PRODUCTION SCENARIO
A Generative AI Engineer wants to score every production trace of a RAG assistant for hallucination against the retrieved context. The team has no labeled expected answers and will not have any for months.
Which built-in judge can the team run now?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: RetrievalGroundedness
RetrievalGroundedness checks whether a response is supported by the retrieved context using only the trace's inputs and outputs, so it needs no expectations and suits production monitoring. Correctness, RetrievalSufficiency, and ToolCallCorrectness each require ground truth supplied as expectations.