also recall@k, retrieval recall
The share of test questions for which the right document appears in the top k retrieved results. Measured before any model writes an answer, it tells you whether a wrong answer is a retrieval miss or a generation problem.
You have done this if
You labelled real questions with the documents that should come back, and tracked how many made the top ten after each index change.
Say it in a review
We track recall at ten on a labelled question set, so we know a bad answer is a retrieval miss before anyone edits the prompt.