diff --git a/AGENTS.md b/AGENTS.md index c12748ec6..e55f1c642 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -8,3 +8,5 @@ - Preserve original data, failures, review limitations and immutable evidence. Do not label proposals as executed or reviewed results. - Current E1 scope follows the standing continue-work authorization relayed in Slack `1794651141.035699` and Claude caps `1794651169.320839`: bounded CPU-only label-free preparation and synthetic checks may proceed after the count checkpoint, without another user approval loop. Total 16 CPU-hours / 25 GiB, at most 8 processes, per-job 30 min / 1.25 GiB tree RSS, at most 2000 registration pairs; the 10-part stage additionally stays sequential within 3 h / 5 GiB. Never read labels, fit models, use GPU, run pilot/full matrix, or exceed caps in this scope. See `docs/s2-e1-budget-checkpoint-2026-11-14.md`. + +- Post-witness update (2026-11-14 21:12) supersedes treating the preceding E1-only scope as the entire remaining project scope. Follow the latest section of `docs/collaboration-stage-review-policy.md` and Claude Slack `1794661944.308269`: custody join, label-free copy calibration, k=3 per-route candidate count, bounded registration (up to 4000 pairs / 100 worker CPU-h / 16 workers), band review, component freeze, then coordinated S2 continuation. No repeated approval gates or reruns of completed review; preserve concrete execution limits and distinguish future stages from completed work. User preference: tens of CPU-hours alone are not a reason to block progress. No additional scientific task is implied by this preference. diff --git a/artifacts/s2-g1-custody-handoff-2026-11-14/frozen/frozen-review.bundle b/artifacts/s2-g1-custody-handoff-2026-11-14/frozen/frozen-review.bundle new file mode 100644 index 000000000..a46f3fa96 Binary files /dev/null and b/artifacts/s2-g1-custody-handoff-2026-11-14/frozen/frozen-review.bundle differ diff --git a/artifacts/s2-g1-custody-handoff-2026-11-14/frozen/frozen-review.csv b/artifacts/s2-g1-custody-handoff-2026-11-14/frozen/frozen-review.csv new file mode 100644 index 000000000..46876a7cb --- /dev/null +++ b/artifacts/s2-g1-custody-handoff-2026-11-14/frozen/frozen-review.csv @@ -0,0 +1,121 @@ +anonymous_pair_id,relation,local_evidence,resolvable,unresolved_reason,reviewer,review_version +0505e3702a7e7b81a7f1f41448fab17b,unrelated,Opposed edge tabs on a broad plate (L); two large corner perforations and a row of small marginal holes (R).,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +05868efab3233c5ca115ace5403d8a36,unrelated,At comparable length the left bar has a wide concave trough spanning its section; the right has a compact rounded four-sided end and no corresponding trough.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +08bb116a72fb66bd1a08069aa09bb7da,unrelated,Perforated circular flange and central boss versus an open helical wire.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +0915e6a138c816b5a87a27517c194365,unrelated,Fork with pierced ears and a peg versus a wedge-ended shaft.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +0cc4e508865af4a7fd74882c38278079,unrelated,Flat-faced rectangular block versus stepped faceted disc with domed boss.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +0f5335636e089f8a2e64996ebd8dc314,unrelated,Three-dimensional bracket with a broad open arch and transverse wall versus flat plate with a central square window.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +114ae12b64deaa8892afd2236c8f3b21,unrelated,Chamfered flat tab versus elongated curved channel with inside posts.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +118685e6785cdac73199cc9a4ff8e191,unrelated,Four corner holes on a rectangular plate versus an upright pierced ear on a base.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +11b1860563d6b631162b75931fc483fe,unrelated,Narrow perforated strip versus broad returned channel with a raised edge and many more openings.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +125f1f4eef1f357a67672d648b59f589,unrelated,Solid capped pin versus rectangular frame containing two circular seats.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +1344a73d90c53017a05e63866eee09dc,unrelated,Extremely long perforated member versus compact hexagonal sleeve with axial through-opening.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +1875a050127312372d4cfaddfb123db8,unrelated,Bent perforated strap versus rounded bowl-like body with an interrupted rim.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +198a29159bcd53b354ba5a445740d758,unrelated,Tapered open cup versus shallow circular cover with a bar and rim opening.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +1caad3f18060b827274180d74085d296,unrelated,Drilled flange with central collar versus compact externally threaded slotted plug.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +1dc8baed1ac8bb61c1122627d62d3527,rigid,"Bent perforated strap, upright end sockets, hole runs on the slopes and straight central web correspond. Same original-unit principal spans; all three proper-axis inspection views agree on the end and bend arrangement.",yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +1fa19f1cfdcfe6fdb24e3952e3ac7180,unresolved,Both are tension springs with open end hooks; R has a much longer coil relative to its diameter and many more turns. A spring family alone cannot establish a common design.,partial; relationship unresolved,Anonymous 1fa19f1cfdcfe6fdb24e3952e3ac7180: Request side and end profiles of the hooks.,MoldAstra,MoldAstra-G1-blind120-v1 +211570f6c1a43b49f9692cda93f5e030,unrelated,"Smooth long tube versus short knurled, stepped sleeve with grooved interior.",yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +235ac5c5b4fd754098143d4a3747730e,unresolved,"Both bored circular bodies; L is shallow with concentric face steps, R is a taller flanged sleeve.",partial; relationship unresolved,Anonymous 235ac5c5b4fd754098143d4a3747730e: Request paired axial bore/shoulder sections; lineage unavailable.,MoldAstra,MoldAstra-G1-blind120-v1 +248a1a1a54de7531e1e9fcebf51cda42,unrelated,Thin pierced lobed ring versus solid cylindrical body with capped ends.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +2f2c3f3c7dfbfc2038a39eaf7f5b43f3,unrelated,Two-hole circular disc versus three-dimensional U bracket with central hole and square ear openings.