Recall@k
Of all the relevant items, what fraction appear in the top-k results — the key metric for the retrieval stage of RAG.
Recall@k measures retrieval coverage: of all the relevant items for a query, what fraction appear in the top-k returned results. It is the metric that matters most for the retrieval stage of RAG, because anything not in the top-k can never reach the model.
Worked example: if a query has 5 relevant documents and 4 appear in the top-10, recall@10 = 0.8; raising k raises recall but feeds more (and noisier) context to the reranker or model. Gotcha: high recall@k with a large k is cheap but pushes the precision problem downstream — you still need a reranker to order the top-k — so tune k to catch the relevant docs without drowning the next stage in distractors.