Routing and fallbacks
Distinguish destination selection from retries and fallback behavior.
Working scaffold: This page describes general routing concepts. It does not claim that ONE provides automatic routing, provider selection, or fallback behavior.
Routing
Routing is the policy that determines which eligible destination handles a request. A policy might consider an explicit selection, capability requirements, availability, latency, or cost. The actual selection inputs and priority rules are product-specific and must be documented precisely.
Fallbacks and retries
A fallback tries another eligible option after a defined failure. A retry repeats an attempt; a fallback may change the destination. Document the conditions that trigger each behavior, the maximum attempts, response semantics, and how the final destination is reported.
Questions a verified guide should answer
- Which failures are eligible for another attempt?
- Can a request be billed more than once when an attempt is retried?
- Are timeouts, validation errors, or user cancellations retried?
- Can applications set an ordered list or routing policy?
- How can an application identify which destination completed the request?
Retries can duplicate side effects in tools or external systems. Explain idempotency and application-level safeguards before recommending automatic retries for non-read-only actions.