/**
 * Statistical functions for evaluation analysis.
 *
 * Assumptions:
 * - All inputs are independent observations from normally distributed populations
 * - Welch's t-test is used (does NOT assume equal variances)
 * - Confidence intervals use the t-distribution for small samples (df ≤ 30)
 *   and normal approximation (z = 1.96) for larger samples
 *
 * Dependencies: simple-statistics (for mean, sampleStandardDeviation, sampleVariance)
 */
/**
 * Arithmetic mean. Returns 0 for empty/null input.
 */
export declare function mean(arr: number[]): number;
/**
 * Sample standard deviation (Bessel's correction: n-1 denominator).
 * Returns 0 for arrays with fewer than 2 elements.
 */
export declare function stddev(arr: number[]): number;
export type PValueCategory = '<0.001' | '<0.01' | '<0.05' | '≥0.05';
/**
 * Bucket a t-statistic into a p-value category by comparing |t| to standard
 * critical-value tables at α = 0.001 / 0.01 / 0.05 (two-tailed).
 *
 * Returns one of four buckets instead of a continuous p value. This avoids
 * implementing the regularized incomplete beta function for an exact CDF
 * while still giving readers enough resolution to distinguish
 * "barely significant" from "strongly significant".
 *
 * For df > 30 falls back to the normal-approximation critical values.
 */
export declare function pValueCategory(tStatistic: number, df: number): PValueCategory;
/**
 * Compute 95% confidence interval for the population mean.
 *
 * Method: t-interval (appropriate for small samples from normal populations)
 * Formula: mean ± t(α/2, n-1) × (s / √n)
 */
export declare function confidenceInterval(arr: number[]): {
    mean: number;
    lower: number;
    upper: number;
    stddev: number;
};
/**
 * Welch's t-test for two independent samples with unequal variances.
 *
 * Tests H₀: μ₁ = μ₂ (no difference in population means)
 * Uses Welch-Satterthwaite approximation for degrees of freedom.
 *
 * Requires at least 2 observations per group. Returns non-significant
 * result for insufficient data or zero variance in both groups.
 */
export interface EffectSizeResult {
    cohensD: number;
    hedgesG: number;
    primary: 'd' | 'g' | 'none';
    magnitude: 'negligible' | 'small' | 'medium' | 'large' | 'none';
    pooledStddev: number;
    n1: number;
    n2: number;
}
/**
 * Effect size for two independent samples.
 *
 * Cohen's d:   (mean_a - mean_b) / pooled_stddev
 * Hedges' g:   J * d, where J = 1 - 3 / (4 * (n1 + n2) - 9)
 *
 * Hedges' g corrects Cohen's d's small-sample bias (Hedges 1981) and is
 * preferred when n1 + n2 < 20. For n1 + n2 >= 20 the correction is < 5%
 * and Cohen's d is the conventional choice.
 *
 * `primary` indicates which to emphasize in UI based on total sample size.
 * `magnitude` uses standard thresholds (|effect| 0.2 / 0.5 / 0.8) applied
 * to the primary metric.
 *
 * Returns effect 0 with primary 'none' for insufficient data or zero variance.
 */
export declare function effectSize(a: number[], b: number[]): EffectSizeResult;
export declare function tTest(a: number[], b: number[]): {
    tStatistic: number;
    df: number;
    significant: boolean;
};