Component-wise evaluation assigns separate metrics to the RAG components responsible for observed behavior, turning vague demo impressions into targeted diagnosis. section 14.0
The evaluation triad separates three questions:
- Faithfulness asks whether the answer is grounded in the retrieved context and primarily evaluates generation. section 14.0
- Answer relevance asks whether the response addresses the query and evaluates generation. section 14.0
- Context relevance asks whether retrieval returned focused, useful context and evaluates the retriever. section 14.0
From score to repair
A low context-relevance score directs attention to missing content, ranking failures, or absent context; a low faithfulness score points toward factual inconsistency in generation. Scores therefore guide retrieval or generation changes instead of merely labeling the whole system as good or bad. section 14.0
