@tangle-network/agent-eval 0.94.0 → 0.95.0

package/dist/{statistics-CCJpTGOS.d.ts → statistics.d.ts}

@@ -1,16 +1,15 @@
-import { d as ContinuousAgreementOptions, C as ContinuousAgreement } from './judge-calibration-0p2QcWNE.js';
-import { J as JudgeScore } from './types-C7DGg5ex.js';
-
+import { type ContinuousAgreement, type ContinuousAgreementOptions } from './judge-calibration';
+import type { JudgeScore } from './types';
 /** Identity: dimensions already follow "higher = better" by prompt convention
  *  (inverted dims like hallucination are scored 10 = best at the source). */
-declare const normalizeScores: (scores: JudgeScore[]) => JudgeScore[];
+export declare const normalizeScores: (scores: JudgeScore[]) => JudgeScore[];
 /** Weighted mean — falls back to uniform weights when omitted */
-declare function weightedMean(scores: {
+export declare function weightedMean(scores: {
     score: number;
     weight?: number;
 }[]): number;
 /** Bootstrap confidence interval */
-declare function confidenceInterval(scores: number[], confidence?: number, opts?: {
+export declare function confidenceInterval(scores: number[], confidence?: number, opts?: {
     seed?: number;
     resamples?: number;
 }): {
@@ -24,24 +23,24 @@ declare function confidenceInterval(scores: number[], confidence?: number, opts?
  * Each inner array is one judge's scores for all items.
  * All arrays must have the same length (same items scored).
  */
-declare function interRaterReliability(judgeScores: JudgeScore[][]): number;
+export declare function interRaterReliability(judgeScores: JudgeScore[][]): number;
 /**
  * Mann-Whitney U test for comparing two independent groups.
  * Returns U statistic and approximate p-value (normal approximation).
  */
-declare function mannWhitneyU(a: number[], b: number[]): {
+export declare function mannWhitneyU(a: number[], b: number[]): {
     u: number;
     p: number;
 };
 /** Partial credit: returns 0-1 ratio of current toward target */
-declare function partialCredit(current: number, target: number): number;
+export declare function partialCredit(current: number, target: number): number;
 /**
  * Paired t-test — before/after measurements on the SAME items.
  * Pairing removes inter-item variance, giving tighter significance than
  * an unpaired test when comparing prompt v1 vs prompt v2 on identical
  * scenarios.
  */
-declare function pairedTTest(before: number[], after: number[]): {
+export declare function pairedTTest(before: number[], after: number[]): {
     t: number;
     df: number;
     p: number;
@@ -50,7 +49,7 @@ declare function pairedTTest(before: number[], after: number[]): {
  * Wilcoxon signed-rank test — paired non-parametric alternative.
  * Use when the differences aren't normally distributed.
  */
-declare function wilcoxonSignedRank(before: number[], after: number[]): {
+export declare function wilcoxonSignedRank(before: number[], after: number[]): {
     w: number;
     p: number;
 };
@@ -59,8 +58,8 @@ declare function wilcoxonSignedRank(before: number[], after: number[]): {
  * Positive d means group b has higher mean than group a.
  * Rule of thumb: |d| < 0.2 negligible, 0.2–0.5 small, 0.5–0.8 medium, > 0.8 large.
  */
-declare function cohensD(a: number[], b: number[]): number;
-type CliffsMagnitude = 'negligible' | 'small' | 'medium' | 'large';
+export declare function cohensD(a: number[], b: number[]): number;
+export type CliffsMagnitude = 'negligible' | 'small' | 'medium' | 'large';
 /**
  * Cliff's delta — a non-parametric effect size for two independent samples.
  * `δ = (#(after > before) − #(after < before)) / (n_before · n_after)`,
@@ -71,14 +70,14 @@ type CliffsMagnitude = 'negligible' | 'small' | 'medium' | 'large';
  * `pairedBootstrap` / `wilcoxonSignedRank` for the non-parametric reporting
  * path. Returns 0 when either sample is empty.
  */
-declare function cliffsDelta(before: number[], after: number[]): number;
+export declare function cliffsDelta(before: number[], after: number[]): number;
 /**
  * Map a Cliff's delta to a qualitative magnitude using the standard
  * Romano et al. thresholds (|δ|): <0.147 negligible, <0.33 small,
  * <0.474 medium, else large.
