_part_ids_ref' in om:
+            ids=arr(safe_ref(p,om['vertex_part_ids_ref']));assert ids.dtype==np.int64
+            assert len(ids)==sum(line.startswith('v ') for line in mesh.read_text().splitlines())
+            assert om['topology_id']=='sha256:'+digest(mesh)
+            if not (op/'pose.pt').exists():
+                assert m.get('omitted_components',{}).get('object_pose')
+            else:
+                root=arr(op/'pose.pt');top=arr(op/'parts/top/pose.pt');joint=arr(op/'joint_state.pt')[:,0]
+                ok=arr(op/'pose_valid.pt');H=np.broadcast_to(np.eye(4),(ok.sum(),4,4)).copy()
+                H[:,:3,:3]=Rotation.from_rotvec(np.c_[np.zeros((ok.sum(),2)),-joint[ok]]).as_matrix()
+                if ok.any():assert np.max(np.abs(root[ok]@H-top[ok]))<2e-6
+        for obj in (op/'assets').glob('*.obj'):
+            for row in obj.read_text().splitlines():
+                if row.startswith('mtllib '):
+                    material=safe_ref(obj.parent,row.split(maxsplit=1)[1])
+                    for r in material.read_text().splitlines():
+                        if r.lower().startswith('map_kd '):safe_ref(material.parent,r.split(maxsplit=1)[1])
+    records=load(DATA_ROOT/m['dataset_id']/'tmp'/RUN/'checks'/f'{m["sequence_id"]}__{m["camera_id"]}.json')['checks']
+    assert records and all(x.get('passed',False) for x in records)
+    media=next(r for r in records if r['test']=='all_source_JPEG_packets_and_RGB_pixels_exact');assert media['frames']==T
+    if 'mano' not in m['components']:
+        assert m.get('omitted_components',{}).get('mano')
+        assert not (p/'annotation/mano/motion.pt').exists()
+        return m,frames,{'camera_id':m['camera_id'],'frames':T,'components':summary,'conversion_checks':records,'structure_and_references':'passed','mano':'source modality absent, omitted'}
+    actual=arr(p/'annotation/mano/motion_valid.pt')
+    if m['dataset_id']=='HOT3D':
+        src=m['components']['mano']['source_locator']
+        with zipfile.ZipFile(DATA_ROOT/'HOT3D'/src['archive']) as z:
+            native={r['timestamp_ns']:r['hand_poses'] for r in map(json.loads,z.read(src['member']).splitlines())}
+        expected=np.array([[str(side) in native.get(f['annotation_matches']['mano']['source_timestamp_ns'],{}) for side in range(2)] for f in frames],bool)
+    else:expected=np.array([[f['source_annotation_index'] is not None]*2 for f in frames],bool)
+    if m['dataset_id']=='HOT3D' and m['components']['mano'].get('valid_policy'):
+        source=m['components']['mano']['valid_policy']['source_locator']
+        with zipfile.ZipFile(DATA_ROOT/'HOT3D'/source['archive']) as z:
+            lookup={int(r['timestamp[ns]']):r['mask']=='True' for r in csv.DictReader(io.StringIO(z.read(source['member']).decode())) if r['stream_id']==m['camera_id']}
+        qa=[lookup[f['timecode_ns']] for f in frames]
+        expected &= np.array(qa,bool)[:,None]
+    if m['dataset_id']=='ARCTIC':
+        rejected=m['components']['mano'].get('valid_policy',{}).get('local_excluded_frame_indices',[])
+        expected[rejected]=False
+    assert np.array_equal(expected,actual),'annotation validity disagrees with declared source and QA/local policy'
+    records.append({'test':'all_frame_source_annotation_and_QA_policy','passed':True,'left_right_valid_counts':actual.sum(0).tolist(),'frames_without_usable_hand':int((~actual.any(1)).sum())})
+    return m,frames,{'camera_id':m['camera_id'],'frames':T,'components':summary,'conversion_checks':records,'structure_and_references':'passed'}
+
+def select_images(p,indices,dest):
+    dest.mkdir(parents=True,exist_ok=True)
+    selection='+'.join(f'eq(n,{i})' for i in indices)
+    command([FF,'-v','error','-y','-threads','2','-noautorotate','-i',p/'observation/rgb.mkv',
+        '-vf',f"select='{selection}',scale=900:900:force_original_aspect_ratio=decrease",'-fps_mode','vfr',dest/'decoded-%03d.png'])
+    output=sorted(dest.glob('decoded-*.png'));assert len(output)==len(indices)
+    return dict(zip(indices,output))
+
+def project_for_view(points,m,params,scale):
+    uv=project_points(points,m['camera']['model'],params)
+    if m['camera']['model']=='fisheye624':
+        angle=np.arctan2(np.linalg.norm(points[:,:2],axis=1),points[:,2])
+        uv[angle>m['camera']['max_solid_angle']]=np.nan
+    return uv*scale
+
+def draw_edges(canvas,uv,edges,color):
+    H,W=canvas.shape[:2]
+    for a,b in edges:
+        xy=uv[[a,b]]
+        if not np.isfinite(xy).all():continue
+        if np.max(np.abs(xy))>max(W,H)*20:continue
+        xy=np.rint(xy).astype(int)
+        if np.linalg.norm(xy[0]-xy[1])>max(W,H)*.35:continue
+        ok,start,end=cv2.clipLine((0,0,W,H),tuple(xy[0]),tuple(xy[1]))
+        if ok:cv2.line(canvas,start,end,color,1,cv2.LINE_AA)
+
+def review_view(p,m,frames,dest):
+    T=m['num_frames'];indices=sorted(set([0,1,T//4,T//2,3*T//4,T-1]));imgs=select_images(p,indices,dest)
+    K=arr(p/'camera/intrinsic.pt');D=arr(p/'camera/distortion.pt') if (p/'camera/distortion.pt').exists() else None
+    motion=arr(p/'annotation/mano/motion.pt');valid=arr(p/'annotation/mano/motion_valid.pt');models=[Mano('left'),Mano('right')]
+    if m['components']['mano']['model_profile']['left_shapedirs_fix']:
+        models[0].data['shapedirs'][:,0]*=-1
+        with torch.no_grad():models[0].layer.shapedirs[:,0]*=-1
+    objects=[]
+    for op in sorted((p/'annotation/object').iterdir()):
+        om=load(op/'metainfo.json');mesh=trimesh.load(p/om['mesh_ref'],process=False,force='mesh')
+        if 'vertex_part_ids_ref' in om:
+            vertices=np.array([[float(x) for x in r.split()[1:4]] for r in (p/om['mesh_ref']).read_text().splitlines() if r.startswith('v ')])
+            # trimesh may split OBJ vertices along UV seams; render exact source-ordered vertices as points.
+            objects.append((op,om,vertices,None,arr(op/'pose.pt'),arr(op/'pose_valid.pt'),arr(p/om['vertex_part_ids_ref']),arr(op/'parts/top/pose.pt')))
+        else:
+            edges=np.asarray(mesh.edges_unique);edges=edges[::max(1,len(edges)//2500)]
+            objects.append((op,om,np.asarray(mesh.vertices),edges,arr(op/'pose.pt'),arr(op/'pose_valid.pt'),None,None))
+    records=[]
+    for i,source in imgs.items():
+        original=np.asarray(Image.open(source).convert('RGB'));overlay=original.copy()
+        scale=np.array([original.shape[1]/m['camera']['image_size'][0],original.shape[0]/m['camera']['image_size'][1]])
+        params=K[i] if m['camera']['model']=='fisheye624' else (K[i],D[i] if D is not None else None)
+        for side,model in enumerate(models):
+            if not valid[i,side]:continue
+            h=motion[i,side];vertices,_,_=model.geometry(h[:48],h[48:58],h[58:]);uv=project_for_view(vertices[0],m,params,scale)
+            faces=np.asarray(model.layer.faces);edges=np.concatenate([faces[:,[0,1]],faces[:,[1,2]],faces[:,[2,0]]]);draw_edges(overlay,uv,edges,(0,235,255) if side==0 else (255,135,20))
+        palette=[(110,255,100),(255,80,210),(240,220,60),(90,180,255),(190,160,255),(255,120,140)]
+        for c,(op,om,vertices,edges,P,ok,parts,top) in enumerate(objects):
+            if not ok[i]:continue
+            xyz=vertices@P[i,:3,:3].T+P[i,:3,3]
+            if parts is not None:xyz[parts==1]=vertices[parts==1]@top[i,:3,:3].T+top[i,:3,3]
+            uv=project_for_view(xyz,m,params,scale)
+            if edges is not None:draw_edges(overlay,uv,edges,palette[c%len(palette)])
+            else:
+                for xy in uv[::max(1,len(uv)//6000)]:
+                    if np.isfinite(xy).all() and 0<=xy[0]<overlay.shape[1] and 0<=xy[1]<overlay.shape[0]:cv2.circle(overlay,tuple(np.rint(xy).astype(int)),1,palette[c%len(palette)],-1)
+        blended=cv2.addWeighted(overlay,.8,original,.2,0);output=dest/f'frame-{i:06d}-overlay.png';raw=dest/f'frame-{i:06d}-rgb.png'
+        Image.fromarray(blended).save(output);source.replace(raw)
+        records.append({'frame_index':i,'source_frame_id':frames[i]['source_frame_id'],'rgb':raw.name,'overlay':output.name,'mano_valid':valid[i].tolist()})
+    return records
+
+def write_reports(dataset,views):
+    base=DATA_ROOT/dataset;root=base/RELEASE;doc=root/'doc';review=doc/'validation';reports=[];cards=[];thumbs=[]
+    rawpaths=set()
+    for p in views:
+        if dataset=='ARCTIC':
+            for assets in (p/'annotation/object').glob('*/assets'):repair_arctic_material_refs(assets)
+        m,frames,summary=validate_view(p);cam=m['camera_id'];samples=review_view(p,m,frames,review/'images'/cam)
+        summary['visual_samples']=samples;reports.append(summary)
+        # All actually consumed source files; camera/mano/object locators name source archives.
+        rawpaths.add(m['components']['rgb']['source_locator'].get('archive',m['components']['rgb']['source_locator'].get('file')))
+        for comp in m['components'].values():
+            src=comp['source_locator']
+            if 'archive' in src:rawpaths.add(src['archive'])
+        if dataset=='ARCTIC':rawpaths.add('raw/data/splits_json.zip')
+        for op in (p/'annotation/object').iterdir():
+            glb=load(op/'metainfo.json').get('original_glb_ref')
+            if glb:rawpaths.add(load(op/'metainfo.json')['original_glb_source_locator']['file'])
+        opts=''.join(f'<option value="{i}"'+(' selected' if s['frame_index']==m['num_frames']//2 else '')+f'>index {s["frame_index"]} · source {s["source_frame_id"]} · valid {s["mano_valid"]}</option>' for i,s in enumerate(samples))
+        samples_js=json.dumps([{k:(f'images/{cam}/'+s[k] if k in ('rgb','overlay') else s[k]) for k in s} for s in samples])
+        cards.append(f'<section class="card" data-samples=\'{html.escape(samples_js,quote=True)}\'><h2>Camera {cam} · {m["camera"]["model"]} · {m["num_frames"]} 帧</h2><select onchange="show(this.closest(\'.card\'))">{opts}</select><label><input type="checkbox" checked onchange="show(this.closest(\'.card\'))"> 显示投影</label><br><a class="zoom" target="_blank"><img alt="共享产物实际帧与投影"></a><p class="caption"></p></section>')
+        middle=samples[len(samples)//2];thumb=Image.open(review/'images'/cam/middle['overlay']).convert('RGB');thumb.thumbnail((400,320));tile=Image.new('RGB',(420,355),'#18202b');tile.paste(thumb,((420-thumb.width)//2,28));ImageDraw.Draw(tile).text((10,8),f'{dataset} camera {cam} / frame {middle["frame_index"]}',fill='white');thumbs.append(tile)
+        print('REVIEWED',dataset,cam,flush=True)
+    cols=3;grid=Image.new('RGB',(cols*420,((len(thumbs)+cols-1)//cols)*355),'#18202b')
+    for i,t in enumerate(thumbs):grid.paste(t,((i%cols)*420,(i//cols)*355))
+    grid.save(review/'overview.png')
+    source_manifest=[]
+    ledger=load(base/'raw/doc'/('archive_checksums.json' if dataset=='ARCTIC' else 'file_checksums.json'))
+    for rel in sorted(x for x in rawpaths if x):
+        file=base/rel;sha=digest(file);record=ledger[rel.removeprefix('raw/data/')]
+        assert record['sha256']==sha and record['size']==file.stat().st_size
+        assert record.get('official_sha256_match',record.get('official_digest_match')) is True
+        if file.suffix=='.zip':assert record['zip_crc']=='passed_all_members'
+        if dataset=='HOT3D':assert record['fresh_download_run']=='2026-09-18-aria-fresh'
+        source_manifest.append({'path':rel,'bytes':file.stat().st_size,'sha256':sha,'matches_verified_official_download':True,'zip_crc_at_download':record.get('zip_crc')})
+    save(doc/'pilot-source-manifest.json',source_manifest)
+    script_manifest=[]
+    for name in ['review_fresh_sequence_pilots.py','fresh_sequence_common.py','convert_fresh_sequence_pilots.py']:
+        shutil.copyfile(WORK/'scripts'/name,root/'script'/name)
+    for p in sorted((root/'script').glob('*.py')):script_manifest.append({'path':p.relative_to(root).as_posix(),'sha256':digest(p)})
+    full={'dataset':dataset,'stage':'pilot_awaiting_human_review','human_review':'pending','full_preprocessing_authorized':False,'upload_authorized':False,
+        'scope':'One complete selected sequence, every released camera frame; not full-dataset acceptance','views':reports,
+        'sources_manifest':'../pilot-source-manifest.json','script_manifest':script_manifest,'preview_selection':'Deterministic boundary and quartile samples per camera; not random or full-dataset review',
+        'preview_rendering':'Derived MANO/object projections from shared tensors, not segmentation ground truth; no depth occlusion test, hidden surfaces may be drawn',
+        'created_at':datetime.now().astimezone().isoformat()}
+    save(review/'report.json',full);save(review/'status.json',{k:full[k] for k in ['dataset','stage','human_review','full_preprocessing_authorized','upload_authorized']})
+    notes=('首帧 RGB 存在，动捕标注无对应项，valid=false；已保留。相机 0 使用 rational8 畸变，外部八相机使用官方针孔投影。关节绕父坐标系 -Z 轴旋转，部件点/网格保持米制。' if dataset=='ARCTIC' else 'Aria 三路相机均保留；RGB 与两路灰度均由原 JPEG 直接封装。device time 经官方映射对齐 timecode；原始 visibility 按百分比转 [0,1]，独立于标注 valid。三路按时间匹配，不假定相同行号自动同步。')
+    if dataset=='HOT3D':notes+=' 每相机双手各806帧有MANO、208帧原始手参数为空；画面中有手但无投影的末尾帧属于源标注缺失，已验证并保留为valid=false。'
+    if dataset=='ARCTIC':notes+=' 原始部件OBJ的材质文件名引用错误已在共享层修正为实际存在的material.mtl，材质标识与纹理不变。'
+    missing=('本次官方 raw 中无独立深度、像素分割、接触标签、实测手关键点数组。保留 MANO，供任务层按需生成关键点；未把预览投影写成新增标注。' if dataset=='ARCTIC' else '本范围无深度流或像素分割；masks/*.csv 是帧筛选条件，不能当作像素 mask。UmeTrack 与 MPS 替代估计保留在 raw；shared 选择 MANO 和官方 GT 相机。没有额外生成关键点或接触标签。')
+    text=f'''<!doctype html><html lang="zh-CN"><meta charset="utf-8"><meta name="viewport" content="width=device-width, initial-scale=1"><title>{dataset} 序列试点审核</title><style>body{{font:16px/1.65 system-ui;background:#f3f5f8;color:#18202b;max-width:1450px;margin:24px auto;padding:0 20px}}header,.card{{background:white;padding:22px;border-radius:14px;margin-bottom:18px}}.status{{color:#9a5800;font-weight:bold}}.grid{{display:grid;grid-template-columns:repeat(auto-fit,minmax(360px,1fr));gap:18px}}.card img{{max-width:100%;max-height:670px;margin-top:12px}}select{{max-width:100%;padding:8px}}a{{color:#1766ab}}label{{display:inline-block;margin:10px}}h1{{margin:0}}.caption{{font-size:14px}}</style><header><h1>{dataset} · 完整序列试点</h1><p class="status">自动校验已完成；人工审核待确认。尚未开始全量预处理或上传。</p><p>共 {len(views)} 路相机、{sum(r['frames'] for r in reports):,} 个相机帧。全部源 JPEG 码流、全部解码 RGB 帧已逐一比较一致；数值转换、有效性和引用检查通过。</p><p>{notes}</p><p>{missing}</p><p>下方是从 shared 实际产物读取的边界及四分位样本。青色=左手，橙色=右手，其余颜色=物体。投影仅用于几何核验，<b>不是源分割，也没有做遮挡剔除</b>。点击图像可放大；取消勾选可查看原图。</p><p><a href="REPORT.md">验收说明</a> · <a href="report.json">机器可读结果</a> · <a href="../RAW_TO_SHARED.md">raw → share 对照</a> · <a href="../CONTRACT.md">约定</a> · <a href="overview.png">全部相机总览</a></p></header><main class="grid">{''.join(cards)}</main><script>function show(card){{let s=JSON.parse(card.dataset.samples)[Number(card.querySelector('select').value)];let path=card.querySelector('input').checked?s.overlay:s.rgb;card.querySelector('img').src=path;card.querySelector('.zoom').href=path;card.querySelector('.caption').textContent='frame_index='+s.frame_index+' / 原帧 '+s.source_frame_id+' / 左右手 valid='+s.mano_valid;}}document.querySelectorAll('.card').forEach(show)</script></html>'''
+    (review/'index.html').write_text(text,encoding='utf-8')
+    rows='\n'.join(f'| {r["camera_id"]} | {r["frames"]} | '+', '.join(str(x['frame_index']) for x in r['visual_samples'])+' | 通过 |' for r in reports)
+    report=f'''# {dataset} 完整序列试点验收\n\n状态：**自动检查通过，人工确认待完成**。本报告不是全量验收或发布许可。\n\n[离线预览](index.html) · [机器结果](report.json) · [原始包摘要](../pilot-source-manifest.json)\n\n| 相机 | 全序列帧数 | 可视化帧索引 | 自动校验 |\n|---|---:|---|---|\n{rows}\n\n## 已验证\n\n- 完整所选序列的每一张源 JPEG 与 MKV 取回的压缩包 SHA256 相同；全部帧在同一 FFmpeg/noautorotate/rgb24 配置下解码字节相同，顺序和数量一致。不是有损重编码。\n- 全部有效 MANO 帧的相机坐标变换与直接变换几何一致；独立源模型配置抽样校验非平手均值/PCA 展开。张量 float32 保存读取一致，invalid 为 NaN，valid 为 bool。\n- SE(3) 正交性、单位与资产引用检查；ARCTIC 部件关节运动与源旋转公式逐帧对照，HOT3D 物体 OBJ 尺度、三角形及纹理像素往返一致。\n- 所有实际组件使用现有注册 schema；没有为缺省模态伪造值。\n\n## 帧对应和缺省范围\n\n{notes}\n\n{missing}\n\n## 人工核验边界\n\n每相机选序列起点、紧邻帧、四分位、中点、四分之三及终点，不是随机抽查。预览只用于核对几何对齐；投影线条/点不是分割真值，遮挡表面可能被画出。数值自洽不等于图像投影正确，请检查所有相机和多个时刻。此处一序列通过不能覆盖全数据集、其他物体或异常序列。全量处理后仍需独立随机抽查和用户发布确认。\n'''
+    (review/'REPORT.md').write_text(report,encoding='utf-8')
+    return full
+
+if __name__=='__main__':
+    raise SystemExit('Static-sample review is retired. Use render_full_sequence_review.py DATASET and finalize_fresh_sequence_review.py; validate_view remains a reusable numeric helper.')
diff --git a/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/script-manifest.json b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/script-manifest.json
new file mode 100644
index 0000000000000000000000000000000000000000..37eb7c6c5fec939233dcff18b3dc16c70d39915a
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/script-manifest.json
@@ -0,0 +1,62 @@
+{
+  "created_at": "2026-09-20T13:24:10.057497+08:00",
+  "scope": "Current production converter and validation scripts; earlier pilot manifests describe their historical run",
+  "files": [
+    {
+      "path": "script/preprocess_fresh_full.py",
+      "sha256": "a28ac6aaf5298a18980ea21ffed28c9537b9b43b239dd01a21e6e4c54146b1bf"
+    },
+    {
+      "path": "script/convert_fresh_sequence_pilots.py",
+      "sha256": "d6f378a15a89d54112543876697710f87a71f4adf0aafb487484f1bab65d5999"
+    },
+    {
+      "path": "script/apply_hot3d_qa.py",
+      "sha256": "7158b0512e59220930d88d04e15f7f6684acdccd0279f3179095d27b66ed3f7b"
+    },
+    {
+      "path": "script/review_fresh_sequence_pilots.py",
+      "sha256": "eb04564e0e2dacaca126b5f9a2333dc7232ef83b178c277b266f81c33bfa5815"
+    },
+    {
+      "path": "script/render_full_sequence_review.py",
+      "sha256": "5b471845025266f8a82dfebb185dda213bfbeaba8e98d0e6cc467826efa9e628"
+    },
+    {
+      "path": "script/fresh_sequence_common.py",
+      "sha256": "17bc18af25ee5c66a24c88d059cfffaea69e62227b7128394f2644f28311713b"
+    },
+    {
+      "path": "script/diagnose_arctic_review.py",
+      "sha256": "75a264b9b2cb4749e7705770a7e1141b860679b2fbee7d710e0414c83814802d"
+    },
+    {
+      "path": "script/pilot_build.py",
+      "sha256": "b45c92503fae4ee1d5163a52179b3bcb136338231096f100e721489e80cb0ac2"
+    },
+    {
+      "path": "script/pilot_fetch.py",
+      "sha256": "05ed8bc88df4492861ba46471906647a95e5392dd93f1a4eb419f2c396511e2d"
+    },
+    {
+      "path": "script/pilot_mano.py",
+      "sha256": "96ceb450b07bff040015072b6572e11af3aec310d54e575d2eb549da8d80bed0"
+    },
+    {
+      "path": "script/sequence_pilot_common.py",
+      "sha256": "9fdce711983c18be88f79fbbace8ed0e42fabb6a375cf167c98041a80a89579e"
+    },
+    {
+      "path": "script/sequence_pilot_world.py",
+      "sha256": "271c4829cae29af87c3fcb1b1668dfcab1df369aaabfc8a333de9965fb580126"
+    },
+    {
+      "path": "script/pilot_hot_normalize.py",
+      "sha256": "fea5bfaaac248a1f33bd799d837760cf23080a9efc1cf715436e2358825f433a"
+    },
+    {
+      "path": "script/audit_source_fields.py",
+      "sha256": "bdcc4818189a8f4bc74f75c2b507d2feec5c2d0b59040c69cd99ab86ba92eac5"
+    }
+  ]
+}
\ No newline at end of file
diff --git a/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/sequence_pilot_common.py b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/sequence_pilot_common.py
new file mode 100644
index 0000000000000000000000000000000000000000..169c4868c9fbba40790ee94c2600fb35e418acef
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/sequence_pilot_common.py
@@ -0,0 +1,53 @@
+"""Whole-selection conversion utilities. Never reads previous pilot outputs."""
