Skip to content

research: certify the full centre sheaf kernel and primitive integer fibre - #1

Merged
Clear1nsight merged 3 commits into
review/centre-sheaf-basefrom
contribution/centre-sheaf-fibre
Oct 18, 2026
Merged

Clear1nsight merged 3 commits into
review/centre-sheaf-basefrom
contribution/centre-sheaf-fibre

Conversation

@Clear1nsight

@Clear1nsight Clear1nsight commented Oct 18, 2026 •

Copy link
Copy Markdown
Owner

Result

The full 112-port common observation kernel on PR3656's 4,080-owner subgroup is exactly one-dimensional. Its previously exhibited (1,-1,-1) vector on ports (0,6,85) spans the whole kernel, not just a restricted three-port family.

This completes one user-assigned sheaf/algebraic-statistics investigation in the PR3147 centre-sharing lineage. There are 456,960 scalar variables, old observation rank 32,640, and exactly 424,319 additional linear observations needed to recover the full later observation. Every proper subset of this particular subgroup, and its left translates, has trivial common kernel and needs 104 observations per owner. This is not a general 4,080-owner support lower bound.

The primitive integer fibre is generated by the same vector, with 4,080 positive and 8,160 negative unit entries. It is not total-count preserving. Adding a fixed total-count row kills the fibre freedom, so the Markov move cannot be advertised as a free scheduling exchange.

Proof and source scope

The cellular-sheaf reduction eliminates 109 independent singleton output-coset classes, leaving one rank-two triple and a connected compatibility graph. It uses all touched output owners. The exact inherited calls-0/4 scalar observation contract, first-call source slice, and active/spectator route split are stated in PROOF.md. Physical acquisition, phases, native payload semantics, dirty restoration, precision, hosting, recursive tariffs and a complete supplier are not certified.

Current public head: 37091843c18491abda68b465ad768180c0176a03.
Base: 20ecf2891d80b5acf18d980b2164fb44ea7faa90, priority revision 6.
Scope: 11 additive files, three commits, one isolated research directory.

Executed verification

Hosted workflow 39167870503 passed planning, bounded replay, and unchanged-input checks on Ubuntu 24.04.5 / CPython 3.12.15. The tested preview merge cd93184c32101bae7a2c85d278b368b8c69d0e16 has the same complete Git tree as the public head: 7d90a18e7099a70b0b36b2e5548180461192a5a6.

Both separately written implementations reconstruct the pinned original archive, normally and with -O. The primary reuses PR3071's authenticated extractor and column-matrix/rational-rank arithmetic. The second independently parses archive members, propagates source coefficients forwards, uses row matrices and a different generating pair, checks modular minors and direct routed residuals, and proves connectivity by union-find. The committed hook compares both substantive results with strict types. Eight generic tests and eight receipt/JSON controls pass in both modes.

Canonical committed-result SHA256:
9e23c75d59502748f930e3bc029d528c684224963113b7fa1d2f46357ae381b9.

python3.12 -B scripts/research_package.py --package research/centre-sheaf-fibre

These are separate same-session implementations, not independent human or proof-kernel review. No full selected cold replay or native supplier execution is claimed.

Publication boundary and preserved work

This is a fork review PR, not an upstream CrocSwap submission. The connector rejected both the attempted upstream claim comment on CrocSwap#3656 and the upstream PR with HTTP403, Resource not accessible by integration. No upstream claim or handoff is represented as published.

The current compatible finalisation arrived while an offline package was being prepared. It has been preserved without a force-push. The separate portable-verification version remains at 4ae21064edc31b348d27a87408184687dcef87e7 on verification/centre-sheaf-offline. It adds frozen derived geometry and standalone corruption controls and is not the same package as this PR. Its derived-data local checks must not be confused with this PR's actual-archive hosted replay.

No main branch, selected frontier, or predecessor source was changed. No PR was merged or closed.