  */
-declare function interpretCliffs(delta: number): CliffsMagnitude;
-interface WeightedCompositeInput {
+export declare function interpretCliffs(delta: number): CliffsMagnitude;
+export interface WeightedCompositeInput {
     /** Per-dimension scores (typically 0..1). */
     dims: Record<string, number>;
     /** Weight per dimension. Every weighted dimension MUST be present in
@@ -89,7 +88,7 @@ interface WeightedCompositeInput {
     /** Optional pass threshold; when set, the result reports `pass`. */
     threshold?: number;
 }
-interface WeightedCompositeResult {
+export interface WeightedCompositeResult {
     composite: number;
     pass?: boolean;
 }
@@ -102,8 +101,8 @@ interface WeightedCompositeResult {
  * weight is negative, or if the weights sum to 0 — none of which can produce
  * a meaningful composite.
  */
-declare function weightedComposite(input: WeightedCompositeInput): WeightedCompositeResult;
-interface CorpusScoreRecord {
+export declare function weightedComposite(input: WeightedCompositeInput): WeightedCompositeResult;
+export interface CorpusScoreRecord {
     /** Stable identifier for the rated item (scenario, span, turn, …). */
     itemId: string;
     /** Identifier for the judge that produced this score. */
@@ -113,14 +112,14 @@ interface CorpusScoreRecord {
     /** Numeric score; must be finite. */
     score: number;
 }
-interface CorpusAgreementPerDimension extends ContinuousAgreement {
+export interface CorpusAgreementPerDimension extends ContinuousAgreement {
     dimension: string;
     /** Item IDs that contributed to this dimension's matrix (every judge scored them). */
     itemIds: string[];
     /** Judge IDs that contributed to this dimension's matrix. */
     judgeIds: string[];
 }
-interface CorpusAgreementReport {
+export interface CorpusAgreementReport {
     /** Per-dimension ICC(2,1) + κ_w + Pearson + Spearman + bootstrap CIs. */
     perDimension: CorpusAgreementPerDimension[];
     /** Mean ICC across dimensions (NaN if no dimension yielded a finite ICC). */
@@ -132,7 +131,7 @@ interface CorpusAgreementReport {
     /** Judges seen across the corpus (sorted). */
     judgeIds: string[];
 }
-interface CorpusAgreementOptions extends ContinuousAgreementOptions {
+export interface CorpusAgreementOptions extends ContinuousAgreementOptions {
     /**
      * Restrict the audit to these dimensions. Default = every dimension
      * that appears in the input. A dimension named here but absent from
@@ -162,7 +161,7 @@ interface CorpusAgreementOptions extends ContinuousAgreementOptions {
  *     another dimension throws (would silently shrink the matrix).
  *   - Duplicate (itemId, judgeName, dimension) records throw.
  */
-declare function corpusInterRaterAgreement(records: CorpusScoreRecord[], opts?: CorpusAgreementOptions): CorpusAgreementReport;
+export declare function corpusInterRaterAgreement(records: CorpusScoreRecord[], opts?: CorpusAgreementOptions): CorpusAgreementReport;
 /**
  * Convenience adapter for `JudgeScore[]` data keyed externally by item.
  *
@@ -171,7 +170,7 @@ declare function corpusInterRaterAgreement(records: CorpusScoreRecord[], opts?:
  * agreement without manually flattening. `itemId` must be unique per
  * row of `itemsScores`.
  */
-declare function corpusInterRaterAgreementFromJudgeScores(itemsScores: Array<{
+export declare function corpusInterRaterAgreementFromJudgeScores(itemsScores: Array<{
     itemId: string;
     scores: JudgeScore[];
 }>, opts?: CorpusAgreementOptions): CorpusAgreementReport;
@@ -181,7 +180,7 @@ declare function corpusInterRaterAgreementFromJudgeScores(itemsScores: Array<{
  *   n = 2 * ( (z_{1-α/2} + z_{1-β}) / d )^2
  * where d is Cohen's d. Returns Infinity for effect ≤ 0.
  */
-declare function requiredSampleSize(opts: {
+export declare function requiredSampleSize(opts: {
     effect: number;
     alpha?: number;
     power?: number;
@@ -193,7 +192,7 @@ declare function requiredSampleSize(opts: {
  * sd(deltas) for score units; treat as a lower bound — Wilcoxon and bootstrap
  * have asymptotic relative efficiency below 1 vs the t-test on heavy tails.
  */
-declare function pairedMde(opts: {
+export declare function pairedMde(opts: {
     nPaired: number;
     alpha?: number;
     power?: number;
@@ -213,7 +212,7 @@ declare function pairedMde(opts: {
  * approximation, treat the result as a lower bound and prefer the discordant-pair
  * floor.
