Skip to content

Audit eight binary senders and their unique correction pattern - #7780

Open
aliromman wants to merge 1 commit into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-090
Open

aliromman wants to merge 1 commit into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-090

Conversation

@aliromman

@aliromman aliromman commented Nov 19, 2026 •

Copy link
Copy Markdown
Contributor

Independent binary audit: every subset of the eight sender replacements in PR #7776 preserves its surviving public operator and has a causal same-chart schedule. After all eight helper restrictions, the eight stated sender corrections are uniquely necessary in the include-or-omit family. Their errors cannot cancel. No new kappa or full ordinary-service admission is claimed. This audits the binary service that limits the source's conditional bound.

Read the simultaneous contraction proof, exact certificate, independent checker, source pins and credits. The scientific source is pinned at 0b86d0f618b8c64be9410a83a417067049076067.

Check Result
Independent word extraction All 34,137 instructions; chart tags and owned lifetimes checked
Restriction subsets All 256; 109,830,912 scalar matrix entries covered in each orientation
Correction patterns All 256 after the full restriction; only all eight corrections pass
Missing-correction errors Eight rank-one errors on eight distinct main outputs
Final native Rank 651 public map; 651 public fields; 2,855 paid packets

The proof projects the original reverse output columns away from the surrendered helpers. A deleted packet has zero projected column; each aliased packet has the same projected column as its retained donor. Adding the deleted sender row to that donor preserves every reverse equation. Acyclicity makes their solution unique. The checker verifies this invariant at every retained label for every subset, with separate forward and reverse propagation.

Each edit removes paid ranks 1 and 3 and surrenders an independent rank-4 helper. With the same 36-call caller, the complete signed margin gains 288*(1+3^(1-C)-4^(1-C)), positive for 0<C<1. The target debit is necessary: the resulting map has only 651 independent public directions. The caller multiplicity and full ordinary integration are inherited, not replayed here. All eight donors occur earlier than their retired partners in the old forward word, so the equivalent old reverse alias misses its deadline. The new causal delivery is part of the construction; fixed-bank exclusions do not apply unchanged.

Python 3.12 and the standard library:

python3.12 -B scripts/research_package.py --package research/bit-eight-sender-audit

Both modes print this prefix and elapsed seconds:

PASS_BIT_EIGHT_SENDER_AUDIT 256 subsets 109830912 entries each way 8 omission controls

All ten retained files are authenticated. Normal/optimized registered replay, syntax/indentation, committed bytes/source history/modes, export and sealed read-only Git-free serial replay pass. Every omitted correction has a recorded one-input witness. The independent graph agrees with the source implementation and binds to its published emitted-word digest. This is a binding check, not a physical-word replay. 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.

Ali Romman (@aliromman) supplied the eight edits in the binary source proof, constructor and receipt. Jerry Chen (@CubeJerry) supplied PR #7001's selected application and retained native word. DreamingOfClouds supplied the paired caller lineage; zxv and retained authors supplied the earlier native source. All original notices remain intact.

AlephNaN (@alephnan) supplied the fixed-delivery restriction theorem in PR #7733. Its hypothesis differs from the changed senders here. This audit uses a new set-based extraction and graph checker; the submitted code runs only for final graph equality and emitted-word binding. It is source-aware implementation independence, not blind or independent human acceptance.

No rational projector, all-width physical native, complete caller, owned-erasure implementation, tape layout, ordinary stock, interval gate or compiler replay ran. No global sender optimum, selected cold replay, Lean, Docker or formal-kernel acceptance is claimed. All five 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