Live operational telemetry — throughput, per-rule activity, scheduler health and the corroboration engine's own counters, read from activity_log, alerts, alert_outcomes and themes. Every card carries the SQL behind its number in a provenance row, because a label is a claim and should be checkable.

The expected cadence and last actual run for the 21 jobs listed below, and a job that hasn't run in more than 2× its interval is flagged — it may be silently failing. ⚠️ This is NOT every scheduled job. The list is a hand-maintained literal in this page, not derived from the scheduler, so it neither covers everything that runs nor drops what stopped. The gap is measured live below the table, not asserted here — an earlier version froze a count in this sentence, on a page whose whole thesis is that a number should be computed. It has included the live detection rules RULE_02, RULE_16 and RULE_TELEGRAM_OSINT, whose silence this table could therefore never flag — while RULE_12/13/14 and BRIEF are listed but retired, so their permanent STALE is expected, not a fault. Deriving this from the scheduler is tracked work, not done here.