Tooling: skip exact zero row updates in cyclic-inverse replay

[thakrarsagar]

The incorporated PR621 verifier performs exact cyclotomic arithmetic even when an elimination coefficient is zero. This source-pinned adapter inserts a two-line guard in memory and runs every original check. The eight-case receipt remains byte-identical. Historical source bytes and default verification stay unchanged.

Status: tooling improvement; no new kappa. Seven alternating full-process pairs on each of two local interpreters measured 24.6–26.2% lower child CPU medians. Wall medians also improved, but one pair was slower; all samples and the preliminary probe are disclosed. This is a scoped result for the existing eight cases, not a full-frontier speedup.

macOS arm64	Reference wall median	Guarded wall median	Reference CPU median	Guarded CPU median
Python 3.13.16	3.709 s	3.156 s	3.149 s	2.374 s
Python 3.14.8	4.205 s	2.649 s	3.212 s	2.371 s

Reproduce with Python 3.11+ and the standard library on macOS or Linux:

python3 -I -B research/cyclic-zero-elimination/verify.py --output /tmp/cyclic-zero-result.json

Use a fresh output path. Expected: PASS_CYCLIC_ZERO_ELIMINATION. The JSON retains every timing, source pin, platform, and control result. replay.py runs only the faster complete replay and accepts the predecessor's --output option.

Validation:

• All 28 measured full replays match the pinned EXPECTED.json byte for byte, including the original unit-root and missing-conjugation controls for every case.
• 36 focused matrix/mode checks cover sparse/dense, row swaps, singular systems, non-rational coefficients, both boundary modes, direct determinant formulas, solution substitution, and input preservation. Changed source bytes and two incorrect guards are rejected. A 2x2 identity solve drops from 12 to 7 field multiplications.
• Isolated optimized replays from /tmp also match on Python 3.11 and 3.14. A separate Codex agent reviewed the guard and ran focused controls under optimized Python.
• make selected-record-check and git diff --check pass. The full repository suite, cold frontier replay, and optional SymPy engine were not rerun for this optional adapter.

Base: 66849e1475c930fd7732b1a0067ffee7e1e95f60. Predecessor: Ali Romman's incorporated #621 at 520774462b19b29af41554b74def31f9d01e997e; exact source and receipt hashes are in SOURCE.json. Scott Woolsey Biggart's pending #668 independently extends resultant coverage through radix 27 using CRT. This work keeps the predecessor's domain and rational-field method and has no dependency on that PR. The PR queue was refreshed through #732; #727's owner-footprint optimization is a separate checker.

All additions are confined to research/cyclic-zero-elimination/. Seed 94, local claim, explanation of the guard, raw evidence, and inherited limits are included. No pinned sprint issue was available. Contributor: thakrarsagar, with substantial OpenAI Codex assistance and a separate AI code review. Apache-2.0; predecessor attribution is retained. This does not prove the full analytic or multiplication theorem, or resolve paid preparation, rounding/restoration, or fixed-tape realization.

[Mister-Meeseeks]

100-PR batch review at eccc3b74767848eb6746f39412d57df6c9544b5d.

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

Reviewed scope: Opt-in zero-elimination adapter; exact finite predecessor receipt retained. Runtime evidence is local, not a new theorem or production default.

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.