Tooling: 5.35× faster production leech_minor via base-256 byte lanes with identical receipts

[VictorLiwentao]

Change and scope

At base 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2, scripts/experiments/affine_vector_screens.py:leech_minor still built each binary row with the historical pairwise sparse-dot genexpr. Jerry Chen’s opt-in adapter research/leech-minor-byte-lanes/ (incorporated CrocSwap/integer-mult-bounds#898) already proved an exact base-256 byte-lane algorithm with identical receipts. This PR wires that equivalent check into the production source used by scripts/audit_affine_vectors.py / make verify-certificates.

Only the source pin for scripts/experiments/affine_vector_screens.py changes inside certificates/affine-vector-screens.json. Every matrix entry is still materialised; Golay inventory, norms, rational span witnesses, binary elimination, polynomial-code screens and the retained historical 2^-31 language are unchanged.

Tooling improvement; no new kappa. Selected frontier untouched.

Result

On Linux / CPython 3.12.3 (Linux-6.12.94+-x86_64-with-glibc2.39), five alternating fresh-process pairs gave median wall time:

• leech-only: 5.579 → 1.043 seconds (5.35×)
• complete audit: 9.465 → 4.213 seconds (2.25×)

Timing is host-descriptive and is not a correctness gate. Every leech_minor receipt matches the baseline checker fieldwise.

Package: research/leech-minor-production/.

Validation

python3.12 -I -B research/leech-minor-production/verify.py
python3.12 -I -O -B research/leech-minor-production/verify.py
python3.12 -B scripts/research_package.py --package research/leech-minor-production
python3.12 -I -B research/leech-minor-production/benchmark.py --rounds 5

Expected:

PASS_LEECH_MINOR_PRODUCTION leech_cases=5 code_cases=4 certs=1 no_new_kappa
PASS_REGISTERED_RESEARCH_PACKAGE research/leech-minor-production outcome=completed
PASS_IDENTICAL_LEECH_MINOR_PRODUCTION_RECEIPTS

Default verify is read-only. Zero-row / corruption controls reject; require gates remain under -O. Full make verify, selected cold replay, Lean, and multiplication-theorem obligations were not rerun for this isolated historical-checker change.

Overlap and credit

Closest priors (different delivery): Jerry Chen’s research/leech-minor-byte-lanes/ / #898 (opt-in byte-lane adapter; this is the production wiring), CrocSwap/integer-mult-bounds#756, CrocSwap/integer-mult-bounds#2194 (same production-promotion pattern on verify_factor). No competing open PR on production leech_minor at refresh.

Douglas R. Colkitt authored the affine/vector screens and original checker. Jerry Chen authored the byte-lane algorithm. Prepared by Wentao Li (VictorLiwentao, @Is__Victor) with substantial AI assistance, under Apache-2.0.