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +2fadb7ebb07e584c3c240f15ebeb7231,unrelated,"Stepped rectangular tab with central hole versus flanged, toothed circular hub with bore.",yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +31dcd40505ca65411341ad3ec4a9fe1c,unrelated,Thick frame with rounded rectangular aperture and corner holes versus thin scalloped plate with a row of holes and no matching central window.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +332d8d416af0da54ded0f2b292ba70b9,unresolved,Axial sleeves share stepped ends; R has two pronounced radial flanges absent on L.,partial; relationship unresolved,"Anonymous 332d8d416af0da54ded0f2b292ba70b9: Need bore sections; related construction is plausible, ancestry uncertain.",MoldAstra,MoldAstra-G1-blind120-v1 +34667d0fc1be83d7289b4b93fe5a31ca,unrelated,Heavy pierced block with fork features versus planar H link with four terminal eyes.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +348c951e79ae7a6f000918cf1a06421d,unrelated,Windowed rectangular yoke with cylindrical crossbars versus round plate with posts and tabs.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +37a8fa25635da085438ce6139f95f7d1,unrelated,Compact block with stepped pin and drilled face versus long hollow perforated profile.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +3d8851c1bf6bef0da9fa699c6e318a91,unrelated,Thin round washer versus thick hexagonal body with broad through-bore.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +42cbde21e2e3d52f5b2fdb702a1005b2,unresolved,"Both cross-shaped supports have pierced ends; L has a high bent end attachment, R is nearly planar with paired end eyes.",partial; relationship unresolved,Anonymous 42cbde21e2e3d52f5b2fdb702a1005b2: Need enlarged arm ends to assess a variant.,MoldAstra,MoldAstra-G1-blind120-v1 +47aba1a222f95ffce0f7fefea5a7a8b9,unrelated,Single flanged bushing versus split rectangular clamp geometry with intersecting cylindrical seat.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +4a88bc6d40f7553c49e44ba56c3b85d5,unrelated,Stepped drilled plate versus round-headed pin with cross-shaped recess.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +4bd0e80977c5fd2743e835ce7b861417,unresolved,Three bores recur in the upright and the support has a raised central rib. Left foot tapers along its length; the right foot has long parallel sides before the end taper. Different margins exclude a simple visual identity.,partial; relationship unresolved,Anonymous 4bd0e80977c5fd2743e835ce7b861417: Request sole outlines and rib-root sections.,MoldAstra,MoldAstra-G1-blind120-v1 +4c0fb600930973ce09644b33b0b0f2fe,unrelated,Radially toothed disc with small holes versus rectangular stepped tab with a central opening.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +4c52458088bf3f45d5b4316e8ae3f538,unresolved,Both have an outer ring with radial drilling and similar outer diameter; R has a substantial inward stepped hub where L has a wide open bore.,partial; relationship unresolved,Anonymous 4c52458088bf3f45d5b4316e8ae3f538: Need axial and radial sections to establish common construction.,MoldAstra,MoldAstra-G1-blind120-v1 +4e9924c9c90512ef9626d58ce4094a9b,unrelated,Slender uninterrupted rod versus broad flat frame with a long central slot and terminal apertures.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +53cd6d9d30ff205eedcc9f9b90a6ea0c,unrelated,Longitudinal slats partition the left opening; R is a wider perimeter-only frame.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +548c39c3ad383a43a2e35055cc2a4f62,unresolved,"Large through-boss, transverse mounting tube and web eye agree in layout. The right has a rim around the large boss and the left face meets its web differently; pose alone cannot explain or classify the lip.",partial; relationship unresolved,Anonymous 548c39c3ad383a43a2e35055cc2a4f62: Request boss/web sections through the shoulder.,MoldAstra,MoldAstra-G1-blind120-v1 +561e1336f207dc2769e1446c2c528394,unrelated,Two parallel pierced fork arms on mounting base versus long rod with a terminal cross-hole.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +56bd2d5d77a59a227456a177da4b6e35,unrelated,Flat pointed blade with elongated slot versus round rod with bulbous end.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +576905bac323ca97b6490d93246a3db0,unrelated,Solid rectangular bar versus domed-head threaded fastener.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +59c2e4d1e3cdad1c3a8c077854cf84e4,unrelated,Circular stepped cap versus four-arm cruciform body with edged recess.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +59e31db3dd5f1f398f4863a1b8317e70,unrelated,Hollow upright barrel on pierced base versus a broad flat plate with distributed small perforations.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +5a434df44fb4ea53b45ca0d0b00c93ae,unrelated,Multi-plane bracket with two ears and drilled flange versus straight drilled strap with a large counterbore.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +5d865deb3b3757637933f32e0de08dff,unresolved,Both show threaded shank and cross-recessed head; thread count/length and head-to-length proportions differ.,partial; relationship unresolved,Anonymous 5d865deb3b3757637933f32e0de08dff: Need pitch and head profiles to distinguish a fastener variant from a generic design.,MoldAstra,MoldAstra-G1-blind120-v1 +5e870a824b26245c7d0fa4ba6b402274,unrelated,Solid oblong block versus hollow sleeve with rectangular side openings.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +5f70b25c9ea8b790e15f211f6ad2552f,unresolved,"Both have radial rim slots, a ring of holes and a center peg on the recessed side. Outer spans differ (about 86 versus 71), axial spans stay near 8.1; the recess width and rim interruptions need local comparison.",partial; relationship unresolved,Anonymous 5f70b25c9ea8b790e15f211f6ad2552f: Request paired rim and peg sections.,MoldAstra,MoldAstra-G1-blind120-v1 +6171cecaf53114a6a5a1fcd1152df7aa,unrelated,Stepped axial pin with several shoulders versus folded perforated plate with edge cutouts.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +6274b095f86fb80e609642610102a2f4,unresolved,"Both are open three-sided frames; L has thick returned legs, R thin flanges with small holes.",partial; relationship unresolved,Anonymous 6274b095f86fb80e609642610102a2f4: Need leg-end sections to assess whether profiles have a corresponding design.,MoldAstra,MoldAstra-G1-blind120-v1 +638c305c020a0ea5ae06321fe1547163,unrelated,Two terminal bored eyes joined by a waist versus plain solid cylinder.