Repository navigation
DONE: T74 raised-entrance wrappers independently bound and checked (#2472) #2463
RossRichardson
started this conversation in
General
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
FINDING / DONE / RELEASED — 2026-10-15T11:56:50.410663+00:00. The scoped independent audit is complete. PR2472, exact head
0709b0fb1ead100a7074378dce5a601922ff8dba, contains the reviewable package. The advisory claim is released; its original expiry was 2026-10-15 13:30 UTC.Established. PR2435's 24 changed entrances in each provider bind to their actual native frames and both complete emitted wrapper streams (48 helper occurrences; 96 orientation bindings). Two retained independent exact kernels verify all 512 columns of the 12 distinct rank-seven residual operators, including global phase and zero coefficients. Each kernel covers 3,145,728 complex coefficients; 72 altered operators fail and 12 equivalent phase redistributions pass. The package includes six event/source mutations and six behavioral tests.
The written derivation explains why the residual is
T_FULL T_E^-1in both orientations, conditional on logical helper identity and no cross terms. The same preR^-1and postL^-1then expose seven active axes with two spectators. The twelve rank-two entrance spaces have Gram rank one, excluding an orthogonal map to a two-coordinate span; the paid two-sided monomial wrappers are valid. No construction defect was found.Evidence. README and commands, exact receipt, mathematical review. Registered isolated/optimized Python3.12 replay passes; an independent Git-free copy with read-only files passes normal Python3.14 in 35.825s with all file hashes unchanged. Reproduction of all six generated producer JSON payloads passes. The donor replay logs are local-only; all source inputs, verifiers and result receipt are public in the linked package. Hosted CI is separate; no CI success is inferred from local replay.
Pins and prior scope. Base/main
0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2, priority revision4; selected-record SHA25610b9e340fd8e354a37a0c4813c90e5de1dd42044b5f6aca4a3fc5a42b685fa32. PR2435324d7d3590dc21003940a44cb0cbdc32b3620f25; PR2029 kernelda1bc730b9963f3e53538964d16364d2076a4e5d; PR2446a3aabff9471dfdc3e85b065616e45a317f38882e. Producer functionality already exists. PR2446 checks scalar/owner constraints, PR2447 widths/owner fill, PR2453 width exchange, and new PR2466 owner-map optimality. This package adds changed-source operator/event binding, not the factorization method or a new kernel.Handoff / remaining obligations. A separately scoped independent complete native replay could reconcile these 48 bindings with the borrow word, global owner windows and both paid wrapper ends at the same source pin. This audit does not certify full scalar restoration, full physical borrow execution, global hosting, complete invoice, precision, or all-size recursive interfaces. No new kappa, full supplier admission, or formal-kernel proof is claimed. The selected record is unchanged; only its consistency check was run, not its full cold replay.
Coordination and attribution. Submission refresh: main unchanged at 11:46 UTC; complete-paginated overlap snapshots at 11:46:59 UTC (666 open PRs including drafts) and 11:50:00 UTC (1,191 Discussion records), plus 1,536 local ledger/package records. Arbitrary code/diffs are outside helper coverage; no novelty certificate. Jeremy Kothe / PR2435, Vlado Vojdanovski / PR2322, and Sagar Thakrar / PR1781, PR2029, PR2446 retain their source credit. New binding and review by Ross Richardson with substantial OpenAI Codex assistance, including one mathematical frame-scheduling specialist. Apache-2.0.
All reactions