API timeouts are not always failures; sometimes the remote system completed the request and only the response was lost. In n8n, I like to model this explicitly:
- assign an idempotency key before the API call
- persist the request state as pending
- retry only when the endpoint is safe to retry
- reconcile ambiguous results with a status lookup
- mark the final state as succeeded, failed, or needs review
This keeps retries from creating duplicate records and gives operators a clear recovery path. What patterns are you using for ambiguous API responses?