Prove a migrated or newly-created alert actually reaches a human BEFORE deleting the
alarm it replaces. Use when: (1) porting CloudWatch alarms to Grafana/Prometheus rules,
(2) about to delete a monitor because "the replacement is live", (3) routing new rules
to an SNS topic or webhook an existing alarm already uses, (4) a migration PR claims
"routing parity" from matching rule counts or matching alarm_actions counts, (5) an
alert has never fired and you are treating that as healthy. Catches the failure where
delivery succeeds, every count matches, and no incident is ever created.
