Skip to content

Audit three interleaved helper bases with direct residual certificates - #7729

Open
aliromman wants to merge 1 commit into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-057
Open

aliromman wants to merge 1 commit into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-057

Conversation

@aliromman

@aliromman aliromman commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor

This scoped audit reproduces the three critical restriction/split comparisons requested for PR #7726. 54 explicit rational certificates confirm the chosen equations and the three published alternate-equation failures. A separate complete invoice count and rational price enclosure agree with the source's claim-point margin and raw bracket signs. No new kappa is claimed.

Read the proof, receipt, full certificate, direct equation checker, independent price implementation, source pins and credit. This answers the algebra-and-price scope in the released handoff; the claim states the shared-constructor limit.

The new checker directly multiplies formal equation residuals. It checks 30 coefficient identities with 10,484 nonzero multipliers, plus 24 response solutions against 176,320 defining equations. It does not use the source witness evaluator to accept these identities. Each restricted and split helper is also checked as an isolated identity input and output at its actual prefix. All local charts, acyclic schedules, normalizations and zero-feedback conditions are checked.

Restriction Fixed split Alternate failure First coordinate and difference
1400 294, forward, chart 288 16 readout coefficients input 10: +1
1401 288, transpose, chart 288 16 response coefficients output 11: -2
1409 302, transpose, chart 306 27 response coefficients output 25: +1

Both chosen and alternate restriction equations pass at their own prefixes. The failures arise later in the fixed split contracts. They do not prove that the chosen bases are unique or optimal. The source constructor remains a shared dependency; this is a separate equation/price implementation, not an independently implemented construction or independent human acceptance.

The final graph digest is ba59d5db17c40185c8608a734ced10135d8c041e432c3840e9034868781c31a2. The new count checks every nested paid/helper route and all caller QUERY, idle and root-target multiplicities. It reproduces 96,512 paid jobs, 17,296 productive jobs and rank surplus 1,328. Rational log-series tails and alternating exponential bounds give positive productive-minus-paid margins at .00378751509478 and .00378751509486, and a negative margin at .00378751509487. The first margin is about 3.0960942589e-8; exact enclosures are in the receipt. The source's physical, uniqueness and compiler claims remain separate.

Base: e54523517d0f03e58352e647c403a5af3e525a57. All 55 scientific files are retained unchanged from PR #7726's scientific commit 3ba9094c54b61f424d2f98cd2c6ca64c01baf9d3, before its separate shared-CI repair. Its construction proof, original witness evaluator and source credit identify the established result. The deposited parent and label map come from PR #7713 and PR #7716. PR #7727 checks earlier sources with a separate toolkit; none of its code is used here.

Run with CPython 3.12 and its standard library:

python3.12 -B scripts/research_package.py --package research/interleaved-basis-residual-audit

Each registered command ends with:

PASS_INTERLEAVED_BASIS_RESIDUAL_AUDIT {'dual_identities': 30, 'primal_solutions': 24, 'primal_equations': 176320, 'nonzero_multipliers': 10484} 96512 paid 17296 productive

Normal and optimized replay pass. Six changed certificates fail, including omission of the exact role549 half-input term. The price checker passes an exact square-root control and the kappa-zero rank-mass control; omitting fixed root targets or caller charges changes the required signs. Exact source bytes, committed history/modes, export, sealed read-only Git-free replay, public links and rendered publication readback pass. 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.

The source constructor, other journal edits, complete caller operator, native phases/all-column physical implementation, stock/reset and outer/compiler/analytic interfaces remain inherited. The original five-command source suite, selected cold replay, Lean and Docker were not run. The selected .00363005817 and all five framework assumptions remain unchanged. This is source-aware author checking with a shared constructor, not blind, human or formal-kernel acceptance. No other workstream or selected-frontier change is included.

Credit Ali Romman and the disclosed OpenAI Codex assistance for the source construction and original witnesses; Clear1nsight for the causal split and banked relay; DreamingOfClouds for literal splits and the signed-curvature correction; SYLforge and Rohan Arun for the word/caller lineage; and the retained native/compiler contributors. The linked source credit verifies these attributions. Original licenses and notices remain. Prepared by Ali Romman with substantial OpenAI Codex assistance, Apache-2.0.

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