+from pilot_build import *
+import hashlib
+
+RELEASE='shared-sequence-v1.1'
+
+class View:
+ def __init__(self,dataset,seq,cam,ids,size,source,kind='video',sampling='dense',split='unknown'):
+  self.root=ROOT/dataset/RELEASE;self.p=self.root/'data'/seq/cam
+  self.p.mkdir(parents=True,exist_ok=True);self.T=len(ids);self.source=source;self.checks=[]
+  self.m={'schema_version':'hand-warehouse/1.1','dataset_id':dataset,'release_id':RELEASE,'sequence_id':seq,'camera_id':cam,'sample_type':kind,'num_frames':self.T,'hand_slots':['left','right'],'sampling':{'type':'single' if kind=='image' else sampling,'source_fps':None,'frames_ref':'camera/frames.jsonl'},'camera':{'model':'unknown','schema_id':None,'image_size':size},'components':{},'split':split}
+  self.frames=[{'frame_index':i,'source_frame_id':str(fid),'timestamp_ns':None,'clock_domain':None,'source_locator':{**source,'source_frame_id':str(fid)}} for i,fid in enumerate(ids)]
+ def add(self,name,path,schema,axes,valid,method='source values retained',**kw):
+  vp=Path(path).with_name(Path(path).stem+'_valid.pt').as_posix()
+  tensor(self.p/vp,valid)
+  self.m['components'][name]={'available':True,'schema_id':schema,'data_ref':path,'axes':axes,'valid_ref':vp,'valid_axes':axes[:valid.ndim],'provenance':{'kind':'format_conversion','method':method},'source_locator':self.source,**kw}
+ def array(self,name,path,schema,a,valid,axes,**kw):
+  a=np.asarray(a,np.float32).copy();valid=np.asarray(valid,bool)
+  assert a.shape[:valid.ndim]==valid.shape and a.shape[0]==self.T
+  assert np.isfinite(a[valid]).all(),name
+  a[~valid]=np.nan;tensor(self.p/path,a);self.add(name,path,schema,axes,valid,**kw)
+  restored=torch.load(self.p/path,map_location='cpu',weights_only=True).numpy()
+  for index in range(self.T):assert np.array_equal(restored[index],a[index],equal_nan=True)
+  self.checks.append({'test':'tensor_serialization_exact','file':path,'shape':list(a.shape),'valid_count':int(valid.sum())})
+ def video(self,name,path,iterator,mask=False,**kw):
+  w,h=self.m['camera']['image_size'];pix='gray16le' if mask else 'rgb24';dtype='<u2' if mask else 'uint8'
+  dst=self.p/path;dst.parent.mkdir(parents=True,exist_ok=True);hashes=[]
+  proc=subprocess.Popen([FF,'-v','error','-y','-f','rawvideo','-pixel_format',pix,'-video_size',f'{w}x{h}','-framerate','1','-i','pipe:0','-c:v','ffv1','-level','3','-g','1',str(dst)],stdin=subprocess.PIPE)
+  try:
+   for frame in iterator:
+    a=np.asarray(frame,dtype=dtype);assert a.shape==((h,w) if mask else (h,w,3));b=a.tobytes();hashes.append(hashlib.sha256(b).hexdigest());proc.stdin.write(b)
+  finally:proc.stdin.close()
+  assert proc.wait()==0 and len(hashes)==self.T
+  proc=subprocess.Popen([FF,'-v','error','-i',str(dst),'-f','rawvideo','-pix_fmt',pix,'pipe:1'],stdout=subprocess.PIPE)
+  nbytes=w*h*(2 if mask else 3)
+  for digest in hashes:
+   b=proc.stdout.read(nbytes);assert len(b)==nbytes and hashlib.sha256(b).hexdigest()==digest,'video roundtrip'
+  assert not proc.stdout.read(1);assert proc.wait()==0
+  self.checks.append({'test':'all_frames_pixel_exact_roundtrip','file':path,'frames':self.T})
+  self.add(name,path,'segment/uint16_labels/v1' if mask else 'rgb/uint8/v1',['T','height','width'] if mask else ['T','height','width','channel'],np.ones(self.T,bool),storage_fps=1,timebase_role='index_only',**kw)
+ def intrinsic(self,K):
+  self.m['camera'].update(model='pinhole',schema_id='camera/pinhole/v1')
+  self.array('intrinsic','camera/intrinsic.pt','camera/pinhole/v1',np.broadcast_to(K,(self.T,3,3)),np.ones(self.T,bool),['T','row','column'])
+ def finish(self):
+  lines(self.p/'camera/frames.jsonl',self.frames);save(self.p/'metainfo.json',self.m)
+  save(self.root/'doc'/f'{self.m["sequence_id"]}__{self.m["camera_id"]}.checks.json',{'checks':self.checks,'frames':self.T,'input':self.source})
+  print('Converted',self.m['dataset_id'],self.m['sequence_id'],self.m['camera_id'],self.T,flush=True)
+
+def snapshot(dataset,scripts):
+ out=ROOT/dataset/RELEASE/'script';out.mkdir(parents=True,exist_ok=True)
+ names={'sequence_pilot_common.py','pilot_build.py','pilot_fetch.py','pilot_mano.py','pilot_hot_normalize.py','audit_source_fields.py',*scripts}
+ names.update(p.name for p in (WORK/'scripts').glob('sequence_pilot_*.py'))
+ for name in names:shutil.copyfile(WORK/'scripts'/name,out/name)
diff --git a/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/sequence_pilot_world.py b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/sequence_pilot_world.py
new file mode 100644
index 0000000000000000000000000000000000000000..1d052448db5f35e32a80867f7dc2a6cda27a8065
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/history/pre-acceleration-20260920/sequence_pilot_world.py
@@ -0,0 +1,55 @@
+"""Convert whole native annotation sequences to camera-space MANO."""
+from sequence_pilot_common import *
+from audit_source_fields import native_npy
+from pilot_mano import Mano
+from pilot_hot_normalize import transform
+
+def camera_mano(model,pose,beta,translation,E):
+ pose=np.asarray(pose).reshape(-1,48);beta=np.asarray(beta).reshape(-1,10);translation=np.asarray(translation).reshape(-1,3)
+ root=np.einsum('v,tvc->tc',model.data['J_regressor'][0],model.data['v_template'][None]+np.einsum('vci,ti->tvc',model.data['shapedirs'],beta))
+ newpose=pose.copy();newpose[:,:3]=np.stack([cv2.Rodrigues(e[:3,:3]@cv2.Rodrigues(p[:3])[0])[0].ravel() for e,p in zip(E,pose)])
+ newt=np.einsum('tij,tj->ti',E[:,:3,:3],root+translation)+E[:,:3,3]-root
+ # Check all frames in modest batches; meshes exist only in memory for verification.
+ errors=[]
+ for begin in range(0,len(pose),64):
+  s=slice(begin,begin+64);world,_,_=model.geometry(pose[s],beta[s],translation[s]);cam,_,_=model.geometry(newpose[s],beta[s],newt[s]);target=np.einsum('tij,tvj->tvi',E[s,:3,:3],world)+E[s,None,:3,3]
+  errors.append(float(np.linalg.norm(cam-target,axis=-1).max()))
+ assert max(errors)<2e-5,max(errors)
+ return np.concatenate([newpose,beta,newt],axis=-1),max(errors)
+
+def arctic():
+ seq='capsulemachine_use_02';src=OLD/'ARCTIC/raw_seqs/s01';a=native_npy(src/(seq+'.mano.npy')).item();c=native_npy(src/(seq+'.egocam.dist.npy')).item();T=len(a['left']['rot']);misc=load(OLD/'ARCTIC/meta/misc.json')['s01'];ids=list(range(T))
+ v=View('ARCTIC','s01_'+seq,'ego',ids,misc['image_size'][0],{'path':str(src.relative_to(WORK)),'sequence':seq},sampling='full_annotation_sequence')
+ E=np.tile(np.eye(4),(T,1,1));E[:,:3,:3]=c['R_k_cam_np'];E[:,:3,3]=c['T_k_cam_np'].reshape(T,3)
+ v.intrinsic(c['intrinsics']);v.m['camera'].update(model='rational8',schema_id='camera/rational8/v1');v.m['components']['intrinsic'].update(schema_id='camera/rational8/v1',distortion_ref='camera/distortion.pt')
+ v.array('distortion','camera/distortion.pt','camera/distortion8/v1',np.broadcast_to(c['dist8'],(T,8)),np.ones(T,bool),['T','parameter'],parameter_names=['k1','k2','p1','p2','k3','k4','k5','k6'])
+ v.array('extrinsic','camera/extrinsic.pt','transform/se3/v1',E,np.ones(T,bool),['T','row','column'],unit='m',convention='T_camera_from_world')
+ motion=np.empty((T,2,61))
+ for s,side in enumerate(['left','right']):
+  b=a[side];model=Mano(side);pose=np.c_[b['rot'],b['pose']].copy();pose[:,3:]+=model.data['hands_mean'];motion[:,s],err=camera_mano(model,pose,np.broadcast_to(b['shape'],(T,10)),b['trans'],E);v.checks.append({'test':'all_frame_camera_MANO_geometry','side':side,'max_m':err,'native_hand_mean':'ARCTIC common/body_models.py flat_hand=False; mean added before normalization'})
+ v.array('mano','annotation/mano/motion.pt','mano/axis_angle61/v1',motion,np.ones((T,2),bool),['T','hand','parameter'],unit='m',angle_unit='rad',coordinate_frame='camera',model_profile={'version':'MANO-v1.2','flat_hand_mean':True,'left_shapedirs_fix':False})
+ for i,row in enumerate(v.frames):row.update(source_annotation_index=i,source_image_frame_id=i+misc['ioi_offset'])
+ v.m['input_gaps']=['full matching RGB image inventory and selected object.npy/object meshes unavailable locally; not source absence'];v.finish();snapshot('ARCTIC',['sequence_pilot_world.py','audit_source_fields.py','pilot_mano.py'])
+
+def oakink2():
+ src=next((OLD/'OakInk2/anno_preview').glob('*.pkl'));a=read_pickle(src);ids=a['frame_id_list'];T=len(ids);cam='104422070969';key=a['cam_def'][cam];images=OLD/'OakInk2/data'/src.stem/cam
+ first=np.asarray(Image.open(images/f'{ids[0]:06d}.png'));v=View('OakInk2',src.stem,cam,ids,[first.shape[1],first.shape[0]],{'annotation':str(src.relative_to(WORK)),'image_root':str(images.relative_to(WORK))},sampling='official_frame_id_list')
+ v.video('rgb','observation/rgb.mkv',(np.asarray(Image.open(images/f'{i:06d}.png').convert('RGB')) for i in ids))
+ K=np.stack([a['cam_intr'][key][i] for i in ids]);E=np.stack([a['cam_extr'][key][i] for i in ids]);v.m['camera'].update(model='pinhole',schema_id='camera/pinhole/v1')
+ v.array('intrinsic','camera/intrinsic.pt','camera/pinhole/v1',K,np.ones(T,bool),['T','row','column'])
+ v.array('extrinsic','camera/extrinsic.pt','transform/se3/v1',E,np.ones(T,bool),['T','row','column'],unit='m',convention='T_camera_from_world',method='official preview applies cam_extr @ world object transform')
+ motion=np.empty((T,2,61))
+ for s,side in enumerate(['left','right']):
+  prefix=['lh','rh'][s];rows=[a['raw_mano'][i] for i in ids];q=np.stack([np.asarray(r[prefix+'__pose_coeffs']).reshape(16,4) for r in rows]);pose=np.stack([cv2.Rodrigues(transform(qq,[0,0,0])[:3,:3])[0].ravel() for qq in q.reshape(-1,4)]).reshape(T,48)
+  beta=np.stack([np.asarray(r[prefix+'__betas']).reshape(10) for r in rows]);trans=np.stack([np.asarray(r[prefix+'__tsl']).reshape(3) for r in rows]);model=Mano(side)
+  root=np.einsum('v,tvc->tc',model.data['J_regressor'][0],model.data['v_template'][None]+np.einsum('vci,ti->tvc',model.data['shapedirs'],beta))
+  motion[:,s],err=camera_mano(model,pose,beta,trans-root,E);v.checks.append({'test':'all_frame_camera_MANO_geometry','side':side,'max_m':err,'source_center':'official manotorch center_idx=0; subtract shape-dependent wrist'})
+ v.array('mano','annotation/mano/motion.pt','mano/axis_angle61/v1',motion,np.ones((T,2),bool),['T','hand','parameter'],unit='m',coordinate_frame='camera',angle_unit='rad',model_profile={'version':'MANO-v1.2','flat_hand_mean':True,'left_shapedirs_fix':False},method='source wxyz quaternions to axis-angle; same mocap frame keys as official preview; root-corrected world to camera')
+ for oid in a['obj_list']:
+  W=np.stack([a['obj_transf'][oid][i] for i in ids]);P=E@W;valid=np.isfinite(P).all((1,2));rel=f'annotation/object/{oid}'
+  v.array('object_'+oid+'_pose',rel+'/pose.pt','object/rigid/v1',P,valid,['T','row','column'],unit='m',convention='T_camera_from_object')
+  save(v.p/rel/'metainfo.json',{'object_id':oid,'identity_scope':'sequence','scope_id':src.stem,'has_model':False,'source_locator':v.source})
+ v.m['input_gaps']=['object asset models not yet downloaded'];v.m['semantic_review_required']=['MANO quaternion hand mean convention must be compared with official renderer; algebraic check alone is insufficient'];v.finish();snapshot('OakInk2',['sequence_pilot_world.py','audit_source_fields.py','pilot_mano.py'])
+
+if __name__=='__main__':
+ p=argparse.ArgumentParser();p.add_argument('dataset',choices=['ARCTIC','OakInk2']);a=p.parse_args();{'ARCTIC':arctic,'OakInk2':oakink2}[a.dataset]()
diff --git a/ARCTIC/shared-v1.0/script/pilot_build.py b/ARCTIC/shared-v1.0/script/pilot_build.py
new file mode 100644
index 0000000000000000000000000000000000000000..2881d0878bd6d3456c2bf84edbf151cb3851a3ac
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/pilot_build.py
@@ -0,0 +1,209 @@
+"""Small, source-traceable dataset pilot. Originals are copied, never edited.
+
+Run with Python + numpy + torch + Pillow + opencv and imageio-ffmpeg.
+Optional runtime installed at tmp/pilot-runtime. Outputs: hand-datasets/.
+"""
+from pathlib import Path
+import argparse, codecs, collections, csv, io, json, pickle, shutil, subprocess, sys, zipfile
+sys.path.insert(0,str(Path(__file__).resolve().parents[1]/'tmp/pilot-runtime'))
+import numpy as np
+import torch
+from PIL import Image, ImageDraw
+import cv2
+from pilot_fetch import ROOT, sha
+
+WORK=ROOT.parent
+OLD=WORK/'datasets'
+sys.path.insert(0,str(WORK/'tmp/pilot-runtime'))
+import imageio_ffmpeg
+FF=imageio_ffmpeg.get_ffmpeg_exe()
+N=16
+
+def save(p,value):
+ p.parent.mkdir(parents=True,exist_ok=True)
+ p.write_text(json.dumps(value,ensure_ascii=False,indent=2,allow_nan=False),encoding='utf-8')
+def lines(p,rows):
+ p.parent.mkdir(parents=True,exist_ok=True)
+ p.write_text(''.join(json.dumps(x,ensure_ascii=False,allow_nan=False)+'\n' for x in rows),encoding='utf-8')
+def tensor(p,a):
+ p.parent.mkdir(parents=True,exist_ok=True);torch.save(torch.as_tensor(np.asarray(a).copy()),p)
+def load(p):return json.loads(p.read_text(encoding='utf-8'))
+
+class Restricted(pickle.Unpickler):
+ def find_class(self,module,name):
+  allowed={('numpy.core.multiarray','_reconstruct'):np.core.multiarray._reconstruct,
+   ('numpy._core.multiarray','_reconstruct'):np.core.multiarray._reconstruct,
+   ('numpy','ndarray'):np.ndarray,('numpy','dtype'):np.dtype,
+   ('numpy.core.numeric','_frombuffer'):np.core.numeric._frombuffer,
+   ('numpy._core.numeric','_frombuffer'):np.core.numeric._frombuffer,
+   ('_codecs','encode'):codecs.encode,
+   ('collections','OrderedDict'):collections.OrderedDict,
+   ('torch._utils','_rebuild_tensor_v2'):torch._utils._rebuild_tensor_v2,
+   ('torch.storage','_load_from_bytes'):lambda b:torch.load(io.BytesIO(b),map_location='cpu',weights_only=True)}
+  if (module,name) not in allowed:raise pickle.UnpicklingError(f'Unsupported class: {module}.{name}')
+  return allowed[(module,name)]
+def read_pickle(p):
+ with p.open('rb') as f:return Restricted(f,encoding='latin1').load()
+
+class Build:
+ def __init__(self,name):
+  self.name=name;self.base=ROOT/name;self.out=self.base/'pilot-v0';self.refs=[];self.views=[];self.checks=[]
+  for p in ['raw/doc','raw/script','raw/data','pilot-v0/doc','pilot-v0/script','pilot-v0/data']:(self.base/p).mkdir(parents=True,exist_ok=True)
+ def copy(self,src):
+  rel=src.relative_to(OLD/self.name);dst=self.base/'raw/data/local_subset'/rel
+  dst.parent.mkdir(parents=True,exist_ok=True)
+  if not dst.exists():shutil.copyfile(src,dst)
+  source_hash=sha(src);assert sha(dst)==source_hash
+  self.refs.append({'source':str(src),'stored':str(dst.relative_to(self.base)), 'sha256':source_hash,'bytes':src.stat().st_size,'kind':'historical_extracted_file_copy','official_archive_verified':False})
+  return dst
+ def view(self,seq,cam,ids,**meta):
+  p=self.out/'data'/seq/cam;p.mkdir(parents=True,exist_ok=True)
+  m={'schema_version':'pilot-v0','dataset':self.name,'sequence_id':seq,'camera_id':cam,'num_frames':len(ids),'fps':None,'split':None,'view_type':None,'time_semantics':'unknown','missing_modalities':{},**meta}
+  self.views.append((p,m))
+  lines(p/'frames.jsonl',[{'frame_index':i,'source_frame_id':str(s),'timestamp_ns':None,'clock_domain':None} for i,s in enumerate(ids)])
+  return p,m
+ def video(self,p,arrays):
+  a=np.stack(arrays);h,w=a.shape[1:3]
+  pixel='rgb24' if a.ndim==4 else ('gray16le' if a.dtype==np.uint16 else 'gray')
+  if pixel=='gray16le':a=a.astype('<u2')
+  a=np.ascontiguousarray(a);p.parent.mkdir(parents=True,exist_ok=True)
+  command=[FF,'-hide_banner','-loglevel','error','-y','-f','rawvideo','-pixel_format',pixel,'-video_size',f'{w}x{h}','-framerate','1','-i','pipe:0','-an','-c:v','ffv1','-level','3','-g','1',str(p)]
+  subprocess.run(command,input=a.tobytes(),check=True,capture_output=True)
+  decoded=subprocess.run([FF,'-v','error','-i',str(p),'-f','rawvideo','-pix_fmt',pixel,'pipe:1'],check=True,capture_output=True).stdout
+  assert decoded==a.tobytes(),f'Lossless roundtrip failed: {p}'
+  self.checks.append({'file':str(p.relative_to(self.out)),'test':'decode_all_frames_pixel_exact','passed':True,'frames':len(a),'dtype':str(a.dtype),'shape':list(a.shape)})
+  return {'codec':'FFV1','pixel_format':pixel,'container_fps':1,'container_time_is_capture_time':False}
+ def finish(self,notes):
+  for p,m in self.views:save(p/'metainfo.json',m)
+  save(self.base/'raw/doc/local_subset_manifest.json',self.refs)
+  report={'dataset':self.name,'views':len(self.views),'observations':sum(m['num_frames'] for _,m in self.views),'checks':self.checks,'notes':notes,'status':'partial' if notes else 'passed'}
+  save(self.out/'doc/report.json',report)
+  for script in ['pilot_build.py','pilot_fetch.py']:
+   shutil.copyfile(WORK/'scripts'/script,self.out/'script'/script)
+  (self.out/'script/README.md').write_text('执行入口：工作区 scripts/pilot_build.py --dataset '+self.name+'。此目录为脚本快照；脚本当前依赖工作区的 datasets 与 tmp/pilot-runtime，尚不是独立发布包。\n',encoding='utf-8')
+  files=[{'path':str(p.relative_to(self.out)), 'bytes':p.stat().st_size,'sha256':sha(p)} for p in sorted((self.out/'data').rglob('*')) if p.is_file()]
+  save(self.out/'doc/output_manifest.json',files)
+  print(json.dumps(report,ensure_ascii=False),flush=True)
+  return report
+
+def h2o():
+ b=Build('H2O');src=OLD/'H2O/subject4_ego/h1/0/cam4';ids=list(range(N))
+ p,m=b.view('subject4_h1_0','cam4',ids,split='test',view_type='ego',time_semantics='ordered_source_frames; acquisition FPS not yet verified')
+ for folder in ['rgb','depth','hand_pose','hand_pose_mano','cam_pose']:
+  for i in ids:b.copy(src/folder/f'{i:06d}.{ "png" if folder in ["rgb","depth"] else "txt"}')
+ intr=np.loadtxt(b.copy(src/'cam_intrinsics.txt'));fx,fy,cx,cy,w,h=intr
+ tensor(p/'camera/intrinsic.pt',np.array([[fx,0,cx],[0,fy,cy],[0,0,1]],np.float32))
+ ext=np.stack([np.loadtxt(src/'cam_pose'/f'{i:06d}.txt').reshape(4,4) for i in ids])
+ tensor(p/'camera/extrinsic.pt',np.linalg.inv(ext).astype(np.float32))
+ m['camera']={'model':'pinhole','width':int(w),'height':int(h),'extrinsic':'T_camera_from_world','source_extrinsic':'T_world_from_camera','unit':'m','distortion':'not supplied in selected intrinsics'}
+ a=np.stack([np.loadtxt(src/'hand_pose_mano'/f'{i:06d}.txt').reshape(2,62) for i in ids])
+ valid=a[:,:,0]==1;motion=np.concatenate([a[:,:,4:52],a[:,:,52:62],a[:,:,1:4]],axis=-1).astype(np.float32)
+ motion[~valid]=np.nan;tensor(p/'annotation/mano/motion.pt',motion);tensor(p/'annotation/mano/valid_flag.pt',valid)
+ m['mano']={'layout':['pose_axis_angle:48','betas:10','translation:3'],'shape':[N,2,61],'slots':['left','right'],'coordinate_frame':'camera','translation_unit':'m','flat_hand_mean':True,'model_version':'not supplied; MANO model not bundled','valid_flag':'source annotation flag, not visibility','geometry_validation':'not_tested; model unavailable'}
+ joints=np.stack([np.loadtxt(src/'hand_pose'/f'{i:06d}.txt').reshape(2,64) for i in ids]);tensor(p/'annotation/keypoints/xyz.pt',joints[:,:,1:].reshape(N,2,21,3).astype(np.float32));tensor(p/'annotation/keypoints/valid_flag.pt',joints[:,:,0]==1)
+ rgb=[np.asarray(Image.open(src/'rgb'/f'{i:06d}.png').convert('RGB')) for i in ids]
+ depth=[np.asarray(Image.open(src/'depth'/f'{i:06d}.png'),dtype=np.uint16) for i in ids]
+ m['rgb']=b.video(p/'observation/rgb.mkv',rgb);m['depth']={**b.video(p/'observation/depth.mkv',depth),'unit':'mm','scale_to_meters':0.001,'invalid_value':0,'rgb_registration':'not_tested'}
+ m['missing_modalities']={'object':'not present in selected historical subset','segmentation':'no native mask in selected subset'}
+ # Independent check: 3D keypoint projections must be finite where the source flags valid.
