Whether this chunk delivered generation the caller can see.
Token counts and provider metadata (a gateway deployment stamp, a thought
signature) reach the caller without being output: they neither start the clock
on time-to-first-token nor give a retry anything to duplicate.
Whether this chunk delivered generation the caller can see.
Token counts and provider metadata (a gateway deployment stamp, a thought signature) reach the caller without being output: they neither start the clock on time-to-first-token nor give a retry anything to duplicate.