Skip to content

Prove that nonplain kind B parts cannot finish by one last call at any width - #7683

Open
aliromman wants to merge 1 commit into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-026
Open

aliromman wants to merge 1 commit into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-026

Conversation

@aliromman

@aliromman aliromman commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor

Result and scope

A nonplain kind B part cannot finish by one last call alone, at any address width and any nonzero rational coefficient. This scoped proof upgrades the finite three-bit observation in PR #7629. It includes arbitrary spare address bits. An explicit linear functional on two or three formal terms kills every available column and is nonzero on the target.

The exact domain is the formal frame game Q[Sp(2m,2)], m>=3, with raw stabilizer P, paid transvections D,W,T, free gamma, M=WD, E=W gamma, and y=M+cE. Assume E is outside PD, c!=0, and level(TWD)=3. If U=span_Q{p,pD,py:p in P}, the result is TM outside U+TU. The proof includes noncommuting calls. It permits free input preparation and additions of earlier arrays, but no free map on the last output itself.

Read the proof, exact receipt and representative matrices, verifier and scope record. The key new step excludes the last possible earlier-input cancellation through incompatible horizontal two-planes. The source's zero finite count is prior evidence; the all-width and all-coefficient theorem is the new contribution. This completes the advance claim.

No new kappa, general two-in-five exclusion, engine-phase lift or selected-assumption discharge is claimed. Post-call free processing, joint use of both last outputs, three-term parts, native realization and a complete physical invoice remain open. All five assumptions and selected PR #7000 remain unchanged.

Evidence and reproduction

Normal and optimized CPython 3.12.16 checks passed on macOS arm64. The complete three-bit census uses all 10752 free elements, excludes 768 plain cases and checks all 32 level-three finishing calls on each remaining part: 319488 part/call pairs. It finds 315392 two-coordinate and 4096 three-coordinate separating certificates. Complete incident columns and all singleton families are checked with polynomial identities in c, so each finite certificate covers every nonzero rational coefficient.

An independent rational elimination route checks the complete target orbit on 80 contracts at widths three through six, both certificate types and five coefficients. Another 240 varied-direction cases cover widths four through six, including 115 noncommuting first/shared pairs. These wider checks are samples; the written proof establishes arbitrary width. Seven positive contracts show why the plain, zero-coefficient and level-three restrictions matter. A corrupted separator and two small linear-system controls check the verifier's failure boundaries.

Metadata, committed bytes, executable modes, source ancestry, the seven-file official export and its integrity check passed. A read-only optimized export replay passed without Git or a usable PATH, with all eight payload/manifest hashes and modes unchanged. The proof and README were checked with GitHub Markdown rendering. No external runtime source is used.

From the repository root, with CPython 3.12 and the standard library:

python3.12 -I -W error -B research/kind-b-last-call-obstruction/verify.py
python3.12 -I -W error -B -O research/kind-b-last-call-obstruction/verify.py
python3.12 -B scripts/research_package.py --package research/kind-b-last-call-obstruction
python3.12 -B scripts/check_research_commit.py --package research/kind-b-last-call-obstruction --check-source-history --check-modes

Expected status markers:

PASS_KIND_B_LAST_CALL_OBSTRUCTION
PASS_REGISTERED_RESEARCH_PACKAGE
PASS_RESEARCH_COMMIT_BYTES

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. Predecessor scientific suites, selected cold replay, native/tape lowering, Lean and Docker were not run. Independent human and formal-kernel review are unclaimed.

Sources and credit

Prepared by @aliromman with substantial OpenAI Codex assistance. Jacob Sussman (@jacobalansussman) supplied the model, finite nonfinishability observation and boundary definitions in PR #7629, pinned source. Nick Petalas (@petalas) supplied the related kind A coset proof in PR #7649, pinned proof. Ali Romman's PR #7679, pinned removal proof and pinned verifier, supplies the related endpoint method and reused binary-matrix/rational-span routines. The finishing proof, separator checks and new census are added here. New work is Apache-2.0.

Base: canonical main e54523517d0f03e58352e647c403a5af3e525a57, priority revision 8. Card complex-rank-shape-decoder is refined to this single-part obstruction. Refreshed open/closed PR and Discussion overlap contains the finite source and the distinct kind A/removal results; sparse local notes and unpublished work remain coverage limits. No source PR or selected artifact is edited. The next separate questions are the exact post-call column family and joint decoding of both last outputs.

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