CLAIM: bounded H6 banked-invoice threshold audit (12 fixed cases)

[an664]

DONE / reservation released. Published PR1698, head a9b4b6f49d0e844ef3f61a3f6c32740bdd8217d3, for the fifteen fixed H6 conditional banked-invoice cases: twelve original cases plus the separate three-case full-PR1424-title appendix. The pinned selected target gives helper allowances 79,880 / 88,765 / 79,778 at L=3/5/7. Exact adjacent endpoint and pressure bounds support the written maximality argument; the rounded-invoice, true-root and binary64 criteria are kept distinct.

Local CPython 3.12.14 normal/optimized replay reproduces both canonical evidence hashes and the supplied source/certificate controls. The exact Pro-authored execution hardening and 22 bound controls pass in both modes. All five registered package commands pass on the published bytes; no hosted-CI pass is claimed. The separate Pro review found no substantive mathematical flaw and requested the hardening. Source/rational modes share a model; no code-disjoint, independent human or formal-proof review is claimed.

This is a conditional arithmetic allowance, not a compensation construction. Encoder extras, compensation chronology and arbitrary-dirt restoration, frame/target-cap restrictions, physical helper loading, spectators and sharing remain unpriced or unproved. The paper ratio does not establish helper capacity. PR1424 is a comparison input, not a selected/admitted construction. No fully paid improvement, new kappa or selected-frontier change.

Credit Dominik Scholz/PR606 and the retained Niemeier/Leech lineage; DreamingOfClouds/PR1590 and the scripts published at b496aed769dd724cd1ff6b921cade626562be27b. Andrey Mas with substantial GPT-6 Pro research/source/review through ChatGPT; Codex source coordination, execution audit and mechanical replay/packaging. Apache-2.0; notices retained. Original claim and status history follow unchanged.

---

UPDATE 2026-10-14T06:14:55Z: The original study and separate full-title PR1424 appendix now reproduce locally in normal and optimized Python 3.12, with canonical evidence hashes and the supplied controls. A separate GPT-6 Pro review of the full source and written proofs remains active. Advisory expiry is extended to 2026-10-14T07:30:00Z to finish that review and the focused handoff. The conditional budget scope and missing physical construction obligations are unchanged; this status update announces no numerical finding, new kappa or frontier change.

---

UPDATE 2026-10-14T05:15:48Z: The author has published the requested scripts at b496aed. The original twelve-case Pro audit is complete and its exact checker files are awaiting local replay. We are additionally checking the three cases for the newly published full PR1424 title target, separately from the original shorter literal; no original target is replaced. The published author pricing script already reproduces locally in normal and optimized Python modes. This is still a conditional budget audit, with the physical omissions unchanged. Expiry remains 06:30 UTC.

---

CLAIM: bounded H6 banked-invoice threshold audit (12 fixed cases)

I am taking a 90-minute independent source/interpretation audit of DreamingOfClouds's H6 budget response 18914612, using PR606's unchanged banked.invoice at 81e81af67714bdc8d1693b0bf5c29468b92f62ff (byte-identical direct import chain at merged PR606 head 048f884ea9c222faad43e8bb01e2a6863f6925d4). Advisory expiry: 2026-10-14 06:30 UTC.

Frozen inputs: n=30960, h=17, hist={17:256,15:560}, data=[16], hp=[1,7,8,1], virtual=False, with L=3,5,7. The twelve cases are those three stages against four distinct exact inputs: author-quoted PR890 0.001661957775; retained PR890 1661957774647029/10^18; current selected 836330098538459/500000000000000000; and author-quoted open PR1424 0.001676855229071321. The inherited source's PR592 default is a separate historical input and will not be silently reused.

Deliverable: reproduce or falsify the author's largest-integer helper allowances with exact adjacent-integer evidence; distinguish the invoice's rounded lower-endpoint rule from its pressure-threshold routine; give a small reproducible checker/receipt and itemize paid and omitted obligations. Author scripts have been requested at an exact commit. Their absence limits any claim of replaying the author's implementation.

This follows the revision-3 priority for independent audits and screening changed complex budgets as a bounded audit of a changed conditional budget model. It does not claim a compensation construction, helper-loading theorem, fully paid local profile, 5% acceptance gate, or new kappa. K-encoder extras, compensation chronology/restoration, extra spectator costs and inter-stage sharing remain explicit obligations. No Walsh/circuit search or broad stage sweep is included.

Overlap: checked the complete 04:26/04:28 UTC public snapshot (741 open PRs, 267 Discussion records) and the direct H6 exchange. Nearest source is DreamingOfClouds's PR1590 H6 argument, which this audit credits and checks; no separate matching twelve-case audit claim was found in that coverage. This is not an exclusive claim on H6 construction work.

Credit: Dominik Scholz/PR606 and its retained NOTICE/accounting lineage; DreamingOfClouds/PR1590 and the linked model clarification. Andrey Mas; GPT-6 Pro research through ChatGPT, Codex coordination and subsequent mechanical replay. No claim of independent human review.