n observations and root samples by dataset, by dataset x station x year, by dataset x year and by
dataset x measurement type (with year and depth spans, and — when the measurement_type table
carries them — the registry's category and variable); the per-station year x month detail is
build_coverage_stations(), a second sidecar fetched on demand — small enough to paint the grid before
DuckDB-WASM wakes up, and the variable-based inventory Task 14 asks for. Since 3.25.0 also taxa[]
(explorer UI plan D14): one row per taxon of the bio realm — key, names, rank, class, n_obs, year
span, life stages and its datasets with n_obs each — so the organism list opens before the engine is
warm and Browse can list organisms by category or dataset. Deterministic: no wall clock, so a
re-run over unchanged inputs writes identical bytes.