List Stats

Every stat, with which boundary levels it may honestly colour.

compatible_levels is computed from geography_relations -- the same two branches geo_match() (databeat/geography.py) uses to decide whether ONE series may honestly answer for ONE requested area, replicated faithfully rather than reimplemented: exact (a stat's series geography IS a geography at this level) or aggregate (a stat's series geography directly contains a geography at this level, one hop only -- geo_match never walks a second hop; that's visible in its SQL, a single non-recursive join, not stated in its module docstring, which is about the geography-identifier model instead). See stat-region-selector-design.md section 2 for the full derivation.

CORRECTED 2026-09-26 (wl:be788e133ac1): this docstring used to say the whole function was "computed live rather than cached... row counts make it a non-issue". That was true for a SINGLE call in isolation, but never measured against the real call pattern -- profiled directly at ~1.2s per call, and a single real page render calls it 4 times (see the cache above). Now cached per country, 60s TTL.

cluster / cluster_label / cluster_sort_order are the product's five subject areas (migration 014), served from here so the map's "Colour by" chips, the panel's tabs, the card ordering and the alert-strip quota all read ONE source of truth rather than four approximations of it. Until this landed the taxonomy existed only in documents, and every one of those surfaces fell back to insertion order or to raw series count -- see unified-design-and-data-plan.md section 2.

A level absent from compatible_levels is not a bug to investigate -- it is the same honest "cannot be shown here" geo_match already enforces per-request for the panel and deep dive, computed once for the whole stat instead of once per click. Restricted to KNOWN_BOUNDARY_LEVELS: a stat's series is always trivially "exact" compatible with its OWN geography's level (crime with garda_division, RPPI's regional facets with cso_rppi_region, any nationally-published series with nuts3), which is a correct geo_match verdict but not a level the front end has a button for -- surfacing it would be noise, not signal.