Tooling: 1.54× faster production data_corners.cpp via cached update columns with identical receipts

[VictorLiwentao]

Change and scope

At base 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2, production research/copied-fixed/data_corners.cpp still paid a fresh update-column rebuild per pivot. The opt-in adapter research/data-corner-performance/ already proved a per-pivot cached-column loop with identical 4,073,300-pair receipts. This PR wires that equivalent loop into the production source.

Only the local_sha256 pin for research/copied-fixed/data_corners.cpp changes inside research/copied-fixed/certificate.json. The opt-in package vendors pairwise-era bytes as baseline.cpp and records the pre-production digest under provenance_sha256, so research-packages CI does not fail working-source drift. Historical inventories without VERIFY matrices are left as provenance.

Tooling improvement; no new kappa. Selected frontier untouched. Does not edit the e8-bit-r420 MANIFEST tree (#2263/#2276).

Result

On Linux / CPython 3.12.3 with taskset -c 0, five alternating fresh-process pairs gave median wall time 170.101 → 110.463 s (1.54×). Timing is host-descriptive and is not a correctness gate. Complete receipts match the baseline checker (4,073,300 pairs; 4,073,290 good / 10 fallback).

Package: research/data-corner-production/.
Tip: 2c7dfd60f7a45d7c6b433afb9d73346f13c59832.

Validation

python3.12 -I -B research/data-corner-production/verify.py
python3.12 -I -O -B research/data-corner-production/verify.py
python3.12 -B scripts/research_package.py --package research/data-corner-production
taskset -c 0 python3.12 -I -B research/data-corner-production/benchmark.py --rounds 5

Expected:

PASS_DATA_CORNER_PRODUCTION pairs=4073300 good=4073290 fallback=10 corruption=subprocess_failure certs=1 no_new_kappa
PASS_REGISTERED_RESEARCH_PACKAGE research/data-corner-production outcome=completed
PASS_IDENTICAL_DATA_CORNER_PRODUCTION_RECEIPTS

Default verify is read-only. Temp-copy column-fill foil rejects via subprocess failure. Package gates under -O still launch asserting C++ children (NDEBUG rejected). 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): research/data-corner-performance/ (opt-in cached columns; this is the production wiring), CrocSwap/integer-mult-bounds#756, CrocSwap/integer-mult-bounds#2257 / #2263 / #2276 (same production-promotion + provenance pattern). No competing open PR on production data_corners.cpp at refresh.

Credit the authors of research/data-corner-performance/ and the copied-fixed lineage. Prepared by Wentao Li (VictorLiwentao, @Is__Victor) with substantial AI assistance, under Apache-2.0.