ReasoningObserver

Sees what one model call REASONED, the moment its response is decoded. A reasoning model answers on two channels: the {@see ChannelObserver} request is what the agent was GIVEN, the answer's `content`/`tool_calls` are what the agent DID, and this is a THIRD fact neither carries — the private deliberation the provider exposes as a separate `reasoning_content`, distinct from the reply it precedes. It is not cost ({@see ReturnObserver}) and it is not the reply; folding it into either would make that seam report two facts as one. So it gets its own seam. Like the other return-side seams this is a SEPARATE, OPTIONAL interface: the gateway reports reasoning only to an observer that opts in (`instanceof ReasoningObserver`), and only when the provider actually spoke it — a model that reasons silently produces no call here, never an empty one. Implementations MUST NOT throw and MUST NOT be slow: observing a channel may not change it.

abstract public function observeReasoning(string $uri, string $reasoning): void

Hands over the reasoning of one call, once its body is decoded on success.

Parameters