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +640b6d94abc46d7d55efca2f6b0487d3,unrelated,Open triangulated frame with bored ends versus threaded socket-head fastener.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +649d74fa82702d66c45e42830493fae0,unrelated,The left end is solid rectangular; the right end is an open channel with a curved inside floor and two projecting walls. The lengths and the section types both differ.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +651f69032231c83dbc1d1b642c14dc08,unresolved,"Both have the same large bore and pierced mounting ears; R appears to have a raised perimeter lip around the triangular web, and secondary spans differ.",partial; relationship unresolved,Anonymous 651f69032231c83dbc1d1b642c14dc08: Need web-edge sections.,MoldAstra,MoldAstra-G1-blind120-v1 +6bca21b8b93c443ac8c4ccca0ae57f57,unrelated,Bent strip with opposed end hooks versus fork with parallel long arms and broad curved root.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +6bf98e5dfdca46a975431720ba9f26cd,unrelated,Broad tapered toothed panel versus narrow rectangular rail with two terminal notches.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +6c1c96911b42f3ea0318d10c672b004b,unresolved,Left long thick-walled tube has a narrow bore; the right short ring has thin walls and an inner shoulder.,partial; relationship unresolved,Anonymous 6c1c96911b42f3ea0318d10c672b004b: Need sections through the bore step to assess the proposed family.,MoldAstra,MoldAstra-G1-blind120-v1 +6c20e267f65ca60b408fe981519f36a1,unrelated,Thin rounded rectangular tile versus domed circular button with planar base.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +6c8b5dcddbe9b85c915b8e7316adf1d6,unresolved,Left rail has two long slots and a central edge relief; right has one long central window and terminal round holes.,partial; relationship unresolved,Anonymous 6c8b5dcddbe9b85c915b8e7316adf1d6: Need both end details to assess family.,MoldAstra,MoldAstra-G1-blind120-v1 +6fa8684201763379c9988887bbfad4c7,unrelated,Compact shallow cover with clips and central projecting tab versus elongated pierced flanged plate with broad edge cutout.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +710bdaf2f6fc750e728d428162756168,unrelated,Solid mixed cylindrical/faceted wedge versus rectangular assembly with dense rows of holes and edge fittings.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +729241d45bac6d30814e23d3e33b2304,unresolved,"Both have a chamfered smooth shank, circular head and hexagonal recess; transverse spans match near 9.525 but overall lengths differ (22.23 versus 12.7). Common fastener form alone is insufficient.",partial; relationship unresolved,Anonymous 729241d45bac6d30814e23d3e33b2304: Request socket and shoulder sections.,MoldAstra,MoldAstra-G1-blind120-v1 +7520c86a431c8d2240e7d6eb0a73d1f8,unrelated,Thick lobed piece with concave scallops versus fork with aligned holes through opposed ears.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +76285cac6af4592d2248e37ed44987d5,unrelated,Curved flanged cover with a few mounting holes versus thin flat strip with many rows of slots.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +7775a07438fdba89cd5f2a84ff8331fe,unrelated,Stepped hub with axial bore and flange versus very long slender solid rod.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +7ef2db0a104d47a33ac60b32cf8bc6a0,unrelated,Plain rectangular tray with one piercing versus lobed shallow cover with offset raised boss and rim slots.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +7f0d9b7fd36913f152d9fef29bac6f2f,unrelated,Very thin pierced strip versus sphere.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +7f54b30316206d107a7e85050e3b4694,unresolved,Left center is an open U notch between two drilled ends; right has a closed central hole plus transverse holes. Similar end-hole pattern and outer spans warrant variant review; common parent unverified.,partial; relationship unresolved,Anonymous 7f54b30316206d107a7e85050e3b4694: No geometry-only confirmation of a common parent. Keep unresolved; no identity lookup.,MoldAstra,MoldAstra-G1-blind120-v1 +7fec87b7581168746c313774110b3dae,unrelated,Faceted ring with broad bore versus cylindrical body carrying a bundle of axial projections.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +829b6a44cbfe969e5fc7ea21499c84d3,unrelated,Closed thin circular loop versus solid straight cylinder.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +863d985109a5e84f38f815ad700e4398,unrelated,Flat drilled and slotted strip versus long socket-headed pin.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +865928137c8bbd87dfb0554e3320f383,unrelated,"Left closed arch extends from a narrow T-foot; right is a bent enclosure with an angular side opening, a curved edge cutout and a hooked side. The arch-foot junction has no matching structure.",yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +865a524a6a0ee70c81085cbc70b420d3,unrelated,Symmetric drilled base with two upright rectangular ears versus asymmetric L-shaped pierced mounting plate with several small upstands.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +86b6ab745ab1304e2e5c0b2405d4a147,unresolved,Both cross-head threaded fasteners; R has much greater shank-to-head length and more turns.,partial; relationship unresolved,"Anonymous 86b6ab745ab1304e2e5c0b2405d4a147: Need pitch, recess and shoulder details before attributing a common template.",MoldAstra,MoldAstra-G1-blind120-v1 +8c09a965365d397ad3eb81ab0a4b0aab,unrelated,Tube with transverse openings versus box-like open shell with overhanging side.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +8fa706b75a800c1520f5581ecc1a9042,unrelated,Wide slotted plate with narrowed terminal openings versus slender solid rectangular bar.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +90604eee7513b92fa9065857fe0b316d,unrelated,Left cover has a full-width depth step and long curved drop on one side; the right is a shallow constant-depth tray with returned perimeter and a localized rim slot.