Validation and science notes

Model completion, numerical convergence, and scientific validation are different claims. Keep them distinct in status pages and in any new results summary.

Three levels of evidence

Execution: the driver and solver finished and wrote an output.
Numerical behavior: the solve converged under its configured criteria, and the saved checkpoint contains the expected fields and final state.
Scientific validation: diagnostics test the intended physical quantity against observations, with suitable spatial coverage, area weighting, and time sampling.

A stage can satisfy one level without satisfying the next. Record what was checked and which output it applies to.

Historical-run metrics

The worklogs caution that a single spatial RMSE against a final-year observation slice does not establish temporal skill or reveal a domain-wide bias. For historical evolution, prefer the documented area-weighted, fixed-coverage, per-basin time series and state the comparison period and observational product. The scripts under config/check_* provide examples of the current checks; see the run and pipeline worklogs for which check applies to each result.

Boundary conditions and inversion choices

Subdomain extraction can introduce new boundaries and therefore change the boundary conditions relative to the full-continent solve. This is especially relevant around the floating calving front and grounding line. The root boundary-condition notes and inversion worklog explain the current choices and evidence.

Likewise, friction law, effective-pressure source, mesh/geometry treatment, cost-function weights, and checkpoint provenance are part of the result—not incidental settings. Link reported diagnostics to the exact model file and configuration used.

Evidence and status records

Inversion worklog: present-day friction/rheology investigation, including corrections and superseded experiments.
Post-inversion pipeline worklog: HO and SSA downstream stages and status history.
Projection experiments: forcing design and projection experiment matrix.
Gadi run guide: stage inputs/outputs and operational caveats.

These records are detailed and date-sensitive. The summary pages here should remain high-level, and should not silently flatten competing results into a single “best” number.