Reciprocal rank fusion (RRF)

Each document scores the sum over result lists of 1/(k + rank), where k is a small constant (~60) that dampens the influence of top positions. A document that both the keyword and the vector retriever agree on rises; one that only one retriever found keeps a respectable score.

RRF needs no tuning per collection, which is why AACSearch uses it as the default hybrid merge before an optional reranking pass reorders the top candidates.