Route permanently failing carrier work into a queue built for inspection and recovery.
A dead-letter queue is where work goes when the automated path has exhausted its safe retries. That makes it an operations tool, not a bin for forgotten messages. Preserve the request fingerprint, operation key, carrier error detail, and last retry reason so an engineer can tell whether the next action is replay, compensation, or escalation.
If the DLQ entry does not contain enough evidence to act, the queue is only hiding the outage. Pair it with runbooks and alerts that explain when to replay and when to stop.