[cmccomb]

Summary

• Keep the unmodified published OWC/OrificeModel physical-motion comparison explicit as a strict expected failure. It uses the existing independently advanced 130 s Python trajectory; the module-scoped fixture prevents a second expensive solve.
• Apply the same full-duration physical limits declared before the corrected-source runs: 20 mm surge, 10 mm heave, 1 mrad pitch, 2 mm flexible displacement, and 1% PTO work. A dedicated PublishedPhysicalMotionGap exception is the only accepted expected-failure outcome, so unrelated setup or source errors still fail CI and an eventual pass becomes a strict XPASS.
• Preserve the corrected-model physical gate and the opt-in published-source force-route gate. The existing same-step MATLAB source control isolates the one changed piston-force connection: original versus corrected heave differs by 1.948 m and flexible displacement by 0.281 m.

The unchanged published source and physical Python share all 26,001 output times, wave samples, 500 PM phases, the pinned HDF5, and orifice parameters. Measured Python/original gaps are 0.1735 m surge, 1.9420 m heave, 0.01071 rad pitch, 0.2798 m flexible displacement, and 99.73% integrated PTO work. This remains a diagnostic of the published model's omitted piston reaction; it does not promote the original input or change any badge or case classification.

Verification

• Local: python -m pytest -q -rx tests/test_owc_orifice_corrected_full.py tests/test_owc_orifice_corrected_piston.py tests/test_owc_orifice_130s_portable.py — 3 passed, 1 expected failure.
• Exact-head Python CI and corrected-source CI: passed (7/7 checks). The fresh-source gate reports 1 passed, 1 expected failure.

Related to #622. This PR leaves that issue closed and does not change #624.