Audit fifteen conditional H6 banked-invoice thresholds

[an664]

For the frozen H6 model, certify the largest integer helper allowance at L = 3, 5, 7 for five distinct exact target inputs. The pinned selected target gives 79,880 / 88,765 / 79,778 helpers. The twelve original cases and separate three-case full-title PR1424 appendix remain distinct. Exact adjacent endpoint and pressure bounds support the written maximality argument; synthetic controls distinguish true-root, rounded-invoice and binary64 predicates.

The package reconstructs the conditional PR606 banked invoice and preserves only its four required dependencies and upstream notices. Source/rational modes share a model and are not code-disjoint verifiers. Local isolated CPython 3.12.14 replay passes normally and under -O; both evidence files regenerate to their canonical hashes. Four source guards, seventeen original controls, ten appendix rejection controls and three positive-control groups pass. The authenticated launcher and deferred appendix output pass twelve local runs plus 22 execution-hardening controls in both modes, bound to the active source. A separate GPT-6 Pro review found no substantive mathematical flaw and requested those hardenings; its full mathematical review report/checker has not been locally recovered or replayed. No human, formal-proof or hosted-CI approval is claimed.

All five registered commands also pass from the final repository package: python3.12 -I -B scripts/research_package.py --package research/h6-banked-invoice-audit.

These are conditional arithmetic allowances. Compensation chronology, arbitrary-dirt restoration, encoder extras, frame-zero precompensation, target cap restrictions, physical helper loading, spectators and inter-stage sharing remain unpriced or unproved. Neither a printed ratio nor these thresholds establishes physical helper capacity. No complete paid construction, 5% gate, new kappa, supplier admission or selected-record change follows. PR1424 remains a comparison target; its parent binding/controller/invoice/geometry review obligations are not discharged here.

Revision-3 independent-moment, refined to this frozen conditional invoice; claim1635. Fresh complete reads of four relevant Discussion threads and the newest 100 open PRs at 06:57:59 UTC supplement the older complete 533-PR/305-Discussion cache. No matching exact audit was identified in that bounded coverage; PR1669 quadratic screening and PR1685 idle packing concern different objects. No all-history novelty claim.

Credit Dominik Scholz / PR606, head 048f884ea9c222faad43e8bb01e2a6863f6925d4, and retained Niemeier/Leech lineage; DreamingOfClouds / PR1590, head 08e134b4a0c7f04869dae703f2a2e688e53c44f0, and published scripts at b496aed769dd724cd1ff6b921cade626562be27b. PR1424 is pinned at bfe423c24555baf35dab991a8d6fe723a038945c. Andrey Mas with substantial GPT-6 Pro research, review and source development through ChatGPT; Codex coordinated sources, audited execution and performed mechanical replay/packaging. Apache-2.0; original notices retained.

[Mister-Meeseeks]

Batch review at a9b4b6f49d0e844ef3f61a3f6c32740bdd8217d3.

Merged to main, preserving the original contributor commits and package attribution.

Reviewed scope: Authenticated H6 invoice audit checks exact rational threshold brackets and the R/R+1 boundary under the declared model. Written true-root monotonicity justifies the largest helper allowance; finite pricing and source agreement do not construct a supplier or admit pending bank provenance.

The pinned scoped command passed; the batch ledger contains its command, input hashes, environment and log. This incorporates the research/tooling contribution and does not change the selected κ or discharge the inherited multiplication-theorem assumptions.

Batch review, attribution and evidence. Thank you for the contribution.