A streaming LLM integration has two independent acceptance gates
A streaming LLM integration has two independent acceptance gates: the model server must emit structured incremental events, and the public ingress must forward SSE without buffering. A healthy model endpoint or authenticated GET does not prove either. For a single-GPU in-place up…