← Back to Glossary
Retrieval Evaluation (Recall@K, MRR, nDCG)
Quantitative metrics measuring how accurately a retriever surfaces correct documents in top search positions.
Why it exists
You cannot improve RAG performance without measuring retrieval accuracy.
How it works
Uses ground-truth evaluation datasets to calculate Recall@K (% of relevant items retrieved), MRR (Mean Reciprocal Rank), and nDCG.
Related terms
Where you'll use this
roadmap