  */
-declare function mcnemarRequiredN(opts: {
+export declare function mcnemarRequiredN(opts: {
     p10: number;
     p01: number;
     alpha?: number;
@@ -225,7 +224,7 @@ declare function mcnemarRequiredN(opts: {
  * of {@link mcnemarRequiredN} (same Lachin asymptotic model, same parameters).
  * Returns a value in [0, 1]; equals `alpha` when there is no effect.
  */
-declare function mcnemarPower(opts: {
+export declare function mcnemarPower(opts: {
     p10: number;
     p01: number;
     nPairs: number;
@@ -233,7 +232,7 @@ declare function mcnemarPower(opts: {
     twoSided?: boolean;
 }): number;
 /** Bonferroni adjustment: multiply every p-value by the test count, clamp at 1. */
-declare function bonferroni(pValues: number[], alpha?: number): {
+export declare function bonferroni(pValues: number[], alpha?: number): {
     adjusted: number[];
     significant: boolean[];
 };
@@ -241,11 +240,11 @@ declare function bonferroni(pValues: number[], alpha?: number): {
  * Benjamini–Hochberg false discovery rate. Returns adjusted q-values and
  * significance at the target FDR; handles ties and preserves q monotonicity.
  */
-declare function benjaminiHochberg(pValues: number[], fdr?: number): {
+export declare function benjaminiHochberg(pValues: number[], fdr?: number): {
     qValues: number[];
     significant: boolean[];
 };
-interface PairedBootstrapResult {
+export interface PairedBootstrapResult {
     /** Number of paired observations. */
     n: number;
     /** Median of paired deltas (after − before). */
@@ -261,7 +260,7 @@ interface PairedBootstrapResult {
     /** Number of bootstrap resamples used. */
     resamples: number;
 }
-interface PairedBootstrapOptions {
+export interface PairedBootstrapOptions {
     /** Confidence level. Default 0.95. */
     confidence?: number;
     /** Bootstrap resample count. Default 2000. */
@@ -277,9 +276,9 @@ interface PairedBootstrapOptions {
  * lower bound is what the promotion gate checks — `low > threshold` means the
  * gain is real at the confidence level. Throws on unequal sample sizes.
  */
-declare function pairedBootstrap(before: number[], after: number[], opts?: PairedBootstrapOptions): PairedBootstrapResult;
+export declare function pairedBootstrap(before: number[], after: number[], opts?: PairedBootstrapOptions): PairedBootstrapResult;
 /** A binomial proportion estimate with a confidence interval. */
-interface ProportionInterval {
+export interface ProportionInterval {
     /** Point estimate successes / n (0 when n = 0). */
     estimate: number;
     /** Lower bound, clamped to [0, 1]. */
@@ -294,9 +293,9 @@ interface ProportionInterval {
  * CI — the continuous `confidenceInterval` assumes the wrong distribution for a
  * proportion. `n = 0 ⇒ {0, 0, 0}`.
  */
-declare function wilson(successes: number, n: number, confidence?: number): ProportionInterval;
+export declare function wilson(successes: number, n: number, confidence?: number): ProportionInterval;
 /** Result of a McNemar paired-binary significance test. */
-interface McNemarResult {
+export interface McNemarResult {
     /** Total paired observations. */
     n: number;
     /** Discordant pairs (b + c) — the only ones that carry signal. */
@@ -322,9 +321,9 @@ interface McNemarResult {
  * reference). Inputs are paired 0/1 (or boolean) arrays, control first to match
  * the module's (before, after) convention. Throws on unequal lengths.
  */
-declare function mcnemar(control: ArrayLike<number | boolean>, treatment: ArrayLike<number | boolean>): McNemarResult;
+export declare function mcnemar(control: ArrayLike<number | boolean>, treatment: ArrayLike<number | boolean>): McNemarResult;
 /** A paired binary effect size (treatment rate − control rate) with a CI. */
-interface RiskDifferenceResult {
+export interface RiskDifferenceResult {
     /** Total paired observations. */
     n: number;
     /** Discordant pairs: treatment-win count. */
@@ -348,7 +347,7 @@ interface RiskDifferenceResult {
  * the interval by ignoring the pairing). Inputs are paired 0/1 (or boolean)
  * arrays, control first. Throws on unequal lengths.
