Backend Retry, Deadlines, and Write Safety

Backend Retry, Deadlines, and Write Safety A workflow diagram generated by Archify. 01 / Selected backend 02 / Shared deadline + evidence Enter and select Classify + budget Dispatch + settle Operation · task · epoch · Selected backend › Enter and select Operation task · epoch Dispatch · remaining budget · Selected backend › Dispatch + settle Dispatch remaining budget Backend Policy · Web registry · Android · Selected backend › Enter and select Backend Policy Web registry · Android Bounded Retry · grant + deadline · Selected backend › Classify + budget Bounded Retry grant + deadline Operation Gate · earlier deadline · journal · Shared deadline + evidence › Classify + budget Operation Gate earlier deadline · journal Typed Outcome · state + recovery · Shared deadline + evidence › Dispatch + settle Typed Outcome state + recovery grant + budget bind owner aggregate Legend Agent logic Policy Context / trace Cloud service External system

Classification is a ceiling

  • • Web uses IdempotencyRegistry; Android uses the total retry manifest
  • • Safe reads replay only with remaining budget; unknown raw methods run once

Mutation replay needs proof

  • • Non-idempotent writes disable blind transport replay
  • • Explicit owners may replay NOT_SENT or REJECTED; UNKNOWN requires inspection

Cancellation stays honest

  • • The aggregate deadline can shorten a phase but never extend it
  • • Caller cancellation stays CancelledError while ordered journal evidence remains