/**
 * POEM — Pareto-Optimal Embedded Modeling, multi-metric ranking.
 *
 * Ranks search candidates across multiple relevance metrics without requiring
 * hand-tuned weights. Uses the TFPR (Top-Fraction Pareto Ranker) approach:
 * vectorized dominance matrix computation with column duplication for
 * query-type-dependent metric weighting.
 *
 * Algorithm:
 *  1. Prune: per-metric top-K → union of survivors
 *  2. Build objectives matrix with column duplication for query-type weighting
 *  3. For each objective column, sort candidates and accumulate pairwise
 *     dominance counts (duplicate columns contribute via weight multiplier)
 *  4. Compute fitness: meanDominance × (numDominating + ε) / (numSubmitting + ε)
 *  5. Sort by fitness, assign ranks
 *
 * References:
 *  - POEM paper: https://iopscience.iop.org/article/10.1088/2632-2153/ab891b
 *  - TFPR: https://github.com/merckgroup/aidd_tfpr
 *  - colourdle: https://github.com/aebrer/colourdle
 */
import type { QueryType } from "./query-classifier.js";
import { type MetricScores } from "./types.js";
export interface RankedCandidate {
    id: number;
    scores: MetricScores;
    rank: number;
}
/**
 * Rank candidates using POEM / TFPR.
 *
 * @param candidates Map of candidateId → MetricScores (all values 0–1)
 * @param queryType  Query type for column duplication weighting
 * @param topK       Per-metric pruning limit (default: 1000)
 * @returns Candidates ordered best-first with assigned ranks (0 = best)
 */
export declare function poemRank(candidates: Map<number, MetricScores>, queryType: QueryType, topK?: number): RankedCandidate[];
//# sourceMappingURL=poem.d.ts.map