  */
-declare function pairedRiskDifference(control: ArrayLike<number | boolean>, treatment: ArrayLike<number | boolean>, confidence?: number): RiskDifferenceResult;
+export declare function pairedRiskDifference(control: ArrayLike<number | boolean>, treatment: ArrayLike<number | boolean>, confidence?: number): RiskDifferenceResult;
 /**
  * Unbiased pass@k for code generation (Chen et al. 2021, "Evaluating Large
  * Language Models Trained on Code"). Given `n` independent samples for one
@@ -359,8 +358,8 @@ declare function pairedRiskDifference(control: ArrayLike<number | boolean>, trea
  * the suite for the corpus pass@k. Computed in the numerically stable product
  * form. Requires 1 ≤ k ≤ n and 0 ≤ c ≤ n.
  */
-declare function passAtK(n: number, c: number, k: number): number;
-interface EProcessOptions {
+export declare function passAtK(n: number, c: number, k: number): number;
+export interface EProcessOptions {
     /** Type-I error budget. The process decides when wealth ≥ 1/alpha
      *  (Ville's inequality). Default 0.05. */
     alpha?: number;
@@ -373,7 +372,7 @@ interface EProcessOptions {
      *  A pre-registered minEffect shifts this — see `sequentialPairedGate`. */
     nullMean?: number;
 }
-interface EProcessStep {
+export interface EProcessStep {
     /** Current wealth W_n — the e-value against H0 after n observations. */
     wealth: number;
     /** Observations consumed so far. */
@@ -381,7 +380,7 @@ interface EProcessStep {
     /** True from the first n where W_n ≥ 1/alpha onward (sticky). */
     decided: boolean;
 }
-interface EProcessState extends EProcessStep {
+export interface EProcessState extends EProcessStep {
     alpha: number;
     maxBet: number;
     nullMean: number;
@@ -390,7 +389,7 @@ interface EProcessState extends EProcessStep {
     /** Observation count at the first threshold crossing; undefined until decided. */
     decidedAtN?: number;
 }
-interface EProcess {
+export interface EProcess {
     /** Consume one observation x ∈ [0,1]. Throws on non-finite / out-of-range
      *  input — a silent clamp would corrupt the type-I guarantee. */
     update(x: number): EProcessStep;
@@ -423,11 +422,10 @@ interface EProcess {
  * wealth keeps updating after the crossing (the e-process remains valid), but
  * the decision time is the first crossing.
  */
-declare function eProcess(opts?: EProcessOptions): EProcess;
+export declare function eProcess(opts?: EProcessOptions): EProcess;
 /** Tiny seedable PRNG (mulberry32) — deterministic resampling/shuffling, not
  *  cryptographic. Exported so e-process shuffles and bootstrap resampling
  *  share ONE PRNG implementation; a seed is REQUIRED (unseeded randomness in
  *  gate verdicts is non-reproducible by construction). */
-declare function mulberry32(seed: number): () => number;
-