+ xyz=joints[:,:,1:].reshape(N,2,21,3);projection=xyz@np.array([[fx,0,cx],[0,fy,cy],[0,0,1]]).T
+ uv=projection[...,:2]/projection[...,2:];assert np.isfinite(uv[joints[:,:,0]==1]).all()
+ image=Image.fromarray(rgb[0]);draw=ImageDraw.Draw(image)
+ for side in range(2):
+  for u,v in uv[0,side]:draw.ellipse((u-3,v-3,u+3,v+3),fill=('red' if side==0 else 'cyan'))
+ image.save(b.out/'doc/keypoints_overlay.png')
+ return b.finish(['MANO geometry not rendered; model asset unavailable. RGB-depth registration not tested. No native segmentation or object input in subset.'])
+
+def visor():
+ b=Build('VISOR');raw=b.base/'raw/data/GroundTruth-SparseAnnotations';jp=raw/'annotations/train/P01_03.json';zp=raw/'rgb_frames/train/P01/P01_03.zip'
+ source=load(jp);annotations=source['video_annotations'][:N]
+ ids=[x['image']['name'] for x in annotations];p,m=b.view('P01_03','ego',ids,split='train',view_type='ego',time_semantics='sparse_annotations; use source frame IDs; MKV timestamps are storage indices')
+ rgbs=[];masks=[];records=[]
+ with zipfile.ZipFile(zp) as z:
+  assert z.testzip() is None
+  for i,frame in enumerate(annotations):
+   rgb=np.asarray(Image.open(io.BytesIO(z.read(frame['image']['name']))).convert('RGB'));rgbs.append(rgb)
+   mask=np.zeros((1080,1920),np.uint16);entities=[]
+   for value,entity in enumerate(frame['annotations'],1):
+    # Same polygon rasterization primitive and integer conversion as official VISOR-VIS.
+    polygons=[np.asarray(poly,dtype=np.int32) for poly in entity['segments'] if len(poly)>=3]
+    if polygons:cv2.fillPoly(mask,polygons,int(value))
+    entities.append({'mask_value':value,'instance_id':entity['id'],'track_id':None,'source':{k:v for k,v in entity.items() if k!='segments'}})
+   if mask.shape!=rgb.shape[:2]:mask=cv2.resize(mask,(rgb.shape[1],rgb.shape[0]),interpolation=cv2.INTER_NEAREST)
+   masks.append(mask);records.append({'frame_index':i,'source_frame_id':frame['image']['name'],'subsequence':frame['image']['subsequence'],'annotation_valid':True,'instances':entities})
+   assert set(np.unique(mask)) <= {0,*[x['mask_value'] for x in entities]}
+ m['rgb']=b.video(p/'observation/rgb.mkv',rgbs);m['segmentation']={**b.video(p/'annotation/segmentation/mask.mkv',masks),'provenance':'native_sparse_polygon_rasterized','background_value':0,'background_semantics':'unlabeled/background; source is not globally exhaustive','ignore_value':None,'polygon_canvas':[1920,1080],'overlap_policy':'source annotation order; last polygon wins','track_id':'unknown; no heuristic tracking'}
+ lines(p/'annotation/segmentation/instances.jsonl',records)
+ m['missing_modalities']={'mano':'not supplied','depth':'not supplied','camera':'not supplied','object_pose_mesh':'not supplied'}
+ im=rgbs[2].copy();mask=masks[2];colors=np.array([[0,0,0],[240,70,70],[70,240,90],[70,100,240],[230,190,60],[150,80,200]],np.uint8)
+ used=mask>0;im[used]=(im[used]*0.55+colors[mask[used]%len(colors)]*0.45).astype(np.uint8)
+ Image.fromarray(im).save(b.out/'doc/segmentation_overlay.png')
+ b.checks.append({'test':'official_zip_crc_all_members','passed':True,'members':len(zipfile.ZipFile(zp).namelist())})
+ b.checks.append({'test':'all_mask_values_resolve_to_frame_instances','passed':True})
+ return b.finish([])
+
+def rhd():
+ b=Build('RHD');src=OLD/'RHD/RHD_published_v2/evaluation';ann=read_pickle(b.copy(src/'anno_evaluation.pickle'))
+ for i in range(N):
+  p,m=b.view(f'evaluation_{i:05d}','camera',[i],split='evaluation',time_semantics='independent_image')
+  rgb=np.asarray(Image.open(b.copy(src/'color'/f'{i:05d}.png')).convert('RGB'))
+  depth=np.asarray(Image.open(b.copy(src/'depth'/f'{i:05d}.png')).convert('RGB'),dtype=np.uint16)
+  code=depth[:,:,0]*256+depth[:,:,1]
+  m['rgb']=b.video(p/'observation/rgb.mkv',[rgb]);m['depth']={**b.video(p/'observation/depth.mkv',[code]),'encoding':'source R*256+G','metric_unit':None,'metric_conversion_status':'not_tested; preserved integer code only'}
+  tensor(p/'camera/intrinsic.pt',ann[i]['K']);tensor(p/'annotation/keypoints/xyz.pt',ann[i]['xyz'].reshape(1,2,21,3));tensor(p/'annotation/keypoints/uv.pt',ann[i]['uv_vis'][:,:2].reshape(1,2,21,2));tensor(p/'annotation/keypoints/visible.pt',(ann[i]['uv_vis'][:,2]==1).reshape(1,2,21))
+  m['keypoints']={'unit':'m','layout':'source 42-joint order reshaped as 2x21','hand_order':['left','right'],'joint_order':'official RHD: wrist then thumb/index/middle/ring/pinky, fingers tip-to-palm','definition_reference':'raw/doc/dataset_README.txt','visibility':'per joint; invisible does not mean unannotated'}
+  m['missing_modalities']={'mano':'not supplied by source','segmentation':'omitted by historical downloader','extrinsic':'not supplied in selected annotations'}
+ return b.finish(['Depth metric conversion pending; segmentation absent from historical input.'])
+
+def freihand():
+ b=Build('FreiHAND');src=OLD/'FreiHAND';K=load(b.copy(src/'training_K.json'));mano=load(b.copy(src/'training_mano.json'));scale=load(b.copy(src/'training_scale.json'))
+ for base in range(4):
+  for variant in range(4):
+   i=base+variant*32560;p,m=b.view(f'training_{base:08d}',f'variant_{variant}',[i],split='training',time_semantics='independent_image',base_sample_id=str(base),variant_id=variant)
+   m['camera_group_semantics']='image_variant_not_physical_camera';m['physical_camera_id']=None
+   rgb=np.asarray(Image.open(b.copy(src/'training/rgb'/f'{i:08d}.jpg')).convert('RGB'));m['rgb']=b.video(p/'observation/rgb.mkv',[rgb])
+   tensor(p/'camera/intrinsic.pt',np.asarray(K[base],np.float32));tensor(p/'annotation/mano/source_theta.pt',np.asarray(mano[base],np.float32));tensor(p/'annotation/mano/source_scale.pt',np.asarray(scale[base],np.float32))
+   m['mano']={'source_theta_layout':['pose48','shape10','uv_root2','scale1'],'motion_status':'not_generated','reason':'To recover MANO translation, official code subtracts posed joint 9 from backprojected root. Requires correct MANO model; last three source values are not translation.'}
+   m['missing_modalities']={'depth':'not supplied','segmentation':'not present in historical input','extrinsic':'not supplied','normalized_mano':'model dependency unresolved'}
+ return b.finish(['Preserved source MANO theta; no misleading motion.pt produced. Correct model needed for normalized translation.'])
+
+def hoi4d():
+ b=Build('HOI4D');rel=Path('ZY20210800001/H1/C20/N18/S286/s01/T1');src=OLD/'HOI4D/Hand_pose/handpose_left_hand'/rel
+ files=sorted(src.glob('*.pickle'),key=lambda p:int(p.stem))[:N];ids=[x.stem for x in files]
+ p,m=b.view('H1_C20_N18_S286_s01_T1','ZY20210800001',ids,view_type='ego',time_semantics='annotation_frame_ids_only')
+ a=np.full((len(ids),2,61),np.nan,np.float32);v=np.zeros((len(ids),2),bool);uv=np.full((len(ids),2,21,2),np.nan,np.float32)
+ for i,frame in enumerate(ids):
+  for side,folder in enumerate(['handpose_left_hand','handpose_right_hand']):
+   f=OLD/'HOI4D/Hand_pose'/folder/rel/(frame+'.pickle')
+   if not f.exists():continue
+   x=read_pickle(b.copy(f));a[i,side]=np.concatenate([x['poseCoeff'].reshape(-1),x['beta'].reshape(-1),x['trans'].reshape(-1)]);v[i,side]=True;uv[i,side]=x['kps2D']
+ tensor(p/'annotation/mano/motion.pt',a);tensor(p/'annotation/mano/valid_flag.pt',v);tensor(p/'annotation/keypoints/uv.pt',uv)
+ tensor(p/'camera/intrinsic.pt',np.load(b.copy(OLD/'HOI4D/camera_params/ZY20210800001/intrin.npy'),allow_pickle=False))
+ m['mano']={'layout':['pose_axis_angle48','betas10','translation3'],'slots':['left','right'],'translation_unit':'m','coordinate_frame':'camera','flat_hand_mean':True,'valid_flag':'source file present and parsed; not visibility','geometry_validation':'not_tested'}
+ m['missing_modalities']={'rgb':'not downloaded','depth':'not downloaded','segmentation':'not downloaded','extrinsic':'not available in selected subset','object':'not available in selected subset'}
+ return b.finish(['Annotation-only mini; no matching RGB/depth/segmentation downloaded. MANO geometry not rendered.'])
+
+def reinterhand():
+ b=Build('ReInterHand');src=next(p for p in sorted((OLD/'ReInterHand').glob('m--*')) if (p/'mano_fits/params').exists());files=sorted((src/'mano_fits/params').glob('*_left.json'))[:N];ids=[x.stem.rsplit('_',1)[0] for x in files]
+ # Capture-level poses have no established camera transform in current local subset.
+ p,m=b.view(src.name,'source_unresolved',ids,time_semantics='capture_annotations_only; not a camera observation')
+ rows=[]
+ for frame in ids:
+  row={'source_frame_id':frame,'hands':{}}
+  for side in ['left','right']:
+   f=src/'mano_fits/params'/f'{frame}_{side}.json'
+   if f.exists():row['hands'][side]=load(b.copy(f))
+  rows.append(row)
+ lines(p/'annotation/mano/source_parameters.jsonl',rows)
+ m['missing_modalities']={'rgb':'not extracted from incomplete archive set','camera':'not available','normalized_mano':'source-to-camera mapping unresolved'}
+ m['not_a_real_camera']=True
+ return b.finish(['Source-only annotation sample. source_unresolved is an explicit staging group, not an invented camera.'])
+
+def main():
+ parser=argparse.ArgumentParser();parser.add_argument('--dataset',nargs='*',default=['VISOR','H2O','RHD','FreiHAND','HOI4D','ReInterHand']);args=parser.parse_args()
+ funcs={'VISOR':visor,'H2O':h2o,'RHD':rhd,'FreiHAND':freihand,'HOI4D':hoi4d,'ReInterHand':reinterhand}
+ for name in args.dataset:
+  try:funcs[name]()
+  except Exception as e:
+   save(ROOT/name/'pilot-v0/doc/build_error.json',{'error':str(e)});raise
+if __name__=='__main__':main()
diff --git a/ARCTIC/shared-v1.0/script/pilot_fetch.py b/ARCTIC/shared-v1.0/script/pilot_fetch.py
new file mode 100644
index 0000000000000000000000000000000000000000..7671f83ac7a97438b95528fe9c13c00768eb7861
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/pilot_fetch.py
@@ -0,0 +1,63 @@
+"""Download bounded official pilot inputs; keep archives intact. No upstream code execution."""
+from pathlib import Path
+import concurrent.futures, hashlib, json, requests
+
+ROOT = Path(__file__).resolve().parents[1] / 'hand-datasets'
+REFS = {
+ 'Dex-YCB': ('NVlabs/dex-ycb-toolkit', ['README.md','dex_ycb_toolkit/dex_ycb.py','dex_ycb_toolkit/sequence_loader.py']),
+ 'VISOR': ('epic-kitchens/VISOR-VIS', ['README.md','vis.py']),
+ 'H2O': ('taeinkwon/h2odataset', ['README.md','download_script.py']),
+ 'HOI4D': ('leolyliu/HOI4D-Instructions', ['README.md','utils/parse_2Dmask_img.py','utils/color_map.py']),
+ 'ARCTIC': ('zc-alexfan/arctic', ['docs/data/README.md','docs/data/data_doc.md']),
+ 'FreiHAND': ('lmb-freiburg/freihand', ['README.md','utils/model.py','utils/fh_utils.py','view_samples.py']),
+ 'RHD': ('lmb-freiburg/hand3d', ['README.md','create_binary_db.py']),
+ 'HOT3D': ('facebookresearch/hot3d', ['README.md','hot3d/data_loaders/loader_hand_poses.py']),
+ 'OakInk2': ('oakink/OakInk2', ['README.md']),
+ 'ReInterHand': ('facebookresearch/InterWild', ['tool/ReInterHand/download/download_checksum_framelist.py','tool/ReInterHand/download/download_ego_per_frame.py','tool/ReInterHand/download/download_mano_fits.py']),
+}
+BASE='https://data.bris.ac.uk/datasets/2v6cgv1x04ol22qp9rm9x2j6a7/'
+
+def sha(p):
+ h=hashlib.sha256()
+ with p.open('rb') as f:
+  for b in iter(lambda:f.read(1024*1024),b''):h.update(b)
+ return h.hexdigest()
+
+def fetch(dataset, url, relative, limit=30_000_000):
+ p=ROOT/dataset/'raw'/relative;p.parent.mkdir(parents=True,exist_ok=True)
+ receipt=p.with_name(p.name+'.receipt.json')
+ if p.exists() and receipt.exists():
+  info=json.loads(receipt.read_text()); assert sha(p)==info['sha256'];return info
+ part=p.with_name(p.name+'.part')
+ with requests.get(url,stream=True,headers={'Accept-Encoding':'identity'},timeout=(15,40)) as r:
+  r.raise_for_status(); expected=int(r.headers.get('Content-Length',0))
+  if expected>limit:raise ValueError(f'File exceeds pilot limit: {expected}')
+  size=0
+  with part.open('wb') as f:
+   for chunk in r.iter_content(1024*1024):
+    size+=len(chunk)
+    if size>limit:raise ValueError('Stream exceeds pilot limit')
+    f.write(chunk)
+  if expected and not r.headers.get('Content-Encoding') and size!=expected:raise ValueError('Incomplete transfer')
+ part.replace(p)
+ info={'url':url,'path':str(p.relative_to(ROOT/dataset)), 'bytes':size,'sha256':sha(p),'official_checksum':None,'verification':'HTTP success; content length where supplied; locally computed SHA256'}
+ receipt.write_text(json.dumps(info,indent=2),encoding='utf-8');return info
+
+def reference(item):
+ name,(repo,paths)=item;results=[]
+ for section in ['raw/doc','raw/script','raw/data','pilot-v0/doc','pilot-v0/script','pilot-v0/data']:(ROOT/name/section).mkdir(parents=True,exist_ok=True)
+ for path in paths:
+  try:results.append(fetch(name,f'https://raw.githubusercontent.com/{repo}/HEAD/{path}',('script/' if path.endswith('.py') else 'doc/')+path))
+  except Exception as e:results.append({'path':path,'error':str(e)})
+ return name,results
+
+def main():
+ results=dict(concurrent.futures.ThreadPoolExecutor(6).map(reference,REFS.items()))
+ for path in ['GroundTruth-SparseAnnotations/annotations/train/P01_03.json','GroundTruth-SparseAnnotations/rgb_frames/train/P01/P01_03.zip','EPIC_100_noun_classes_v2.csv','frame_mapping.json']:
+  try:results['VISOR'].append(fetch('VISOR',BASE+path,'data/'+path))
+  except Exception as e:results['VISOR'].append({'path':path,'error':str(e)})
+ for name,rows in results.items():
+  (ROOT/name/'raw/doc/download_manifest.json').write_text(json.dumps(rows,indent=2),encoding='utf-8')
+  print(name,[(r.get('path'),r.get('bytes',r.get('error'))) for r in rows],flush=True)
+
+if __name__=='__main__':main()
diff --git a/ARCTIC/shared-v1.0/script/pilot_hot_normalize.py b/ARCTIC/shared-v1.0/script/pilot_hot_normalize.py
new file mode 100644
index 0000000000000000000000000000000000000000..3b224e4d405082be63eab19b8461328423626a66
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/pilot_hot_normalize.py
@@ -0,0 +1,39 @@
+from pilot_mano import *
+
+def transform(q,t):
+ q=np.asarray(q,float);q=q/np.linalg.norm(q);w,x,y,z=q
+ R=np.array([[1-2*(y*y+z*z),2*(x*y-z*w),2*(x*z+y*w)],[2*(x*y+z*w),1-2*(x*x+z*z),2*(y*z-x*w)],[2*(x*z-y*w),2*(y*z+x*w),1-2*(x*x+y*y)]])
+ T=np.eye(4);T[:3,:3]=R;T[:3,3]=t;return T
+def csv_transform(row):return transform([float(row['q_wo_'+a]) for a in ['w','x','y','z']],[float(row[f't_wo_{a}[m]']) for a in ['x','y','z']])
+def main():
+ base=ROOT/'HOT3D/pilot-v0';p=next((base/'data').rglob('metainfo.json')).parent;m=load(p/'metainfo.json');frames=[json.loads(l) for l in (p/'frames.jsonl').read_text().splitlines()];source=OLD/'HOT3D/P0001_10a27bf7'
+ camera=next(c for c in load(source/'camera_models.json') if c['stream_id']=='214-1');dc=camera['T_Device_Camera'];Tdc=transform(dc['quaternion_wxyz'],dc['translation_xyz'])
+ device={int(r['timestamp[ns]']):csv_transform(r) for r in csv.DictReader((source/'headset_trajectory.csv').open())}
+ Tcws=np.stack([np.linalg.inv(device[f['timecode_ns']]@Tdc) for f in frames]);tensor(p/'camera/extrinsic.pt',Tcws.astype(np.float32));tensor(p/'camera/intrinsic.pt',np.array(camera['projectionParams'],np.float32))
+ m['camera']={'projection_model':'FISHEYE624','intrinsic_layout':'15 native projectionParams, not a 3x3 K','width':camera['imageWidth'],'height':camera['imageHeight'],'extrinsic':'T_camera_from_world','unit':'m','calibration_source':'static camera_models.json + exact timecode headset trajectory; online calibration not substituted'}
+ models=[Mano('left'),Mano('right')]
+ # Match the explicitly documented fix in HOT3D official mano_layer.py.
+ fixed=np.abs(models[0].data['shapedirs'][:,0]-models[1].data['shapedirs'][:,0]).sum()<1
+ if fixed:
+  models[0].data['shapedirs'][:,0]*=-1
+  with torch.no_grad():models[0].layer.shapedirs[:,0]*=-1
+ rows=[json.loads(l) for l in (p/'annotation/mano/source_parameters.jsonl').read_text().splitlines()];motion=np.full((len(rows),2,61),np.nan,np.float32);valid=np.zeros((len(rows),2),bool);errors=[]
+ for i,row in enumerate(rows):
+  for side,hand in row['source_annotation']['hand_poses'].items():
+   side=int(side);model=models[side];beta=np.array(hand['betas']);wrist=hand['wrist_xform'];Tw=transform(wrist['q_wxyz'],wrist['t_xyz']);pose=np.r_[cv2.Rodrigues(Tw[:3,:3])[0].ravel(),np.array(hand['pose'])@model.data['hands_components'][:15]+model.data['hands_mean']]
+   world_vertices,_,_=model.geometry(pose,beta,Tw[:3,3]);root=model.data['J_regressor'][0]@(model.data['v_template']+np.einsum('vci,i->vc',model.data['shapedirs'],beta))
+   Tc=Tcws[i];camera_pose=pose.copy();camera_pose[:3]=cv2.Rodrigues(Tc[:3,:3]@Tw[:3,:3])[0].ravel();camera_t=Tc[:3,:3]@(root+Tw[:3,3])+Tc[:3,3]-root
+   camera_vertices,_,_=model.geometry(camera_pose,beta,camera_t);transformed=world_vertices@Tc[:3,:3].T+Tc[:3,3];error=float(np.linalg.norm(camera_vertices-transformed,axis=-1).max());errors.append(error);assert error<1e-5
+   motion[i,side]=np.r_[camera_pose,beta,camera_t];valid[i,side]=True
+ tensor(p/'annotation/mano/motion.pt',motion);tensor(p/'annotation/mano/valid_flag.pt',valid)
+ m['mano']={'layout':['pose_axis_angle48','betas10','translation3'],'slots':['left','right'],'coordinate_frame':'camera','translation_unit':'m','flat_hand_mean':True,'model_version':'MANO v1.2','left_shapedirs_fix':bool(fixed),'conversion':'official 15-PCA expansion; world-to-camera rotation and shape-dependent root correction','valid_flag':'source annotation present, not visibility'}
+ objects=[json.loads(l) for l in (p/'annotation/object/source_world_poses.jsonl').read_text().splitlines()];object_ids=sorted({r['object_uid'] for r in objects})
+ for oid in object_ids:
+  lookup={int(r['timestamp[ns]']):r for r in objects if r['object_uid']==oid};poses=np.full((len(frames),4,4),np.nan,np.float32);flags=np.zeros(len(frames),bool)
+  for i,frame in enumerate(frames):
+   if frame['timecode_ns'] in lookup:poses[i]=Tcws[i]@csv_transform(lookup[frame['timecode_ns']]);flags[i]=True
+  tensor(p/'annotation/object'/oid/'pose.pt',poses);tensor(p/'annotation/object'/oid/'valid_flag.pt',flags)
+ m['object']={'pose':'T_camera_from_object','unit':'m','mesh_status':'not downloaded','object_ids':object_ids};save(p/'metainfo.json',m)
+ save(base/'doc/mano_validation.json',{'test':'camera-space vertices equal transformed world-space vertices','max_vertex_error_m':max(errors),'passed':True,'limitation':'algebraic transform check; no independent fisheye reprojection evaluation','left_shapedirs_fix':bool(fixed)})
+ print('HOT3D normalized',motion.shape,'max transform error',max(errors))
+if __name__=='__main__':main()
diff --git a/ARCTIC/shared-v1.0/script/pilot_mano.py b/ARCTIC/shared-v1.0/script/pilot_mano.py
new file mode 100644
index 0000000000000000000000000000000000000000..d4596f225a22b61f642a2ef0193588172a5b581e
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/pilot_mano.py
@@ -0,0 +1,87 @@
+"""Local licensed MANO model dependency; never copies models into shared outputs."""
