Found (pre-existing) during the #365 landing gates (PR #369, 2026-07-16): tests/parallel/test_1017_custom_mg_parallel_mpi.py::test_parallel_custom_fmg_stokes_constrained fails at np4 with rel-L2 error 0.00562 against the 0.005 tolerance — identical on unmerged development, so it is not caused by the landing. Passes at np2.
Either the tolerance is marginally too tight for the np4 partition (a 12% overshoot smells like partition-dependent iteration count rather than a wrong answer), or there is a genuine small np4-specific accuracy loss in the custom-FMG constrained path. Related territory: #291 / FO-06 (Stokes_Constrained 0.4% 3D velocity, release-blocking list).
Underworld development team with AI support from Claude Code
Found (pre-existing) during the #365 landing gates (PR #369, 2026-07-16):
tests/parallel/test_1017_custom_mg_parallel_mpi.py::test_parallel_custom_fmg_stokes_constrainedfails at np4 with rel-L2 error 0.00562 against the 0.005 tolerance — identical on unmerged development, so it is not caused by the landing. Passes at np2.Either the tolerance is marginally too tight for the np4 partition (a 12% overshoot smells like partition-dependent iteration count rather than a wrong answer), or there is a genuine small np4-specific accuracy loss in the custom-FMG constrained path. Related territory: #291 / FO-06 (Stokes_Constrained 0.4% 3D velocity, release-blocking list).
Underworld development team with AI support from Claude Code