Retrieval relevance
Test the evidence path and the generated response against explicit expectations.
RAG EVALUATION · PARIMI
RAG Evaluation and Grounding Testing
Evaluate retrieval quality and generated answers together so teams can distinguish retrieval failures from generation failures.
Context precision
Test the evidence path and the generated response against explicit expectations.
Context recall
Test the evidence path and the generated response against explicit expectations.
Answer groundedness
Test the evidence path and the generated response against explicit expectations.
Faithfulness
Test the evidence path and the generated response against explicit expectations.
Hallucination detection
Test the evidence path and the generated response against explicit expectations.
Missing or stale knowledge
Test the evidence path and the generated response against explicit expectations.
RAG regression
Test the evidence path and the generated response against explicit expectations.
Find out whether the failure is retrieval or generation.
A useful RAG evaluation strategy tests the retrieval layer, context construction and final answer rather than treating the system as a single black box.