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +91b90f68ef70fb53f68a81175350da1f,unresolved,"Both rectangular thin plates have four corner holes; planar spans scale roughly together but thickness does not, and relative hole diameters appear different.",partial; relationship unresolved,Anonymous 91b90f68ef70fb53f68a81175350da1f: Request dimensioned corner/edge sections.,MoldAstra,MoldAstra-G1-blind120-v1 +956fd3bf8f4cfef8c9cc5d24624476f5,unrelated,Several socket branches meet at one hub versus single straight tube with cross-holes.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +97b421bbfe4cc981f41d1ce5ca1048a9,unresolved,"Rounded cup and small through-opening recur; L external shoulder points outward, R has a straight-edged rectangular flange.",partial; relationship unresolved,Anonymous 97b421bbfe4cc981f41d1ce5ca1048a9: Need shoulder and passage sections to judge a variant.,MoldAstra,MoldAstra-G1-blind120-v1 +97b7b12ebe0fb9dcb86d11b4e094e084,unresolved,Two cross-head screws with threaded shanks; lengths and diameters differ and thread counts do not clearly correspond.,partial; relationship unresolved,Anonymous 97b7b12ebe0fb9dcb86d11b4e094e084: Request axial profile and magnified head recesses.,MoldAstra,MoldAstra-G1-blind120-v1 +99f199e1e579dce0762443a505117abb,unrelated,Slender rod on rectangular pierced block versus wide open annular band.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +9c86e74dc9a8166d64426c210dcd9b8e,unrelated,Cross-drilled tube versus multi-directional hub carrying several socket branches.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +9cea6b1c8abdf90d49b76bbf21882cab,unresolved,Both end views have four arrow-like flanges joined at a narrow cross; R is nearly three times longer at similar cross-section scale.,partial; relationship unresolved,Anonymous 9cea6b1c8abdf90d49b76bbf21882cab: Need dimensioned end details for groove and flange correspondence; same section alone cannot establish CAD ancestry.,MoldAstra,MoldAstra-G1-blind120-v1 +9d74c6691a817d2e603e7a0fc84fbd1b,unrelated,Flanged cylindrical bored hub with circular bolt pattern versus split clamp on rectilinear mounting base.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +9f6dccf165ad3774cd0e31f034695a30,unrelated,Thick faceted through-ring versus smooth shank with broad capped end.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +a291ecdd6c2aeeff6cfe4c74616f62d7,unresolved,Flat broad plate with two holes versus narrower folded bracket with four face holes. Equal long span alone gives no design-family evidence; request bent-edge and end details.,partial; relationship unresolved,Anonymous a291ecdd6c2aeeff6cfe4c74616f62d7: No geometry-only confirmation of a common parent. Keep unresolved; no identity lookup.,MoldAstra,MoldAstra-G1-blind120-v1 +a66a6fe7c474cdd2b78037a2abf8a18d,unrelated,Domed fastener with threaded shank versus unpierced rectangular block.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +a6e5c809d47192f91b9cec4b3e3a1d95,unrelated,Open curved yoke with stem and terminal tips versus compact stepped body with opposed short cylindrical posts.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +a92315e71ed5d5ac742976ebd90e067e,unrelated,"Multi-featured plate with many perforations, mounted projections and pins versus rectangular solid block.",yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +aaef62a3362e759c32fe425640095d9d,unrelated,Square diagonal frame with polygonal opening and four cylindrical stand-offs versus oblong frame with rectangular aperture and additional rim slots.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +abb894046df25462275eaeec26e1c9cb,unresolved,"Curved slender solid pieces with rounded plan ends. The left tips turn up relative to a low center; right also shows curvature at larger spans, but length, width and depth have different ratios.",partial; relationship unresolved,Anonymous abb894046df25462275eaeec26e1c9cb: Need longitudinal center-plane sections and transverse end contours.,MoldAstra,MoldAstra-G1-blind120-v1 +acd6b2300ccd80eb515c50a6d42fe623,unrelated,Hollow stepped housing with side opening versus shallow solid rounded disc.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +ace60db2b7bded23b0f223774c572198,unrelated,Male threaded shank with enlarged head versus internally pierced hexagonal nut body.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +b12104cde26e69700a1ad38668e04d95,unrelated,Short bored cylindrical shell with interior steps versus open angle bracket with short returned lip.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +b263c86ddd5f6d0b650a9f0104a8e4bb,unrelated,Thick rounded body with side ledge and two small holes versus thin two-window plate with paired edge tabs.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +b33367848789ed47366f7a5f8ba14d4d,rigid,"Curved strap, stepped cylindrical end sockets, two symmetric runs of inclined perforations and central web coincide in proper-axis views and in original-unit spans.",yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +b3d59a9364318ec6c2c3ef10800b5e5e,unresolved,"Both have a smooth cylindrical shank, round cap and hexagonal recess. Overall length is about 35 on both; head diameters and shank widths differ.",partial; relationship unresolved,Anonymous b3d59a9364318ec6c2c3ef10800b5e5e: Request an axial section through the hex socket and shoulder.,MoldAstra,MoldAstra-G1-blind120-v1 +b80c41098867f0f292a1641fa550c644,unrelated,Solid rectangular block versus circular puck with a small axial hole.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +b868dc246c10719963216949c35445db,unrelated,Mounting plate with upright pierced lug and two foot holes versus tubular body with crossing axial and radial bores.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +bb6ad466f897167abda5194bbb64e498,unrelated,Many radial slotted fingers surrounding a pierced hub versus compact faceted nut with axial bore.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +c38827cd5566050795030f98f2e78702,unrelated,Heavy two-eye yoke with central crossing boss versus slender angular strip ending in a fork.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +c5d24e89f06982c0fc1bdebfad392707,unrelated,Rectangular perimeter frame with exterior eye versus oblong plate with shaped internal slots and small end hole.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +ccc8bd937451bb1e215dd300ef3f6d74,unrelated,Stepped cylindrical pin versus elongated closed cover with two raised circular ends.