[cmccomb]

Change

Expose the existing opt-in RM3 End Stops wecsim_ode45 solver through public WEC.run for regular-wave floating joints. The physical implicit default is unchanged. The lower-level solver still requires hard stops and delayed added mass.

Independent MATLAB comparison

A new end-to-end public-builder test uses only the pinned RM3 HDF5, regular wave, and configured PTO as inputs. It compares the independently advanced 120 s result against the archived published R2025b fixture on both bodies' active positions and velocities, PTO stroke/speed/force, and ordinary damper work under the existing 120 s gates. The public run observed 5.12e-5 m peak stroke gap, 2.08e-4 m/s speed gap, 2.50 kN force gap, and 5.14e-7 relative work gap. The pinned fixture is not fed into the solver.

A local full-duration check on this head confirms the limit: the independent solver omits MATLAB accepted steps at 120.6836 and 120.7450 s, and the existing full-run stroke, speed, and force limits first fail at 121.1, 121.0, and 123.8 s. Peak 400 s gaps are 0.277 m stroke, 1.179 m/s speed, and 8.36 MN force. A diagnostic that supplies only those two MATLAB internal times reduces the 130 s stroke gap from 0.0591 to 0.00418 m, speed gap from 0.1279 to 0.0147 m/s, and force gap from 1.514 to 0.251 MN. The 400 s trajectory still diverges. This isolates solver scheduling at the onset but supplies source timing and is not an independent full published-case gate. A fresh local R2025b 130 s state-logging run reproduced the archived PTO stroke and speed exactly and logged the same 1,325 accepted times as the existing full-run fixture. It has eight active continuous joint states, matching the Python source-state map; the 24 logged converter states remain zero. With only those accepted times prescribed, Python advances its own state and forces through 130 s within 6.52e-8 m PTO stroke and 3.65e-7 m/s PTO speed of MATLAB. This confirms close force-law and RK-stage behavior under the source schedule; selecting that schedule independently remains the blocker. A local accepted-step trace pinpoints the first step-count difference to 120.6–120.7 s: MATLAB accepts an extra 120.68355556 s step, whereas Python advances directly to 120.7 s with an embedded error ratio of 0.001319. The missing step follows prior step-size proposals rather than rejection of the direct Python step at 120.6 s. A second local source run setting scalar AbsTol=1e-6 instead of auto produced bit-identical accepted times and states through 130 s. Replacing Python's history-scaled error norm with a current-state absolute-plus-relative norm instead lost the passing early trajectory, so no tolerance change is proposed here.

End_Stops remains qualified; strict case badges are unchanged.

Local verification

• 69 passed across Python API, physical boundaries, End Stops source timing, and standalone End Stops tests.
• tools/render_parity_badge.py --check and git diff --check pass.