Ordered Retries in Kafka: Why The Retry Topic Is Breaking Downstream

I was on-call when the P1 hit. Slack lighting up, a downstream team's database rejecting writes. Constraint violations everywhere. I pulled up the logs. The events looked fine. The schema was correct.