+from pilot_build import *
+from types import SimpleNamespace
+import smplx
+
+class StoredState:
+ def __setstate__(self,state):self.__dict__.update(state)
+class ModelReader(Restricted):
+ def find_class(self,module,name):
+  if (module,name) in [('chumpy.ch','Ch'),('chumpy.reordering','Select'),('scipy.sparse.csc','csc_matrix')]:return StoredState
+  if (module,name)==('__builtin__','set'):return set
+  return super().find_class(module,name)
+
+def model_data(side):
+ p=WORK/'tmp/pilot-models'/f'MANO_{side.upper()}.pkl'
+ with p.open('rb') as f:d=ModelReader(f,encoding='latin1').load()
+ # Materialize only the two known serialized structures; execute no Chumpy code.
+ s=d['shapedirs']
+ if isinstance(s,StoredState):d['shapedirs']=s.a.x.reshape(-1)[s.idxs].reshape(s.preferred_shape)
+ sparse=d['J_regressor'];dense=np.zeros(sparse._shape)
+ for col in range(sparse._shape[1]):
+  a,z=sparse.indptr[col:col+2];dense[sparse.indices[a:z],col]=sparse.data[a:z]
+ d['J_regressor']=dense
+ assert d['shapedirs'].shape==(778,3,10) and dense.shape==(16,778)
+ return d
+
+class Mano:
+ def __init__(self,side):
+  self.side=side
+  self.data=model_data(side)
+  self.layer=smplx.MANO(model_path=str(WORK/'tmp/pilot-models'),data_struct=SimpleNamespace(**self.data),is_rhand=side=='right',use_pca=False,flat_hand_mean=True)
+ def geometry(self,pose,betas,translation=None,keypoint_convention='freihand'):
+  pose=torch.as_tensor(np.asarray(pose).copy(),dtype=torch.float32).reshape(-1,48);betas=torch.as_tensor(np.asarray(betas).copy(),dtype=torch.float32).reshape(-1,10)
+  with torch.no_grad():o=self.layer(global_orient=pose[:,:3],hand_pose=pose[:,3:],betas=betas)
+  v=o.vertices.numpy();j=o.joints.numpy()[:,:16]
+  if translation is not None:
+   t=np.asarray(translation).reshape(-1,1,3);v=v+t;j=j+t
+  # Standard wrist + thumb,index,middle,ring,pinky proximal-to-distal order.
+  coords=[]
+  for k in [0,13,14,15,1,2,3,4,5,6,10,11,12,7,8,9]:coords.append(j[:,k])
+  out=np.stack([j[:,0],j[:,13],j[:,14],j[:,15],v[:,744],j[:,1],j[:,2],j[:,3],v[:,320],j[:,4],j[:,5],j[:,6],v[:,443],j[:,10],j[:,11],j[:,12],v[:,555],j[:,7],j[:,8],j[:,9],v[:,672]],axis=1)
+  if keypoint_convention=='manopth':out[:,[4,8,12,16,20]]=v[:,[745,317,444 if self.side=='right' else 445,556,673]]
+  return v,out,j
+
+def freihand_normalize():
+ base=ROOT/'FreiHAND/pilot-v0';model=Mano('right');reports=[]
+ for meta_file in sorted((base/'data').rglob('metainfo.json')):
+  p=meta_file.parent;m=load(meta_file);theta=torch.load(p/'annotation/mano/source_theta.pt',weights_only=True).numpy().reshape(1,61)
+  K=torch.load(p/'camera/intrinsic.pt',weights_only=True).numpy();focal=(K[0,0]+K[1,1])/2
+  root=np.r_[(theta[0,58:60]-K[:2,2]),focal]/theta[0,60]
+  pose=theta[:,:48].copy();pose[:,3:]+=model.data['hands_mean']
+  _,_,j=model.geometry(pose,theta[:,48:58]);translation=root-j[0,4]
+  motion=np.full((1,2,61),np.nan,np.float32);motion[0,1]=np.r_[pose[0],theta[0,48:58],translation]
+  tensor(p/'annotation/mano/motion.pt',motion);tensor(p/'annotation/mano/valid_flag.pt',[[False,True]])
+  _,keypoints,_=model.geometry(pose,theta[:,48:58],translation)
+  assert np.allclose(keypoints[0,9],root,atol=1e-6)
+  m['mano']={'layout':['pose_axis_angle48','betas10','translation3'],'slots':['left','right'],'coordinate_frame':'camera','translation_unit':'m','flat_hand_mean':True,'model_version':'MANO v1.2','conversion':'source non-flat mean added to finger pose; official orthographic root recovery minus posed MANO joint 4 (21-joint index9)','valid_flag':'right-hand source annotation available; left not supplied'}
+  m['missing_modalities'].pop('normalized_mano',None);save(meta_file,m)
+  reports.append({'path':str(p.relative_to(base)),'root_recovery_error_m':float(np.linalg.norm(keypoints[0,9]-root))})
+ save(base/'doc/mano_validation.json',{'status':'passed_algebraic_root_check','model_sha256':sha(WORK/'tmp/pilot-models/MANO_RIGHT.pkl'),'samples':reports,'limit':'No independent 3D ground truth currently in input; algebraic check does not prove fit accuracy.'})
+ print('FreiHAND normalized',len(reports))
+
+def validate_mano(name):
+ base=ROOT/name/'pilot-v0';models=[Mano('left'),Mano('right')];summaries=[]
+ for mp in (base/'data').rglob('metainfo.json'):
+  p=mp.parent;motionp=p/'annotation/mano/motion.pt'
+  if not motionp.exists():continue
+  a=torch.load(motionp,weights_only=True).numpy();valid=torch.load(p/'annotation/mano/valid_flag.pt',weights_only=True).numpy();K=torch.load(p/'camera/intrinsic.pt',weights_only=True).numpy();pred=np.full((len(a),2,21,3),np.nan,np.float32)
+  for side,model in enumerate(models):
+   ix=np.flatnonzero(valid[:,side])
+   if len(ix):_,pred[ix,side],_=model.geometry(a[ix,side,:48],a[ix,side,48:58],a[ix,side,58:],keypoint_convention='manopth')
+  tensor(p/'annotation/mano/derived_keypoints.pt',pred)
+  uv=pred@K.T;uv=uv[...,:2]/uv[...,2:]
+  row={'sequence':str(p.relative_to(base)),'keypoint_convention':'official manopth, including handedness-specific fingertip vertices','finite_valid_geometry':bool(np.isfinite(pred[valid]).all())}
+  target=p/('annotation/keypoints/xyz.pt' if name=='H2O' else 'annotation/keypoints/uv.pt')
+  if target.exists():
+   gt=torch.load(target,weights_only=True).numpy();err=np.linalg.norm((pred if name=='H2O' else uv)-gt,axis=-1);good=np.isfinite(err)&valid[:,:,None]
+   row['mean_error']=float(err[good].mean());row['max_error']=float(err[good].max());row['unit']='m' if name=='H2O' else 'pixels'
+  summaries.append(row)
+  meta=load(p/'metainfo.json');meta['mano']['geometry_validation']='measured against source keypoints; see doc/mano_validation.json';meta['mano']['model_version']='MANO v1.2'
+  meta['mano']['derived_keypoints_convention']='official manopth 21 joints, including side-specific fingertip vertices'
+  save(p/'metainfo.json',meta)
+ save(base/'doc/mano_validation.json',{'status':'measured_not_auto_accepted','samples':summaries,'model_sha256':{s:sha(WORK/'tmp/pilot-models'/f'MANO_{s.upper()}.pkl') for s in ['left','right']}})
+ print(name,summaries)
+
+if __name__=='__main__':
+ freihand_normalize();validate_mano('H2O');validate_mano('HOI4D')
diff --git a/ARCTIC/shared-v1.0/script/preprocess_fresh_full.py b/ARCTIC/shared-v1.0/script/preprocess_fresh_full.py
new file mode 100644
index 0000000000000000000000000000000000000000..31a051dc543bdb9cd4e89b7cd8aa0ebdeb7e1b1a
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/preprocess_fresh_full.py
@@ -0,0 +1,354 @@
+"""Authorized ARCTIC/HOT3D production queue. One heavy converter, immutable raw.
+
+Checkpoints and logs live under dataset/tmp; distributable validation lives in doc.
+No upload or raw deletion. All output is independently checked before completion.
+"""
+from fresh_sequence_common import *
+from convert_fresh_sequence_pilots import arctic,hot3d,verify_hot3d_frame_order_and_metadata
+from review_fresh_sequence_pilots import validate_view,arr
+from apply_hot3d_qa import apply_qa
+from diagnose_arctic_review import official_object
+from datetime import datetime
+import os,ctypes,msvcrt,traceback,html,random
+import fresh_sequence_common as common
+from accelerated_media import POLICY as MEDIA_POLICY,revalidate as revalidate_media
+
+JOB='full-preprocess-v1.0'
+CONTROL=WORK/'tmp'/JOB
+RESERVE=400_000_000_000
+SCRIPT=Path(__file__).resolve()
+TEST_IDS={'P0004','P0005','P0006','P0008','P0016','P0020'}
+torch.set_num_threads(4)
+
+def now():return datetime.now().astimezone().isoformat()
+
+def atomic(p,value):
+    p.parent.mkdir(parents=True,exist_ok=True);q=p.with_name(p.name+'.writing')
+    q.write_text(json.dumps(value,ensure_ascii=False,indent=2,allow_nan=False),encoding='utf-8');os.replace(q,p)
+
+def temporary(ds):return DATA_ROOT/ds/'tmp'/JOB
+
+PRODUCTION_SCRIPTS=['preprocess_fresh_full.py','accelerated_media.py','convert_fresh_sequence_pilots.py','apply_hot3d_qa.py','review_fresh_sequence_pilots.py','render_full_sequence_review.py','fresh_sequence_common.py','diagnose_arctic_review.py','pilot_build.py','pilot_fetch.py','pilot_mano.py','sequence_pilot_common.py','sequence_pilot_world.py','pilot_hot_normalize.py','audit_source_fields.py']
+
+def code_manifest():
+    return [{'path':'script/'+f,'sha256':digest(WORK/'scripts'/f)} for f in PRODUCTION_SCRIPTS]
+
+def verify_completed(ds,row,source):
+    root=DATA_ROOT/ds/RELEASE;record=temporary(ds)/'completed'/f'{row["sequence_id"]}.json';done=load(record)
+    migration_file=temporary(ds)/'acceleration-migrations'/f'{row["sequence_id"]}.json'
+    migration=load(migration_file) if migration_file.exists() else None
+    migration_matches=bool(migration and migration['checkpoint_sha256']==digest(record) and migration['new_scripts']==code_manifest() and migration['passed'] and migration['source']==source)
+    assert done['source']==source or migration_matches,'Completed source identity changed without verified migration'
+    if done.get('scripts')!=code_manifest():
+        assert migration_matches,'Code changed without a matching successful explicit revalidation'
+    verified={f['path']:f for f in migration['verified_files']} if migration_matches else {}
+    for f in done['files']:
+        path=root/f['path'];stat=path.stat();assert stat.st_size==f['size'],path
+        cached=verified.get(f['path'])
+        if not cached or cached['mtime_ns']!=stat.st_mtime_ns or cached['sha256']!=f['sha256']:
+            assert digest(path)==f['sha256'],path
+    return done
+
+def migrate_completed(ds):
+    """Revalidate existing outputs once; preserve historical code and full-pixel proof."""
+    rows={r['sequence_id']:r for r in load(CONTROL/'plan.json')['datasets'][ds]['sequences']};records=inputs(ds)
+    for checkpoint in sorted((temporary(ds)/'completed').glob('*.json')):
+        done=load(checkpoint);row=rows[done['sequence_id']];source_files=list(row['inputs'])
+        if ds=='HOT3D':source_files+=sorted(rel for rel in records if rel.startswith('assets/'))
+        source=[check_input(ds,rel,records) for rel in source_files]
+        # Older smoke records did not yet list assets; validate their recorded inputs
+        # and add the verified complete dependency list to this migration record.
+        by_path={x['file']:x for x in source};assert all(by_path[x['file']]==x for x in done['source'])
+        legacy=load(CONTROL/'acceleration/legacy-script-manifest.json')['files']
+        assert done.get('scripts') in [None,legacy,code_manifest()],'Unknown historical code revision'
+        views=[]
+        for camera in row['cameras']:
+            _,_,report=validate_view(DATA_ROOT/ds/RELEASE/'data'/row['sequence_id']/camera);views.append(report)
+        verified=[];root=DATA_ROOT/ds/RELEASE
+        for file in done['files']:
+            path=root/file['path'];stat=path.stat();assert stat.st_size==file['size'] and digest(path)==file['sha256'],path
+            verified.append(dict(file,mtime_ns=stat.st_mtime_ns))
+        result={'passed':True,'dataset':ds,'sequence':row['sequence'],'checked_at':now(),'checkpoint_sha256':digest(checkpoint),'new_scripts':code_manifest(),'source':source,'verified_files':verified,'view_checks':views,'numerical_proof':'Every historical output byte/digest rechecked and current schema/validity validators rerun; unchanged source geometry proofs retained; no relabelling or tensor conversion','media_proof':'Historical complete JPEG and RGB equality proof retained for byte-identical files'}
+        atomic(temporary(ds)/'acceleration-migrations'/checkpoint.name,result)
+        atomic(root/'doc/validation/full/acceleration-revalidation'/checkpoint.name,result)
+        print('MIGRATED_VERIFIED',ds,row['sequence_id'],len(verified),flush=True)
+
+def snapshot_scripts(ds):
+    root=DATA_ROOT/ds/RELEASE
+    for f in PRODUCTION_SCRIPTS:shutil.copyfile(WORK/'scripts'/f,root/'script'/f)
+    shutil.copyfile(CONTROL/'acceleration-config.json',root/'script/acceleration-config.json')
+    atomic(root/'doc/validation/full/script-manifest.json',{'created_at':now(),'scope':'Current production converter and validation scripts; earlier pilot manifests describe their historical run','files':code_manifest()})
+
+def inputs(ds):
+    base=DATA_ROOT/ds
+    name='archive_checksums.json' if ds=='ARCTIC' else 'file_checksums.json'
+    return load(base/'raw/doc'/name)
+
+def check_input(ds,rel,records):
+    p=DATA_ROOT/ds/'raw/data'/rel;r=records[rel];st=p.stat()
+    assert st.st_size==r['size'],f'Raw size changed: {p}'
+    assert r.get('official_sha256_match',r.get('official_digest_match',False))
+    if st.st_mtime_ns!=r['mtime_ns']:
+        assert digest(p)==r['sha256'],f'Raw digest changed: {p}'
+        if p.suffix=='.zip':
+            with zipfile.ZipFile(p) as z:assert z.testzip() is None
+    return {'file':'raw/data/'+rel,'size':st.st_size,'sha256':r['sha256'],'download_checksum_and_crc_record_reused':st.st_mtime_ns==r['mtime_ns']}
+
+def plan():
+    CONTROL.mkdir(parents=True,exist_ok=True);out={'created_at':now(),'order':['ARCTIC','HOT3D'],'datasets':{},'reserve_bytes':RESERVE}
+    for ds in out['order']:
+        base=DATA_ROOT/ds;records=inputs(ds)
+        for rel in records:check_input(ds,rel,records)
+        rows=[];auxiliary=[]
+        if ds=='ARCTIC':
+            with zipfile.ZipFile(base/'raw/data/splits_json.zip') as z:splits=json.loads(z.read('splits_json/protocol_p1.json'))
+            for p in sorted((base/'raw/data/images_zips').rglob('*.zip')):
+                seq=p.relative_to(base/'raw/data/images_zips').with_suffix('').as_posix()
+                if '/' not in seq:
+                    assert seq=='backgrounds',seq
+                    auxiliary.append({'file':'raw/data/images_zips/backgrounds.zip','handling':'retain original archive in raw; background reference photographs are not hand-object capture sequences'})
+                    continue
+                with zipfile.ZipFile(p) as z:
+                    im=[i for i in z.infolist() if re.fullmatch(r'[0-8]/[0-9]+\.jpg',i.filename)]
+                    extra=[i.filename for i in z.infolist() if i.filename.endswith('.jpg') and not re.fullmatch(r'[0-8]/[0-9]+\.jpg',i.filename)]
+                    assert all(n.startswith('__MACOSX/') or Path(n).name.startswith('._') for n in extra),(p,extra[:5])
+                    cams=sorted({i.filename.split('/')[0] for i in im},key=int)
+                    counts={c:sum(i.filename.startswith(c+'/') for i in im) for c in cams}
+                    b=sum(i.file_size for i in im)
+                assert cams,(seq,counts)
+                rows.append({'sequence':seq,'sequence_id':seq.replace('/','_'),'object':seq.split('/')[1].split('_')[0],
+                    'cameras':cams,'frames_per_camera':counts,'jpeg_bytes':b,'split':next((k for k,v in splits.items() if seq in v),'unknown'),
+                    'inputs':['raw_seqs.zip','meta.zip','splits_json.zip','images_zips/'+seq+'.zip']})
+            budget=sum(r['jpeg_bytes'] for r in rows)+160_000_000_000
+        else:
+            for p in sorted((base/'raw/data/Aria').iterdir()):
+                if not p.is_dir():continue
+                gt=next(p.glob('*ground_truth.zip'));hz=next(p.glob('*hand_data.zip'));vrs=next(p.glob('*.vrs'))
+                with zipfile.ZipFile(gt) as z:
+                    m=json.loads(z.read('metadata.json'));cams=json.loads(z.read('camera_models.json'))
+                    assert m['headset']=='Aria'
+                    hasgt=m['have_hand_object_pose_gt'];split='test' if m['participant_id'] in TEST_IDS else 'train';assert hasgt==(split=='train')
+                    if hasgt:assert 'masks/mask_qa_pass.csv' in z.namelist()
+                with zipfile.ZipFile(hz) as z:mano_bytes=z.getinfo('mano_hand_pose_trajectory.jsonl').file_size
+                assert hasgt or mano_bytes==0
+                rows.append({'sequence':p.name,'sequence_id':p.name,'cameras':[c['stream_id'] for c in cams],'split':split,
+                    'source_gt_available':hasgt,'vrs_bytes':vrs.stat().st_size,'inputs':[q.relative_to(base/'raw/data').as_posix() for q in [gt,hz,vrs]]})
+            # Includes persistent extracted JPEGs, current media spool, meshes, tensors and reports.
