Independent QQ16 completed-defect audit and feedback guard refinement

[Arham-Tahir64]

DONE / claim released — 2026-10-15, priority revision 5.

Public artifact: PR2670, exact commit 0ce9c0f1f2b42b2e403801ea79f55a21cdf6e035, package research/qq16-defect-independent-audit/.

Established: a separately written JSON-only reverse-covector interpreter independently corroborates PR2628's five all-cut envelopes at all 2,680 initialized changed-field boundaries of the unchanged 2,463-event PR2540 word, and all five envelopes of all 464 final fields. P=240, A=1529, K=2298, native=192, cover=280; the original signed-error 841 interior witness also reproduces in a separate literal numeric execution. No predecessor Python verifier is imported or executed.

The exact feedback threshold K(P+1)=553818 proves u <= 2^(-b-20) sufficient for every integer b>=0 under the same declared contract. PR2628's conservative 21-bit guard remains correct. Twenty is minimal only for this sufficient uniform-in-b inequality; a 19-bit majorant failure is not physical impossibility or an optimal precision theorem.

Evidence: CPython 3.12.14 registered package replay, CPython 3.14.7 optimized replay and nine behavioral control groups pass. A Git-free read-only export passes with eleven unchanged files. Source corruption, chronology/owned dirt, noncommuting route order, capture persistence, escaped scratch errors and approximate borrowed restoration are covered. Author/AI-checked, no independent human or proof-assistant review; hosted CI is not claimed. The worktree is clean.

Before submission main remained b4560d9 and both exact predecessor heads were unchanged. Final overlap fully paginated 840 open PRs (16:52:25 UTC) and 1,305 Discussion records (16:55:16 UTC), no warnings. Relevant overlaps are deliberately audited PR2628 and the credited stability lineage, not a claim of original envelopes.

Next useful task: realize this completed-instruction defect allowance in one actual rounded native child and convert component errors to the stated maximum-per-owner payload Hilbert norm. Child internals, entry/F0/B1 bounds, exact control/CLEAR, retention, movement, hosting and supplier admission remain separate obligations. No changed word, invoice, selected frontier or new kappa.

Credit ywz65's QQ16/native/precision sources, Ali Romman, thakrarsagar, starship-s, Ross Richardson, PR2096/2166/2503 and retained lineage. Arham Tahir with substantial OpenAI Codex assistance; no subagents.

---

Original claim retained below:

CLAIM — deliberate independent QQ16 completed-defect audit and feedback guard refinement

Status: investigating, 2026-10-15 16:44 UTC, priority revision 5. Evidence LOCAL-ONLY. Claim expires 2026-10-15 18:44 UTC.

Question: independently reconstruct PR2628's all-cut P/A/K/native/cover envelopes from PR2540's declarative word and the pinned main owner-map registry, without importing or executing either predecessor verifier. Check every changed-field boundary by reverse covectors, rather than reusing the author's forward sensitivity code. If the coefficients agree, determine the smallest uniform integer guard implied by the stated feedback inequality itself; its existing 21-bit guard is sufficient and is not alleged false.

Base: canonical main b4560d9. PR2628 e64a7ba, PR2540 630d557. Exact owner registry is main research/joint-query/certificate.json.gz. Original selector seed 7895d4b19cc7dc36ccded3c9bb293ae6, recursive-interfaces/joint-query-controller; scope changed to the newly completed precision implication because PR2603/2628 already cover the initially considered new bound.

Method/domain: exact rational arithmetic, all 2,463 events and initialized changed-field cuts, separate 240 temporal inputs and all 2,136 completed-instruction defect sites. Preserve ordered owner actions, exact owned CLEAR and temporal capture semantics. Independently test semantic mutations, source binding and the finite guard calculation. Bounded to this source and error contract; no child-internal, physical component-rounding, B1/entry, retention, supplier or kappa claim.

Overlap: complete 824 open PRs including drafts (16:38:47 UTC), 1,283 Discussion records (16:41:28 UTC), local notes/ledger, plus current direct reads. PR2628 is the result being deliberately audited, and PR2603 is the separate exact-grid predecessor; this is not a claim to originate their envelope or method. No matching independent all-cut audit/guard refinement was found; metadata coverage is not a novelty guarantee.

Expected artifact: research/qq16-defect-independent-audit with standalone JSON-only checker, exact source pins and rejecting controls, or a precise discrepancy/handoff. About 40 minutes plus verification. Credit ywz65, Ali Romman, thakrarsagar, starship-s, Ross Richardson and the retained QQ16/precision lineage. Arham Tahir with substantial OpenAI Codex assistance; no subagents or independent human review.