Backend is the leg that contributed this match: "index" (lexical), "vector" (semantic) or "code" (the org's repositories). It is the same name that leg reports itself under in Fusion.Backends, so a hit can be traced to a status.
Rank is this document's 1-based position in THAT leg's own result list, before fusion — 1 is the leg's best hit. It is the only input to the fused score: RRF adds 1/(60+rank) per leg, which is why a document two legs ranked second beats one a single leg ranked first.
Score is the leg's NATIVE score, on that leg's own scale, reported for explanation and never used in ranking — the scales are incomparable (a cosine similarity against a term-match count), which is why fusion works on ranks. The vector leg reports Qdrant's cosine similarity; the lexical leg exposes no per-row score and reports 0, meaning "unscored", not "scored zero".
