BM25 length normalization b in [0, 1]; the default scores exactly as before (0.75).
bm25K1?
number
BM25 term-frequency saturation k1; the default scores exactly as before this option existed (1.2).
candidateLimit?
number
-
documentIds?
string[]
-
embedding?
number[]
-
filter?
any
-
fusion?
string
How the lexical and semantic lists combine: "blend" (default) adds each list's min-max normalized scores over its top candidateLimit, weighted by lexicalWeight and semanticWeight; "rrf" is reciprocal rank fusion with k = 60.