Tooling: accelerate historical Leech minors with exact byte lanes

[CubeJerry]

The historical affine-vector audit spends most of its runtime reconstructing five Leech minors with pairwise Python dot products. This opt-in, source-pinned adapter packs signed dot products into exact integer byte lanes, preserves every row and original check, and reproduces the complete nine-screen certificate byte for byte. Tooling improvement; no new kappa or negative result.

Three alternating fresh-process pairs on Linux x86_64 / CPython 3.12.14 measured median complete-audit wall time 18.946560 → 8.988220 seconds (2.107932×). This includes source admission, all five Leech and four unchanged polynomial-code screens, exact ranks, certificate comparison and output. Raw samples and implementation hashes are retained in BENCHMARK.json; timings varied on the shared runner and are descriptive, with no acceptance threshold. A preliminary run that overlapped tests is explicitly excluded.

Only research/leech-minor-byte-lanes/ is added. The original checker, default commands, selected frontier and upstream/ remain unchanged. The byte-lane proof, 41 input pins, complete expected receipt, benchmark, corruption controls and attribution are included.

Reproduce and validation

python3 -I -B research/leech-minor-byte-lanes/verify.py
python3 -I -B research/leech-minor-byte-lanes/test_lanes.py -v

Expected verification marker: PASS_LEECH_BYTE_LANES all_frozen_rows_equal=true cases=5.

• 5,927 full rows / 7,531,037 matrix entries compare exactly against the pinned pairwise implementation; the complete nine-screen certificate and all six timed certificates retain their original SHA256.
• 11 test groups pass, including all three unchanged predecessor tests in both modes. Literal integer oracles add 684,513 entry comparisons across exhaustive small signed vectors and seed-86 cases from every Leech generator family.
• Controls cover sign/shift errors, omitted diagonals, column permutation despite equal rank, dependent rows, faulty kernels, invalid lane domains, source/fixture/certificate/helper corruption, missing files, incomplete receipts, disabled assertions and output preservation.
• Minimal offline relocation to a path containing spaces passes with unchanged input bytes. All 41 pins match base Git objects. Selected-record integrity and its five existing tests pass; whitespace checks pass.
• Full make verify, cold selected-supplier replay and Lean were not rerun for this isolated historical-checker adapter. No hosted CI pass or full multiplication theorem is claimed.

Pins, overlap and credit

Base f567884aeb8851b6e794af1184119a3bfbe1e631; OS-random assignment seed 86, verification engineering. The local claim records three candidates and the completed result. No pinned sprint coordination issue existed. Final metadata checks covered arrivals through #894 and targeted all-state Leech/affine-checker searches; no duplicate accelerator was found.

Credit Douglas R. Colkitt for the original audit and fixtures at d55bcd62ceb8f322bcf311a5a66e94c3ac7724f0, Curtis Dransfield's #752 for the related carry-safe integer-lane technique on different quadratic-affine minors, and merramil's #778 for related Gram-packing work. Their exact heads and the upstream source pin are retained in NOTICE.md. No pending PR code is imported; #598's separate Leech/Turyn negative searches are not reopened.

Prepared byCubeJerry)with substantial OpenAI Codex assistance, under Apache-2.0. The selected conditional κ remains 0.001661634653669567, with all inherited construction, all-size, realization, hosting, precision/tape and analytic assumptions.

[Mister-Meeseeks]

100-PR batch review at 521c86e7fd4d0fbd1ed4e0524aa26a1e45861477.

Merged to main, preserving the original contributor commits and package attribution.

Reviewed scope: Opt-in byte-lane Leech-minor accelerator with complete row and certificate comparisons; default checker retained.

The pinned scoped command passed; the batch ledger contains its command, input hashes, environment and log. This incorporates the research/tooling contribution and does not change the selected κ or discharge the inherited multiplication-theorem assumptions.

Batch review, attribution and evidence. Thank you for the contribution.