Skip to content

Research (no κ claim): exact ten-H1 floor for an eight-field helper block - #5471

Open
apoursh wants to merge 1 commit into
CrocSwap:mainfrom
apoursh:research/bound-improvement-screen
Open

apoursh wants to merge 1 commit into
CrocSwap:mainfrom
apoursh:research/bound-improvement-screen

Conversation

@apoursh

@apoursh apoursh commented Oct 25, 2026

Copy link
Copy Markdown
Contributor

Change and mathematical scope

Proves an exact 10 H1 minimum for the dependency-closed eight-field backward helper-300 block at zero-based events [9044,9116) of the selected #5413 provider. The current block attains the minimum. This rules out the proposed H1 + H1 → H2 saving in this particular neighborhood; no κ improvement is claimed.

The proof permits arbitrary finite rational ADD/SCALE factorizations, temporary basis changes, scalar order and intermediate binary subspaces. It fixes the eight arbitrary entering fields and their entry/exit frames, and requires monotone climbs. Auxiliary fields, new warps, descents, chart-changing exchanges, cross-column conversions and changed boundary contracts are outside scope.

The obstruction is a transfer-rank cut: if either of the two zero-start helpers jumps directly to its terminal plane, only one carrier can transmit their two-dimensional entering information to other terminal planes, whereas the exact backward outputs require rank two. Both helpers therefore need two H1 climbs; the six line-start fields need one each, giving 2 + 2 + 6 = 10 H1.

Adds only research/helper300-eight-field-floor/: written proof, source-bound exact certificate, two separately authored checkers, and registered research metadata. Physical all-width interpretation remains conditional on the inherited native chart contract. This is neither a whole-provider lower bound nor a formal-kernel verification.

Task and overlap

Validation

CPython 3.12.14 on macOS arm64, standard library only:

python3 -B scripts/research_package.py --package research/helper300-eight-field-floor
python3 -B scripts/check_research_commit.py --package research/helper300-eight-field-floor --check-source-history

Passed at the publication head:

  • All four registered executions: both checkers in ordinary and optimized Python.
  • Primary checker: exact transfer and source binding, invertible prefix grammar, all 16 monotone frame-path assignments per orientation, rejection of all seven cost-saving assignments, five controls, protected continuation checkpoints, and equality with committed RESULT.json. The inherited forward window is an orientation control; the selected-provider claim is backward.
  • Separately authored checker: identical backward matrix, boundary frames, field set, 28 retained events and 10-H1 invoice; determinant 1, basis-vector replay, and nine rejected semantic mutations.
  • Committed-byte and source-history preflight; committed-package export integrity check; all four direct checks in a fresh read-only export with Git absent from PATH. Ordinary/optimized outputs agree.

Finite checks establish the source-bound hypotheses and support the written carrier lemma; they are not an enumeration of arbitrary programs. No complete provider/caller or absolute-phase replay, compiler-funding calculation, Lean audit, independent human review, or new exponent calculation was performed. The selected construction is unchanged.

Credit

Credit Cao Yifan's #4610 native word, Dillon Amburgey's #5402 coupled provider, Ali Romman's cached-centre relations, Chris R/Clear1nsight's #5413 completion and source validation, GamingPuzzled's #5465 bottleneck analysis, tomdif's historical #1286 helper-basis screen, and the retained native/compiler lineage. No priority claim is made for elementary cut-rank arguments.

Prepared by apoursh, with substantial OpenAI Codex assistance from a coordinator, synthesis worker, separate mathematical auditor and separately authored source checker. AI-worker separation is not independent human review. New work is Apache-2.0; predecessor notices are retained.

Open next questions

The package's README next questions and TASK.json identify the next bounded problem: extending the same seed through [9044,9137) closes over twenty fields. First derive its entry/exit geometry, rank cuts and complete paid profile; a larger interface could alter the one-carrier obstruction, but no profitable rewrite is supplied. Changed transport or auxiliary fields require an explicitly priced new contract. These follow-ups are outside this PR.

@apoursh

apoursh commented Oct 25, 2026

Copy link
Copy Markdown
Contributor Author

See also #5478

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