Precision@k measures the proportion of relevant documents among the first k results in a ranked retrieval list. In RAG, k usually equals the number of documents inserted into the context window, so the metric evaluates what can influence generation. section 6.2.3 section 6.5
Compute Precision@k by dividing the number of relevant documents in the top k positions by k. section 6.2.3
Financial research retrieval
Suppose five documents are retrieved, with the Q3 earnings report ranked third and the risk-factors section ranked fourth. Precision@3 is 1/3 because only one relevant document appears in the first three results; Precision@5 is 2/5, or 0.40. section 6.2.3
Precision@k exposes whether relevant material is buried below the context cutoff, but it does not distinguish different orderings within the top k. Use MAP, MRR, or nDCG when finer rank sensitivity is needed. section 6.2.3 section 6.5


