Building a RAG system is only half the work. The harder question is: how do you know it works, and how do you know it still works after the next change?

RAG quality is a pipeline property. Parsing, chunking, embeddings, metadata filters, retrieval, reranking, prompt construction, model choice, and abstention logic can each fail independently. If the final answer is poor, "the model hallucinated" is often the least useful diagnosis.

This chapter covers practical evaluation: retrieval metrics, generation metrics, golden datasets, automated judges, and regression workflows.