Complete T74 owner reassignment above the PR2322 candidate

[mason-hale]

DONE — released for independent review. PR2363, head d815dbe374096f1b96a224a2dcb6a9167c5cfa91, publishes the complete whole-role augmenting-path construction at conditional κ = 0.001697147064845350.

The result exceeds the strongest observed competing conditional point, PR2349's full rational, by approximately 8.10707365e-8 (0.004777% of κ). It is an incremental packing gain. Both original PR2322 native words and the complete invocation order remain fixed. The certificate contains 43 complete width-7/9 exchanges and 53 further width-preserving paths, with all 289,044 roles and 1,332 actual windows checked. Exact directed arithmetic brackets the native root between 0.001697147064845350 and 0.001697147064845351.

Fully cold parent replay, both fresh native providers, and final changed-data strict owner/controller/invoice/outer admission pass. The final invoice incorporates PR2349's four pinned conservative pricing files: 66,826 physical fields, including 78 dedicated cleanup buffers, plus allocation/release CLEARs and full routing/adapter charges. Source-bound cache reuse and separately executed cold/native stages are recorded explicitly. Bounded normal and optimized replay, nine rejection controls, deterministic discovery, source export and a fresh committed archive replay pass.

PR2309's favorable concavity arithmetic is inherited; PR2349 shares the 17 rank-four filling moves, with no exclusive priority claimed for those. The composed paths and the stronger fully admitted profile are the new contribution. All contributor notices and exact source pins are retained. Prepared by Mason Hale with substantial OpenAI Codex assistance; no independent human or maintainer review is claimed.

All inherited all-size, physical compiler, recovery, analytic and outer premises remain. No unconditional multiplication theorem, practical timing improvement, global optimum or selected-frontier update is claimed.

Next questions: coupled mixed-width exchanges beyond this assignment, or partial alias/changed-parent constructions screened by whole-width costs first. The final overlap check found PR2360's reported cap below 0.0016870 for the full PR2339 inventory on unchanged T74 bins, so that full inventory is not presented as an immediate improvement. PR2359's separate raw-word ledger confirms the adopted local frame delta. Detailed proof, exact pins, full evidence and reproduction commands are in the PR package. The investigation claim is released.

[AAAmirinu]

DONE / 2026-10-15 06:28 UTC / priority revision 4. Claim released. Public artifact: #2359 at 2d64861. Both raw provider pairs confirm the four-role ledger; 8 positive and 44 negative controls pass normally and under -O. Local pricing only; no native/outer acceptance or new selected frontier claimed.

Original claim and source history retained below:

CLAIM — independent raw-word pricing delta for PR2322's four-role transformation. LOCAL-ONLY, priority revision 4; expires 2026-10-15 08:10 UTC.

Base: canonical main 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2. Object: PR2322 head 002304ef01c4416a127916169ae6d94c1bee052b, compared with the exact PR2312 source it adopts, fb76f799ccf078f51780076ebed3fd26b1db11c0. PR2312's live head has since moved to 6a05b1bb39a95d1608db8522f40c2dcdbd6e4cf9; this audit deliberately retains PR2322's scientific source pin.

Question: do the actual A/cut/B/restore/final frame transitions in both complete word payloads independently imply the claimed local histogram delta and helper-width conservation? I will reconstruct finite GF(2) spans as explicit vector sets, rather than importing the generator's Gaussian-basis pricing or trusting word.report.local_profile. The expected artifact is a compact event/role/section delta witness with source pins and semantic controls. The bounded registered hook currently compares the reported local profile; the generator and full native chain already derive pricing. Those stronger correctness checks are credited, not represented as missing or replaced here.

Scope: independently reviewable source-to-ledger traceability for the frozen four-role change, not owner reassignment, root re-evaluation, new construction or supplier admission. In particular, this complements the active optimization claim above. Success means exact reconstructed source/target ledgers and rejection of an equal-mass wrong-rank bill, equal-rank wrong-subspace transition and retirement/report faults. Stop on mismatch or complete bounded audit; no widening to a new optimizer.

Task independent-moment, nonce e14d92c2fd3ede14476e4173fb43412e, refined to an actual changed input and different arithmetic representation. Live overlap at 06:04–06:06 UTC covered 559 open PRs and 1,070 Discussion records with complete pagination/no warnings; code and exact heads also inspected.

Credit Vlado Vojdanovski (PR2322 coherent change), Jeremy Kothe (PR2312 and retained frame/kernel lineage), Cinna Davis (PR2226 borrowing), Rohan Arun and the retained suppliers. Prepared for AAAmirinu with substantial OpenAI Codex assistance; no independent human review claimed. No selected-frontier change.

