/**
 * Scoring — weighted aggregation, multi-trial metrics, and verdicts.
 */
import type { GraderResult } from "../graders/types.js";
export interface MultiTrialMetrics {
    perTrialPassRate: number;
    passAtK: number;
    passToTheK: number;
    k: number;
}
export interface StimulusScore {
    stimulusName: string;
    trialResults: GraderResult[][];
    aggregateScore: number;
    multiTrial: MultiTrialMetrics;
    unscored: boolean;
    flaky: boolean;
    flakinessPercent: number;
}
export interface SkillScore {
    skillName: string;
    stimulusScores: StimulusScore[];
    overallScore: number;
    overallMultiTrial: MultiTrialMetrics;
    passed: boolean;
}
/** Returns true when a threshold is a finite number in the supported [0, 1] range. */
export declare function isValidThreshold(threshold: number): boolean;
/**
 * Resolve whether a graded result counts as a pass.
 *
 * When a threshold is configured, the verdict is driven by the aggregate
 * (possibly weighted) score — `score >= threshold` — so reporters and
 * `vally grade` agree with pass@k and the eval-level verdict. Without a
 * threshold, it falls back to the binary all-graders-passed flag.
 *
 * This is the single source of truth for the per-trial pass rule; the plan
 * summarizer, the reporters, and `vally grade` MUST use it so the glyphs,
 * counts, pass@k, and verdicts never disagree.
 */
export declare function resolveGradePass(grade: {
    passed: boolean;
    score: number;
}, threshold: number | undefined): boolean;
/**
 * Unbiased pass@k estimator (Chen et al., 2021 — Codex paper).
 * pass@k = 1 - C(n-c, k) / C(n, k)
 * where n = total trials, c = successes, k = target attempts.
 *
 * More accurate than the naive 1-(1-p)^k at small sample sizes (k=3..10),
 * which is exactly the regime evals operate in.
 */
export declare function passAtK(n: number, c: number, k: number): number;
/**
 * Naive pass@k for comparison/testing.
 * Formula: 1 - (1 - p)^k where p = per-trial pass rate.
 */
export declare function passAtKNaive(perTrialPassRate: number, k: number): number;
/**
 * Compute pass^k: probability that ALL k trials succeed.
 * Formula: p^k where p = per-trial pass rate.
 */
export declare function passToTheK(perTrialPassRate: number, k: number): number;
/** Compute multi-trial metrics from a set of trial pass/fail results. */
export declare function computeMultiTrialMetrics(trialPassed: boolean[]): MultiTrialMetrics;
/** Compute a StimulusScore from trial summaries. */
export declare function computeStimulusScore(stimulusName: string, trialGrades: Array<{
    grade: GraderResult;
    passed: boolean;
}>, hasGraders: boolean): StimulusScore;
/**
 * Aggregate per-stimulus scores into a per-skill score.
 * Skill-level multi-trial metrics are the mean of per-stimulus metrics —
 * NOT a pool of raw trial outcomes across different stimuli.
 */
export declare function computeSkillScore(skillName: string, stimulusScores: StimulusScore[], threshold: number): SkillScore;
//# sourceMappingURL=scorer.d.ts.map