AI-300 Practice Question 8: Groundedness Evaluator
Microsoft Machine Learning Operations Engineer Associate (AI-300) · Test 1 · question 8 of 10
Testers reviewing a retrieval-augmented HR assistant report that some answers are well written and clearly on topic, but state leave entitlements that do not appear anywhere in the policy documents the retriever returned. Which built-in Microsoft Foundry evaluator directly measures this problem?
Show answer and explanation
Answer:
Groundedness is the RAG system evaluator for precisely this failure: it measures how well the generated response aligns with the given context without fabricating content. Microsoft frames it as the *precision* aspect of a response; the answer contains nothing outside the grounding context. It takes response and context (with query optional but recommended) and returns a 1-5 score with a pass/fail label (option D).
- A is wrong because Relevance asks whether the response addresses the user's query. An invented leave entitlement can be perfectly on topic and still score well on relevance. - B is wrong because Coherence judges logical organization and readability. A fabricated answer is often the most coherent one in the set. - C is wrong because Fluency judges grammar and linguistic quality only. Testers already said the answers are well written, so fluency would not flag them.
Reference: Retrieval-Augmented Generation (RAG) Evaluators for Generative AI - Microsoft Foundry