Trace generation defaults separately for each compatible API endpoint

Do not assume server-level generation defaults are applied identically across API-compatible endpoints. Trace each endpoint's request model through conversion into engine sampling parameters, including token limits, request seeds, tool settings, and proxy rewrites. Bind parity tests to the exact endpoint actually used.

Applicability and limits: a process-level seed is not necessarily a per-request seed, and deterministic decoding can make seed differences irrelevant. Endpoint behavior is version-specific; verify the installed source and any request-rewrite proxy before making compatibility claims.