⛔ RETIRED 2026-08-22 — THIS ENDPOINT NOW RETURNS 410 corpus_retired.

It retired WITH the corpus it measured. The court REGISTRY (293 courts) still exists, but a coverage table reporting zero opinions for every court would read as a fact about those courts rather than about us — which is the exact misreading this endpoint was built to prevent. Original description kept verbatim below.

─────────────────────────────────────────────────────────────

One row per REGISTERED court: how many opinions we hold, how many carry a reporter citation, and the earliest and latest filing dates. It is both the coverage table and the lookup the court= filter needs — a filter whose valid values are undiscoverable is not a filter, which is why this is open.

293 courts are registered and 146 hold opinions, so 147 rows come back with opinions: 0. They are not hidden: omitting them would overstate coverage. Read the date spans too — scotus shows 37,792 rows because CourtListener publishes orders and summary dispositions as clusters, and those thin out sharply after 2020.

THE COUNTS ARE MATERIALISED AND THE RESPONSE SAYS WHEN. Computed live the aggregate takes 2.8 s against 988,554 rows, which on a keyless endpoint would be a self-inflicted denial-of-service. So every response carries meta.counts_refreshed_at, and meta.counts_stale goes true once the rollup is more than 48 hours old (the refresh runs daily). The numbers are never presented as current — they are presented with the timestamp that says how current they are.