+            budget=sum(r['vrs_bytes'] for r in rows)*3+150_000_000_000
+        assert len(rows)==(339 if ds=='ARCTIC' else 198)
+        out['datasets'][ds]={'sequences':rows,'count':len(rows),'views':sum(len(r['cameras']) for r in rows),'additional_budget_bytes':budget,'raw_files_verified':len(records),'auxiliary_raw_only':auxiliary}
+        print('PLAN',ds,len(rows),out['datasets'][ds]['views'],budget,flush=True)
+    out['free_bytes_at_plan']=shutil.disk_usage(DATA_ROOT).free
+    out['additional_peak_budget_bytes']=sum(v['additional_budget_bytes'] for v in out['datasets'].values())
+    assert out['free_bytes_at_plan']-out['additional_peak_budget_bytes']>=RESERVE,out
+    atomic(CONTROL/'plan.json',out)
+    return out
+
+def approve():
+    p=load(CONTROL/'plan.json');confirmation={'confirmed_at':now(),'user_statement':'HOT3D和ARCTIC我都确认了，可以开始做预处理了。',
+        'scope':'Reviewed pilots accepted; full preprocessing authorized for ARCTIC and HOT3D Aria only',
+        'full_preprocessing_authorized':True,'full_output_review':'pending','upload_authorized':False,'raw_retention':True}
+    atomic(CONTROL/'authorization.json',confirmation)
+    state=load(WORK/'tmp/dataset-work-state.json')
+    for ds in p['order']:
+        root=DATA_ROOT/ds/RELEASE;doc=root/'doc/validation'
+        atomic(doc/'pilot-human-confirmation.json',confirmation)
+        for rp in [doc/'report.json',*sorted((doc/'objects').glob('*/report.json')),*([doc/'objects/report.json'] if (doc/'objects/report.json').exists() else [])]:
+            r=load(rp);r.update(human_review='approved',full_preprocessing_authorized=True,upload_authorized=False,pilot_confirmation=confirmation)
+            atomic(rp,r)
+        for page in [doc/'index.html',*sorted((doc/'objects').rglob('index.html'))]:
+            if not page.exists():continue
+            s=page.read_text(encoding='utf-8')
+            s=s.replace('人工确认待完成','试点人工确认已通过；全量产物待单独验收').replace('用户人工确认待完成','用户试点确认已通过').replace('用户确认前不全量处理','用户已确认试点，允许全量预处理')
+            if 'id="production-authorization"' not in s:
+                s=s.replace('<header>','<header><p id="production-authorization" style="color:#176b3a"><strong>用户已确认本页试点，授权全量预处理。</strong> 全量完成后另行随机抽查；尚未授权上传。</p>',1)
+            page.write_text(s,encoding='utf-8')
+        if (doc/'objects/artifact-manifest.json').exists():
+            mp=doc/'objects/artifact-manifest.json';a=load(mp)
+            for f in a['files']:
+                file=mp.parent/f['path'];f.update(size=file.stat().st_size,sha256=digest(file))
+            atomic(mp,a)
+        for meta in (root/'data').glob('*/*/metainfo.json'):
+            m=load(meta);m['pilot_human_review']='approved';m['full_preprocessing_authorized']=True
+            if 'acceptance' in m:m['acceptance'].update(status='pilot_user_approved',training_release_approved=False,reason='Pilot confirmed; full outputs and release still require separate review')
+            atomic(meta,m)
+        state['active'][ds].update(stage='full_preprocessing_queued',human_sample_review='approved',full_preprocessing_authorized=True,upload_authorized=False,
+            production_plan=str(CONTROL/'plan.json'),full_sequence_count=p['datasets'][ds]['count'],full_view_count=p['datasets'][ds]['views'])
+        atomic(doc/'status.json',{'stage':'full_preprocessing_queued','pilot_review':'approved','full_output_review':'pending','upload_authorized':False})
+        docprod=doc/'full';docprod.mkdir(parents=True,exist_ok=True)
+        atomic(docprod/'plan.json',p['datasets'][ds])
+        (docprod/'README.md').write_text('# 全量预处理\n\n试点已获用户确认。逐序列转换、逐帧码流/像素和几何校验，完成后另建随机抽查报告。原始包保留，不自动上传。\n\n每个序列的正式自动检查在本目录sequences/；运行日志和进度只在数据集tmp。断点以输入官方摘要及全部输出文件摘要核验。\n\nHOT3D测试集未发布手物姿态真值：保留RGB、相机和物体资产，姿态及可见性缺省。ARCTIC仅保留已确认s09/laptop的曝光排除，其它序列不新增帧号屏蔽。\n',encoding='utf-8')
+        for f in ['preprocess_fresh_full.py','convert_fresh_sequence_pilots.py','apply_hot3d_qa.py','review_fresh_sequence_pilots.py','render_full_sequence_review.py','fresh_sequence_common.py','diagnose_arctic_review.py']:
+            shutil.copyfile(WORK/'scripts'/f,root/'script'/f)
+    atomic(WORK/'tmp/dataset-work-state.json',state)
+
+def official_arctic_check(row,views):
+    base=DATA_ROOT/'ARCTIC';seq=row['sequence'];oid=row['object'];sid=seq.split('/')[0]
+    if row['split']=='test':
+        with zipfile.ZipFile(base/'raw/data/raw_seqs.zip') as z:
+            assert all(f'raw_seqs/{seq}.{x}.npy' not in z.namelist() for x in ['mano','object','egocam.dist'])
+        for p in views:
+            m=load(p/'metainfo.json');assert 'mano' not in m['components'] and not list((p/'annotation/object').glob('*/pose.pt'))
+            if p.name=='0':assert 'intrinsic' not in m['components'] and 'extrinsic' not in m['components'] and m['camera']['schema_id'] is None
+        return {'passed':True,'test':'official_held_out_test_ground_truth_omitted; static_mesh_metres_and_part_ids_checked_per_view','source_gt_available':False}
+    with zipfile.ZipFile(base/'raw/data/meta.zip') as z:
+        pre=f'meta/object_vtemplates/{oid}/';verts=np.array([[float(x) for x in s.split()[1:4]] for s in z.read(pre+'mesh.obj').decode().splitlines() if s.startswith('v ')])/1000
+        parts=np.asarray(json.loads(z.read(pre+'parts.json')),np.int64);params=json.loads(z.read(pre+'object_params.json'));offset=json.loads(z.read('meta/misc.json'))[sid]['ioi_offset']
+    with zipfile.ZipFile(base/'raw/data/raw_seqs.zip') as z:objects=native_npy_bytes(z.read(f'raw_seqs/{seq}.object.npy'))
+    world=[];kp=[]
+    for start in range(0,len(objects),32):
+        v,k=official_object(objects[start:start+32],verts,parts,params,oid);world.append(v);kp.append(k)
+    world=np.concatenate(world);kp=np.concatenate(kp);worst=0.;checked=0
+    for p in views:
+        frames=[json.loads(l) for l in (p/'camera/frames.jsonl').read_text().splitlines()];ix=np.array([int(f['source_frame_id'])-offset for f in frames]);op=p/'annotation/object'/oid
+        E=arr(p/'camera/extrinsic.pt');P=arr(op/'pose.pt');top=arr(op/'parts/top/pose.pt');valid=arr(op/'pose_valid.pt');ids=arr(op/'vertex_part_ids.pt');points=arr(op/'keypoints/xyz.pt')
+        assert np.array_equal(ids,1-parts)
+        for i in np.flatnonzero(valid):
+            actual=verts@P[i,:3,:3].T+P[i,:3,3];sel=ids==1;actual[sel]=verts[sel]@top[i,:3,:3].T+top[i,:3,3]
+            expected=world[ix[i]]@E[i,:3,:3].T+E[i,:3,3];expected_kp=kp[ix[i]]@E[i,:3,:3].T+E[i,:3,3]
+            worst=max(worst,float(np.linalg.norm(actual-expected,axis=-1).max()),float(np.linalg.norm(points[i]-expected_kp,axis=-1).max()));checked+=1
+    assert worst<2e-5,(seq,worst)
+    return {'test':'all_valid_frames_all_vertices_and_keypoints_vs_pinned_official_ObjectTensors','passed':True,'camera_frames':checked,'vertices_per_frame':len(verts),'max_error_m':worst}
+
+def worker(ds,seq):
+    plan=load(CONTROL/'plan.json');row=next(r for r in plan['datasets'][ds]['sequences'] if r['sequence']==seq)
+    assert load(CONTROL/'authorization.json')['full_preprocessing_authorized']
+    base=DATA_ROOT/ds;root=base/RELEASE;tmp=temporary(ds);result=tmp/'completed'/f'{row["sequence_id"]}.json';records=inputs(ds)
+    source_files=list(row['inputs'])
+    if ds=='HOT3D':source_files+=sorted(rel for rel in records if rel.startswith('assets/'))
+    source=[check_input(ds,rel,records) for rel in source_files]
+    if result.exists():
+        verify_completed(ds,row,source)
+        print('VERIFIED_RESUME',ds,seq,flush=True);return
+    assert shutil.disk_usage(base).free>RESERVE+15_000_000_000,'Disk reserve reached'
+    reused={c for c in row['cameras'] if (root/'data'/row['sequence_id']/c/'metainfo.json').exists() and load(root/'data'/row['sequence_id']/c/'metainfo.json').get('pilot_complete')}
+    config=load(CONTROL/'acceleration-config.json');assert config['media_policy']==MEDIA_POLICY
+    conversion_start=time.perf_counter()
+    media_workers=min(config['workers_by_dataset'][ds],len(row['cameras']))
+    print('ACCELERATED_MEDIA',ds,seq,media_workers,MEDIA_POLICY,flush=True)
+    with common.media_pipeline(media_workers):
+        if ds=='ARCTIC':arctic(seq,initialize=False)
+        else:hot3d(seq,initialize=False)
+    conversion_seconds=time.perf_counter()-conversion_start
+    if ds=='HOT3D':
+        verify_hot3d_frame_order_and_metadata(seq)
+        if row['source_gt_available']:apply_qa(seq,publish=False)
+    views=[root/'data'/row['sequence_id']/cam for cam in row['cameras']]
+    results=[]
+    for p in views:
+        m,frames,r=validate_view(p);m.update(pilot=False,production_complete=True,pilot_human_review='approved',full_output_human_review='pending',upload_authorized=False)
+        if p.name in reused:
+            r['reused_media_revalidation']=revalidate_media(p)
+        if ds=='HOT3D' and not row['source_gt_available']:
+            assert 'mano' not in m['components'] and not list((p/'annotation/object').glob('*/pose.pt'))
+        atomic(p/'metainfo.json',m);results.append(r)
+    independent=official_arctic_check(row,views) if ds=='ARCTIC' else {'passed':True,'test':'all_frame_camera_MANO_geometry_PCA_and_source_QA_per_view' if row['source_gt_available'] else 'all_frame_media_camera_geometry_and_omitted_pose_components','source_gt_available':row['source_gt_available']}
+    files=[]
+    for p in sorted((root/'data'/row['sequence_id']).rglob('*')):
+        if p.is_file():files.append({'path':p.relative_to(root).as_posix(),'size':p.stat().st_size,'sha256':digest(p)})
+    done={'dataset':ds,'sequence':seq,'sequence_id':row['sequence_id'],'completed_at':now(),'source':source,'scripts':code_manifest(),'views':results,'independent_geometry':independent,
+        'files':files,'bytes':sum(f['size'] for f in files),'frames':sum(v['frames'] for v in results),'processing':{'media_policy':MEDIA_POLICY,'media_workers':media_workers,'conversion_seconds':conversion_seconds,'total_seconds':time.perf_counter()-conversion_start},'full_output_human_review':'pending','upload_authorized':False}
+    atomic(root/'doc/validation/full/sequences'/f'{row["sequence_id"]}.json',done);atomic(result,done)
+    print('SEQUENCE_COMPLETE',ds,seq,done['frames'],done['bytes'],flush=True)
+
+def random_review(ds,rows):
+    from render_full_sequence_review import render_view,asset_preview
+    doc=DATA_ROOT/ds/RELEASE/'doc/validation/full_random';doc.mkdir(parents=True,exist_ok=True);rng=random.Random(20260920);selected=[]
+    if ds=='ARCTIC':
+        for oid in sorted({r['object'] for r in rows}):
+            group=[r for r in rows if r['object']==oid and r['split']!='test'];r=rng.choice(group);selected.append((r,rng.choice(r['cameras']),'labelled_object:'+oid,len(group)))
+        if not any(c=='0' for _,c,_,_ in selected):
+            group=[r for r in rows if '0' in r['cameras'] and r['split']!='test'];r=rng.choice(group);selected.append((r,'0','labelled_ego_camera_coverage',len(group)))
+        for split in ['train','val','test']:
+            if not any(r['split']==split for r,_,_,_ in selected):
+                group=[r for r in rows if r['split']==split];r=rng.choice(group);selected.append((r,'0' if split=='test' else rng.choice(r['cameras']),'split_coverage:'+split,len(group)))
+    else:
+        for split in ['train','test']:
+            group=[r for r in rows if r['split']==split];r=rng.choice(group)
+            for cam in r['cameras']:selected.append((r,cam,'split:'+split,len(group)))
+    sampling={'seed':20260920,'method':'ARCTIC: equal allocation by object among labelled train/val, uniform sequence then available camera; add labelled ego and missing train/val/test strata. Held-out test uses ego to show missing calibration/pose. HOT3D: equal allocation train/test, uniform whole sequence per split, every camera. Not frame-weighted; whole selected sequence is rendered.',
+        'human_review':'pending','selections':[{'sequence_id':r['sequence_id'],'camera_id':c,'stratum':s,'stratum_sequences':n,'frame_indices':'all'} for r,c,s,n in selected]}
+    atomic(doc/'sampling.json',sampling);cards=[];results=[];assets={}
+    for row,cam,_,_ in selected:
+        st={'stage':'generating_full_random_review','dataset':ds,'sequence':row['sequence'],'camera_id':cam,'pid':os.getpid(),'time':now(),'completed_previews':len(results),'total_previews':len(selected)}
+        atomic(temporary(ds)/'status.json',st);atomic(CONTROL/'status.json',st)
+        p=DATA_ROOT/ds/RELEASE/'data'/row['sequence_id']/cam;sub=doc/row['sequence_id'];tmp=temporary(ds)/'random-review'/row['sequence_id'];qa=None
+        qfile=DATA_ROOT/ds/'tmp'/RUN/'qa'/f'{row["sequence_id"]}.json'
+        if ds=='HOT3D' and row['source_gt_available']:
+            if not qfile.exists():apply_qa(row['sequence'],publish=False)
+            qa=load(qfile)
+        pixel_proof=revalidate_media(p,full_pixels=True)
+        atomic(sub/'pixel-validation'/f'{cam}.json',pixel_proof)
+        result=render_view(p,ds,qa,doc=sub,tmp=tmp);result['whole_sequence_source_RGB_validation']=pixel_proof;results.append(result)
+        for op in (p/'annotation/object').glob('*'):
+            out=sub/'assets'/f'{op.name}.mp4'
+            if not out.exists():asset_preview(op,out)
+            assets[out.relative_to(doc).as_posix()]=row['sequence_id']+' / '+op.name
+        cards.append(f'<section><h2>{row["sequence_id"]} · camera {cam} · {result["frames"]}帧</h2><video controls preload="metadata" src="{row["sequence_id"]}/{result["video_ref"]}"></video></section>')
+    for ref,label in assets.items():cards.append(f'<section><h2>物体资产 · {label}</h2><video controls preload="metadata" src="{ref}"></video></section>')
+    atomic(doc/'report.json',{'dataset':ds,'sampling':sampling,'views':results,'asset_previews':assets,'human_review':'pending','upload_authorized':False})
+    (doc/'index.html').write_text('<!doctype html><meta charset="utf-8"><title>'+ds+' 全量随机抽查</title><style>body{font:16px/1.7 system-ui;max-width:1536px;margin:25px auto;background:#f3f5f9}section,header{background:white;padding:20px;margin:20px}video{width:100%}</style><header><h1>'+ds+' 全量随机抽查</h1><p>从全部完成产物读取，选中序列全部帧已渲染；用户人工确认待完成，不自动上传。</p><a href="sampling.json">随机种子与抽样范围/权重</a> · <a href="report.json">逐模态检查</a></header>'+''.join(cards),encoding='utf-8')
+    (doc/'REPORT.md').write_text('# 全量随机抽查\n\n[离线完整序列](index.html) · [抽样定义](sampling.json) · [模态检查](report.json)\n\n人工确认待完成。源缺失模态缺省，不以数值通过替代视觉质量确认。\n',encoding='utf-8')
+
+def run():
+    assert not (CONTROL/'maintenance.json').exists() or not load(CONTROL/'maintenance.json')['active'],'Maintenance in progress; production start disabled'
+    CONTROL.mkdir(parents=True,exist_ok=True);guard=(CONTROL/'process.lock').open('a+b');guard.seek(0)
+    try:msvcrt.locking(guard.fileno(),msvcrt.LK_NBLCK,1)
+    except OSError:raise RuntimeError('Another production queue owns the lock')
+    assert load(CONTROL/'authorization.json')['full_preprocessing_authorized'];p=load(CONTROL/'plan.json')
+    power=ctypes.windll.kernel32.SetThreadExecutionState(0x80000001)
+    try:
+        for ds in p['order']:
+            snapshot_scripts(ds)
+            rows=p['datasets'][ds]['sequences'];tmp=temporary(ds);tmp.mkdir(parents=True,exist_ok=True)
+            records=inputs(ds)
+            # Smaller full sequences first; sequence boundaries remain unchanged.
+            rows=sorted(rows,key=lambda r:r.get('jpeg_bytes',r.get('vrs_bytes',0)))
+            for n,row in enumerate(rows):
+                st={'stage':'converting','dataset':ds,'sequence':row['sequence'],'completed_sequences':n,'total_sequences':len(rows),'pid':os.getpid(),'time':now(),'media_policy':MEDIA_POLICY,'media_workers':load(CONTROL/'acceleration-config.json')['workers_by_dataset'][ds]}
+                atomic(tmp/'status.json',st);atomic(CONTROL/'status.json',st)
+                registry=load(WORK/'tmp/dataset-work-state.json');registry['active'][ds].update(stage='full_preprocessing',production_status=str(tmp/'status.json'),production_process_id=os.getpid());atomic(WORK/'tmp/dataset-work-state.json',registry)
+                assert shutil.disk_usage(DATA_ROOT).free>RESERVE+15_000_000_000,'Disk reserve reached'
+                checkpoint=tmp/'completed'/f'{row["sequence_id"]}.json'
+                if checkpoint.exists():
+                    dependencies=list(row['inputs'])
+                    if ds=='HOT3D':dependencies+=sorted(rel for rel in records if rel.startswith('assets/'))
+                    verify_completed(ds,row,[check_input(ds,rel,records) for rel in dependencies])
+                    print('VERIFIED_RESUME',ds,row['sequence'],flush=True)
+                    continue
+                log=tmp/'logs'/f'{row["sequence_id"]}.log';log.parent.mkdir(parents=True,exist_ok=True)
+                with log.open('ab') as out:
+                    child=subprocess.Popen([sys.executable,'-u',str(SCRIPT),'worker','--dataset',ds,'--sequence',row['sequence']],stdout=out,stderr=subprocess.STDOUT)
+                    while child.poll() is None:
+                        st.update(worker_pid=child.pid,time=now(),log=str(log));atomic(tmp/'status.json',st);atomic(CONTROL/'status.json',st)
+                        if shutil.disk_usage(DATA_ROOT).free<RESERVE:child.terminate();child.wait();raise RuntimeError('Disk reserve reached; active sequence stopped, raw retained')
+                        time.sleep(20)
+                    if child.returncode:raise RuntimeError(f'Worker failed: {ds} {row["sequence"]}; see {log}')
+                print('QUEUE_PROGRESS',ds,n+1,len(rows),row['sequence'],flush=True)
+            st={'stage':'generating_full_random_review','dataset':ds,'pid':os.getpid(),'time':now(),'completed_sequences':len(rows),'total_sequences':len(rows)}
+            atomic(tmp/'status.json',st);atomic(CONTROL/'status.json',st)
+            state=load(WORK/'tmp/dataset-work-state.json');state['active'][ds].update(stage='generating_full_random_review');atomic(WORK/'tmp/dataset-work-state.json',state)
+            random_review(ds,rows)
+            state=load(WORK/'tmp/dataset-work-state.json');state['active'][ds].update(stage='full_random_awaiting_human_review',full_output_human_review='pending',upload_authorized=False);atomic(WORK/'tmp/dataset-work-state.json',state)
+            atomic(tmp/'status.json',{'stage':'full_random_awaiting_human_review','time':now(),'completed_sequences':len(rows),'total_sequences':len(rows),'report':str(DATA_ROOT/ds/RELEASE/'doc/validation/full_random/index.html')})
+        atomic(CONTROL/'status.json',{'stage':'all_preprocessing_complete_awaiting_full_random_review','time':now(),'upload_authorized':False})
+    except BaseException as e:
+        error={'stage':'failed_needs_attention','time':now(),'error':str(e),'dataset':locals().get('ds'),'sequence':locals().get('row',{}).get('sequence')}
+        atomic(CONTROL/'status.json',error)
+        if 'ds' in locals():atomic(temporary(ds)/'status.json',error)
+        raise
+    finally:ctypes.windll.kernel32.SetThreadExecutionState(0x80000000);guard.close()
+
+if __name__=='__main__':
+    ap=argparse.ArgumentParser();ap.add_argument('mode',choices=['plan','approve','worker','run','migrate']);ap.add_argument('--dataset',choices=['ARCTIC','HOT3D']);ap.add_argument('--sequence');a=ap.parse_args()
+    if a.mode in ['worker','migrate']:
+        CONTROL.mkdir(parents=True,exist_ok=True)
+        with (CONTROL/'worker.lock').open('a+b') as lock:
+            lock.seek(0)
+            try:msvcrt.locking(lock.fileno(),msvcrt.LK_NBLCK,1)
+            except OSError:raise RuntimeError('A production worker is already active; refusing overlapping conversion')
+            if a.mode=='migrate':migrate_completed(a.dataset)
+            else:worker(a.dataset,a.sequence)
+    elif a.mode=='plan':plan()
+    elif a.mode=='approve':approve()
+    else:run()
diff --git a/ARCTIC/shared-v1.0/script/render_full_sequence_review.py b/ARCTIC/shared-v1.0/script/render_full_sequence_review.py
new file mode 100644
index 0000000000000000000000000000000000000000..a29dc18687e4ebe26f4db71a514ae4c34220e0e6
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/render_full_sequence_review.py
@@ -0,0 +1,256 @@
+"""Full-frame, all-camera diagnostic videos read from actual shared tensors/media.
+
+Preview MP4 is explicitly display-only; warehouse MKV packets remain untouched.
