/**
 * Forgen Meta-Learning — Matcher Weight Tuner (Feature 2)
 *
 * Analyzes which scoring component (TF-IDF, BM25, Bigram) best discriminates
 * reflected vs. non-reflected solutions and adjusts ensemble weights.
 *
 * Algorithm:
 *   1. Load all non-retired solutions with evidence.injected > 0
 *   2. Partition into "effective" (reflected/injected > median) vs "ineffective"
 *   3. For each component: compute discrimination ratio (effective_mean / ineffective_mean)
 *   4. Shift weights toward the component with highest discrimination
 *   5. Apply guardrails: clamp [floor, ceiling], max delta per cycle, normalize to 1.0
 *
 * Cold-start: requires 10+ solutions with injected > 0, 3+ with reflected > 0.
 */
import type { MatcherWeights, MetaLearningConfig } from './types.js';
/**
 * Tune matcher ensemble weights based on solution effectiveness data.
 * Returns null if cold-start conditions are not met.
 */
export declare function tuneMatcherWeights(config: MetaLearningConfig): MatcherWeights | null;