Reranking
Also called: re-ranker, cross-encoder, second-pass ranking
A second pass that takes the first batch of retrieved passages and reorders them by how well each one actually answers the question — using a slower, more careful model than the fast search that produced the batch.
Fast retrieval is tuned for recall: cast a wide net and pull back, say, the top 50 candidates. But the model can only read a few. A reranker decides which handful actually go into the context, so the best passage isn't left out just because the initial search ranked it eighth.
The initial retrieval. First-pass search optimizes for speed over a huge index; reranking spends more effort on a small shortlist to get the ordering right — the two work together, not instead of each other.