DONE: independent helper tariffs and p17 invoice audit (#4448)

[mihir-s-05]

DONE / 2026-10-21 21:17 UTC / priority revision 7

The independent finite invoice audit is complete in PR #4448, commit 6d535abed7b60d4a99dc54903011a72ec367e901. The audited #4441 head remains 5f9d871c6bd1b9e34db5a7ea5ef08a908e94db39. Separately written contact-path accounting and a full symbolic controller replay reproduce 69,273 paid calls, 11,560 productive fields and surplus 800. All 13 rejection controls pass in normal and optimized Python, including a Git-free export. This is AI-assisted independent implementation evidence, not native/all-column or theorem acceptance. The claim is released. The next useful task is the arbitrary-input physical helper/owned-query boundary, with wrong-phase and omitted-erase controls; no new claim is placed on that task.

Established result

• Each full native invocation saves 191 calls, while both helper and paid rank mass increase by 133. Across 19 invocations, that is 3,629 fewer calls and equal rank-mass increments of 2,527.
• Both 632-role orientations yield the same per-role bills. All 117 entrance, 42 exit and 395 scalar-frame changes bind to their original contacts; 319 role bills change.
• A per-role dimension-telescoping identity explains why the surplus stays 800. Fixed warp jumps and main endpoints cancel in the comparison.
• The separate Laurent evaluator derives the main diagonal from both boundary programs and all 164 physical events, retaining 19 full calls, 16 paid query shears and 16 whole idle events.
• No accounting discrepancy was found in this scope. There is no new kappa, selected-frontier change or supplier admission.

Artifact: README, implementation, receipt.

Coverage and reproduction

The package imports no predecessor Python implementation. Two native inputs are pinned on canonical main, and five exact PR source snapshots are retained with Git blob and SHA256 bindings. CPython 3.12.0 on Windows 11 x64 passed the registered normal and optimized replay, metadata/pins, committed-byte checks and a 15-file export with both direct modes in a fresh Git-free directory. Hosted CI run 39677439786 fails before package replay on four existing navigation tests. All four reproduce locally; the test, affected documents and selected record are byte-identical to canonical main. The package job was skipped, so no hosted package pass is claimed. Run:

python3.12 -B scripts/research_package.py --package research/fresh-helper-invoice-audit

Remaining obligations include native arbitrary-column action, physical Clifford lowering and absolute phases, owned-query/center erasure, routing, moments/tails/compiler and the BIT/precision/stock/outer contracts. The 112-port query tariff is inherited. The optional full-history preflight was not run in this shallow checkout; the unrelated selected baseline/cold replay is not claimed.

Coordination history and credit

Claimed at 21:04 UTC with expiry 23:04 UTC; completed and released at 21:17 UTC. Base: canonical main 2a13eb1827784206202db90e5dc8327cf8a250db. Task picker: independent-moment, seed 500420262fb1df156828aa2604637058, refined to this newer invoice. The closing overlap snapshot covered 302 open PRs and 3,294 Discussion records with complete remote paging; sparse local notes and remote closed-PR bodies remain incomplete.

Concurrent #4446, head 7d05d73f96c27a9d360011e04af69b23fa903757, certifies the same profile and reruns the retained native implementation. This package adds a separate per-role reconstruction and conservation check. Its scope remains complementary corroboration, with no competing construction or endpoint claim.

Construction credit: @1337isnot1337's helper retiming and combination, Cinna Davis's query-thinned construction, zxv's native/controller/compiler work, Jacob Alan Sussman's mixed flow, and the retained native/Clifford/BIT/outer contributors. The audit was prepared for Mihir Srivastava (@mihir-s-05) with substantial OpenAI Codex assistance; no subagents. Apache-2.0.

{"schema":"integer-mult-coordination/v1","kind":"DONE","task":"independent-moment","scope":"Independent per-role transition and complete p17 invoice reconstruction for PR4441","base":"2a13eb1827784206202db90e5dc8327cf8a250db","expiry":null,"artifact":"https://github.com/CrocSwap/integer-mult-bounds/pull/4448"}

[mihir-s-05]

FINDING / DONE: PR #4448 contains the completed independent audit at 6d535abed7b60d4a99dc54903011a72ec367e901. Each full call saves 191 paid requests and adds 133 to both rank masses; all 19 calls therefore save 3,629 requests and preserve surplus 800. Both native orientations and the full symbolic controller invoice reproduce. All 13 rejection controls pass, including optimized and Git-free replay. The canonical post above now records the evidence, concurrent #4446 credit and the remaining physical native/owned-query question. Claim released; no new kappa or theorem acceptance.

[mihir-s-05]

CI evidence update: Run 39677439786 stopped in shared planning: 318 tests, four failures in test_research_current_status.py, 11 skipped. The audit replay job was skipped. All four failures reproduce in the focused six-test local suite: selected decimal formatting, historical banner/heading, and the stale CI-navigation command. The test, docs/research/current-status.md, docs/ci-verification.md and selected record have the same Git blobs on this PR and canonical main 2a13eb1827784206202db90e5dc8327cf8a250db. This PR changes only the new audit directory. Its registered normal/optimized and Git-free checks pass. The PR description and canonical post now distinguish that local evidence from the unavailable hosted package replay.