[cmccomb]

Change

Pin the exact HDF5 and a compact 225 s MATLAB R2025b split-step trace from run 39153812298. The source HDF5 differs at floating-point precision from the previously bundled WaveStar HDF5, so the new portable test uses the artifact's exact file. Its 500 wave components match the published baseline component bank exactly.

The test independently advances the current physical WaveStar plant at 1 ms with the 50 ms sampled NMPC and source rate-boundary delay. It preserves the existing seven <0.5% aggregate limits for full-duration motion, force, command, and PTO work. The largest local relative error is 0.2392% (requested-torque RMS). The strict pointwise autonomous gate still fails: pitch differs by 0.007289 rad against 0.0005 rad, and speed, stroke, force, and requested torque also exceed their unchanged limits. WaveStar remains qualified; no solver default, pointwise gate, or badge count changes.

Verification

• New portable full-duration test: 1 passed locally on current master.
• Existing full split-step comparison with the exact source HDF5: aggregate 1 passed, pointwise 1 xfailed as expected.
• Existing prescribed-command plant gate: 1 passed; maximum pitch gap 0.0002427 rad, speed gap 0.001438 rad/s, stroke gap 0.0484 mm, and force gap 0.00423 N. Source torque requests are supplied for this diagnostic.
• Source timing/actuator test, parity-badge check, and git diff --check pass.

Pointwise gap localization

A separate local rerun using the exact pinned HDF5 and phase bank found the first breach of the unchanged pitch gate at 17.55 s and pitch-speed gate at 17.80 s, shortly after NMPC activates at 15 s. The command gate first fails at 24.15 s and axial-force gate at 43.80 s. Before control starts, the maximum pitch gap is 4.29e-5 rad over 10–15 s. The mismatch grows under closed-loop feedback, rather than appearing as a late isolated failure. This diagnostic does not establish pointwise parity.