Round 2 verification audit for C024

Model: gpt-5

Claim: The final fidelity is independent of all global errors (e.g. X_e X_e, X_e Z_e), while all local errors (e.g. X_e I_e, I_e Z_e) become amplified.

Source alignment: source/main.tex:138-145 (Theoretical error analysis)

Prior official verdict: verified 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: Simulate steady-state error for each of the 16 channels and confirm global (both-ion) operators leave fidelity unchanged while single-ion operators degrade it.. 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 verified with failure_reason None and limitations []. Notes: Paper (main.tex l.145): final fidelity independent of all global errors (e.g. X_eX_e, X_eZ_e); all local errors (e.g. X_eI_e, I_eZ_e) amplified. Reproduced in the discrete cycle by inserting each of the 16 M_{i,j}=exp(-i(pi/2)sigma_i(x)sigma_j) channels (sigma_0 = FULL identity, per the paper) once per cycle at prob p and taking the power-iteration steady state. RESULT: all 9 global channels give d(1-F)/dp < 4e-11 (fidelity unaffected); all 6 local channels are amplified (bit/Y local slope 5.60, phase local slope 4.35). Dichotomy CONFIRMED exactly as claimed. [Round-1 correction: the original run used a buggy shared the self-contained model where PAULI_DE['I'] was the {down,e}-restricted projector (zeroing |up>), which broke the global/local split and gave a false 'mismatch'. The self-contained code was fixed (sigma_0 = full identity; M_error() helper; power-iteration steady_state) and this claim re-run.]