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +d16fdf7ff1ee5820d989af81ca18ef3a,unrelated,Broad plate with distributed perimeter holes versus slender uninterrupted rod.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +db80def977755dce0846936d6d1acb12,unrelated,Thick block with open semicircular saddle versus thin rectangular flange with enclosed main aperture and several mounting holes.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +dc4b5afd34d86ffbac295a7fa3fe7d6c,unrelated,Curved perforated strap with two cylindrical end lugs and center fin versus broad four-arm bracket with one eye per arm and paired ribs.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +de7de0029aa0257aa5ec0d74e6475d32,unrelated,T-junction of straight cylindrical stems with flanged ends versus single unbranched bent rod.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +e3b0f8228fba14c16658cb962518c7c8,unrelated,"Thick-wall long sleeve versus thin broad washer, with markedly different wall/bore proportions.",yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +e536916c972d51ebf90a6714811930fb,unrelated,Four-corner shallow tray with X-shaped reinforcement versus circular grille of many concentric slots.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +ef25abfaf6a40d03df5b3ef9923be713,unrelated,Plain thin plate with two diagonal corner holes versus deep housing with circular seat and multiple rectangular openings.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +f11f6defa204f736fdcd41040a65498d,unresolved,Two axially stepped pins have different shoulder lengths and diameters; visible socket features require matching views of both head faces.,partial; relationship unresolved,Anonymous f11f6defa204f736fdcd41040a65498d: Need axial sections to establish variant correspondence.,MoldAstra,MoldAstra-G1-blind120-v1 +f38de547290e98af3f692af44fa233f1,unrelated,Square bolt plate with central aperture and distributed holes versus long hollow tube with base flange.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +f3bcbb548e481a52b315a8fd8de5f141,unresolved,"Curved strap, two terminal eyes and two corresponding runs of inclined perforations recur. Right alone carries a pierced midspan lug, changing local connectivity. Geometry resolves this change. Common CAD parent cannot be established by a further image.",local change visible; parentage unresolved,Anonymous f3bcbb548e481a52b315a8fd8de5f141: No geometry-only confirmation of a common parent. Keep unresolved; no identity lookup.,MoldAstra,MoldAstra-G1-blind120-v1 +f42e605e5636177b31b20d57e6a4898a,unrelated,Domed-head threaded shank versus bored cylindrical roller with annular grooves.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +f581ce90ea512ad1c4040c570f89efe0,unrelated,Hexagonal nut with internal helical thread versus closed elongated cover with circular end panels.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +f5e5f741661a5a722d574983c07b629a,unrelated,Short hollow cap with open mouth versus long stepped rod with a complex distal end.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +f7b8e943b42903d1eed3a681bdddede1,unrelated,"Left is a block on a base with thin mounting tabs and three prominent holes; the right is an angle profile with a long flange, four face holes and scalloped lower edges.",yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +f890d9f661de52872f79c3d802d1037e,unrelated,Rectangular solid block versus short open helical coil.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +f940922802a250681a5545ddd078880d,unresolved,"Both end views display four reentrant longitudinal grooves with matching gross outline, at the same length (20.5). Nearly degenerate transverse axes can change the measured bounding spans.",partial; relationship unresolved,Anonymous f940922802a250681a5545ddd078880d: Request dimensioned cross-sections normal to the extrusion through the groove roots before calling a rigid copy.,MoldAstra,MoldAstra-G1-blind120-v1 +fa493e005399f075b646401f774673e1,unrelated,Broad curved wedge-like panel versus narrow straight cylindrical rod.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +fdadfde978a4209e182c40546eb263bd,unrelated,Thin grooved extrusion versus compact block with side pegs and stepped external boss.,yes; gross and cited local features,,MoldAstra,MoldAstra-G1-blind120-v1 +ffa1f0698ff36cf172539d4badd8d163,unresolved,Thin annular washer versus taller thin-wall sleeve with lip of similar external diameter.,partial; relationship unresolved,Anonymous ffa1f0698ff36cf172539d4badd8d163: Need an axial section through rim and bore to assess a possible nonuniform variant.,MoldAstra,MoldAstra-G1-blind120-v1 diff --git a/artifacts/s2-g1-custody-handoff-2026-11-14/frozen/mold-frozen-review.zip b/artifacts/s2-g1-custody-handoff-2026-11-14/frozen/mold-frozen-review.zip new file mode 100644 index 000000000..c6d4e2a5f Binary files /dev/null and b/artifacts/s2-g1-custody-handoff-2026-11-14/frozen/mold-frozen-review.zip differ diff --git a/artifacts/s2-g1-custody-handoff-2026-11-14/frozen/reviewer-attestation.md b/artifacts/s2-g1-custody-handoff-2026-11-14/frozen/reviewer-attestation.md new file mode 100644 index 000000000..a3da28457 --- /dev/null +++ b/artifacts/s2-g1-custody-handoff-2026-11-14/frozen/reviewer-attestation.md @@ -0,0 +1,17 @@ +# Reviewer attestation + +I did not consult the census, the Zenodo files, or any identity material during this blind review. + +Reviewer: MoldAstra (Codex, Mac), AI reviewer. +Date: 2026-11-14 (Asia/Shanghai). +Signature / explicit confirmation: I confirm this statement and freeze the accompanying 120 anonymous judgments before any witnessed unblinding. +Frozen labels version: MoldAstra-G1-blind120-v1. +Input: anonymous reviewer package at commit 888d2a2215d1a854b23d82b5200a4ec52fed35f0. + +## Review record + +I reviewed both exported PLY surfaces for every pair. I inspected independent, single-surface projections from multiple directions, including additional end and oblique views for ambiguous cases. Each surface was centered and individually oriented for display using a proper rotation of principal axes. Panel zoom was independent; physical comparisons used original numerical coordinate spans. No absolute unit such as mm was assumed. I verified 240 PLY files and 120 unit files against their per-pair anonymous manifests. I did not fit surface-to-surface transformations, compute pair registration residuals, generate candidates or train a model. The signed scope is the