Skip to content

Audit the final P104 helper and its uniquely required pulse weights - #7784

Open
aliromman wants to merge 1 commit into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-092
Open

aliromman wants to merge 1 commit into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-092

Conversation

@aliromman

@aliromman aliromman commented Nov 19, 2026 •

Copy link
Copy Markdown
Contributor

Scoped independent audit: the final helper split in PR #7781 preserves all 1,279 prescribed public outputs. Its three fixed pulse weights are uniquely required over the rationals. A partial omission can corrupt old owners even though omitting all three fresh-input terms loses only the fresh return. No new kappa or complete supplier admission is claimed.

Read the proof, exact certificate, checker, source pins and credits. The audited source is 913fb6febe28cc3ccad1c16de8e45b8bce0da063.

Check Result
Exact final split 48 changed rows; both published graph hashes recovered
Arbitrary-input maps Old forward, new forward and new adjoint; 4,904,966 coefficient positions
Fixed pulse family Rank-three response; a three-row minor has determinant 25/4
Negative controls All eight fresh-term omission subsets and twelve rational pulse controls
Native rank change 5,714 paid MOVEs unchanged; one rank-eight target becomes two rank-four targets

For sends 92740, 98411 and 98796, let R_i be the output response to the stated pulse divided by eight. The checker shows that each pulse is invisible to both local probes. Their response columns are independent. If their complete pulse weights are lambda_i, the full error is (y-C) sum_i (lambda_i-1) R_i, where y is the independent fresh input and C is the exact calibration. Hence only the stated weights preserve every output. This conclusion fixes these sends and the rest of the graph; it is not a global supplier optimum.

With the retained 48-call multiplicity, the local signed-margin change is 48*(2*4^(1-kappa)-8^(1-kappa)), positive for 0<kappa<1. The full finite caller and increased ordinary work are not repriced here.

Use Python 3.12 and the standard library:

python3.12 -B scripts/research_package.py --package research/p104-final-helper-audit

Both modes print this prefix and elapsed seconds:

PASS_P104_FINAL_HELPER_AUDIT 1279 owners 3 pulse rank

Normal/optimized registered replay, syntax/indentation, committed bytes/source history/modes, export and sealed read-only Git-free serial replay pass. All 19 retained files are authenticated. Hosted CI did not certify this package. The plan job failed in the shared test_current_ci_section_matches_selected_record: the base guide names frontier6030 while the selected record requires make frontier7000-verify. The log reports 760 tests, one failure and 24 skipped; package replay was skipped. This package changes no shared CI file. Local replay remains the scientific evidence.

Clear1nsight (@Clear1nsight) supplied the new construction, final journal and proof. The retained overlay constructs the preimage. Ali Romman (@aliromman) supplied PR #7776's split algebra and carrier. Its split routine is used for a forward cross-check. PR #7777's exact rational expander supplies generic coefficient and chart checks. The new inverse, adjoint and error-law checks retain all original notices and source lineage.

The earlier fourteen births, physical word, full caller, binary service, ordinary costs, stock, compiler and selected cold replay did not run. No Lean, Docker, independent human or proof-kernel acceptance is claimed. This is source-aware computational corroboration. All five inherited assumptions and the selected frontier remain unchanged.

Base: canonical main e54523517d0f03e58352e647c403a5af3e525a57, priority revision 8. Prepared by Ali Romman with substantial OpenAI Codex assistance in research, implementation, verification and writing. Apache-2.0. Canonical claim.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant