Tooling: 3.12× faster inplace-side prefix replay with identical receipts

[VictorLiwentao]

Change and scope

At base 0b1cccf86e2f5132ed286a26ef0a681a495740ea, the historical inplace-side prefix checker (scripts/experiments/inplace_side.py:verify_prefix) rejects a specified n-role Gauss–Jordan side family using three pinned witnesses (h = 42, 46, 50). Each elimination column previously rebuilt sorted(unused) to locate the least unused row with the pivot bit set.

This PR replaces that search with a least-unused flag scan and column masks. The lexicographic pivot rule, XOR updates, switch witnesses, deficit arithmetic and every receipt field are unchanged. Only the source pin for scripts/experiments/inplace_side.py changes inside certificates/inplace-side.json.

Tooling improvement; no new kappa. The selected frontier, upstream/, and the mathematical rejection in docs/research/inplace-side.md are not reopened.

Result

On Linux / CPython 3.12.3, five alternating fresh-process pairs of the three pinned witnesses gave median 17.061 s → 5.466 s (3.12×). Timing is host-descriptive and is not a correctness gate. Receipts match the byte-identical baseline checker fieldwise; a greatest-unused mutant is rejected on all three witnesses.

Package: research/inplace-side-performance/.

Validation

Python 3.12, standard library only:

python3.12 -I -B research/inplace-side-performance/verify.py
python3.12 -I -O -B research/inplace-side-performance/verify.py
python3.12 -B scripts/research_package.py --package research/inplace-side-performance
python3.12 -I -B research/inplace-side-performance/benchmark.py --rounds 5

Expected:

PASS_INPLACE_SIDE_PREFIX_FAST witnesses=3 h=[42, 46, 50] paths=[151, 415, 767] ... no_new_kappa
PASS_REGISTERED_RESEARCH_PACKAGE research/inplace-side-performance outcome=completed
PASS_IDENTICAL_INPLACE_PREFIX_RECEIPTS

Default verify is read-only. Use --write-certificate only after editing the production source. 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 objects): CrocSwap/integer-mult-bounds#756 packed balanced-Fano replay; CrocSwap/integer-mult-bounds#1584 finite-bit block-core frame interning; CrocSwap/integer-mult-bounds#1076 / CrocSwap/integer-mult-bounds#1075 quadratic tooling. No competing open PR on this inplace-side object at refresh.

Douglas R. Colkitt authored the inplace-side rejection, witnesses and original checker. Prepared by Wentao Li (VictorLiwentao, @Is__Victor) with substantial AI assistance, under Apache-2.0.

[Mister-Meeseeks]

Batch review at 7c16aea5074ae673e58bc38c8d99a948e3937853.

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

Reviewed scope: Ascending pivot scan replaces repeated unused-column sorting without changing the deterministic side routine. Exact predecessor outputs and certificate regeneration agree; combined current-main tests pass. Author speedup ratio is not independently timed.

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.