Tooling: 2.26× faster production emitted-local audit via integer rows with identical receipts

[VictorLiwentao]

Change and scope

At base 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2, production research/complex-k492-owner-repair/code/hpe_audit_local.py still stored each sparse scalar row as a dictionary of Fraction values. woklerx’s opt-in adapter research/emitted-local-integer-replay-seed90/ (incorporated #1184) already proved shared-denominator integer rows with identical complete local-audit receipts. This PR wires that equivalent rewrite into the production k492 checker.

SELECTED-SOURCE-bound research/complex-sw4-frame8/local/hpe_audit_local.py copies are deliberately left pairwise so the selected inventory is not refreshed by this tip.

No certificate κ pins change. Same-tip provenance hygiene:

• research/complex-signed-prefix-audit/ and research/selected1094-wire-prefix-audit/ demote the k492 hpe_audit_local.py pin to provenance and retarget live VERIFY inputs to the still-pairwise sw4 checker.
• research/emitted-local-integer-replay-seed90/ vendors pairwise-era bytes as local baseline.py.
• Non-VERIFY hygiene inventories (stage-diagonal / center-quarter / dyadic) retarget off the rewritten k492 path onto the pairwise sw4 checker.

Tooling improvement; no new kappa. Selected frontier untouched.

Result

On Linux / CPython 3.12.3 with taskset -c 0, five alternating fresh-process pairs gave median complete local-audit wall time 7.080 → 3.131 s (2.26×). Timing is host-descriptive and is not a correctness gate. Complete receipts match the baseline checker (assertions=31, controls=25).

Package: research/emitted-local-integer-production/.
Tip: 3bc76f12b29732e901f13b19d341253379f04cb1.

Validation

python3.12 -I -B research/emitted-local-integer-production/verify.py
python3.12 -I -O -B research/emitted-local-integer-production/verify.py
python3.12 -B scripts/research_package.py --package research/emitted-local-integer-production
python3.12 -B scripts/research_package.py --package research/complex-signed-prefix-audit --check-only
python3.12 -B scripts/research_package.py --package research/selected1094-wire-prefix-audit --check-only
python3.12 -B -m unittest tests.test_emitted_local_integer_production
taskset -c 0 python3.12 -I -B research/emitted-local-integer-production/benchmark.py --rounds 5

Expected:

PASS_EMITTED_LOCAL_INTEGER_PRODUCTION assertions=31 controls=25 corruption=subprocess_failure certs=1 no_new_kappa
PASS_REGISTERED_RESEARCH_PACKAGE research/emitted-local-integer-production outcome=completed
PASS_PACKAGE_METADATA_AND_PINS
PASS_IDENTICAL_EMITTED_LOCAL_INTEGER_PRODUCTION_RECEIPTS

Default verify is read-only. Temp-copy ADD-drop foil rejects via subprocess failure. Package gates under -O still launch assertion-enabled children. Full make verify, selected cold replay, Lean, and multiplication-theorem obligations were not rerun for this isolated checker change.

Overlap and credit

Closest priors (different delivery): woklerx’s research/emitted-local-integer-replay-seed90/ / #1184 (opt-in integer rows; this is the production wiring), CrocSwap/integer-mult-bounds#756, CrocSwap/integer-mult-bounds#2329 / #2348 (same production-promotion + provenance pattern). No competing open PR on production k492 hpe_audit_local.py at refresh.

Credit woklerx, icekylinx, and the retained local-checker lineage. Prepared by Wentao Li (VictorLiwentao, @Is__Victor) with substantial AI assistance, under Apache-2.0.

[Mister-Meeseeks]

Maintainer review at 3bc76f12b29732e901f13b19d341253379f04cb1: the integer-row implementation and historical-baseline approach are promising, but the downstream migration is incomplete.

1. Two registered checks (complex-signed-prefix-audit and selected1094-wire-prefix-audit) change their source key without updating EXPECTED.json. complex-stage-diagonal-normalization-seed24/CERTIFICATE.json has the same issue in source_hashes. I reproduced the first receipt mismatch, made local repairs that only rename the key to the byte-identical complex-sw4-frame8 checker, and replayed the affected checks. The diagonal certificate and historical opt-in integer checker both pass after that correction. These local corrections do not change mathematical fields and will be retained in the review evidence.

2. Other live-bound consumers still use the replaced k492 path. In particular, I executed complex-ck-target-forest/verify.py:check_sources at this head: it rejects source pin: research/complex-k492-owner-repair/code/hpe_audit_local.py. Restoring only the original checker bytes makes that complete source check pass. This is a regression introduced by the production rewrite, not an already-stale selected-result pointer. replica-prefix-audit and supplier-circuit-consolidation also have live source-check loops over manifests containing this path, and the JP8 native replay helpers still execute it.

Please complete the consumer inventory and preserve authenticated historical execution for those consumers as well, using the retained byte-identical checker/baseline where appropriate. Update each affected source-key receipt through its generator and retain its semantic fields and rejection coverage. A source-hash demotion alone is not sufficient for executable dependencies.

The selected frontier remains unchanged. This review does not dispute the scalar-row equivalence or independently certify the reported host-specific speedup; it is holding production integration until downstream replay compatibility is complete.