Credit and handoff

Matarener/CrocSwap#3656 retains the subgroup and original witness; dcharb78/CrocSwap#3147 the programme; Ross Richardson/CrocSwap#3071 the observation model and reused extractor; Sagar Thakrar/CrocSwap#3291, Gonzalo Mier/CrocSwap#3097 and Sudarsh3301/CrocSwap#3168 the corrected graph/census lineage; Aaron Wang/navinor, ywz65 and the selected geometry/compiler contributors retain credit. Classical sheaf and Markov-basis theory are credited in the proof. Prepared by Clear1nsight with substantial OpenAI assistance.

The completed result supports stopping further proper-subblock searches inside this particular coset. A genuinely different owner block or output contract remains open, but needs an explicit information target and favourable full cost allowance before physical construction. The global support minimum and physical sharing remain unresolved; no new kappa is claimed.

Draft bounded follow-up to PR3147/3656. Reuse the frozen PR3071 source extractor; do not change selected sources or claim physical sharing.
Add a separate archive parser, sparse forward coefficient propagation, row-matrix closure, modular rank certificate and direct residual controls.
…fibre

Record the source-dependent 112-port, 4080-owner kernel result, its sharp proper-subblock and linear-observation consequences, and the primitive fibre limitation. Add strict cross-implementation receipt replay. Preserve all original source bytes and the selected frontier. No native supplier or kappa claim.
@Clear1nsight Clear1nsight changed the title Draft contribution: centre sheaf kernel and integer-fibre certificate research: classify the full 4080-owner centre kernel using sheaves and integer fibres Oct 18, 2026
@Clear1nsight
Clear1nsight marked this pull request as ready for review October 18, 2026 21:34

Copy link
Copy Markdown
Owner Author

Completed handoff

The fixed-block sheaf/integer-fibre investigation is complete at 37091843c18491abda68b465ad768180c0176a03. The final registered workflow https://github.com/Clear1nsight/integer-mult-bounds/actions/runs/39167870503 passed both jobs. Normal and optimised replays agree on canonical receipt SHA256 9e23c75d59502748f930e3bc029d528c684224963113b7fa1d2f46357ae381b9. The preview-merge tree is exactly the published head's tree, 7d90a18e7099a70b0b36b2e5548180461192a5a6.

For all 112 ports on the fixed 4,080-owner H block, the common kernel is exactly the span of CrocSwap#3656's known witness. Proper subsets of H have no common-kernel direction. The integer coefficient fibre has one primitive move, but that move does not preserve total count. These are information/representation results, not a native or scheduling construction.

No further direction search within this same H is warranted under the unchanged observations. A useful successor must specify a different owner set, such as a union of cosets, or an actual weaker consumer, and then provide its compatibility equations and complete physical budget. The global minimum-support problem, native acquisition/restoration, chronology, precision, hosting and supplier admission remain open.

This fork PR is ready for review and remains unmerged. The contribution is not submitted upstream: current integration permissions rejected both upstream claim and PR writes. No upstream claim is reserved. The exact branch for a future authorised upstream submission is Clear1nsight:contribution/centre-sheaf-fibre, targeting CrocSwap main.

Credit Matarener/CrocSwap#3656 for the subgroup and witness, David Charbonneau/CrocSwap#3147 for the investigation, Ross Richardson/CrocSwap#3071 for the observation model/extractor, Sagar Thakrar/CrocSwap#3291 and Sudarsh Chaturvedi/CrocSwap#3168 for the graph/census lineage, and all retained source contributors. Classical sheaf and Markov-basis theory is explicitly credited. Substantial OpenAI assistance is disclosed; separate same-author implementations are not independent human or formal-kernel acceptance.

@Clear1nsight Clear1nsight changed the title research: classify the full 4080-owner centre kernel using sheaves and integer fibres research: certify the full centre sheaf kernel and primitive integer fibre Oct 18, 2026
@Clear1nsight
Clear1nsight merged commit ae52586 into review/centre-sheaf-base Oct 18, 2026
2 checks passed
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