Skip to content

Prove sharp three-rectangle canonical rows and optimal output work - #7740

Open
aliromman wants to merge 1 commit into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-063
Open

aliromman wants to merge 1 commit into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-063

Conversation

@aliromman

@aliromman aliromman commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor

Every canonical padded-rotation row has an O(h)-bit description using at most three rectangles in quotient/remainder wire coordinates. The bound is sharp. First and center rows need at most one rectangle. The descriptor gives a complete row stream with Theta(2^h) arithmetic bit work and O(h) temporary bits, while preserving the exact canonical routing rule. This is an enabling proof and algorithm. No new kappa, literal tape row producer or complete CRT caller is claimed.

Read the proof, descriptor and stream, receipt, verifier, source pins and credits.

The closed descent in PR #7734 makes both child lengths constant between two remainder thresholds. The retained endpoint formulas then give one quotient interval per strip. For (h,s,a,j)=(6,53,25,7), three explicit one entries form a fooling set: no all-one rectangle can cover two. Thus even overlapping rectangle covers need three. Summed rectangle areas also give the exact active-switch count for that canonical row; this is not a global swap optimum.

Canonical shrinking countdowns emit each fixed run pattern. Their exact total bit visits are 2n-popcount(n) for a countdown from n. Template copies, phase preparation and output writes are charged; the proof does not assume constant-cost full-width integer operations per output.

Evidence Coverage
Complete canonical census 2,793 networks, 28,797 rows, 1,633,182 entries; all legal inputs through height 6
Separate canonical oracle Unchanged PR4100 compiler; every row also matches the retained PR7734 point query and new stream
Wide descriptor checks 280 descriptors and 4,937 boundary/random points through height 65,536
Structural checks Rectangle caps, exact areas, switch pairing and the explicit three-rectangle lower bound
Cost and failure controls 131,328 counter decrements; all-row work bounds; 14 invalid domains, 3 rectangle mutations and 1 coordinate-order mutation

Python 3.12 and its standard library:

python3.12 -B scripts/research_package.py --package research/rotation-row-rectangles

Both registered normal and optimized commands print:

PASS_ROTATION_ROW_RECTANGLES sharp three-rectangle row and output-linear stream

Normal/optimized replay, committed bytes/source history/modes, export and sealed read-only Git-free serial 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 written proof covers every height; the executable descriptor cap is 65,536. Large rows are not materialized. Python instrumentation is not a literal transition bill or a runtime speedup.

Credit Ali Romman for the closed descent, Nick Petalas for the endpoint and child formulas in PR #7504, and Cinna Davis for the canonical compiler in PR #4100. Their source bytes and notices remain intact. The physical point services in PR #7593 and PR #7735, and the changed Waksman routing in PR #6629, have different contracts.

A physical row producer still needs output layout, private ownership, restoration, head return and complete time/history bills. A fair ROM comparison needs the actual caller's query/reuse order, current-state offsets, transport, payload use and retirement. Those obligations remain open. All five framework assumptions and the selected frontier are unchanged. Historical full suites, selected cold replay, Lean and Docker were not run.

Base: canonical main e54523517d0f03e58352e647c403a5af3e525a57, revision 8. Consumed source heads: PR7734 57bb6efed1be1dda9eb1a0f622ec2773bc8b0047 and PR4100 337d04939571a637aeab54a0f58f54fde811663a. Six retained files have exact byte pins; no unmerged ancestry is asserted. Prepared by Ali Romman with substantial OpenAI Codex assistance in research, implementation, verification and writing. Separate finite oracles do not imply independent human or formal-kernel acceptance. 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