Match each RAGAS metric to the specific RAG quality problem it detects

A **RAG** system has two moving parts: a retriever that fetches context chunks and a generator that writes an answer from them. RAGAS gives each part its own metrics, so a low score tells you which part to fix. The two generator-side metrics judge the written answer. Faithfulness asks whether every claim in the answer is actually supported by the retrieved context. A low score means the model hallucinated, it invented facts the context never stated. Answer relevance asks a different question: does the answer actually address what the user asked? An answer can be perfectly grounded in the context yet drift off-topic, and that is exactly what answer relevance catches. The two retriever-side metrics judge the fetched chunks. Context precision asks whether the chunks you retrieved were relevant, so a low score means you pulled in noisy distractors. Context recall asks whether you fetched all the chunks needed to answer, so a low score means the retriever missed key evidence. The clean way to remember it: two metrics blame the writer, two blame the librarian. Sort the four into those buckets first, then the matching is easy.

**RAGAS** decomposes RAG quality into a 2x2 grid: answer-side versus context-side. The two answer-side metrics judge what the generator wrote. Faithfulness checks that every claim in the answer is entailed by the retrieved context. A low score means hallucination. Answer relevance checks that the answer addresses the user's actual question. A low score means the model wrote something grounded but off-topic. The two context-side metrics judge what the retriever fetched. Context precision asks if the retrieved chunks were relevant and well-ranked. A low score means noisy distractors crowded the context window. Context recall asks if the retriever fetched all the evidence needed. A low score means key source chunks were missed entirely. The mapping follows directly. Faithfulness pairs with hallucinated claims. Answer relevance pairs with the grounded but off-topic answer. Context precision pairs with noisy irrelevant chunks. Context recall pairs with missing evidence. The diagnostic value is the split. If faithfulness drops while context recall stays high, the bug is in the generator. If context recall drops, the retriever is starving the generator and no prompt fix will help.