Round 2 verification audit for C045

Model: gpt-5

Claim: In the alternating-drive scheme the protocol converges optimally at gamma^opt ~= 0.23 pi with rate 1/N_0^opt = -(1/2) log(lambda_+^opt) ~= 1/7.92 per cycle; for the experimentally measured gamma ~= 0.31 pi the rate is 1/10.35 per cycle.

Source alignment: source/supp_content.tex:56-56 (Supp. Mat. S1)

Prior official verdict: partial with failure_reason mismatch.

Executable evidence: run.py. Sandbox rerun logs: run.log.

Independent audit: I checked the code against supp_content.tex:56. It uses the paper closed form for lambda_+ and cross-checks it against the full two-cycle superoperator. The optimum matches. The only discrepancy is the measured-gamma rate, and a direct sensitivity check shows gamma=0.305pi nearly reproduces the printed 1/10.35. I therefore remove failure_reason=mismatch and record rounded_input_sensitivity.

Decision: Round 2 verdict is partial with failure_reason None and limitations ['paper_text_only_reimplementation', 'rounded_input_sensitivity']. Notes: Round-2 audit preserves partial but removes the hard mismatch flag. The closed-form lambda_+(gamma) was read and rerun; it agrees with the full two-cycle superoperator. The optimum is gamma=0.2310pi and N0=7.9165, matching the paper. At the literal rounded gamma=0.31pi the rate is 1/10.6855, but gamma=0.305pi gives N0=10.3096, consistent with the printed 1/10.35. Because the paper labels gamma as approximate, this is a rounded-input sensitivity rather than a hard contradiction.