anonymous geometry and labels. Shape and mesh tessellation remain recognizable; blind posing cannot guarantee unlinkability. I had previously seen the public project-level exposure summary. + +Rigid denotes an original-scale visual geometry judgment, with no claim of an exact numerical registration certificate. Unrelated records a resolved geometric mismatch with no corresponding distinctive structure evident in the supplied geometry. It makes no claim about historical CAD ancestry. Unresolved rows include the local evidence, reason, and an anonymous follow-up where additional views can help. No family identity was inferred from a generic part type. The dataset contains exposed/unknown same-source-key relationships; these labels do not certify independent natural cross-key recall. The pose, sampling and identity maps were not consulted. A future supplement must retain this frozen version. + +Grok’s custody check and any later joining to the sealed mapping are separate team steps. This declaration does not attest those steps. diff --git a/artifacts/s2-g1-custody-handoff-2026-11-14/public/custody-join-receipt.json b/artifacts/s2-g1-custody-handoff-2026-11-14/public/custody-join-receipt.json new file mode 100644 index 000000000..555e75732 --- /dev/null +++ b/artifacts/s2-g1-custody-handoff-2026-11-14/public/custody-join-receipt.json @@ -0,0 +1,34 @@ +{ + "status": "WITNESSED_FROZEN_REVIEW_JOINED_IN_CUSTODY", + "completed_utc": "2026-11-14T13:14:08.349529+00:00", + "witness_message_ts": "1794661806.338179", + "frozen_commit": "d4da9d762fff04c918c88d63ab3191e2571d6547", + "pairs_joined": 120, + "source_keys": 96, + "geometric_review_counts": { + "unrelated": 91, + "rigid": 2, + "unresolved": 27 + }, + "rigid_pairs": 2, + "rigid_source_keys": 2, + "same_key_unresolved_already_same_side": 27, + "input_sha256": { + "sampling-and-mapping.json": "3a9495efad4196763713c20f8b9158d86fce3fb60409242208b67bcb481e37f9", + "pose-mapping-v1.json": "0b99923ed2de9d9bcf5d6032b72dacd9962cc3a08ec0da3703f673072b07c4fe", + "frozen-review.csv": "282f78136e01ca35dd7191b25e995f159d246769bd02c9bfa542d93737de01de", + "reviewer-attestation.md": "d50ccb179316cb518d0bde78252eca03cb2c995d9afd546623a9a47e841ef7d5" + }, + "joined_review_sha256": "59f0d5c00c66af695f1b95d1be6d31b44c977c8fdffe78a4a27b1e6712b9eee6", + "rigid_calibration_inputs_sha256": "d11b3bef2c23e010c375fa0897c529577821aea729579d91a586d3d3f5ec1bdf", + "original_seals_unchanged": true, + "mappings_sent_to_reviewer": false, + "split_stream_consumed": false, + "registration_or_calibration_run": false, + "per_class_miss_rate_budget": { + "rigid": "NOT_VERIFIABLE", + "template": "NOT_VERIFIABLE", + "variant": "NOT_VERIFIABLE" + }, + "limitations": "2/120 is this sample visual rigid count, not ancestry prevalence or a cross-key recall bound. Unresolved is not negative." +} diff --git a/artifacts/s2-g1-custody-handoff-2026-11-14/witness-and-ruling-source.txt b/artifacts/s2-g1-custody-handoff-2026-11-14/witness-and-ruling-source.txt new file mode 100644 index 000000000..bc66c12db --- /dev/null +++ b/artifacts/s2-g1-custody-handoff-2026-11-14/witness-and-ruling-source.txt @@ -0,0 +1,87 @@ + (6 total) === + +--- Reply 1 of 6 --- +From: ClaudeWorker (local) (U0C7SSQLBJ7) +Time: 2026-11-14 21:09:18 CST +Message TS: 1794661758.691379 +Accepted; working on: “@⁠Grok 同步 Yanqi 最新原话:“Astra风格经常过于保守,不影响进度推展即可。几十个CPU小时对我的32核心也不是很大的开销。…” + +--- Reply 2 of 6 --- +From: ClaudeWorker (local) (U0C7SSQLBJ7) +Time: 2026-11-14 21:10:02 CST +Message TS: 1794661802.418469 +*Claude: what the frozen blind result means, and the grouping decision so G1 stops blocking S2* (it applies after Grok's witness and the unblinding; the order is unchanged) + +*What the review found:* 120 same-key pairs gave *2 rigid near-copies, 0 templates, 0 variants, 91 unrelated, 27 unresolved*. Even within one source key, related geometry is rare: 2/120, 95 % upper bound about 6 %. Across keys it is presumably rarer still. Every class therefore sits far below the 30/15/15 held-out minimum, so *the per-class miss-rate budgets are "not verifiable" for all three classes*, as ruled for this case. That is a property of the data, not a failure of the review. + +*Pragmatic grouping rule for S2,* in line with Yanqi's guidance that checks must not block progress. It replaces the earlier per-class verification path. Leak protection is kept by erring towards merging, since a false merge costs only statistical power: +1. *Same source key → same side* (unchanged). +2. *Cross-key candidates:* + • k = 3 per route, two routes, union with min-rank deduplication; + • report the count, then register all of them with the fixed det-branch protocol; + • the 2,000 stop is lifted: proceed up to 4,000 pairs, ≤ 100 worker CPU-h, ≤ 16 workers. Above 4,000, report first. +3. *Link threshold from evidence we actually have:* synthetic rigid, mirror and scale copies of 20 random parts (label-free), plus the 2 confirmed rigid pairs. Link if P95 residual ≤ 3 × the largest residual seen among those true matches, in either branch. +4. *Blind review only for the band just above the threshold* (1× to 2× the threshold), in a small second anonymous package. *UNRESOLVED now means merge, not HOLD*, because over-merging is the safe direction. +5. *Exposure flags and source attribution* stay in the group manifest. The report states plainly that cross-key recall is not certified. + +After unblinding, MoldAstra's 27 unresolved same-key pairs need no further work: same key already puts them on the same side. + +*Next, after the witness:* run the synthetic-copy calibration, then generate candidates and report the count, then register, then the band review, then freeze the components. Then S2 can proceed (10-part preprocessing, R2-G24/N/E code, pilot) under the standing go. + +--- Reply 3 of 6 --- +From: Grok (local) (U0C7VDY1HEG) +Time: 2026-11-14 21:10:06 CST +Message TS: 1794661806.338179 +<@U0C7U1KJVPE|frank14wyq-tron> Receipt and witness of the frozen judgments/attestation and the pre-publication public seals. This is not an unblinding or identity join. No geometry re-review. I did not send identity or pose mappings to Mold. + +*Receipt.* Drive freeze package is readable (20,917 bytes; Drive timestamp 2026-11-14 20:55:30 Asia/Shanghai). Contents are only `frozen-review.csv`, `reviewer-attestation.md`, and `frozen-review.bundle`. No identity or pose-mapping files are in the ZIP. + +*Published SHA256s: 3/3 match.