Round 2 verification audit for C017

Model: gpt-5

Claim: The drive (C) angle is alternated: theta_1 = pi in odd cycles (t_C = 6.4 us) and theta = pi/2 in even cycles (t_C = 3.2 us), acting as a spin-echo; one cycle takes ~= 165 us on average.

Source alignment: source/main.tex:122-122 (Experiment)

Prior official verdict: partial with failure_reason None.

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

Independent audit: I scanned the copied script for imports/shared helper dependencies and reran it through the sandbox. The code is self-contained in this attempt directory and targets the claim strategy: Check t_C scales linearly with theta (pi -> 6.4 us, pi/2 -> 3.2 us) and that the per-cycle time is consistent with the drive (A) and drive (C) durations.. I checked the relevant family model rather than relying only on exit status; the rerun is treated as one reproducibility input.

Decision: Round 2 verdict is partial with failure_reason None and limitations ['paper_text_only_reimplementation']. Notes: Paper: theta_1=pi (t_C=6.4us), theta=pi/2 (t_C=3.2us), cycle ~165us avg. With Omega_C=2pi*78kHz, t_C=theta/Omega_C gives 6.41us and 3.21us (linear in theta, ratio exactly 2) — matches to 0.2%. Per-cycle reconstruction = drive A(150us)+gapA(3.2)+repump B(5us)+drive-C avg(4.8us) = 163.0us vs paper 165us (~1%). The ~165us requires summing sequence components from paper text -> partial.