Measured, before it changes

Changes since 2.24.0.

#New

• What each dataset costs to serve. Every read of a statistics table from its publisher is recorded: why it was read (a map, a download, a sample, a placement check or a harvest), whether it came from the cache, how many rows and bytes it brought and how long it took. Staff see the totals and the busiest datasets for the last 30 days.
• A test set for every statistics table. Each table in the catalogue is listed with a snapshot of how it is placed today and the problems found in it (drawn on the country outline although its rows name finer areas, a total shown as a place, no unit, error pages read as tables, duplicates). Nothing changes for users: every table keeps its current version until a re-processed version passes the new checks.
• A reference set and scorer. Hand-checked answers for statistics tables from ten countries, taken from each publisher's own metadata, score how well tables are read and placed. Later changes to processing are measured against it and may not lower any score.