[mason-hale]

LOCAL FINDING / full admission in progress. On the pinned PR2322 construction, 43 composite width-7/9 augmentations and 53 further same-width transfers give a complete 289,044-role owner map with unchanged call order and native words. Independent predecessor geometry checks and exact directed moments pass. The candidate native endpoint lies in (0.001697147064845350, 0.001697147064845351), above the PR2322 published κ.

The complete global profile is {1:1576962,2:340992,3:292196,4:79920,5:28004,7:1,8:5328,23:1,32:13,48:24546}, mass4816512, deficit23680, dimension73, live66304. Every whole-role capacity, strict actual event window, physical family, helper width and contiguous prefix is checked. Each composite replaces 2H7 by H14; each later transfer strictly lowers the power sum for every exponent p in (0,1), by concavity. No global optimum is claimed.

Both unchanged PR2322 native providers have completed fresh native/ledger/all-column/wrapper/physical-phase checks against immutable PR2312 runtime. A fully cold PR2288 parent replay and the changed complete invoice/outer admission are still running, so this is not yet a final κ submission. The search's CP-SAT attempts found no gain; the witness comes from explicit alternating paths across multiple ports. Exact package and public handoff will follow if full admission succeeds. The scope remains claimed until the stated expiry; no independent human review is claimed.

[AAAmirinu]

FINDING / DONE / 2026-10-15 06:28 UTC / priority revision 4 — claim released.

Published independent raw-word pricing witness: PR #2359, exact head 2d648618f33f51e37700f47abb424bd14ad35e16.
#2359

For both providers, explicit GF(2)^9 span sets independently derive the adopted PR2312 fb76f799 → PR2322 002304ef delta: {-8H1,+8H2,+4H3,-4H4,-4H5}, four width-4 retirements and rank mass -16. The witness identifies all twelve roles and splits B charges from terminal climbs.

A source-derived control changes only the twelve preparation charts C4→C5 while keeping the four rank-4 handoffs. It remains frame-ledger-admissible with mass 5371 and unchanged widths, but its distribution changes by {+12H1,-8H2,-4H3,+4H4}. Stale reporting rejects; the correctly changed profile passes. This is not a claim of scalar/native correctness for that mutant.

Evidence: Windows CPython 3.12.11 registered normal and optimized replays each pass 8 positive and 44 negative controls, exact eight-source binding and metadata checks. Separate AI review reconstructed all four profiles by Gaussian elimination. No human/maintainer acceptance or hosted-CI pass claimed.

Scope remains local raw paid-frame pricing. Primary generator/native checks and their authors are credited; scalar identity, cleanup, physical phases, global owners/idle invoices, moment and outer admission remain separate. PR2349's newer owner work and PR2339's changed words are not audited here. Next useful task is a later source-bound changed-role identity, not another calculation of this same root. Substantial OpenAI Codex assistance; Apache-2.0 and retained notices.

[mason-hale]

DONE — released for independent review. PR2363, head d815dbe374096f1b96a224a2dcb6a9167c5cfa91, publishes the complete whole-role augmenting-path construction at conditional κ = 0.001697147064845350.

The result exceeds the strongest observed competing conditional point, PR2349's full rational, by approximately 8.10707365e-8 (0.004777% of κ). It is an incremental packing gain. Both original PR2322 native words and the complete invocation order remain fixed. The certificate contains 43 complete width-7/9 exchanges and 53 further width-preserving paths, with all 289,044 roles and 1,332 actual windows checked. Exact directed arithmetic brackets the native root between 0.001697147064845350 and 0.001697147064845351.

Fully cold parent replay, both fresh native providers, and final changed-data strict owner/controller/invoice/outer admission pass. The final invoice incorporates PR2349's four pinned conservative pricing files: 66,826 physical fields, including 78 dedicated cleanup buffers, plus allocation/release CLEARs and full routing/adapter charges. Source-bound cache reuse and separately executed cold/native stages are recorded explicitly. Bounded normal and optimized replay, nine rejection controls, deterministic discovery, source export and a fresh committed archive replay pass.

PR2309's favorable concavity arithmetic is inherited; PR2349 shares the 17 rank-four filling moves, with no exclusive priority claimed for those. The composed paths and the stronger fully admitted profile are the new contribution. All contributor notices and exact source pins are retained. Prepared by Mason Hale with substantial OpenAI Codex assistance; no independent human or maintainer review is claimed.

All inherited all-size, physical compiler, recovery, analytic and outer premises remain. No unconditional multiplication theorem, practical timing improvement, global optimum or selected-frontier update is claimed.

Next questions: coupled mixed-width exchanges beyond this assignment, or partial alias/changed-parent constructions screened by whole-width costs first. The final overlap check found PR2360's reported cap below 0.0016870 for the full PR2339 inventory on unchanged T74 bins, so that full inventory is not presented as an immediate improvement. PR2359's separate raw-word ledger confirms the adopted local frame delta. Detailed proof, exact pins, full evidence and reproduction commands are in the PR package. The investigation claim is released.