+"""
+from fresh_sequence_common import *
+from review_fresh_sequence_pilots import arr, validate_view, project_for_view
+from datetime import datetime
+import html
+
+VIDEO_RUN='full-video-review-v2'
+CELL=512
+WIDTH=3*CELL
+HEIGHT=800
+COLORS=[(110,255,100),(255,80,210),(240,220,60),(90,180,255),(190,160,255),(255,120,140)]
+
+def text(im,txt,xy,color=(235,235,235),size=.5):cv2.putText(im,txt,xy,cv2.FONT_HERSHEY_SIMPLEX,size,color,1,cv2.LINE_AA)
+
+def fast_edges(im,uv,edges,color):
+    xy=uv[edges];good=np.isfinite(xy).all((1,2)) & (np.max(np.nan_to_num(np.abs(xy),nan=CELL*20),axis=(1,2))<CELL*12)
+    xy=xy[good];xy=xy[np.linalg.norm(xy[:,0]-xy[:,1],axis=1)<CELL*.7]
+    if len(xy):cv2.polylines(im,np.rint(xy).astype(np.int32),False,color,1,cv2.LINE_AA)
+
+def dots(im,uv,color,radius=2):
+    good=np.isfinite(uv).all(1)&(uv[:,0]>=0)&(uv[:,0]<im.shape[1])&(uv[:,1]>=0)&(uv[:,1]<im.shape[0])
+    xy=np.rint(uv[good]).astype(int);xy[:,0]=np.clip(xy[:,0],0,im.shape[1]-1);xy[:,1]=np.clip(xy[:,1],0,im.shape[0]-1)
+    for d in range(max(1,radius)):
+        im[np.clip(xy[:,1]+d,0,im.shape[0]-1),xy[:,0]]=color
+        im[xy[:,1],np.clip(xy[:,0]+d,0,im.shape[1]-1)]=color
+
+def model_vertices(motion,valid,side,fix):
+    model=Mano(['left','right'][side])
+    if side==0 and fix:
+        model.data['shapedirs'][:,0]*=-1
+        with torch.no_grad():model.layer.shapedirs[:,0]*=-1
+    verts=np.full((len(motion),778,3),np.nan,np.float32);joints=np.full((len(motion),16,3),np.nan,np.float32)
+    ids=np.flatnonzero(valid[:,side])
+    for start in range(0,len(ids),64):
+        ix=ids[start:start+64];a=motion[ix,side];verts[ix],_,joints[ix]=model.geometry(a[:,:48],a[:,48:58],a[:,58:])
+    faces=np.asarray(model.layer.faces);edges=np.unique(np.sort(np.concatenate([faces[:,[0,1]],faces[:,[1,2]],faces[:,[2,0]]]),axis=1),axis=0)
+    return verts,joints,edges
+
+def encoder(path,fps,log,size=(WIDTH,HEIGHT)):
+    path.parent.mkdir(parents=True,exist_ok=True)
+    return subprocess.Popen([str(FF),'-v','error','-y','-f','rawvideo','-pix_fmt','rgb24','-s',f'{size[0]}x{size[1]}','-r',str(fps),'-i','pipe:0',
+        '-an','-c:v','libx264','-preset','veryfast','-crf','19','-pix_fmt','yuv420p','-threads','2','-bf','0','-g','60','-movflags','+faststart',str(path)],stdin=subprocess.PIPE,stderr=log)
+
+def read_exact(pipe,n):
+    chunks=[];count=0
+    while count<n:
+        b=pipe.read(n-count)
+        if not b:break
+        chunks.append(b);count+=len(b)
+    return b''.join(chunks)
+
+def trajectory_plot(E,width=265,height=145):
+    good=np.isfinite(E).all((1,2));xyz=np.full((len(E),3),np.nan)
+    xyz[good]=-(np.swapaxes(E[good,:3,:3],1,2)@E[good,:3,3,None])[:,:,0]
+    base=np.full((height,width,3),24,np.uint8);xy=xyz[:,[0,2]];valid=np.isfinite(xy).all(1)
+    if not valid.any():return base,np.zeros((len(E),2),int),xyz
+    lo=np.nanmin(xy,0);hi=np.nanmax(xy,0);span=np.maximum(hi-lo,.05);scale=min((width-24)/span[0],(height-35)/span[1]);pix=(xy-lo)*scale+12
+    pix[:,1]=height-12-pix[:,1];pixels=np.nan_to_num(pix).astype(int)
+    for i in range(1,len(E)):
+        if valid[i-1] and valid[i]:cv2.line(base,tuple(pixels[i-1]),tuple(pixels[i]),(110,110,110),1)
+    text(base,'camera world trajectory X/Z (m)',(7,15),size=.38)
+    return base,pixels,xyz
+
+def object_specs(p,source):
+    result=[]
+    for k,op in enumerate(sorted((p/'annotation/object').glob('*'))):
+        if not (op/'pose.pt').exists():continue
+        om=load(op/'metainfo.json');rel=op.relative_to(p);mesh=trimesh.load(p/om['mesh_ref'],process=False,force='mesh')
+        rawpose=arr(source/rel/'pose.pt');rawvalid=arr(source/rel/'pose_valid.pt');valid=arr(op/'pose_valid.pt')
+        spec={'id':om['object_id'],'name':om.get('category_name',om['object_id']),'color':COLORS[k%len(COLORS)],'P':rawpose,'rawvalid':rawvalid,'valid':valid,'op':op}
+        if 'vertex_part_ids_ref' in om:
+            xyz=np.array([[float(x) for x in r.split()[1:4]] for r in (p/om['mesh_ref']).read_text().splitlines() if r.startswith('v ')])
+            part=arr(p/om['vertex_part_ids_ref']);idx=np.arange(0,len(xyz),max(1,len(xyz)//4500))
+            spec.update(vertices=xyz[idx],parts=part[idx],top=arr(source/rel/'parts/top/pose.pt'),edges=None,
+                keypoints=arr(source/rel/'keypoints/xyz.pt'),joint=arr(source/rel/'joint_state.pt')[:,0])
+        else:
+            edges=np.asarray(mesh.edges_unique);edges=edges[::max(1,len(edges)//1100)]
+            unique,inv=np.unique(edges,return_inverse=True);spec.update(vertices=np.asarray(mesh.vertices)[unique],edges=inv.reshape(-1,2),parts=None)
+        spec['visibility']=arr(op/'visibility.pt') if (op/'visibility.pt').exists() else None
+        spec['raw_visibility']=arr(source/rel/'visibility.pt') if (source/rel/'visibility.pt').exists() else None
+        result.append(spec)
+    return result
+
+def render_view(p,dataset,qa_data,doc=None,tmp=None,source=None,snapshot_indices=None):
+    m,frames,structural=validate_view(p);cam=m['camera_id'];T=m['num_frames'];root=DATA_ROOT/dataset/RELEASE
+    doc=doc or root/'doc/validation';tmp=tmp or DATA_ROOT/dataset/'tmp'/VIDEO_RUN;tmp.mkdir(parents=True,exist_ok=True)
+    if source is None:
+        source=p
+        for candidate in [DATA_ROOT/dataset/'tmp'/VIDEO_RUN/'diagnostic'/m['sequence_id']/cam,DATA_ROOT/dataset/'tmp'/VIDEO_RUN/'diagnostic'/cam]:
+            if (candidate/'snapshot.json').exists() and Path(load(candidate/'snapshot.json')['source_view']).resolve()==p.resolve():source=candidate;break
+    qa=np.array(qa_data[cam]['qa_pass'],bool) if dataset=='HOT3D' and qa_data and qa_data[cam].get('qa_pass') is not None else None
+    if 'mano' in m['components']:
+        motion=arr(source/'annotation/mano/motion.pt');present=arr(source/'annotation/mano/motion_valid.pt');valid=arr(p/'annotation/mano/motion_valid.pt')
+    else:motion=np.full((T,2,61),np.nan);present=np.zeros((T,2),bool);valid=present.copy()
+    visibility=arr(p/'annotation/mano/visibility.pt') if (p/'annotation/mano/visibility.pt').exists() else None
+    rawvis=arr(source/'annotation/mano/visibility.pt') if (source/'annotation/mano/visibility.pt').exists() else None
+    hands=[model_vertices(motion,present,s,m['components'].get('mano',{}).get('model_profile',{}).get('left_shapedirs_fix',False)) for s in range(2)]
+    objs=object_specs(p,source);K=arr(p/'camera/intrinsic.pt') if (p/'camera/intrinsic.pt').exists() else np.full((T,3,3),np.nan);D=arr(p/'camera/distortion.pt') if (p/'camera/distortion.pt').exists() else None;E=arr(p/'camera/extrinsic.pt') if (p/'camera/extrinsic.pt').exists() else np.full((T,4,4),np.nan)
+    W,H=m['camera']['image_size'];scale0=min(CELL/W,CELL/H);w=int(W*scale0)//2*2;h=int(H*scale0)//2*2;scale=np.array([w/W,h/H])
+    fps=m['sampling']['source_fps'] or 30;path=doc/'videos'/f'{cam}-full.mp4';logpath=tmp/f'render-{cam}.log';log=logpath.open('wb')
+    dec=subprocess.Popen([str(FF),'-v','error','-threads','2','-noautorotate','-i',str(p/'observation/rgb.mkv'),'-vf',f'scale={w}:{h}','-pix_fmt','rgb24','-fps_mode','passthrough','-f','rawvideo','pipe:1'],stdout=subprocess.PIPE,stderr=log)
+    enc=encoder(path,fps,log);traj,tp,world=trajectory_plot(E);rendered=0;exposure=[]
+    snapshots=sorted(set(snapshot_indices or [0,1,26,27,179,180,T//4,T//2,3*T//4,T-1]))
+    timeline=np.zeros((12,WIDTH,3),np.uint8)
+    for x in range(WIDTH):timeline[:,x]=(40,135,65) if qa is not None and qa[min(T-1,x*T//WIDTH)] else ((170,42,50) if qa is not None else (75,80,95))
+    for i in range(T):
+        b=read_exact(dec.stdout,w*h*3);assert len(b)==w*h*3,(cam,i,'decoder ended early')
+        original=np.frombuffer(b,np.uint8).reshape(h,w,3).copy();raw=original.copy();usable=original.copy()
+        gray=cv2.cvtColor(cv2.resize(original,(96,96)),cv2.COLOR_RGB2GRAY)
+        exposure.append({'frame_index':i,'p95_luma':float(np.percentile(gray,95)),'median_luma':float(np.median(gray))})
+        params=K[i] if m['camera']['model']=='fisheye624' else (K[i],D[i] if D is not None else None)
+        for side,(verts,joints,edges) in enumerate(hands):
+            if not present[i,side]:continue
+            uv=project_for_view(verts[i],m,params,scale);color=(0,235,255) if side==0 else (255,135,20)
+            fast_edges(raw,uv,edges,color)
+            if valid[i,side]:fast_edges(usable,uv,edges,color)
+            # White MANO joints are a diagnostic derivation, not a new keypoint component.
+            dots(raw,project_for_view(joints[i],m,params,scale),(255,255,255),2)
+            if valid[i,side]:dots(usable,project_for_view(joints[i],m,params,scale),(255,255,255),2)
+        for obj in objs:
+            if not obj['rawvalid'][i]:continue
+            P=obj['P'][i];xyz=obj['vertices']@P[:3,:3].T+P[:3,3]
+            if obj['parts'] is not None:
+                top=obj['top'][i];sel=obj['parts']==1;xyz[sel]=obj['vertices'][sel]@top[:3,:3].T+top[:3,3]
+            uv=project_for_view(xyz,m,params,scale)
+            for image,enabled in [(raw,True),(usable,obj['valid'][i])]:
+                if not enabled:continue
+                if obj['edges'] is not None:fast_edges(image,uv,obj['edges'],obj['color'])
+                else:
+                    dots(image,uv[obj['parts']==0],(100,240,90));dots(image,uv[obj['parts']==1],(230,80,230))
+                    dots(image,project_for_view(obj['keypoints'][i],m,params,scale),(255,255,255),3)
+                # Local object axes in metres make pose orientation visible.
+                axes=np.vstack([np.zeros(3),np.eye(3)*.045])@P[:3,:3].T+P[:3,3];auv=project_for_view(axes,m,params,scale)
+                for j,c in enumerate([(255,60,60),(60,255,60),(60,120,255)]):fast_edges(image,auv,np.array([[0,j+1]]),c)
+        canvas=np.full((HEIGHT,WIDTH,3),20,np.uint8)
+        for col,(im,label) in enumerate([(original,'SOURCE RGB'),(raw,'NATIVE LABEL DIAGNOSTIC'),(usable,'SHARED USABLE SUPERVISION')]):
+            canvas[40+(CELL-h)//2:40+(CELL-h)//2+h,col*CELL+(CELL-w)//2:col*CELL+(CELL-w)//2+w]=im
+            text(canvas,label,(col*CELL+12,25),size=.58)
+        status='NO OFFICIAL QA PROVIDED' if qa is None else ('OFFICIAL QA PASS' if qa[i] else 'OFFICIAL QA FAIL - NOT SUPERVISION')
+        if dataset=='ARCTIC' and i in m['components'].get('mano',{}).get('valid_policy',{}).get('local_excluded_frame_indices',[]):status='LOCAL REVIEW REJECT - NOT HAND SUPERVISION'
+        color=(235,195,65) if qa is None else ((80,240,110) if qa[i] else (255,90,90))
+        text(canvas,f'{dataset} {cam}  index={i}/{T-1}  source={frames[i]["source_frame_id"]}   {status}',(12,578),color,.6)
+        text(canvas,f'MANO source L/R={int(present[i,0])}/{int(present[i,1])}  shared valid L/R={int(valid[i,0])}/{int(valid[i,1])}  (m/rad)',(12,603),size=.49)
+        def vals(a):return 'n/a' if a is None else ','.join('unknown' if not np.isfinite(x) else f'{100*x:.1f}%' for x in np.atleast_1d(a))
+        text(canvas,f'visibility native L/R: {vals(rawvis[i] if rawvis is not None else None)}  usable: {vals(visibility[i] if visibility is not None else None)}',(12,626),size=.48)
+        text(canvas,f'camera={m["camera"]["model"]}; intrinsic '+np.array2string(K[i].ravel()[:5],precision=2),(12,649),size=.45)
+        text(canvas,'camera world position (m): '+np.array2string(world[i],precision=3),(12,672),size=.45)
+        clock=frames[i]['timestamp_ns'];text(canvas,f'timestamp_ns={clock}  clock={frames[i]["clock_domain"]}',(12,695),size=.44)
+        text(canvas,f'preview FPS={fps}; '+('capture FPS unknown, playback only' if m['sampling']['source_fps'] is None else 'source capture time is stored separately'),(12,718),size=.44)
+        text(canvas,'cyan=left; orange=right; white=derived MANO joints / source object points',(12,741),size=.41)
+        text(canvas,'Projection shows hidden surfaces; it is NOT native segmentation.',(12,762),(180,180,180),.43)
+        for j,obj in enumerate(objs):
+            a=obj['raw_visibility'];v=obj['visibility'];s=f'{obj["name"]}: valid={int(obj["valid"][i])}'
+            if a is not None:s+=f' vis={vals(v[i])} (raw {vals(a[i])})'
+            if 'joint' in obj:s+=f' hinge={obj["joint"][i]:.3f}rad  keypoints=32'
+            text(canvas,s,(660,603+j*24),obj['color'],.44)
+        tr=traj.copy()
+        if np.isfinite(world[i]).all():cv2.circle(tr,tuple(tp[i]),4,(255,220,70),-1)
+        canvas[610:755,WIDTH-277:WIDTH-12]=tr
+        canvas[-12:]=timeline;cursor=min(WIDTH-1,int(i*WIDTH/T));canvas[-12:,cursor:cursor+2]=255
+        enc.stdin.write(canvas.tobytes());rendered+=1
+        if i in snapshots:
+            snap=doc/'video_snapshots'/cam;snap.mkdir(parents=True,exist_ok=True);Image.fromarray(canvas).save(snap/f'{i:06d}.jpg',quality=92)
+        if i%200==0:print('VIDEO_FRAME',dataset,cam,i,T,flush=True)
+    assert not dec.stdout.read(1),'extra source frames';dec.stdout.close();assert dec.wait()==0
+    enc.stdin.close();assert enc.wait()==0;log.close();assert rendered==T
+    packets=framehash(['-i',str(path)],copy=True);assert len(packets)==T
+    result={'sequence_id':m['sequence_id'],'camera_id':cam,'frames':T,'rendered_frames':rendered,'video_ref':path.relative_to(doc).as_posix(),'video_sha256':digest(path),'preview_fps':fps,
+        'preview_is_training_data':False,'preview_encoding':'H264 CRF19, scaled, display only; shared original MKV unchanged',
+        'source_labels_from':'raw-derived diagnostic cache' if source!=p else 'shared unfiltered source-compatible labels',
+        'all_source_and_shared_valid_masks_used':True,'structure_checks':structural,'mano_valid_counts':valid.sum(0).tolist(),
+        'official_qa_pass_frames':int(qa.sum()) if qa is not None else None,'official_qa_fail_frames':int((~qa).sum()) if qa is not None else None,
+        'exposure_diagnostics':{'method':'scaled shared decoded RGB to 96x96 gray; p95<35 flags a candidate only, does not modify valid','suspect_dark_frame_indices':[x['frame_index'] for x in exposure if x['p95_luma']<35]}}
+    save(tmp/f'{cam}-exposure.json',exposure)
+    save(tmp/f'{cam}-video-result.json',result);print('VIDEO_COMPLETE',dataset,cam,T,flush=True)
+    return result
+
+def asset_preview(op,out):
+    """Textured point-splat turntable plus the source texture; diagnostic, not photoreal rendering."""
+    p=op.parents[2];m=load(op/'metainfo.json');meshpath=p/m['mesh_ref']
+    if (op/'assets/mesh_tex.obj').exists():meshpath=op/'assets/mesh_tex.obj'
+    mesh=trimesh.load(meshpath,process=False,force='mesh');xyz=np.asarray(mesh.vertices);idx=np.arange(0,len(xyz),max(1,len(xyz)//35000));xyz=xyz[idx]
+    texture=getattr(mesh.visual.material,'image',None) if hasattr(mesh.visual,'material') else None
+    texture_source=None
+    if meshpath.name=='mesh_tex.obj' and op.relative_to(DATA_ROOT).parts[0]=='ARCTIC':
+        # The loader can silently return a 2x2 default material. Resolve the
+        # official single-texture MTL explicitly instead of accepting that placeholder.
+        mtl=[r.split(maxsplit=1)[1] for r in meshpath.read_text().splitlines() if r.startswith('mtllib ')]
+        diffuse=[]
+        for name in mtl:
+            material=meshpath.parent/name
+            diffuse.extend(material.parent/r.split(maxsplit=1)[1] for r in material.read_text().splitlines() if r.startswith('map_Kd '))
+        diffuse=list(set(diffuse));assert len(diffuse)==1,(meshpath,diffuse)
+        assert diffuse[0].resolve().is_relative_to(meshpath.parent.resolve())
+        texture=Image.open(diffuse[0]).convert('RGB');assert min(texture.size)>2
+        texture_source={'file':diffuse[0].relative_to(p).as_posix(),'sha256':digest(diffuse[0]),'size':list(texture.size),'resolution_method':'explicit official MTL map_Kd; reject loader placeholder'}
+    if texture is not None and getattr(mesh.visual,'uv',None) is not None:
+        tex=np.asarray(texture.convert('RGB'));uv=np.asarray(mesh.visual.uv)[idx]
+        rgb=tex[np.clip(np.rint((1-uv[:,1])*(len(tex)-1)).astype(int),0,len(tex)-1),np.clip(np.rint(uv[:,0]*(tex.shape[1]-1)).astype(int),0,tex.shape[1]-1)]
+    else:rgb=np.broadcast_to([150,210,245],xyz.shape).astype(np.uint8)
+    extent=np.ptp(xyz,axis=0);xyz=xyz-(xyz.min(0)+xyz.max(0))/2;scale=300/max(np.linalg.norm(extent),.001)
+    logpath=DATA_ROOT/op.relative_to(DATA_ROOT).parts[0]/'tmp'/VIDEO_RUN/'assets.log'
+    logpath.parent.mkdir(parents=True,exist_ok=True)
+    out.parent.mkdir(parents=True,exist_ok=True)
+    with logpath.open('ab') as log:
+        enc=encoder(out,12,log,size=(800,480))
+        for i in range(72):
+            image=np.full((480,800,3),24,np.uint8);R=Rotation.from_euler('xy',[.35,i*2*np.pi/72]).as_matrix();v=xyz@R.T
+            order=np.argsort(v[:,2]);xy=np.rint(v[:,:2]*scale+[220,220]).astype(int)
+            good=(xy[:,0]>=0)&(xy[:,0]<430)&(xy[:,1]>=35)&(xy[:,1]<400);order=order[good[order]]
+            for dx,dy in [(0,0),(1,0),(0,1)]:image[np.clip(xy[order,1]+dy,0,479),np.clip(xy[order,0]+dx,0,429)]=rgb[order]
+            if texture is not None:
+                thumb=texture.convert('RGB');thumb.thumbnail((350,345));image[40:40+thumb.height,440:440+thumb.width]=np.asarray(thumb)
+            text(image,m.get('category_name',m['object_id']),(12,25),size=.65)
+            text(image,'TEXTURED 3D POINT PREVIEW',(12,420),size=.55);text(image,'SOURCE MATERIAL TEXTURE',(445,420),size=.5)
+            text(image,'OBJ extent (m): '+np.array2string(extent,precision=4),(12,450),size=.5);enc.stdin.write(image.tobytes())
+        enc.stdin.close();assert enc.wait()==0
+    assert len(framehash(['-i',str(out)],copy=True))==72
+    return {'object_id':m['object_id'],'video_ref':'assets/'+out.name,'frames':72,'texture_present':texture is not None,'texture_source':texture_source,'extent_m':extent.tolist(),'method':'diagnostic textured point turntable, not photoreal material rendering'}
+
+def publish_review(dataset,results,assets):
+    root=DATA_ROOT/dataset/RELEASE;doc=root/'doc/validation'
+    scope=[('RGB','全部帧原图与逐帧码流/像素校验'),('相机内参/畸变','投影使用声明模型和全部参数；HUD显示相机模型'),('相机外参','全帧姿态、米制相机位置与世界XZ轨迹'),('MANO','逐帧网格线框、派生关节点、左右手valid'),('物体姿态/几何','逐帧轮廓、局部XYZ轴、物体ID'),('材质/纹理','每个物体的原纹理与完整旋转预览')]
+    scope+=([('部件/关节','底座/顶盖分别着色，逐帧关节角'),('物体关键点','源32个物体点逐帧叠加'),('visibility','原数据未提供，明确缺省')] if dataset=='ARCTIC' else [('visibility','原值0–1已修正，原生/可用比例同时显示'),('官方QA','完整红绿时间带；QA失败帧保留诊断、停用监督'),('部件/关节、独立物体关键点','本序列无该源模态，不虚构')])
+    scope+=[('深度/像素分割/接触','本次raw范围未提供，不以投影或帧筛选CSV冒充')]
+    table=''.join(f'<tr><td>{a}</td><td>{b}</td></tr>' for a,b in scope)
+    cards=''.join(f'<section><h2>Camera {r["camera_id"]} · {r["frames"]} 帧 · 有效MANO左右 {r["mano_valid_counts"]}</h2><video controls preload="metadata" src="{r["video_ref"]}"></video><p>全部{r["rendered_frames"]}帧已渲染，无码帧跳过。官方QA通过/未通过：{r["official_qa_pass_frames"]}/{r["official_qa_fail_frames"]}（null表示来源无此QA）。</p></section>' for r in results)
+    gallery=''.join(f'<section><h3>物体 {a["object_id"]}</h3><video controls loop preload="metadata" src="{a["video_ref"]}"></video><p>纹理点云旋转预览及原始材质图；此预览不证明完整PBR渲染等价。</p></section>' for a in assets)
+    known=('已修正转换器颠倒顶盖/底座编码的错误：原parts.json的0是顶盖、1是底座，共享层重新映射为0底座/1顶盖。9路相机前两帧均为源欠曝图，已排除对应几何监督；camera 6 index 270的手部偏移作为已知问题记录，按用户要求撤销单帧屏蔽，恢复原标注和valid。整套试点仍待验收。中栏显示源参数诊断，右栏按当前valid绘制。详见 <a href="ARCTIC_FINDINGS.md">问题与修复证据</a>、<a href="diagnosis/camera6-270-before-after.jpg">修复前后对照</a>。预览30fps仅用于播放，不代表真实采集帧率；没有源准确性QA。' if dataset=='ARCTIC' else '原始开头帧存在明显错位，官方QA为False；源头戴位姿开头27帧保持不变，帧号/时间/包顺序均未人为移动。中栏保留原始诊断投影，右栏只绘制当前valid监督。已修正visibility误除100。QA按相机整帧保守筛选，不能断言具体哪个实体错。')
+    page=f'''<!doctype html><html lang="zh-CN"><meta charset="utf-8"><meta name="viewport" content="width=device-width, initial-scale=1"><title>{dataset} 完整视频审核 v2</title><style>body{{font:16px/1.7 system-ui;background:#edf1f7;color:#172136;max-width:1550px;margin:25px auto;padding:0 20px}}header,section{{background:white;padding:22px;border-radius:12px;margin-bottom:20px}}video{{width:100%;background:#151515}}h1{{margin:0}}.warn{{color:#a71d2b;font-weight:bold}}table{{border-collapse:collapse;width:100%}}td,th{{border:1px solid #ccd4df;padding:8px;text-align:left}}.assets{{display:grid;grid-template-columns:repeat(auto-fit,minmax(500px,1fr));gap:18px}}a{{color:#1567a3}}</style><header><h1>{dataset} · 完整序列审核 v2</h1><p class="warn">人工确认待完成。旧“抽帧投影通过”结论撤回，本页替代旧报告；尚未全量或发布。</p><p>视频三栏：原图 ｜ 原始标注诊断 ｜ 共享层有效监督。所有相机、所有帧都已渲染；青=左手，橙=右手，其他色=物体/部件，白点=MANO派生关节点或源物体关键点。投影不做遮挡剔除，也不是分割真值。</p><p>{known}</p><p>标注 valid=实际标注存在且数值合法，并通过适用官方QA及已明确登记的本地排除规则；可见性不代替valid，剩余帧不因此自动认定人工通过。诊断中被排除的参数不进入有效监督，raw保留。MP4经过缩放与展示编码，仅为doc预览，训练输入仍是无新增损失的shared MKV。</p><p><a href="REPORT.md">正式说明</a> · <a href="report.json">逐相机检查结果</a> · <a href="../RAW_TO_SHARED.md">字段映射</a> · <a href="../CONTRACT.md">更新后的valid定义</a></p><h2>逐模态检查覆盖</h2><table><tr><th>模态</th><th>检查/可视化方式</th></tr>{table}</table></header>{cards}<h2>物体资产检查</h2><div class="assets">{gallery}</div><script>document.querySelectorAll('video').forEach(v=>v.addEventListener('play',()=>document.querySelectorAll('video').forEach(o=>{{if(o!==v)o.pause()}})));</script></html>'''
+    (doc/'index.html').write_text(page,encoding='utf-8')
+    old=load(doc/'report.json');old.update(stage='full_video_review_awaiting_human',human_review='pending',review_revision='full-video-v2',
+        old_sample_projection_conclusion='withdrawn',full_video_views=results,asset_previews=assets,modality_coverage=[{'modality':a,'check':b} for a,b in scope],known_issues=known,
+        agent_visual_inspection={'status':'pending_new_full_video_review','not_equivalent_to_user_approval':True},created_at=datetime.now().astimezone().isoformat())
+    old.pop('views',None);old.pop('offline_artifact_check',None);save(doc/'report.json',old)
+    save(doc/'status.json',{'stage':'full_video_review_awaiting_human','human_review':'pending','full_preprocessing_authorized':False,'upload_authorized':False})
+    rows='\n'.join(f'| {r["camera_id"]} | {r["frames"]} | {r["rendered_frames"]} | {r["mano_valid_counts"]} | {r["official_qa_pass_frames"]} |' for r in results)
+    (doc/'REPORT.md').write_text(f'# {dataset} 完整序列审核 v2\n\n[离线完整视频](index.html) · [机器结果](report.json)\n\n人工确认待完成；旧抽帧投影通过结论撤回。{known}\n\n|相机|源帧数|渲染帧数|左右手有效标注帧数|官方QA通过帧数|\n|---|---:|---:|---|---:|\n{rows}\n\n## 逐模态覆盖\n\n'+''.join(f'- {a}：{b}。\n' for a,b in scope)+'\n完整视频表示每帧已渲染，并不等于每帧人工审核通过。被官方QA排除的源标注仅在诊断栏显示；不挪帧、不隐藏原始异常。所有有效性筛选、可见性与图像编码验收分开记录。源标注质量问题无法通过格式转换修好；通过valid屏蔽，并保留raw溯源。\n',encoding='utf-8')
+    for f in ['CONTRACT.md','schema.json']:shutil.copyfile(WORK/'distribution/hand-warehouse-1.0'/f,root/'doc'/f)
+    for f in ['render_full_sequence_review.py','apply_hot3d_qa.py','fresh_sequence_common.py','review_fresh_sequence_pilots.py','convert_fresh_sequence_pilots.py']:shutil.copyfile(WORK/'scripts'/f,root/'script'/f)
+    print('FULL_VIDEO_REVIEW_READY',dataset,flush=True)
+
+if __name__=='__main__':
+    ap=argparse.ArgumentParser();ap.add_argument('dataset',choices=['HOT3D','ARCTIC']);a=ap.parse_args();torch.set_num_threads(3)
+    root=DATA_ROOT/a.dataset/RELEASE;qa=load(root/'doc/validation/source_qa.json') if a.dataset=='HOT3D' else None
+    # This entry point owns the original single-sequence report. Other review
+    # sequences have their own output directories in arctic_object_review.py.