-export { mulberry32 as A, normalizeScores as B, type CorpusAgreementReport as C, pairedMde as D, type EProcessState as E, pairedRiskDifference as F, pairedTTest as G, partialCredit as H, passAtK as I, requiredSampleSize as J, weightedComposite as K, weightedMean as L, type McNemarResult as M, wilson as N, type PairedBootstrapOptions as P, type RiskDifferenceResult as R, type WeightedCompositeInput as W, type PairedBootstrapResult as a, benjaminiHochberg as b, type CliffsMagnitude as c, type CorpusAgreementOptions as d, type CorpusAgreementPerDimension as e, type CorpusScoreRecord as f, type EProcess as g, type EProcessOptions as h, type EProcessStep as i, type ProportionInterval as j, type WeightedCompositeResult as k, bonferroni as l, cliffsDelta as m, cohensD as n, confidenceInterval as o, pairedBootstrap as p, corpusInterRaterAgreement as q, corpusInterRaterAgreementFromJudgeScores as r, eProcess as s, interRaterReliability as t, interpretCliffs as u, mannWhitneyU as v, wilcoxonSignedRank as w, mcnemar as x, mcnemarPower as y, mcnemarRequiredN as z };
+export declare function mulberry32(seed: number): () => number;
+//# sourceMappingURL=statistics.d.ts.map

package/dist/statistics.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"statistics.d.ts","sourceRoot":"","sources":["../src/statistics.ts"],"names":[],"mappings":"AACA,OAAO,EACL,KAAK,mBAAmB,EACxB,KAAK,0BAA0B,EAEhC,MAAM,qBAAqB,CAAA;AAC5B,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,SAAS,CAAA;AAEzC;6EAC6E;AAC7E,eAAO,MAAM,eAAe,GAAI,QAAQ,UAAU,EAAE,KAAG,UAAU,EAAY,CAAA;AAE7E,iEAAiE;AACjE,wBAAgB,YAAY,CAAC,MAAM,EAAE;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,MAAM,CAAC,EAAE,MAAM,CAAA;CAAE,EAAE,GAAG,MAAM,CAUjF;AAED,oCAAoC;AACpC,wBAAgB,kBAAkB,CAChC,MAAM,EAAE,MAAM,EAAE,EAChB,UAAU,SAAO,EACjB,IAAI,GAAE;IAAE,IAAI,CAAC,EAAE,MAAM,CAAC;IAAC,SAAS,CAAC,EAAE,MAAM,CAAA;CAAO,GAC/C;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,CA8BhD;AAED;;;;;GAKG;AACH,wBAAgB,qBAAqB,CAAC,WAAW,EAAE,UAAU,EAAE,EAAE,GAAG,MAAM,CAkDzE;AAED;;;GAGG;AACH,wBAAgB,YAAY,CAAC,CAAC,EAAE,MAAM,EAAE,EAAE,CAAC,EAAE,MAAM,EAAE,GAAG;IAAE,CAAC,EAAE,MAAM,CAAC;IAAC,CAAC,EAAE,MAAM,CAAA;CAAE,CA4C/E;AAED,iEAAiE;AACjE,wBAAgB,aAAa,CAAC,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,GAAG,MAAM,CAGrE;AAED;;;;;GAKG;AACH,wBAAgB,WAAW,CACzB,MAAM,EAAE,MAAM,EAAE,EAChB,KAAK,EAAE,MAAM,EAAE,GACd;IAAE,CAAC,EAAE,MAAM,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,CAAC,EAAE,MAAM,CAAA;CAAE,CAmBtC;AAED;;;GAGG;AACH,wBAAgB,kBAAkB,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,GAAG;IAAE,CAAC,EAAE,MAAM,CAAC;IAAC,CAAC,EAAE,MAAM,CAAA;CAAE,CA8B9F;AAED;;;;GAIG;AACH,wBAAgB,OAAO,CAAC,CAAC,EAAE,MAAM,EAAE,EAAE,CAAC,EAAE,MAAM,EAAE,GAAG,MAAM,CAWxD;AAED,MAAM,MAAM,eAAe,GAAG,YAAY,GAAG,OAAO,GAAG,QAAQ,GAAG,OAAO,CAAA;AAEzE;;;;;;;;;GASG;AACH,wBAAgB,WAAW,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,GAAG,MAAM,CAWrE;AAED;;;;GAIG;AACH,wBAAgB,eAAe,CAAC,KAAK,EAAE,MAAM,GAAG,eAAe,CAM9D;AAED,MAAM,WAAW,sBAAsB;IACrC,6CAA6C;IAC7C,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC5B;;;+CAG2C;IAC3C,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC/B,oEAAoE;IACpE,SAAS,CAAC,EAAE,MAAM,CAAA;CACnB;AAED,MAAM,WAAW,uBAAuB;IACtC,SAAS,EAAE,MAAM,CAAA;IACjB,IAAI,CAAC,EAAE,OAAO,CAAA;CACf;AAED;;;;;;;;GAQG;AACH,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,sBAAsB,GAAG,uBAAuB,CA2BxF;AAiBD,MAAM,WAAW,iBAAiB;IAChC,sEAAsE;IACtE,MAAM,EAAE,MAAM,CAAA;IACd,yDAAyD;IACzD,SAAS,EAAE,MAAM,CAAA;IACjB,uDAAuD;IACvD,SAAS,EAAE,MAAM,CAAA;IACjB,qCAAqC;IACrC,KAAK,EAAE,MAAM,CAAA;CACd;AAED,MAAM,WAAW,2BAA4B,SAAQ,mBAAmB;IACtE,SAAS,EAAE,MAAM,CAAA;IACjB,sFAAsF;IACtF,OAAO,EAAE,MAAM,EAAE,CAAA;IACjB,6DAA6D;IAC7D,QAAQ,EAAE,MAAM,EAAE,CAAA;CACnB;AAED,MAAM,WAAW,qBAAqB;IACpC,yEAAyE;IACzE,YAAY,EAAE,2BAA2B,EAAE,CAAA;IAC3C,6EAA6E;IAC7E,UAAU,EAAE,MAAM,CAAA;IAClB,8DAA8D;IAC9D,oBAAoB,EAAE,MAAM,CAAA;IAC5B,qCAAqC;IACrC,UAAU,EAAE,MAAM,EAAE,CAAA;IACpB,8CAA8C;IAC9C,QAAQ,EAAE,MAAM,EAAE,CAAA;CACnB;AAED,MAAM,WAAW,sBAAuB,SAAQ,0BAA0B;IACxE;;;;OAIG;IACH,UAAU,CAAC,EAAE,MAAM,EAAE,CAAA;IACrB;;;;OAIG;IACH,MAAM,CAAC,EAAE,MAAM,EAAE,CAAA;CAClB;AAED;;;;;;;;;;;;;;;GAeG;AACH,wBAAgB,yBAAyB,CACvC,OAAO,EAAE,iBAAiB,EAAE,EAC5B,IAAI,GAAE,sBAA2B,GAChC,qBAAqB,CAgHvB;AAED;;;;;;;GAOG;AACH,wBAAgB,wCAAwC,CACtD,WAAW,EAAE,KAAK,CAAC;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,UAAU,EAAE,CAAA;CAAE,CAAC,EAC5D,IAAI,GAAE,sBAA2B,GAChC,qBAAqB,CAoBvB;AAqFD;;;;;GAKG;AACH,wBAAgB,kBAAkB,CAAC,IAAI,EAAE;IACvC,MAAM,EAAE,MAAM,CAAA;IACd,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,QAAQ,CAAC,EAAE,OAAO,CAAA;CACnB,GAAG,MAAM,CAUT;AAED;;;;;GAKG;AACH,wBAAgB,SAAS,CAAC,IAAI,EAAE;IAC9B,OAAO,EAAE,MAAM,CAAA;IACf,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,QAAQ,CAAC,EAAE,OAAO,CAAA;CACnB,GAAG,MAAM,CAQT;AAED;;;;;;;;;;;;;GAaG;AACH,wBAAgB,gBAAgB,CAAC,IAAI,EAAE;IACrC,GAAG,EAAE,MAAM,CAAA;IACX,GAAG,EAAE,MAAM,CAAA;IACX,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,QAAQ,CAAC,EAAE,OAAO,CAAA;CACnB,GAAG,MAAM,CAiBT;AAED;;;;GAIG;AACH,wBAAgB,YAAY,CAAC,IAAI,EAAE;IACjC,GAAG,EAAE,MAAM,CAAA;IACX,GAAG,EAAE,MAAM,CAAA;IACX,MAAM,EAAE,MAAM,CAAA;IACd,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,QAAQ,CAAC,EAAE,OAAO,CAAA;CACnB,GAAG,MAAM,CAcT;AAED,mFAAmF;AACnF,wBAAgB,UAAU,CACxB,OAAO,EAAE,MAAM,EAAE,EACjB,KAAK,SAAO,GACX;IAAE,QAAQ,EAAE,MAAM,EAAE,CAAC;IAAC,WAAW,EAAE,OAAO,EAAE,CAAA;CAAE,CAKhD;AAED;;;GAGG;AACH,wBAAgB,iBAAiB,CAC/B,OAAO,EAAE,MAAM,EAAE,EACjB,GAAG,SAAO,GACT;IAAE,OAAO,EAAE,MAAM,EAAE,CAAC;IAAC,WAAW,EAAE,OAAO,EAAE,CAAA;CAAE,CAgB/C;AAID,MAAM,WAAW,qBAAqB;IACpC,qCAAqC;IACrC,CAAC,EAAE,MAAM,CAAA;IACT,gDAAgD;IAChD,MAAM,EAAE,MAAM,CAAA;IACd,6BAA6B;IAC7B,IAAI,EAAE,MAAM,CAAA;IACZ,+DAA+D;IAC/D,GAAG,EAAE,MAAM,CAAA;IACX,+DAA+D;IAC/D,IAAI,EAAE,MAAM,CAAA;IACZ,yCAAyC;IACzC,UAAU,EAAE,MAAM,CAAA;IAClB,0CAA0C;IAC1C,SAAS,EAAE,MAAM,CAAA;CAClB;AAED,MAAM,WAAW,sBAAsB;IACrC,sCAAsC;IACtC,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,8CAA8C;IAC9C,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,gDAAgD;IAChD,SAAS,CAAC,EAAE,QAAQ,GAAG,MAAM,CAAA;IAC7B,0DAA0D;IAC1D,IAAI,CAAC,EAAE,MAAM,CAAA;CACd;AAED;;;;;GAKG;AACH,wBAAgB,eAAe,CAC7B,MAAM,EAAE,MAAM,EAAE,EAChB,KAAK,EAAE,MAAM,EAAE,EACf,IAAI,GAAE,sBAA2B,GAChC,qBAAqB,CAqDvB;AAcD,iEAAiE;AACjE,MAAM,WAAW,kBAAkB;IACjC,mDAAmD;IACnD,QAAQ,EAAE,MAAM,CAAA;IAChB,sCAAsC;IACtC,KAAK,EAAE,MAAM,CAAA;IACb,sCAAsC;IACtC,KAAK,EAAE,MAAM,CAAA;CACd;AAED;;;;;;GAMG;AACH,wBAAgB,MAAM,CAAC,SAAS,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,EAAE,UAAU,SAAO,GAAG,kBAAkB,CAgB1F;AAED,2DAA2D;AAC3D,MAAM,WAAW,aAAa;IAC5B,iCAAiC;IACjC,CAAC,EAAE,MAAM,CAAA;IACT,kEAAkE;IAClE,WAAW,EAAE,MAAM,CAAA;IACnB,4EAA4E;IAC5E,CAAC,EAAE,MAAM,CAAA;IACT,0EAA0E;IAC1E,CAAC,EAAE,MAAM,CAAA;IACT,sFAAsF;IACtF,SAAS,EAAE,MAAM,CAAA;IACjB,yEAAyE;IACzE,MAAM,EAAE,MAAM,CAAA;CACf;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,OAAO,CACrB,OAAO,EAAE,SAAS,CAAC,MAAM,GAAG,OAAO,CAAC,EACpC,SAAS,EAAE,SAAS,CAAC,MAAM,GAAG,OAAO,CAAC,GACrC,aAAa,CAgBf;AAED,6EAA6E;AAC7E,MAAM,WAAW,oBAAoB;IACnC,iCAAiC;IACjC,CAAC,EAAE,MAAM,CAAA;IACT,6CAA6C;IAC7C,CAAC,EAAE,MAAM,CAAA;IACT,2CAA2C;IAC3C,CAAC,EAAE,MAAM,CAAA;IACT,sEAAsE;IACtE,cAAc,EAAE,MAAM,CAAA;IACtB,iDAAiD;IACjD,KAAK,EAAE,MAAM,CAAA;IACb,iDAAiD;IACjD,KAAK,EAAE,MAAM,CAAA;IACb,6BAA6B;IAC7B,UAAU,EAAE,MAAM,CAAA;CACnB;AAED;;;;;;;GAOG;AACH,wBAAgB,oBAAoB,CAClC,OAAO,EAAE,SAAS,CAAC,MAAM,GAAG,OAAO,CAAC,EACpC,SAAS,EAAE,SAAS,CAAC,MAAM,GAAG,OAAO,CAAC,EACtC,UAAU,SAAO,GAChB,oBAAoB,CA6BtB;AAED;;;;;;;;;GASG;AACH,wBAAgB,OAAO,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,GAAG,MAAM,CAW/D;AAuBD,MAAM,WAAW,eAAe;IAC9B;8CAC0C;IAC1C,KAAK,CAAC,EAAE,MAAM,CAAA;IACd;;uBAEmB;IACnB,MAAM,CAAC,EAAE,MAAM,CAAA;IACf;;+EAE2E;IAC3E,QAAQ,CAAC,EAAE,MAAM,CAAA;CAClB;AAED,MAAM,WAAW,YAAY;IAC3B,wEAAwE;IACxE,MAAM,EAAE,MAAM,CAAA;IACd,oCAAoC;IACpC,CAAC,EAAE,MAAM,CAAA;IACT,iEAAiE;IACjE,OAAO,EAAE,OAAO,CAAA;CACjB;AAED,MAAM,WAAW,aAAc,SAAQ,YAAY;IACjD,KAAK,EAAE,MAAM,CAAA;IACb,MAAM,EAAE,MAAM,CAAA;IACd,QAAQ,EAAE,MAAM,CAAA;IAChB,qCAAqC;IACrC,SAAS,EAAE,MAAM,CAAA;IACjB,kFAAkF;IAClF,UAAU,CAAC,EAAE,MAAM,CAAA;CACpB;AAED,MAAM,WAAW,QAAQ;IACvB;qEACiE;IACjE,MAAM,CAAC,CAAC,EAAE,MAAM,GAAG,YAAY,CAAA;IAC/B,KAAK,IAAI,aAAa,CAAA;CACvB;AAED;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AACH,wBAAgB,QAAQ,CAAC,IAAI,GAAE,eAAoB,GAAG,QAAQ,CAoD7D;AA8DD;;;0DAG0D;AAC1D,wBAAgB,UAAU,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,MAAM,CASrD"}

package/dist/statistics.test.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=statistics.test.d.ts.map

package/dist/statistics.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"statistics.test.d.ts","sourceRoot":"","sources":["../src/statistics.test.ts"],"names":[],"mappings":""}

package/dist/steering-optimizer.d.ts

@@ -0,0 +1,58 @@
+import { type RunScore, type RunScoreWeights } from './run-score';
+import type { SteeringBundle } from './steering';
+export type SteeringOptimizerBackend = 'pairwise' | 'ax-gepa';
+export interface SteeringOptimizationRow {
+    variantId: string;
+    scenarioId: string;
+    bundle: SteeringBundle;
+    score: RunScore;
+    metadata?: Record<string, unknown>;
+}
