Study state is typed
A study is a typed set of related runs: parameter sweep, mesh refinement, time-step independence, design comparison, or validation benchmark. The StudySpec records variants, controlled variables, QoIs, stop criteria, and aggregation method. Without it, a 12-case sweep is just a folder of one-offs.
Postprocess recipes are repeatable
Recipes bind fields, patches or named selections, extraction methods, plots, figures, and output files. PyVista and viewer sessions are renderers over typed artifacts — they don't replace the recipe. Re-running a study regenerates the figures from the same recipe.
Compare by typed artifacts
Compare views line up variants by run identity, mesh, solver settings, QoI definition, and artifact lineage. Two visually similar figures are not comparable unless the extraction recipe and units match — and the platform will tell you so.
Mesh-convergence series
ADR 0021 freezes the MeshConvergenceSeries record and its applicability check. A GCI or Richardson extrapolation result must cite three monotonically refined meshes with matching QoI definitions — the platform refuses to publish a single-mesh "converged" claim.
Dashboards
Auto-dashboards and database search present QoI tables, residual summaries, and run metadata as queryable projections over durable records — not as manually curated screenshots that age out the moment the case re-runs.