What metrics would you use to evaluate a retrieval-augmented generation (RAG) system, and why is accuracy alone insufficient?
