/**
 * Reciprocal Rank Fusion (RRF) — combine multiple ranked lists into one.
 *
 * Reference: Cormack, Clarke, Büttcher (2009), "Reciprocal Rank Fusion
 * outperforms Condorcet and individual Rank Learning Methods."
 *
 * Formula: score(d) = Σ 1 / (k + rank_i(d))
 * where k is a constant (default 60) that dampens high-rank dominance.
 *
 * Used to combine BM25 keyword ranking and dense-vector cosine ranking
 * without score normalization (scales are incomparable).
 */
export interface RankedItem<T> {
    /** Stable identifier used to dedupe across lists. */
    id: string;
    /** Original item payload, returned in fused result. */
    item: T;
}
export interface FusedResult<T> {
    id: string;
    item: T;
    score: number;
    /** Rank (1-based) in each input list, or undefined if absent. */
    ranks: Record<string, number>;
}
/**
 * Fuse multiple ranked lists. Lists are keyed by name (e.g., "semantic", "keyword")
 * for debuggability. Items ordered best-first in each list.
 *
 * Returns items sorted by fused score descending.
 */
export declare function reciprocalRankFusion<T>(lists: Record<string, RankedItem<T>[]>, opts?: {
    k?: number;
    topK?: number;
}): FusedResult<T>[];
//# sourceMappingURL=rrf.d.ts.map