Diagram — where a RAG pipeline is measured
SVG · code-native
QUESTION
RETRIEVE
GENERATE
ANSWER
RETRIEVAL METRICS
context precision · context recall
GENERATION METRICS
faithfulness · answer relevancy
A wrong answer is almost always one of two bugs: retrieval never found the right context, or generation ignored it. The metrics split cleanly along that line.
via
Jason Teixeira — RAG Evaluation Metrics Explained