Reranking

A second scoring pass applied to a small set of retrieved candidates, using a model that reads the query and passage together rather than comparing precomputed vectors. It is slower per pair than retrieval-augmented generation's first-stage lookup, which is exactly why it only runs on a shortlist rather than the whole index.

Why exams ask this

Tested as the fix for "retrieval returned the right passage at rank six instead of rank one." A reranker is a second, slower, more accurate pass over a small candidate set, not a replacement for the first-stage retriever. The distractor proposes retrieving more candidates instead of scoring them better.

Relevant to

Related concepts

Resources

No resources linked to this concept yet.