- Alert fires (new rows in core.dead_letters). Triage by queue + last_error grouping in Softcrum Ops.
- Poison (schema-invalid): fix producer, mark discarded with reason.
- Transient exhausted (provider outage): bulk replay from Ops (re-publishes with SAME job_id — processed_jobs guarantees no double effects).
- Logic bug: fix + test, replay, verify downstream effects via correlation_id trace.
- Weekly: dead_letters must trend to zero; standing categories become engineering tickets.
Runbook — Dead Letter Review & Replay
- Alert fires (new rows in core.dead_letters). Triage by queue + last_error grouping in Softcrum Ops. 2. Poison (schema-invalid): fix producer, mark discarded with reason. 3.
⌘I