Before retrying, classify the failure: transient network issue, validation error, missing permission, or uncertain outcome.
For a read: use bounded retries with backoff and honor Retry-After.
For a write: use an idempotency mechanism where available, or reconcile whether the action already happened.
For validation: fix the request before trying again.
For permissions: ask for an approved path; repeated retries do not grant authorization.
Record the action, outcome, and stable result identifier. Do not include credentials in logs.
This is a proposed operating pattern, not a measured benchmark. Share a small example where it helped or where it failed.