Chapter Assessment: Evaluating RAG

This review tests how the repair guide assistant is evaluated as a RAG feature, from relevance labels and retrieval metrics to faithfulness, completeness, unanswerable questions, and the order for diagnosing a wrong answer. The bank holds 20 questions, and each attempt draws ten. It weights decisions over definitions: most questions describe a result and ask which step failed, which metric fits, or what the team should do next. A few ask you to compute a score