Definition
Plain language
A two-step search where a fast method grabs a shortlist and a smarter model reorders it.
As stated in the literature
A retrieval setup where a cheap retriever narrows a corpus to a handful of candidates and a stronger model reorders them; contrasted with full in-context retrieval because it sidesteps million-token corpora and generalization to unseen corpus sizes.
Why it matters: It combines speed and accuracy by using a cheap method to shortlist and an expensive one only on the finalists, avoiding the need to run the smart model over everything.
For example, a fast search grabs the twenty most likely documents, then a slower, smarter model reads those twenty and pushes the truly best one to the top.
Heard on the show
“… Prior evaluations were either proprietary systems with no controlled comparison, or reranking setups where a real retriever narrows the corpus to a handful of candidates first and the model …”Episode 198 — The Model That Knows the Answer and Can't Say It