Skip to content

Use the minimal physical odd scale; reduce one scalar-rate bound 128-fold - #7718

Open
aliromman wants to merge 1 commit into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-051
Open

aliromman wants to merge 1 commit into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-051

Conversation

@aliromman

@aliromman aliromman commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor

The fixed physical supplier in PR #7717 can use odd scale 11025 instead of 2083725 in every native invocation, with the same binary allowance, signed caller, storage and recursive jobs. The conservative scalar-service rate falls from 2^29 to 2^22 steps per padded cell. This factor of 128 applies to one charged upper-bound coefficient; it is not a measured runtime or total-cost improvement. No new kappa or smaller selected root format is claimed. Proof · Exact certificate and physical witnesses · Checker · Source pins

The normalized minimum product grid 90316800 was already reported by the source. The new result proves its physical odd-scale substitution and complete finite cost consequence. For Gaussian-dyadic invertible charts on the complete root factor, a normalized scalar coefficient has its odd denominator cleared exactly when its physical chart block does. Independent owner inputs give necessity. The source's per-call scale/unscale and nested debt induction then retain the original root bounds. Arbitrary global phases, nonunit chart maps and a restricted correlated-input contract are explicitly excluded, with exact counterexamples to those extensions.

All 9,966,426 coefficient products are checked. Separate Fraction columns and exact physical chart components give sharp witnesses:

Omitted prime Independent input Product chart Actual under-scaled component
3 1277 118 -2360575/1536
5 1207 324 441/20
7 278 595 -750/7

Each value has an odd denominator that extra binary precision cannot repair. Thus every sufficient odd scale for this arbitrary-input physical contract is divisible by 11025=3^2*5^2*7^2. The implementation retains binary height 20, although the normalized coefficient minimum is 13. No smaller full physical width is asserted.

The full smaller-grid replay checks all 1,470 endpoints and addressed bank operations. Every ordinary count and layout formula equals the source receipt; only the scalar rate changes. The complete caller retains 124,816,384 scalar services and 54,112 odd conversions. Each native invocation retains 8,999 numerical homes, 1,526,938 reset services and 414,942 transfer legs. The 796 factors now need at most 14 bits, and all 4,776 signed bit-service cases pass. Missing prime factors, a live-sender overwrite and a partial payload clear fail. The source's separate caller scale 2025, W+26 padding, 338/588-bit root envelope and compiler H remain valid; the depth guard is checked again.

The 100,416 paid jobs and 20,656 productive jobs are unchanged. This preserves the source's conditional application under the same five native/owned/compiler/precision/analytic interfaces. It does not discharge the whole precision_tape assumption or prove a materialized all-size rounded multiplier. The selected frontier is unchanged.

Use CPython 3.12 and the standard library:

python3.12 -B scripts/research_package.py --package research/quotient-split-minimal-odd-scale

Both registered normal and optimized commands print:

PASS_MINIMAL_ODD_PHYSICAL_SUBSTITUTION 11025 128

Normal/optimized replay, retained source bytes, committed bytes/modes, source history, export and sealed read-only Git-free replay 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 full phase catalogue, compiler search, predecessor suites, selected cold replay, Lean and Docker were not rerun. The inherited source evidence is pinned, and the exact graph is unchanged. This is author checking with reused engines, not blind, independent human or formal-kernel review.

Source head is 586f4f81a8bd09a65a8e6d0bfff57a0fec8925e8, graph 5183479ecac34a1cb5c046f202e19d7161679265fe121b4a0af922fe344d8fc1. Base is refreshed canonical main e54523517d0f03e58352e647c403a5af3e525a57, priority revision 8. The source's complete application and the existing common-root proof are material dependencies. No frozen checker is patched or disabled.

Credit Gareth Davidson / @bitplane for the PR #866 prefix-grid criterion, Ali Romman's PR #7680 for the integer-gcd and rational witness structure, icekylinx and the credited fixed-root authors for odd clearing and nested precision, Clear1nsight for the physical realization, DreamingOfClouds for the split witnesses, and SYLforge, Rohan Arun, Jerry Chen / CubeJerry, Douglas Colkitt, mightyOil and the retained contributors for the source and framework. Full credit and pinned predecessors. All source notices are retained. Prepared by Ali Romman with substantial OpenAI Codex assistance, Apache-2.0.

The canonical claim is limited to this scale substitution. Complete open/closed PR and Discussion snapshots, exact source-reference searches and fresh prepublication deltas were checked. Sparse local and unpublished work remain overlap limits. No helper/paid-bank search or other contributor's PR is changed. No follow-up is reserved.

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