+    sequence={'ARCTIC':'s09_laptop_use_03','HOT3D':'P0015_179e1b84'}[a.dataset]
+    views=sorted(p.parent for p in (root/'data'/sequence).glob('*/metainfo.json'));results=[]
+    for p in views:results.append(render_view(p,a.dataset,qa))
+    assets=[]
+    for op in sorted((views[0]/'annotation/object').iterdir()):assets.append(asset_preview(op,root/'doc/validation/assets'/f'{op.name}.mp4'))
+    publish_review(a.dataset,results,assets)
diff --git a/ARCTIC/shared-v1.0/script/repair_arctic_pilot.py b/ARCTIC/shared-v1.0/script/repair_arctic_pilot.py
new file mode 100644
index 0000000000000000000000000000000000000000..bc1e5bcaf3a87be9711702603cdb21192cf467fa
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/repair_arctic_pilot.py
@@ -0,0 +1,57 @@
+"""Repair ARCTIC pilot part IDs and exclude confirmed unusable supervision.
+
+These exclusions are specific to the reviewed sequence; no general dark-image
+threshold or blanket first-two-frame rule is applied to other sequences.
+"""
+from fresh_sequence_common import *
+from review_fresh_sequence_pilots import arr
+
+def repair():
+    base=DATA_ROOT/'ARCTIC';root=base/RELEASE;seq='s09_laptop_use_03'
+    exposure=load(root/'doc/validation/exposure.json');evidence=load(root/'doc/validation/diagnosis.json')
+    with zipfile.ZipFile(base/'raw/data/meta.zip') as z:
+        rawids=np.array(json.loads(z.read('meta/object_vtemplates/laptop/parts.json')),np.int64)
+    assert set(np.unique(rawids))=={0,1}
+    records={}
+    for meta in sorted((root/'data'/seq).glob('*/metainfo.json')):
+        p=meta.parent;m=load(meta);cam=m['camera_id'];T=m['num_frames']
+        cache=base/'tmp/full-video-review-v2/diagnostic'/cam
+        if not (cache/'snapshot.json').exists():
+            cache.mkdir(parents=True,exist_ok=True)
+            for f in p.rglob('*.pt'):
+                target=cache/f.relative_to(p);target.parent.mkdir(parents=True,exist_ok=True);shutil.copyfile(f,target)
+            save(cache/'metainfo.json',m);save(cache/'snapshot.json',{'purpose':'before local supervision exclusion; raw-compatible parameter diagnostics','source_view':str(p)})
+        obj=p/'annotation/object/laptop';tensor(obj/'vertex_part_ids.pt',1-rawids)
+        om=load(obj/'metainfo.json');om['part_mapping']={'0':'bottom','1':'top'}
+        om['part_id_conversion']={'formula':'shared_id = 1 - raw_parts_json_value','source_mapping':{'0':'top','1':'bottom'},'upstream_code':evidence['upstream_code']}
+        save(obj/'metainfo.json',om)
+        # Selection is explicitly reviewed. Statistics only verify the observed prefix.
+        first=exposure[cam]['frames'];assert all(first[i]['p95_luma']<35 for i in [0,1]) and first[2]['p95_luma']>100
+        image_ok=np.ones(T,bool);image_ok[[0,1]]=False
+        hand_ok=image_ok.copy()
+        for key,c in m['components'].items():
+            if key!='mano' and c['schema_id'] not in ['object/rigid/v1','object/joint_state/v1','object_keypoints/camera/v1']:continue
+            values=arr(cache/c['data_ref']).copy();present=arr(cache/c['valid_ref']);gate=hand_ok if key=='mano' else image_ok
+            valid=present & gate.reshape((T,)+(1,)*(present.ndim-1));values[~valid]=np.nan
+            tensor(p/c['data_ref'],values);tensor(p/c['valid_ref'],valid)
+            c['valid_policy']={'semantics':'usable_annotation','formula':'source_annotation_present AND source_numerically_valid AND local_review_pass_for_known_exclusions',
+                'official_accuracy_qa':'not provided by selected raw; ARCTIC geometric framing validity is not image alignment QA',
+                'local_excluded_frame_indices':np.flatnonzero(~gate).tolist(),
+                'local_exclusion_reasons':{'0,1':'confirmed severe source underexposure; both raw and shared images preserved'},
+                'source_locator':{'review':'doc/validation/ARCTIC_FINDINGS.md','exposure':'doc/validation/exposure.json','geometry':'doc/validation/diagnosis.json'},
+                'limitation':'Not a claim that remaining frames pass manual review; entire pilot remains blocked pending all-camera video review'}
+        m['validity_revision']='usable-annotation-2026-09-20';m['acceptance']={'status':'blocked_pending_visual_review','training_release_approved':False,'reason':'known source dark frames excluded; parts fixed; residual source-compatible hand misalignment requires review'}
+        save(meta,m)
+        checksfile=base/'tmp'/RUN/'checks'/f'{seq}__{cam}.json';checks=load(checksfile)
+        checks['checks']=[r for r in checks['checks'] if r['test'] not in ['all_frame_articulated_geometry_independent_scipy_rotation','pinned_official_ObjectTensors_articulation_after_part_remap']]
+        checks['checks'].append({'test':'pinned_official_ObjectTensors_articulation_after_part_remap','passed':True,**next(x for x in evidence['tests'] if x['camera_id']==cam)})
+        save(checksfile,checks)
+        records[cam]={'image_exclusions':[0,1],'mano_exclusions':np.flatnonzero(~hand_ok).tolist(),'usable_mano_frames':arr(p/'annotation/mano/motion_valid.pt').sum(0).tolist()}
+        print('ARCTIC_REPAIRED',cam,records[cam],flush=True)
+    evidence['repair_applied']=True;evidence['local_exclusions']=records;save(root/'doc/validation/diagnosis.json',evidence)
+    findings=WORK/'docs/ARCTIC-pilot-findings-2026-09-20.md'
+    if findings.exists():shutil.copyfile(findings,root/'doc/validation/ARCTIC_FINDINGS.md')
+    for f in ['repair_arctic_pilot.py','diagnose_arctic_review.py','convert_fresh_sequence_pilots.py','review_fresh_sequence_pilots.py','render_full_sequence_review.py']:
+        shutil.copyfile(WORK/'scripts'/f,root/'script'/f)
+
+if __name__=='__main__':repair()
diff --git a/ARCTIC/shared-v1.0/script/review_fresh_sequence_pilots.py b/ARCTIC/shared-v1.0/script/review_fresh_sequence_pilots.py
new file mode 100644
index 0000000000000000000000000000000000000000..f78fca1aa5b127d142d052ea202004f7b0ce31a0
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/review_fresh_sequence_pilots.py
@@ -0,0 +1,225 @@
+"""Validate actual shared outputs and create portable, offline pilot reviews.
+
+Run after convert_fresh_sequence_pilots.py. Never authorizes full processing.
+"""
+from fresh_sequence_common import *
+import html
+from datetime import datetime
+
+def arr(p):
+    value=torch.load(p,map_location='cpu',weights_only=True)
+    assert isinstance(value,torch.Tensor) and value.device.type=='cpu'
+    return value.numpy()
+
+def safe_ref(root,ref):
+    path=(root/ref).resolve()
+    assert path.is_relative_to(root.resolve()) and path.is_file(),ref
+    return path
+
+def validate_view(p):
+    m=load(p/'metainfo.json');T=m['num_frames'];registry=load(WORK/'distribution/hand-warehouse-1.0/schema.json')
+    assert m['schema_version']==registry['schema_version'] and m['pilot_complete']
+    frames=[json.loads(line) for line in (p/'camera/frames.jsonl').read_text().splitlines()]
+    assert len(frames)==T and [f['frame_index'] for f in frames]==list(range(T))
+    assert len({f['source_frame_id'] for f in frames})==T
+    for frame in frames:
+        for match in frame.get('annotation_matches',{}).values():
+            if 'available' in match:match['source_record_present']=match.pop('available')
+    lines(p/'camera/frames.jsonl',frames)
+    summary=[]
+    for key,c in m['components'].items():
+        assert c['schema_id'] in registry['components'],(key,c['schema_id'])
+        assert c['available'] and c['source_locator'] and c['provenance']['kind'] in ['source','derived','format_conversion']
+        data=safe_ref(p,c['data_ref']);valid=arr(safe_ref(p,c['valid_ref']))
+        assert valid.dtype==bool and valid.ndim==len(c['valid_axes'])
+        if data.suffix=='.pt':
+            a=arr(data);assert a.dtype==np.float32 and a.shape[0]==T
+            assert a.shape[:valid.ndim]==valid.shape and len(c['axes'])==a.ndim
+            assert np.isfinite(a[valid]).all() and np.isnan(a[~valid]).all()
+            if c['schema_id'] in ['object/rigid/v1','transform/se3/v1']:assert_se3(a[valid])
+            if c['schema_id']=='mano/axis_angle61/v1':assert a.shape==(T,2,61)
+        elif data.suffix=='.mkv':
+            assert c['storage']['profile'] in registry['storage_profiles']
+            assert valid.shape==(T,) and valid.all()
+        for attr in c.get('attributes',{}).values():
+            a=arr(safe_ref(p,attr['data_ref']));b=arr(safe_ref(p,attr['valid_ref']))
+            assert b.dtype==bool and a.shape==b.shape
+            assert np.isfinite(a[b]).all() and ((a[b]>=0)&(a[b]<=1)).all() and np.isnan(a[~b]).all()
+        if 'distortion_ref' in c:
+            d=arr(safe_ref(p,c['distortion_ref']));assert d.shape==(T,8) and np.isfinite(d).all()
+        summary.append({'component':key,'schema':c['schema_id'],'valid':int(valid.sum()),'total':int(valid.size)})
+    for op in sorted((p/'annotation/object').glob('*')):
+        om=load(op/'metainfo.json');mesh=safe_ref(p,om['mesh_ref'])
+        if 'vertex_part_ids_ref' in om:
+            ids=arr(safe_ref(p,om['vertex_part_ids_ref']));assert ids.dtype==np.int64
+            assert len(ids)==sum(line.startswith('v ') for line in mesh.read_text().splitlines())
+            assert om['topology_id']=='sha256:'+digest(mesh)
+            if not (op/'pose.pt').exists():
+                assert m.get('omitted_components',{}).get('object_pose')
+            else:
+                root=arr(op/'pose.pt');top=arr(op/'parts/top/pose.pt');joint=arr(op/'joint_state.pt')[:,0]
+                ok=arr(op/'pose_valid.pt');H=np.broadcast_to(np.eye(4),(ok.sum(),4,4)).copy()
+                H[:,:3,:3]=Rotation.from_rotvec(np.c_[np.zeros((ok.sum(),2)),-joint[ok]]).as_matrix()
+                if ok.any():assert np.max(np.abs(root[ok]@H-top[ok]))<2e-6
+        for obj in (op/'assets').glob('*.obj'):
+            for row in obj.read_text().splitlines():
+                if row.startswith('mtllib '):
+                    material=safe_ref(obj.parent,row.split(maxsplit=1)[1])
+                    for r in material.read_text().splitlines():
+                        if r.lower().startswith('map_kd '):safe_ref(material.parent,r.split(maxsplit=1)[1])
+    records=load(DATA_ROOT/m['dataset_id']/'tmp'/RUN/'checks'/f'{m["sequence_id"]}__{m["camera_id"]}.json')['checks']
+    assert records and all(x.get('passed',False) for x in records)
+    media=next(r for r in records if r['test'] in ['all_source_JPEG_packets_and_RGB_pixels_exact','all_source_JPEG_packets_exact']);assert media['frames']==T
+    if media['test']=='all_source_JPEG_packets_exact':
+        pixels=next(r for r in records if r['test']=='source_RGB_pixels_exact')
+        assert pixels['scope']=='deterministic_sample' and pixels['sampled_frames']==len(pixels['frame_indices']) and pixels['sampled_frames']==min(T,16)
+        assert pixels['decoded_from_container'] and pixels['frame_indices']==sorted(set(pixels['frame_indices']))
+    if 'mano' not in m['components']:
+        assert m.get('omitted_components',{}).get('mano')
+        assert not (p/'annotation/mano/motion.pt').exists()
+        return m,frames,{'camera_id':m['camera_id'],'frames':T,'components':summary,'conversion_checks':records,'structure_and_references':'passed','mano':'source modality absent, omitted'}
+    actual=arr(p/'annotation/mano/motion_valid.pt')
+    if m['dataset_id']=='HOT3D':
+        src=m['components']['mano']['source_locator']
+        with zipfile.ZipFile(DATA_ROOT/'HOT3D'/src['archive']) as z:
+            native={r['timestamp_ns']:r['hand_poses'] for r in map(json.loads,z.read(src['member']).splitlines())}
+        expected=np.array([[str(side) in native.get(f['annotation_matches']['mano']['source_timestamp_ns'],{}) for side in range(2)] for f in frames],bool)
+    else:expected=np.array([[f['source_annotation_index'] is not None]*2 for f in frames],bool)
+    if m['dataset_id']=='HOT3D' and m['components']['mano'].get('valid_policy'):
+        source=m['components']['mano']['valid_policy']['source_locator']
+        with zipfile.ZipFile(DATA_ROOT/'HOT3D'/source['archive']) as z:
+            lookup={int(r['timestamp[ns]']):r['mask']=='True' for r in csv.DictReader(io.StringIO(z.read(source['member']).decode())) if r['stream_id']==m['camera_id']}
+        qa=[lookup[f['timecode_ns']] for f in frames]
+        expected &= np.array(qa,bool)[:,None]
+    if m['dataset_id']=='ARCTIC':
+        rejected=m['components']['mano'].get('valid_policy',{}).get('local_excluded_frame_indices',[])
+        expected[rejected]=False
+    assert np.array_equal(expected,actual),'annotation validity disagrees with declared source and QA/local policy'
+    records.append({'test':'all_frame_source_annotation_and_QA_policy','passed':True,'left_right_valid_counts':actual.sum(0).tolist(),'frames_without_usable_hand':int((~actual.any(1)).sum())})
+    return m,frames,{'camera_id':m['camera_id'],'frames':T,'components':summary,'conversion_checks':records,'structure_and_references':'passed'}
+
+def select_images(p,indices,dest):
+    dest.mkdir(parents=True,exist_ok=True)
+    selection='+'.join(f'eq(n,{i})' for i in indices)
+    command([FF,'-v','error','-y','-threads','2','-noautorotate','-i',p/'observation/rgb.mkv',
+        '-vf',f"select='{selection}',scale=900:900:force_original_aspect_ratio=decrease",'-fps_mode','vfr',dest/'decoded-%03d.png'])
+    output=sorted(dest.glob('decoded-*.png'));assert len(output)==len(indices)
+    return dict(zip(indices,output))
+
+def project_for_view(points,m,params,scale):
+    uv=project_points(points,m['camera']['model'],params)
+    if m['camera']['model']=='fisheye624':
+        angle=np.arctan2(np.linalg.norm(points[:,:2],axis=1),points[:,2])
+        uv[angle>m['camera']['max_solid_angle']]=np.nan
+    return uv*scale
+
+def draw_edges(canvas,uv,edges,color):
+    H,W=canvas.shape[:2]
+    for a,b in edges:
+        xy=uv[[a,b]]
+        if not np.isfinite(xy).all():continue
+        if np.max(np.abs(xy))>max(W,H)*20:continue
+        xy=np.rint(xy).astype(int)
+        if np.linalg.norm(xy[0]-xy[1])>max(W,H)*.35:continue
+        ok,start,end=cv2.clipLine((0,0,W,H),tuple(xy[0]),tuple(xy[1]))
+        if ok:cv2.line(canvas,start,end,color,1,cv2.LINE_AA)
+
+def review_view(p,m,frames,dest):
+    T=m['num_frames'];indices=sorted(set([0,1,T//4,T//2,3*T//4,T-1]));imgs=select_images(p,indices,dest)
+    K=arr(p/'camera/intrinsic.pt');D=arr(p/'camera/distortion.pt') if (p/'camera/distortion.pt').exists() else None
+    motion=arr(p/'annotation/mano/motion.pt');valid=arr(p/'annotation/mano/motion_valid.pt');models=[Mano('left'),Mano('right')]
+    if m['components']['mano']['model_profile']['left_shapedirs_fix']:
+        models[0].data['shapedirs'][:,0]*=-1
+        with torch.no_grad():models[0].layer.shapedirs[:,0]*=-1
+    objects=[]
+    for op in sorted((p/'annotation/object').iterdir()):
+        om=load(op/'metainfo.json');mesh=trimesh.load(p/om['mesh_ref'],process=False,force='mesh')
+        if 'vertex_part_ids_ref' in om:
+            vertices=np.array([[float(x) for x in r.split()[1:4]] for r in (p/om['mesh_ref']).read_text().splitlines() if r.startswith('v ')])
+            # trimesh may split OBJ vertices along UV seams; render exact source-ordered vertices as points.
+            objects.append((op,om,vertices,None,arr(op/'pose.pt'),arr(op/'pose_valid.pt'),arr(p/om['vertex_part_ids_ref']),arr(op/'parts/top/pose.pt')))
+        else:
+            edges=np.asarray(mesh.edges_unique);edges=edges[::max(1,len(edges)//2500)]
+            objects.append((op,om,np.asarray(mesh.vertices),edges,arr(op/'pose.pt'),arr(op/'pose_valid.pt'),None,None))
+    records=[]
+    for i,source in imgs.items():
+        original=np.asarray(Image.open(source).convert('RGB'));overlay=original.copy()
+        scale=np.array([original.shape[1]/m['camera']['image_size'][0],original.shape[0]/m['camera']['image_size'][1]])
+        params=K[i] if m['camera']['model']=='fisheye624' else (K[i],D[i] if D is not None else None)
+        for side,model in enumerate(models):
+            if not valid[i,side]:continue
+            h=motion[i,side];vertices,_,_=model.geometry(h[:48],h[48:58],h[58:]);uv=project_for_view(vertices[0],m,params,scale)
+            faces=np.asarray(model.layer.faces);edges=np.concatenate([faces[:,[0,1]],faces[:,[1,2]],faces[:,[2,0]]]);draw_edges(overlay,uv,edges,(0,235,255) if side==0 else (255,135,20))
+        palette=[(110,255,100),(255,80,210),(240,220,60),(90,180,255),(190,160,255),(255,120,140)]
+        for c,(op,om,vertices,edges,P,ok,parts,top) in enumerate(objects):
+            if not ok[i]:continue
+            xyz=vertices@P[i,:3,:3].T+P[i,:3,3]
+            if parts is not None:xyz[parts==1]=vertices[parts==1]@top[i,:3,:3].T+top[i,:3,3]
+            uv=project_for_view(xyz,m,params,scale)
+            if edges is not None:draw_edges(overlay,uv,edges,palette[c%len(palette)])
+            else:
+                for xy in uv[::max(1,len(uv)//6000)]:
+                    if np.isfinite(xy).all() and 0<=xy[0]<overlay.shape[1] and 0<=xy[1]<overlay.shape[0]:cv2.circle(overlay,tuple(np.rint(xy).astype(int)),1,palette[c%len(palette)],-1)
+        blended=cv2.addWeighted(overlay,.8,original,.2,0);output=dest/f'frame-{i:06d}-overlay.png';raw=dest/f'frame-{i:06d}-rgb.png'
+        Image.fromarray(blended).save(output);source.replace(raw)
+        records.append({'frame_index':i,'source_frame_id':frames[i]['source_frame_id'],'rgb':raw.name,'overlay':output.name,'mano_valid':valid[i].tolist()})
+    return records
+
+def write_reports(dataset,views):
+    base=DATA_ROOT/dataset;root=base/RELEASE;doc=root/'doc';review=doc/'validation';reports=[];cards=[];thumbs=[]
+    rawpaths=set()
+    for p in views:
+        if dataset=='ARCTIC':
+            for assets in (p/'annotation/object').glob('*/assets'):repair_arctic_material_refs(assets)
+        m,frames,summary=validate_view(p);cam=m['camera_id'];samples=review_view(p,m,frames,review/'images'/cam)
+        summary['visual_samples']=samples;reports.append(summary)
+        # All actually consumed source files; camera/mano/object locators name source archives.