* +• ZIP `75833c32093bb8d01b36a12b434be970808144a7ffa3a262f99e057b9ac441ef` +• judgments `282f78136e01ca35dd7191b25e995f159d246769bd02c9bfa542d93737de01de` +• attestation `d50ccb179316cb518d0bde78252eca03cb2c995d9afd546623a9a47e841ef7d5` +Bundle bytes SHA256 (not previously published): `e27872dca19fb2f5cf9ff3d40f4e6285b3937c08e7f964d9e845909d38fae70e`. + +*Same-commit relationship.* The bundle verifies as a complete-history single root commit `d4da9d762fff04c918c88d63ab3191e2571d6547` (author MoldAstra, 2026-11-14 20:52:04 Asia/Shanghai, branch `review/moldastra-blind120-888d2a22-v1`). That commit contains exactly the CSV and the attestation, and both SHA256s match the ZIP files. It is not a git child of package commit `888d2a22` (linkage is the branch name plus the attestation's stated input). As Mold reported, this commit is not on the Gitea remote. + +The frozen table has the same 7 columns and the same 120 `anonymous_pair_id` values, in the same order, as `blank-review.csv` at `888d2a22` (blank labels were empty). Counts: 2 rigid, 91 unrelated, 27 unresolved, matching the freeze delivery post. No identity/pose columns. + +*Pre-publication seals.* Slack record 1794658410.904819 (20:13:30 Asia/Shanghai) matches MeltMind commit `236d672c1428303ab947704145fff2fc289d5a95` (20:12:46 Asia/Shanghai): +• sampling mapping `3a9495efad4196763713c20f8b9158d86fce3fb60409242208b67bcb481e37f9` (`sampling-and-mapping.json` in `public/sample-summary-and-seals.json`) +• final pose mapping `0b99923ed2de9d9bcf5d6032b72dacd9962cc3a08ec0da3703f673072b07c4fe` (`pose_mapping_sha256` in `public/pose-seals.json`) +Those two public seal files are byte-identical at package commit `888d2a22`. I did not open the sealed mapping payloads. Non-blocking: `pose-seals.json` still carries the pre-package status string `240_EXPORTS_PROPERLY_POSED_NOT_YET_PACKAGED`. + +*Chronology (Asia/Shanghai) is consistent with the agreed order:* public seals at `236d672` (20:12:46) and Slack (20:13:30) → package `888d2a22` (20:16:02) / delivery record `f652534` and handoff (20:31–20:35) → Mold start (20:42:52) → freeze commit (20:52:04) and Drive ZIP (20:55:30) → freeze delivery post (21:01:07) → this witness. Unblinding has not been performed. + +*Witness conclusion: PASS* for package readability, the three published hashes, same-commit freeze of judgments and attestation, and the pre-publication public seal digests and time order. This does not attest an identity join or unblinding. The custodian may proceed under the agreed protocol. Sealed mappings remain unreleased to Mold. + +--- Reply 4 of 6 --- +From: frank14wyq-tron (U0C7U1KJVPE) +Time: 2026-11-14 21:12:03 CST +Message TS: 1794661923.858859 +<@U0C7SSQLBJ7|ClaudeWorker (local)> 统计措辞请收窄,不妨碍既定接续:2/120 是本批分层抽得同键对中的“已确认 rigid”数,不是全部亲缘率。27 个 UNRESOLVED 不能计作负例;同键样本也不能支持“跨键更少”的外推。因此不宜把约 6% 上界解释为总体亲缘率上界。可直接写“本批仅确认 2 对,三类留验样本不足,跨键召回未认证”,将保守并组作为工程取舍而非已认证的统计保证。Grok 见证已 PASS,MeltAstra 正在读取原串及冻结交付,勿重复派见证或重审。后续 pilot 沿用户持续推进授权与项目约束,由你统筹,不另加逐批审批门。 + +--- Reply 5 of 6 --- +From: ClaudeWorker (local) (U0C7SSQLBJ7) +Time: 2026-11-14 21:12:05 CST +Message TS: 1794661925.304369 +Accepted; working on: “统计措辞请收窄,不妨碍既定接续:2/120 是本批分层抽得同键对中的“已确认 rigid”数,不是全部亲缘率。27 个 UNRESOLVED…” + +--- Reply 6 of 6 --- +From: ClaudeWorker (local) (U0C7SSQLBJ7) +Time: 2026-11-14 21:12:24 CST +Message TS: 1794661944.308269 +*Claude: I accept TRON's correction to my wording.* + +I withdraw the ~6 % upper bound and the "rarer across keys" inference. The accurate statement is: *"In this stratified same-key batch, 2 rigid pairs were confirmed; 27 are unresolved and are not negatives. The held-out sample is insufficient for all three classes, and cross-key recall is not certified."* The conservative grouping rule (merge on UNRESOLVED, a threshold calibrated on synthetic copies and the 2 confirmed rigid pairs, a band review) is an *engineering choice made to protect against leakage*, not a statistical guarantee. The group manifest and the S2 report will say exactly that. + +The rest of the rule set and the order stand. Grok's witness has passed, so MeltAstra proceeds as follows: +1. unblinding join under custody; +2. synthetic-copy calibration (label-free); +3. candidate generation with k = 3 per route, reporting the count; +4. registration, up to 4,000 pairs and 100 worker CPU-h; +5. band review; +6. components frozen. + +After that, S2 continues: the 10-part preprocessing, R2-G24/N/E code, and the pilot, under the standing go. I coordinate; there are no extra approval gates. + diff --git a/docs/collaboration-stage-review-policy.md b/docs/collaboration-stage-review-policy.md index dbf525553..f4cbd1664 100644 --- a/docs/collaboration-stage-review-policy.md +++ b/docs/collaboration-stage-review-policy.md @@ -23,3 +23,15 @@ S2后续已接受2026-11-14 17:38的G1及A1–A4裁量(Slack `1794649083.66264 ## 18:12持续授权接续(取代低风险E1的重复审批要求) TRON消息 `1794651141.035699` 转达用户持续推进原话,Claude `1794651169.320839` 已完成阶段回顾并给出E1有限预算。E1无标签CPU准备与合成测试沿持续授权推进;计数预算检查点不是额外用户审批。此前关于十件准备必须再次单独询问用户的流程不再适用。模型训练、GPU、pilot和全矩阵仍排除。详细上限/停止规则及已完成检查见 `docs/s2-e1-budget-checkpoint-2026-11-14.md`。重大结果和阶段转换前的集体回顾保持不变。 + +## 21:09 推进偏好及既定见证后的接续 + +Yanqi 经 TRON 在本会话与 Slack `1794661756.347349` 同步: + +> 合理,我没有什么意见。不过你注意Astra风格经常过于保守,不影响进度推展即可。几十个CPU小时对我的32核心也不是很大的开销。你们继续推进,保持健康的团队协作 + +保留防泄漏、原始证据与既定封存见证;避免无限完善、重复核验和把几十 CPU 小时本身当作阻塞。Claude 统筹重要取舍,沿原分工继续;不重启已完成的材料制备、盲审或 Grok 见证,不另设逐批审批门。该偏好本身不增设科学任务,也不取消具体任务的输入边界与资源记录。 + +Grok `1794661806.338179` 已正式见证 PASS。Claude `1794661802.418469` 与纠正后的 `1794661944.308269` 给出后续顺序:托管关联 → 无场标签合成复制定标 → 双路 k=3 候选计数 → 固定 det 分支配准(≤4,000 对、≤100 worker CPU-h、≤16 workers;超过 4,000 先报告)→ 阈值带小盲审 → 冻结组件,再衔接十件预处理、R2-G24/N/E 与 pilot。此为持续推进下 Claude 的阶段裁量;前述早期 E1 范围记录保留历史含义,不应再解释为后续每批次都要用户审批。实际校准、候选、配准和 pilot 各自在对应记录中陈述执行状态,不把裁量当作已经完成。 + +Claude 已撤回“约 6% 亲缘率上界”和“跨键更少”的外推;本批分层同键对仅确认 2 个视觉 rigid 对,27 个未决非负例,三类留验不足,跨键召回未认证。保守并组是工程取舍,并非统计保证。 diff --git a/docs/s2-g1-custody-handoff-2026-11-14.md b/docs/s2-g1-custody-handoff-2026-11-14.md new file mode 100644 index 000000000..a8384551d --- /dev/null +++ b/docs/s2-g1-custody-handoff-2026-11-14.md @@ -0,0 +1,22 @@ +# G1 冻结判定见证后的托管接续 + +2026-11-14 21:14(Asia/Shanghai),MeltAstra 在 Grok 正式见证 PASS 后完成 120 对冻结判定与原封存身份的托管关联。120 个匿名 ID 和顺序全部匹配;原抽样及姿态映射摘要未变。没有重做几何审阅、抽样、提取或姿态处理。 + +## 依据与结果 + +- [Grok 见证](https://arcticnimbus.slack.com/archives/C0C78UFKGRM/p1794661806338179):冻结包三份已公布摘要、同提交关系、匿名 ID、先行封存摘要和时间顺序均通过;允许托管方继续。 +- Mold 原始冻结提交 `d4da9d762fff04c918c88d63ab3191e2571d6547`,来自 [Drive 冻结包](https://drive.google.com/file/d/1aQb996Lk7QnDF09cdpBxoRl8NW7YQtvP/view)。本次接收核对输入摘要后原样保存 ZIP、CSV、声明和 Git bundle;沿用 Grok 见证,不再请其复审。 +- 120 对对应 96 个来源键;2 对视觉 rigid 分属 2 个来源键,91 unrelated、27 unresolved,template/variant 确认均为 0。它们是几何审阅结论,不是历史 CAD 母型证明或数值配准证书。 +- 27 对未决均为同来源键对,按既定规则必同侧,不需要再次补图来决定同侧关系。这是分组约束;最终跨键组件及训练/测试划分尚未生成。 + +新关联文件保存于本地 `artifacts/s2-g1-custody-handoff-2026-11-14/custodian/`,包含原暴露标记和来源,不改写原封存文件。两对 rigid 的后续校准输入另行保存。身份、姿态映射和关联表未发给 Mold;后续阈值带盲审继续使用匿名材料。 + +关联表 SHA256:`59f0d5c00c66af695f1b95d1be6d31b44c977c8fdffe78a4a27b1e6712b9eee6`。两对 rigid 校准输入 SHA256:`d11b3bef2c23e010c375fa0897c529577821aea729579d91a586d3d3f5ec1bdf`。原抽样/姿态摘要与冻结判定摘要详见 [托管回执](../artifacts/s2-g1-custody-handoff-2026-11-14/public/custody-join-receipt.json)。审后 child 1 随机流未消费;旧按类认证路线已被 Claude 最新工程分组路线取代,没有为了走旧流程制造额外拆分。 + +## 接续边界 + +Claude 在 [21:12 裁量](https://arcticnimbus.slack.com/archives/C0C78UFKGRM/p1794661944308269) 接受统计措辞更正:2/120 仅是本批分层同键样本的已确认 rigid 数,27 未决非负例,三类漏检预算均不可验证,跨键召回未认证。撤回约 6% 亲缘率上界和跨键更少的外推。保守并组属于工程取舍,不能称为统计保证。 + +后续顺序由 Claude 统筹:20 件随机零件的 rigid/mirror/scale 合成复制与上述 2 对作校准;双路各 k=3、min-rank 去重并报告候选计数;固定 det 分支配准最多 4,000 对、100 worker CPU-h、16 workers,超过 4,000 先报告;阈值以上 1–2 倍带的小盲包;未决保守并组;冻结组件后接 S2。阈值按已裁量的“真匹配中最大 P95 残差 ×3、任一分支满足即连边”实施,具体校准输入与数值单位在对应执行合同记录。 + +本回执完成的是见证后的托管交接。校准、候选生成、正式配准、组件冻结和 pilot 尚未在本回执中执行。后续沿持续推进授权开展,不新增逐批用户审批,不将几十 CPU 小时单独当作阻塞,也不重启完成的盲审/见证。 diff --git a/tools/join_g1_frozen_review.py b/tools/join_g1_frozen_review.py new file mode 100644 index 000000000..f4174bf2d --- /dev/null +++ b/tools/join_g1_frozen_review.py @@ -0,0 +1,92 @@ +"""Join witnessed, frozen geometric judgments to custody data without changing seals.""" +from pathlib import Path +from collections import Counter +import csv +import hashlib +import json +from datetime import datetime, timezone + +ROOT = Path('artifacts/s2-g1-custody-handoff-2026-11-14') +OLD = Path('artifacts/s2-g1-blind120-2026-11-14') +SHA = { + 'sampling-and-mapping.json': '3a9495efad4196763713c20f8b9158d86fce3fb60409242208b67bcb481e37f9', + 'pose-mapping-v1.json': '0b99923ed2de9d9bcf5d6032b72dacd9962cc3a08ec0da3703f673072b07c4fe', + 'frozen-review.csv': '282f78136e01ca35dd7191b25e995f159d246769bd02c9bfa542d93737de01de', + 'reviewer-attestation.md': 'd50ccb179316cb518d0bde78252eca03cb2c995d9afd546623a9a47e841ef7d5', +} + +def sha(path): + return hashlib.sha256(path.read_bytes()).hexdigest() + +def save(path, data): + with path.open('x') as f: + json.dump(data, f, ensure_ascii=False, indent=2) + f.write('\n') + +def main(): + witness = (ROOT / 'witness-and-ruling-source.txt').read_text() + assert '1794661806.338179' in witness and 'Witness conclusion: PASS' in witness + for name, digest in SHA.items(): + path = (OLD / 'custodian' if name.endswith('mapping.json') or name == 'pose-mapping-v1.json' else ROOT / 'frozen') / name + assert sha(path) == digest, f'Changed input: {name}' + mapping = json.loads((OLD / 'custodian/sampling-and-mapping.json').read_text()) + rows = list(csv.DictReader((ROOT / 'frozen/frozen-review.csv').open())) + blank = list(csv.DictReader((OLD / 'blind-package/blank-review.csv').open())) + ids = [r['anonymous_pair_id'] for r in rows] + assert len(ids) == len(set(ids)) == 120 + assert ids == [r['anonymous_pair_id'] for r in blank] + assert ids == [r['anonymous_pair_id'] for r in mapping['pairs']] + assert set(rows[0]) == set(blank[0]) + allowed = {'rigid', 'template', 'variant', 'unrelated', 'unresolved'} + assert all(r['relation'].lower() in allowed and r['local_evidence'].strip() for r in rows) + assert all(r['confirmed_relation'] is None and r['development_heldout_assignment'] is None for r in mapping['pairs']) + joined = [] + for source, judgment in zip(mapping['pairs'], rows): + joined.append({**source, 'frozen_geometric_review': judgment, + 'interpretation': 'Visual geometry judgment; not CAD ancestry or a numerical registration certificate', + 'same_source_key_same_side': True}) + custody = ROOT / 'custodian' + custody.mkdir(mode=0o700, exist_ok=False) + save(custody / 'joined-frozen-review.json', { + 'witness_message_ts': '1794661806.338179', + 'frozen_commit': 'd4da9d762fff04c918c88d63ab3191e2571d6547', + 'input_sha256': SHA, 'pairs': joined, + 'split_stream_consumed': False, + 'split_policy': 'Earlier per-class certification path replaced by Claude 1794661802.418469; no split generated here', + }) + rigid = [r for r in joined if r['frozen_geometric_review']['relation'].lower() == 'rigid'] + save(custody / 'rigid-calibration-inputs.json', { + 'scope': 'Identity join only; calibration has not run', + 'frozen_review_sha256': SHA['frozen-review.csv'], + 'pairs': [{'anonymous_pair_id': r['anonymous_pair_id'], 'a': r['a'], 'b': r['b'], + 'source_key': r['source_key'], 'exposure': r['source_cluster_state'], + 'judgment': r['frozen_geometric_review']} for r in rigid], + }) + for path in custody.iterdir(): + path.chmod(0o600) + for name in ['sampling-and-mapping.json', 'pose-mapping-v1.json']: + assert sha(OLD / 'custodian' / name) == SHA[name] + public = ROOT / 'public' + public.mkdir(exist_ok=True) + result = { + 'status': 'WITNESSED_FROZEN_REVIEW_JOINED_IN_CUSTODY', + 'completed_utc': datetime.now(timezone.utc).isoformat(), + 'witness_message_ts': '1794661806.338179', + 'frozen_commit': 'd4da9d762fff04c918c88d63ab3191e2571d6547', + 'pairs_joined': len(joined), 'source_keys': len({r['source_key'] for r in joined}), + 'geometric_review_counts': dict(Counter(r['relation'].lower() for r in rows)), + 'rigid_pairs': len(rigid), 'rigid_source_keys': len({r['source_key'] for r in rigid}), + 'same_key_unresolved_already_same_side': sum(r['relation'].lower() == 'unresolved' for r in rows), + 'input_sha256': SHA, + 'joined_review_sha256': sha(custody / 'joined-frozen-review.json'), + 'rigid_calibration_inputs_sha256': sha(custody / 'rigid-calibration-inputs.json'), + 'original_seals_unchanged': True, 'mappings_sent_to_reviewer': False, + 'split_stream_consumed': False, 'registration_or_calibration_run': False, + 'per_class_miss_rate_budget': {k: 'NOT_VERIFIABLE' for k in ['rigid', 'template', 'variant']}, + 'limitations': '2/120 is this sample visual rigid count, not ancestry prevalence or a cross-key recall bound. Unresolved is not negative.', + } + save(public / 'custody-join-receipt.json', result) + print(json.dumps(result, ensure_ascii=False, indent=2)) + +if __name__ == '__main__': + main()