Starting Nov 6 around 20:00 UTC, an internal database our check scheduler relies on stopped updating after its disk filled up. As a result, some monitors failed to record recoveries — appearing degraded or down even though checks were passing — and sent repeated degraded/recovered notifications. Monitors created during this window also weren't checked. We expanded storage and rebuilt the database on Nov 8 at 12:55 UTC; all monitors have recovered and we're adding safeguards to prevent this from happening silently again.