[cmccomb]

Change

Record a full-duration physical-step diagnostic for the published OSWEC irregular passive-yaw case. The ledger and case-status row now state that both MATLAB and Python separate after heading events. Add a measured JSON record and a yaw/heading-event plot. The case remains mixed_modes; badges and solver defaults are unchanged.

Evidence

• The pinned 600 s MATLAB source trajectory matches the archived 0.01 s source prefix exactly through 250 s. A fresh local R2025b 0.005 s replay reproduces the archived 160 s half-step yaw within 1.1e-15 rad with identical heading choices.
• Independent physical WEC.run at 0.0025 s passes the unchanged yaw, speed, torque, and work limits through 160 s, then fails at 250 and 600 s. Full-run yaw and PTO-work gaps against published MATLAB are 0.1154 rad and 12.55%.
• Independently halving Python's step to 0.00125 s passes the same limits through 160 s, but its own 600 s yaw and PTO-work changes are 0.07350 rad and 6.15%. Both implementations are step-sensitive on the full event-driven path.
• The first 0.0006 rad yaw-limit breaches occur at 142.93 s for MATLAB 0.01→0.005, 166.98 s for Python 0.0025 versus published MATLAB, and 167.73 s for Python 0.0025→0.00125.

The inputs, hashes, limits, and window results are in docs/evidence/oswec-variable-yaw-refinement.json; the plot is in docs/figures/oswec-variable-yaw-step-gap.png. The separate matched MATLAB 0.005→0.0025 s source audit is still running and will be evaluated independently. This PR makes no full-duration physical parity claim.

Verification

• python3 tools/render_parity_badge.py --check
• 63 CSV cases and unchanged evidence-status counts verified
• First-breach times recomputed from saved trajectories
• git diff --check; plot visually inspected