[cmccomb]

Summary

• Archive the exact pinned MATLAB HDF5 and both unchanged 30 s, 0.1 s PTO Extension body/PTO histories from successful reference run 38282150325. The archived time, active motion, and PTO columns are identical to the earlier published-step fixture.
• Gate independent public Python motion with physical implicit added mass against every inherited original threshold for both float and spar inputs: all six body position/velocity axes, PTO stroke/speed, and zero force/power. The run receives no MATLAB motion, force, or solver-step histories as inputs.
• Add an opt-in source-ode4 negative control and a focused R2025b workflow that regenerates both unchanged inputs on the PR head and compares the complete source histories and generated hydrodynamic data.

Local evidence

Both physical 301-sample runs pass the inherited limits. Largest active gaps are float 85.849 mm heave/stroke and 202.600 mm/s speed against 90 mm and 210 mm/s, and spar 14.877 mm and 6.522 mm/s against 18 mm and 8 mm/s. The float speed margin is 7.400 mm/s (3.52% of its limit). The source-ode4 control matches the source active heave within 7e-12 m, while the qualifying physical trajectory remains distinct.

python -m pytest -q tests/test_rm3_pto_extension_published_physical.py: 3 passed locally. The optional fresh-source checks also passed against the downloaded original MATLAB artifact; actionlint passed for the new workflow.

Interpretation

These are formal passes under inherited thresholds, not a claim that the coarse published MATLAB trajectories are source-converged or pointwise reproduced. The original limits preceded this source-step investigation but were not chosen prospectively blind to all earlier errors. A source-only FixedStep refinement still removes over 99% of the float and over 99.99% of the spar squared gap to a refined trace. The corrected fine-step baselines remain separate. This PR leaves all case status, badge counts, and issue states unchanged pending review and exact-head Linux CI.

Related: #624, #618.