Which statements correctly describe the four RAGAS metrics?

**RAGAS** is a framework for evaluating retrieval-augmented generation (RAG) systems. A RAG pipeline has two halves to grade: the retriever, which fetches context, and the generator, which writes the answer over that context. The four metrics split across those halves. **Faithfulness** checks whether the claims in the answer are actually supported by the retrieved context, so it catches hallucination. Answer relevance checks whether the answer addresses the user's question, regardless of context. The other two grade the retriever. Context precision asks: of the chunks you retrieved, what fraction were actually relevant? Low precision means you pulled in noise. Context recall asks: of everything you needed, how much did you retrieve? Low recall means you missed important chunks. The trick is option E. Faithfulness, answer relevance, and context precision can all be computed without a gold answer, using only the question, the retrieved context, and the generated answer. But context recall needs a reference answer to measure coverage against. So RAGAS is mostly reference-free, not entirely. Option F is the correct reading of the precision-recall pattern: high precision plus low recall means clean but incomplete retrieval.

**RAGAS** evaluates a RAG pipeline on four metrics that split across its two stages. Two grade generation. Two grade retrieval. Generation metrics. Faithfulness checks whether every claim in the answer is entailed by the retrieved context. Answer relevance checks whether the answer addresses the user's question, independent of context. They catch different failures: hallucination versus off-topic drift. Retrieval metrics. Context precision is the fraction of retrieved chunks that are relevant. Context recall is how much of the needed information was retrieved at all. They trade off: a precise but incomplete retriever scores high precision and low recall. That makes options A through D correct descriptions, and option F the correct reading of the precision-recall split. Option E is the trap. Three of the four metrics are reference-free, computed from the question, context, and answer alone. Context recall is the exception. To measure how much of the answer was covered, you compare against ground-truth claims from a reference answer. No reference, no recall. So the suite is mostly reference-free, but calling it fully reference-free is wrong.