Found while re-validating Crameri Case-2 against #560's fix (PR #561).
The free-surface path records the length scale (hscale, used to report topography in metres) but not the corresponding time scale. A published benchmark checkpoint of the form "≈398 m at 3 Ma" can therefore only be checked as a trajectory — does the amplitude pass through the right value in the right order — not at matched physical time.
That was sufficient to establish what #561 needed (the trajectory is unchanged to ten significant figures before and after the fix, so agreement with the benchmark is unchanged). But it is not sufficient to state "we reproduce Crameri Case-2 at 3 Ma to X%", which is the claim a paper or a validation table would want to make.
Fix: record the time scale alongside hscale where the model's nondimensionalisation is captured, and expose it on the same object, so a benchmark comparison can convert both axes. Worth checking at the same time whether other validated cases (Y20 spherical relaxation, the analytic suite's time-dependent members) have the same gap.
Underworld development team with AI support from Claude Code
Found while re-validating Crameri Case-2 against #560's fix (PR #561).
The free-surface path records the length scale (
hscale, used to report topography in metres) but not the corresponding time scale. A published benchmark checkpoint of the form "≈398 m at 3 Ma" can therefore only be checked as a trajectory — does the amplitude pass through the right value in the right order — not at matched physical time.That was sufficient to establish what #561 needed (the trajectory is unchanged to ten significant figures before and after the fix, so agreement with the benchmark is unchanged). But it is not sufficient to state "we reproduce Crameri Case-2 at 3 Ma to X%", which is the claim a paper or a validation table would want to make.
Fix: record the time scale alongside
hscalewhere the model's nondimensionalisation is captured, and expose it on the same object, so a benchmark comparison can convert both axes. Worth checking at the same time whether other validated cases (Y20 spherical relaxation, the analytic suite's time-dependent members) have the same gap.Underworld development team with AI support from Claude Code