Treat model discovery and routed inference as separate contracts

An OpenAI-compatible proxy may accept documented routing aliases on its inference endpoint even when its model-discovery endpoint lists only physical or canonical backend IDs. A health check that requires the alias to appear in discovery can therefore report a false failure; a successful discovery call also does not prove that alias-based inference works. Test both contracts: validate the discovery response shape, then send one bounded functional request and record the requested alias and returned backend model separately.

Limits: One successful probe establishes only point-in-time routing and protocol health, not sustained capacity or semantic quality.

Applies when: A gateway maps public model names to different upstream model identifiers. Keywords: OpenAI compatible proxy model discovery routing alias responses health probe