Round 2 verification audit for C028

Model: gpt-5

Claim: Fig. 3b: comparison of steady-state errors for uncorrelated (solid) versus correlated (dashed) errors, showing correlations increase fidelity for similar constituent operators.

Source alignment: source/main.tex:139-145 (Fig. 3b)

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: Reproduce the correlated-vs-uncorrelated steady-state error curves and confirm the ordering shown in Fig. 3b.. 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 ['visual_match_only', 'paper_text_only_reimplementation']. Notes: Reproduced uncorrelated (solid) vs correlated (dashed) steady-state error curves. Correlated bit-flip ~3.07p (paper ~3.2p) < uncorrelated bit-flip ~6.03p (paper ~5.2p): confirms correlations increase fidelity for similar constituent operators (the claim). Correlated phase-flip ~0 (DFS). Uncorrelated bit-flip coeff ~16% above paper due to channel parametrisation, but the qualitative ordering claimed is reproduced. Theory-reproducible plot; Tier-A visual match only -> caps at partial.