+        rawpaths.add(m['components']['rgb']['source_locator'].get('archive',m['components']['rgb']['source_locator'].get('file')))
+        for comp in m['components'].values():
+            src=comp['source_locator']
+            if 'archive' in src:rawpaths.add(src['archive'])
+        if dataset=='ARCTIC':rawpaths.add('raw/data/splits_json.zip')
+        for op in (p/'annotation/object').iterdir():
+            glb=load(op/'metainfo.json').get('original_glb_ref')
+            if glb:rawpaths.add(load(op/'metainfo.json')['original_glb_source_locator']['file'])
+        opts=''.join(f'<option value="{i}"'+(' selected' if s['frame_index']==m['num_frames']//2 else '')+f'>index {s["frame_index"]} · source {s["source_frame_id"]} · valid {s["mano_valid"]}</option>' for i,s in enumerate(samples))
+        samples_js=json.dumps([{k:(f'images/{cam}/'+s[k] if k in ('rgb','overlay') else s[k]) for k in s} for s in samples])
+        cards.append(f'<section class="card" data-samples=\'{html.escape(samples_js,quote=True)}\'><h2>Camera {cam} · {m["camera"]["model"]} · {m["num_frames"]} 帧</h2><select onchange="show(this.closest(\'.card\'))">{opts}</select><label><input type="checkbox" checked onchange="show(this.closest(\'.card\'))"> 显示投影</label><br><a class="zoom" target="_blank"><img alt="共享产物实际帧与投影"></a><p class="caption"></p></section>')
+        middle=samples[len(samples)//2];thumb=Image.open(review/'images'/cam/middle['overlay']).convert('RGB');thumb.thumbnail((400,320));tile=Image.new('RGB',(420,355),'#18202b');tile.paste(thumb,((420-thumb.width)//2,28));ImageDraw.Draw(tile).text((10,8),f'{dataset} camera {cam} / frame {middle["frame_index"]}',fill='white');thumbs.append(tile)
+        print('REVIEWED',dataset,cam,flush=True)
+    cols=3;grid=Image.new('RGB',(cols*420,((len(thumbs)+cols-1)//cols)*355),'#18202b')
+    for i,t in enumerate(thumbs):grid.paste(t,((i%cols)*420,(i//cols)*355))
+    grid.save(review/'overview.png')
+    source_manifest=[]
+    ledger=load(base/'raw/doc'/('archive_checksums.json' if dataset=='ARCTIC' else 'file_checksums.json'))
+    for rel in sorted(x for x in rawpaths if x):
+        file=base/rel;sha=digest(file);record=ledger[rel.removeprefix('raw/data/')]
+        assert record['sha256']==sha and record['size']==file.stat().st_size
+        assert record.get('official_sha256_match',record.get('official_digest_match')) is True
+        if file.suffix=='.zip':assert record['zip_crc']=='passed_all_members'
+        if dataset=='HOT3D':assert record['fresh_download_run']=='2026-09-18-aria-fresh'
+        source_manifest.append({'path':rel,'bytes':file.stat().st_size,'sha256':sha,'matches_verified_official_download':True,'zip_crc_at_download':record.get('zip_crc')})
+    save(doc/'pilot-source-manifest.json',source_manifest)
+    script_manifest=[]
+    for name in ['review_fresh_sequence_pilots.py','fresh_sequence_common.py','convert_fresh_sequence_pilots.py']:
+        shutil.copyfile(WORK/'scripts'/name,root/'script'/name)
+    for p in sorted((root/'script').glob('*.py')):script_manifest.append({'path':p.relative_to(root).as_posix(),'sha256':digest(p)})
+    full={'dataset':dataset,'stage':'pilot_awaiting_human_review','human_review':'pending','full_preprocessing_authorized':False,'upload_authorized':False,
+        'scope':'One complete selected sequence, every released camera frame; not full-dataset acceptance','views':reports,
+        'sources_manifest':'../pilot-source-manifest.json','script_manifest':script_manifest,'preview_selection':'Deterministic boundary and quartile samples per camera; not random or full-dataset review',
+        'preview_rendering':'Derived MANO/object projections from shared tensors, not segmentation ground truth; no depth occlusion test, hidden surfaces may be drawn',
+        'created_at':datetime.now().astimezone().isoformat()}
+    save(review/'report.json',full);save(review/'status.json',{k:full[k] for k in ['dataset','stage','human_review','full_preprocessing_authorized','upload_authorized']})
+    notes=('首帧 RGB 存在，动捕标注无对应项，valid=false；已保留。相机 0 使用 rational8 畸变，外部八相机使用官方针孔投影。关节绕父坐标系 -Z 轴旋转，部件点/网格保持米制。' if dataset=='ARCTIC' else 'Aria 三路相机均保留；RGB 与两路灰度均由原 JPEG 直接封装。device time 经官方映射对齐 timecode；原始 visibility 按百分比转 [0,1]，独立于标注 valid。三路按时间匹配，不假定相同行号自动同步。')
+    if dataset=='HOT3D':notes+=' 每相机双手各806帧有MANO、208帧原始手参数为空；画面中有手但无投影的末尾帧属于源标注缺失，已验证并保留为valid=false。'
+    if dataset=='ARCTIC':notes+=' 原始部件OBJ的材质文件名引用错误已在共享层修正为实际存在的material.mtl，材质标识与纹理不变。'
+    missing=('本次官方 raw 中无独立深度、像素分割、接触标签、实测手关键点数组。保留 MANO，供任务层按需生成关键点；未把预览投影写成新增标注。' if dataset=='ARCTIC' else '本范围无深度流或像素分割；masks/*.csv 是帧筛选条件，不能当作像素 mask。UmeTrack 与 MPS 替代估计保留在 raw；shared 选择 MANO 和官方 GT 相机。没有额外生成关键点或接触标签。')
+    text=f'''<!doctype html><html lang="zh-CN"><meta charset="utf-8"><meta name="viewport" content="width=device-width, initial-scale=1"><title>{dataset} 序列试点审核</title><style>body{{font:16px/1.65 system-ui;background:#f3f5f8;color:#18202b;max-width:1450px;margin:24px auto;padding:0 20px}}header,.card{{background:white;padding:22px;border-radius:14px;margin-bottom:18px}}.status{{color:#9a5800;font-weight:bold}}.grid{{display:grid;grid-template-columns:repeat(auto-fit,minmax(360px,1fr));gap:18px}}.card img{{max-width:100%;max-height:670px;margin-top:12px}}select{{max-width:100%;padding:8px}}a{{color:#1766ab}}label{{display:inline-block;margin:10px}}h1{{margin:0}}.caption{{font-size:14px}}</style><header><h1>{dataset} · 完整序列试点</h1><p class="status">自动校验已完成；人工审核待确认。尚未开始全量预处理或上传。</p><p>共 {len(views)} 路相机、{sum(r['frames'] for r in reports):,} 个相机帧。全部源 JPEG 码流、全部解码 RGB 帧已逐一比较一致；数值转换、有效性和引用检查通过。</p><p>{notes}</p><p>{missing}</p><p>下方是从 shared 实际产物读取的边界及四分位样本。青色=左手，橙色=右手，其余颜色=物体。投影仅用于几何核验，<b>不是源分割，也没有做遮挡剔除</b>。点击图像可放大；取消勾选可查看原图。</p><p><a href="REPORT.md">验收说明</a> · <a href="report.json">机器可读结果</a> · <a href="../RAW_TO_SHARED.md">raw → share 对照</a> · <a href="../CONTRACT.md">约定</a> · <a href="overview.png">全部相机总览</a></p></header><main class="grid">{''.join(cards)}</main><script>function show(card){{let s=JSON.parse(card.dataset.samples)[Number(card.querySelector('select').value)];let path=card.querySelector('input').checked?s.overlay:s.rgb;card.querySelector('img').src=path;card.querySelector('.zoom').href=path;card.querySelector('.caption').textContent='frame_index='+s.frame_index+' / 原帧 '+s.source_frame_id+' / 左右手 valid='+s.mano_valid;}}document.querySelectorAll('.card').forEach(show)</script></html>'''
+    (review/'index.html').write_text(text,encoding='utf-8')
+    rows='\n'.join(f'| {r["camera_id"]} | {r["frames"]} | '+', '.join(str(x['frame_index']) for x in r['visual_samples'])+' | 通过 |' for r in reports)
+    report=f'''# {dataset} 完整序列试点验收\n\n状态：**自动检查通过，人工确认待完成**。本报告不是全量验收或发布许可。\n\n[离线预览](index.html) · [机器结果](report.json) · [原始包摘要](../pilot-source-manifest.json)\n\n| 相机 | 全序列帧数 | 可视化帧索引 | 自动校验 |\n|---|---:|---|---|\n{rows}\n\n## 已验证\n\n- 完整所选序列的每一张源 JPEG 与 MKV 取回的压缩包 SHA256 相同；全部帧在同一 FFmpeg/noautorotate/rgb24 配置下解码字节相同，顺序和数量一致。不是有损重编码。\n- 全部有效 MANO 帧的相机坐标变换与直接变换几何一致；独立源模型配置抽样校验非平手均值/PCA 展开。张量 float32 保存读取一致，invalid 为 NaN，valid 为 bool。\n- SE(3) 正交性、单位与资产引用检查；ARCTIC 部件关节运动与源旋转公式逐帧对照，HOT3D 物体 OBJ 尺度、三角形及纹理像素往返一致。\n- 所有实际组件使用现有注册 schema；没有为缺省模态伪造值。\n\n## 帧对应和缺省范围\n\n{notes}\n\n{missing}\n\n## 人工核验边界\n\n每相机选序列起点、紧邻帧、四分位、中点、四分之三及终点，不是随机抽查。预览只用于核对几何对齐；投影线条/点不是分割真值，遮挡表面可能被画出。数值自洽不等于图像投影正确，请检查所有相机和多个时刻。此处一序列通过不能覆盖全数据集、其他物体或异常序列。全量处理后仍需独立随机抽查和用户发布确认。\n'''
+    (review/'REPORT.md').write_text(report,encoding='utf-8')
+    return full
+
+if __name__=='__main__':
+    raise SystemExit('Static-sample review is retired. Use render_full_sequence_review.py DATASET and finalize_fresh_sequence_review.py; validate_view remains a reusable numeric helper.')
diff --git a/ARCTIC/shared-v1.0/script/sequence_pilot_common.py b/ARCTIC/shared-v1.0/script/sequence_pilot_common.py
new file mode 100644
index 0000000000000000000000000000000000000000..169c4868c9fbba40790ee94c2600fb35e418acef
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/sequence_pilot_common.py
@@ -0,0 +1,53 @@
+"""Whole-selection conversion utilities. Never reads previous pilot outputs."""
+from pilot_build import *
+import hashlib
+
+RELEASE='shared-sequence-v1.1'
+
+class View:
+ def __init__(self,dataset,seq,cam,ids,size,source,kind='video',sampling='dense',split='unknown'):
+  self.root=ROOT/dataset/RELEASE;self.p=self.root/'data'/seq/cam
+  self.p.mkdir(parents=True,exist_ok=True);self.T=len(ids);self.source=source;self.checks=[]
+  self.m={'schema_version':'hand-warehouse/1.1','dataset_id':dataset,'release_id':RELEASE,'sequence_id':seq,'camera_id':cam,'sample_type':kind,'num_frames':self.T,'hand_slots':['left','right'],'sampling':{'type':'single' if kind=='image' else sampling,'source_fps':None,'frames_ref':'camera/frames.jsonl'},'camera':{'model':'unknown','schema_id':None,'image_size':size},'components':{},'split':split}
+  self.frames=[{'frame_index':i,'source_frame_id':str(fid),'timestamp_ns':None,'clock_domain':None,'source_locator':{**source,'source_frame_id':str(fid)}} for i,fid in enumerate(ids)]
+ def add(self,name,path,schema,axes,valid,method='source values retained',**kw):
+  vp=Path(path).with_name(Path(path).stem+'_valid.pt').as_posix()
+  tensor(self.p/vp,valid)
+  self.m['components'][name]={'available':True,'schema_id':schema,'data_ref':path,'axes':axes,'valid_ref':vp,'valid_axes':axes[:valid.ndim],'provenance':{'kind':'format_conversion','method':method},'source_locator':self.source,**kw}
+ def array(self,name,path,schema,a,valid,axes,**kw):
+  a=np.asarray(a,np.float32).copy();valid=np.asarray(valid,bool)
+  assert a.shape[:valid.ndim]==valid.shape and a.shape[0]==self.T
+  assert np.isfinite(a[valid]).all(),name
+  a[~valid]=np.nan;tensor(self.p/path,a);self.add(name,path,schema,axes,valid,**kw)
+  restored=torch.load(self.p/path,map_location='cpu',weights_only=True).numpy()
+  for index in range(self.T):assert np.array_equal(restored[index],a[index],equal_nan=True)
+  self.checks.append({'test':'tensor_serialization_exact','file':path,'shape':list(a.shape),'valid_count':int(valid.sum())})
+ def video(self,name,path,iterator,mask=False,**kw):
+  w,h=self.m['camera']['image_size'];pix='gray16le' if mask else 'rgb24';dtype='<u2' if mask else 'uint8'
+  dst=self.p/path;dst.parent.mkdir(parents=True,exist_ok=True);hashes=[]
+  proc=subprocess.Popen([FF,'-v','error','-y','-f','rawvideo','-pixel_format',pix,'-video_size',f'{w}x{h}','-framerate','1','-i','pipe:0','-c:v','ffv1','-level','3','-g','1',str(dst)],stdin=subprocess.PIPE)
+  try:
+   for frame in iterator:
+    a=np.asarray(frame,dtype=dtype);assert a.shape==((h,w) if mask else (h,w,3));b=a.tobytes();hashes.append(hashlib.sha256(b).hexdigest());proc.stdin.write(b)
+  finally:proc.stdin.close()
+  assert proc.wait()==0 and len(hashes)==self.T
+  proc=subprocess.Popen([FF,'-v','error','-i',str(dst),'-f','rawvideo','-pix_fmt',pix,'pipe:1'],stdout=subprocess.PIPE)
+  nbytes=w*h*(2 if mask else 3)
+  for digest in hashes:
+   b=proc.stdout.read(nbytes);assert len(b)==nbytes and hashlib.sha256(b).hexdigest()==digest,'video roundtrip'
+  assert not proc.stdout.read(1);assert proc.wait()==0
+  self.checks.append({'test':'all_frames_pixel_exact_roundtrip','file':path,'frames':self.T})
+  self.add(name,path,'segment/uint16_labels/v1' if mask else 'rgb/uint8/v1',['T','height','width'] if mask else ['T','height','width','channel'],np.ones(self.T,bool),storage_fps=1,timebase_role='index_only',**kw)
+ def intrinsic(self,K):
+  self.m['camera'].update(model='pinhole',schema_id='camera/pinhole/v1')
+  self.array('intrinsic','camera/intrinsic.pt','camera/pinhole/v1',np.broadcast_to(K,(self.T,3,3)),np.ones(self.T,bool),['T','row','column'])
+ def finish(self):
+  lines(self.p/'camera/frames.jsonl',self.frames);save(self.p/'metainfo.json',self.m)
+  save(self.root/'doc'/f'{self.m["sequence_id"]}__{self.m["camera_id"]}.checks.json',{'checks':self.checks,'frames':self.T,'input':self.source})
+  print('Converted',self.m['dataset_id'],self.m['sequence_id'],self.m['camera_id'],self.T,flush=True)
+
+def snapshot(dataset,scripts):
+ out=ROOT/dataset/RELEASE/'script';out.mkdir(parents=True,exist_ok=True)
+ names={'sequence_pilot_common.py','pilot_build.py','pilot_fetch.py','pilot_mano.py','pilot_hot_normalize.py','audit_source_fields.py',*scripts}
+ names.update(p.name for p in (WORK/'scripts').glob('sequence_pilot_*.py'))
+ for name in names:shutil.copyfile(WORK/'scripts'/name,out/name)
diff --git a/ARCTIC/shared-v1.0/script/sequence_pilot_world.py b/ARCTIC/shared-v1.0/script/sequence_pilot_world.py
new file mode 100644
index 0000000000000000000000000000000000000000..1d052448db5f35e32a80867f7dc2a6cda27a8065
--- /dev/null
+++ b/ARCTIC/shared-v1.0/script/sequence_pilot_world.py
@@ -0,0 +1,55 @@
+"""Convert whole native annotation sequences to camera-space MANO."""
+from sequence_pilot_common import *
+from audit_source_fields import native_npy
+from pilot_mano import Mano
+from pilot_hot_normalize import transform
+
+def camera_mano(model,pose,beta,translation,E):
+ pose=np.asarray(pose).reshape(-1,48);beta=np.asarray(beta).reshape(-1,10);translation=np.asarray(translation).reshape(-1,3)
+ root=np.einsum('v,tvc->tc',model.data['J_regressor'][0],model.data['v_template'][None]+np.einsum('vci,ti->tvc',model.data['shapedirs'],beta))
+ newpose=pose.copy();newpose[:,:3]=np.stack([cv2.Rodrigues(e[:3,:3]@cv2.Rodrigues(p[:3])[0])[0].ravel() for e,p in zip(E,pose)])
+ newt=np.einsum('tij,tj->ti',E[:,:3,:3],root+translation)+E[:,:3,3]-root
+ # Check all frames in modest batches; meshes exist only in memory for verification.
+ errors=[]
+ for begin in range(0,len(pose),64):
+  s=slice(begin,begin+64);world,_,_=model.geometry(pose[s],beta[s],translation[s]);cam,_,_=model.geometry(newpose[s],beta[s],newt[s]);target=np.einsum('tij,tvj->tvi',E[s,:3,:3],world)+E[s,None,:3,3]
+  errors.append(float(np.linalg.norm(cam-target,axis=-1).max()))
+ assert max(errors)<2e-5,max(errors)
+ return np.concatenate([newpose,beta,newt],axis=-1),max(errors)
+
+def arctic():
+ seq='capsulemachine_use_02';src=OLD/'ARCTIC/raw_seqs/s01';a=native_npy(src/(seq+'.mano.npy')).item();c=native_npy(src/(seq+'.egocam.dist.npy')).item();T=len(a['left']['rot']);misc=load(OLD/'ARCTIC/meta/misc.json')['s01'];ids=list(range(T))
+ v=View('ARCTIC','s01_'+seq,'ego',ids,misc['image_size'][0],{'path':str(src.relative_to(WORK)),'sequence':seq},sampling='full_annotation_sequence')
+ E=np.tile(np.eye(4),(T,1,1));E[:,:3,:3]=c['R_k_cam_np'];E[:,:3,3]=c['T_k_cam_np'].reshape(T,3)
+ v.intrinsic(c['intrinsics']);v.m['camera'].update(model='rational8',schema_id='camera/rational8/v1');v.m['components']['intrinsic'].update(schema_id='camera/rational8/v1',distortion_ref='camera/distortion.pt')
+ v.array('distortion','camera/distortion.pt','camera/distortion8/v1',np.broadcast_to(c['dist8'],(T,8)),np.ones(T,bool),['T','parameter'],parameter_names=['k1','k2','p1','p2','k3','k4','k5','k6'])
+ v.array('extrinsic','camera/extrinsic.pt','transform/se3/v1',E,np.ones(T,bool),['T','row','column'],unit='m',convention='T_camera_from_world')
+ motion=np.empty((T,2,61))
+ for s,side in enumerate(['left','right']):
+  b=a[side];model=Mano(side);pose=np.c_[b['rot'],b['pose']].copy();pose[:,3:]+=model.data['hands_mean'];motion[:,s],err=camera_mano(model,pose,np.broadcast_to(b['shape'],(T,10)),b['trans'],E);v.checks.append({'test':'all_frame_camera_MANO_geometry','side':side,'max_m':err,'native_hand_mean':'ARCTIC common/body_models.py flat_hand=False; mean added before normalization'})
+ v.array('mano','annotation/mano/motion.pt','mano/axis_angle61/v1',motion,np.ones((T,2),bool),['T','hand','parameter'],unit='m',angle_unit='rad',coordinate_frame='camera',model_profile={'version':'MANO-v1.2','flat_hand_mean':True,'left_shapedirs_fix':False})
+ for i,row in enumerate(v.frames):row.update(source_annotation_index=i,source_image_frame_id=i+misc['ioi_offset'])
+ v.m['input_gaps']=['full matching RGB image inventory and selected object.npy/object meshes unavailable locally; not source absence'];v.finish();snapshot('ARCTIC',['sequence_pilot_world.py','audit_source_fields.py','pilot_mano.py'])
+
+def oakink2():
+ src=next((OLD/'OakInk2/anno_preview').glob('*.pkl'));a=read_pickle(src);ids=a['frame_id_list'];T=len(ids);cam='104422070969';key=a['cam_def'][cam];images=OLD/'OakInk2/data'/src.stem/cam
+ first=np.asarray(Image.open(images/f'{ids[0]:06d}.png'));v=View('OakInk2',src.stem,cam,ids,[first.shape[1],first.shape[0]],{'annotation':str(src.relative_to(WORK)),'image_root':str(images.relative_to(WORK))},sampling='official_frame_id_list')
+ v.video('rgb','observation/rgb.mkv',(np.asarray(Image.open(images/f'{i:06d}.png').convert('RGB')) for i in ids))
+ K=np.stack([a['cam_intr'][key][i] for i in ids]);E=np.stack([a['cam_extr'][key][i] for i in ids]);v.m['camera'].update(model='pinhole',schema_id='camera/pinhole/v1')
+ v.array('intrinsic','camera/intrinsic.pt','camera/pinhole/v1',K,np.ones(T,bool),['T','row','column'])
+ v.array('extrinsic','camera/extrinsic.pt','transform/se3/v1',E,np.ones(T,bool),['T','row','column'],unit='m',convention='T_camera_from_world',method='official preview applies cam_extr @ world object transform')
+ motion=np.empty((T,2,61))
+ for s,side in enumerate(['left','right']):
+  prefix=['lh','rh'][s];rows=[a['raw_mano'][i] for i in ids];q=np.stack([np.asarray(r[prefix+'__pose_coeffs']).reshape(16,4) for r in rows]);pose=np.stack([cv2.Rodrigues(transform(qq,[0,0,0])[:3,:3])[0].ravel() for qq in q.reshape(-1,4)]).reshape(T,48)
+  beta=np.stack([np.asarray(r[prefix+'__betas']).reshape(10) for r in rows]);trans=np.stack([np.asarray(r[prefix+'__tsl']).reshape(3) for r in rows]);model=Mano(side)
+  root=np.einsum('v,tvc->tc',model.data['J_regressor'][0],model.data['v_template'][None]+np.einsum('vci,ti->tvc',model.data['shapedirs'],beta))
+  motion[:,s],err=camera_mano(model,pose,beta,trans-root,E);v.checks.append({'test':'all_frame_camera_MANO_geometry','side':side,'max_m':err,'source_center':'official manotorch center_idx=0; subtract shape-dependent wrist'})
+ v.array('mano','annotation/mano/motion.pt','mano/axis_angle61/v1',motion,np.ones((T,2),bool),['T','hand','parameter'],unit='m',coordinate_frame='camera',angle_unit='rad',model_profile={'version':'MANO-v1.2','flat_hand_mean':True,'left_shapedirs_fix':False},method='source wxyz quaternions to axis-angle; same mocap frame keys as official preview; root-corrected world to camera')
+ for oid in a['obj_list']:
+  W=np.stack([a['obj_transf'][oid][i] for i in ids]);P=E@W;valid=np.isfinite(P).all((1,2));rel=f'annotation/object/{oid}'
+  v.array('object_'+oid+'_pose',rel+'/pose.pt','object/rigid/v1',P,valid,['T','row','column'],unit='m',convention='T_camera_from_object')
+  save(v.p/rel/'metainfo.json',{'object_id':oid,'identity_scope':'sequence','scope_id':src.stem,'has_model':False,'source_locator':v.source})
+ v.m['input_gaps']=['object asset models not yet downloaded'];v.m['semantic_review_required']=['MANO quaternion hand mean convention must be compared with official renderer; algebraic check alone is insufficient'];v.finish();snapshot('OakInk2',['sequence_pilot_world.py','audit_source_fields.py','pilot_mano.py'])
+
+if __name__=='__main__':
+ p=argparse.ArgumentParser();p.add_argument('dataset',choices=['ARCTIC','OakInk2']);a=p.parse_args();{'ARCTIC':arctic,'OakInk2':oakink2}[a.dataset]()