How would you answer an interview scenario involving message retries and dead-letter queues?

In an interview, I would first define message retries and dead-letter queues and the problem it solves, then explain how I would use bounded retries with backoff, distinguish transient from permanent failures, preserve diagnostic context, monitor dead-letter volume, and provide a controlled replay process. I would also call out the main failure mode: retrying forever without backoff can overload dependencies and repeatedly execute a permanently invalid message. Finally, I would describe how I would test, monitor, and safely roll back or recover the solution.