Recall@K
A retrieval metric measuring how much of the relevant result set is present within the top K returned candidates.
Recall@K makes the quality cost of approximate retrieval visible by measuring how many relevant results survive into the first K candidates. Because the value changes with K, it should be reported with the dataset and candidate budget rather than as an isolated score.
Higher recall can require more graph traversal, memory access, or a wider candidate set. Retrieval quality therefore has to be evaluated together with latency and resource cost.