+export interface SteeringOptimizationSelector {
+    backend: SteeringOptimizerBackend;
+    signature?: string;
+    labels?: string[];
+    rationale?: string;
+}
+export interface SteeringOptimizationResult {
+    backend: SteeringOptimizerBackend;
+    recommendedVariantId: string;
+    rationale: string;
+    rankings: Array<{
+        variantId: string;
+        mean: number;
+        runs: number;
+    }>;
+    selector?: SteeringOptimizationSelector;
+    /** Runnable handle on the trained classifier. Present only when the
+     *  ax-gepa backend completed training; calls the optimized selector
+     *  program via ax's `forward`. */
+    selectVariant?: (row: {
+        task: string;
+        split: string;
+        seedPreview: string;
+    }) => Promise<{
+        variantId: string;
+        rationale: string;
+    }>;
+    skipped?: boolean;
+}
+export interface SteeringOptimizerConfig {
+    weights?: Partial<RunScoreWeights>;
+}
+export interface AxSteeringOptimizerConfig extends SteeringOptimizerConfig {
+    provider: 'openai' | 'anthropic';
+    apiKey: string;
+    model: string;
+    teacherModel?: string;
+    minScenarioWinners?: number;
+}
+export declare class PairwiseSteeringOptimizer {
+    optimize(rows: SteeringOptimizationRow[], config?: SteeringOptimizerConfig): SteeringOptimizationResult;
+}
+export declare class AxGepaSteeringOptimizer {
+    private readonly config;
+    constructor(config: AxSteeringOptimizerConfig);
+    optimize(rows: SteeringOptimizationRow[]): Promise<SteeringOptimizationResult>;
+}
+//# sourceMappingURL=steering-optimizer.d.ts.map

package/dist/steering-optimizer.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"steering-optimizer.d.ts","sourceRoot":"","sources":["../src/steering-optimizer.ts"],"names":[],"mappings":"AACA,OAAO,EAAqB,KAAK,QAAQ,EAAE,KAAK,eAAe,EAAE,MAAM,aAAa,CAAA;AACpF,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,YAAY,CAAA;AAEhD,MAAM,MAAM,wBAAwB,GAAG,UAAU,GAAG,SAAS,CAAA;AAE7D,MAAM,WAAW,uBAAuB;IACtC,SAAS,EAAE,MAAM,CAAA;IACjB,UAAU,EAAE,MAAM,CAAA;IAClB,MAAM,EAAE,cAAc,CAAA;IACtB,KAAK,EAAE,QAAQ,CAAA;IACf,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,4BAA4B;IAC3C,OAAO,EAAE,wBAAwB,CAAA;IACjC,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,MAAM,CAAC,EAAE,MAAM,EAAE,CAAA;IACjB,SAAS,CAAC,EAAE,MAAM,CAAA;CACnB;AAED,MAAM,WAAW,0BAA0B;IACzC,OAAO,EAAE,wBAAwB,CAAA;IACjC,oBAAoB,EAAE,MAAM,CAAA;IAC5B,SAAS,EAAE,MAAM,CAAA;IACjB,QAAQ,EAAE,KAAK,CAAC;QAAE,SAAS,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;IAClE,QAAQ,CAAC,EAAE,4BAA4B,CAAA;IACvC;;sCAEkC;IAClC,aAAa,CAAC,EAAE,CAAC,GAAG,EAAE;QACpB,IAAI,EAAE,MAAM,CAAA;QACZ,KAAK,EAAE,MAAM,CAAA;QACb,WAAW,EAAE,MAAM,CAAA;KACpB,KAAK,OAAO,CAAC;QAAE,SAAS,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;IACvD,OAAO,CAAC,EAAE,OAAO,CAAA;CAClB;AAED,MAAM,WAAW,uBAAuB;IACtC,OAAO,CAAC,EAAE,OAAO,CAAC,eAAe,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,yBAA0B,SAAQ,uBAAuB;IACxE,QAAQ,EAAE,QAAQ,GAAG,WAAW,CAAA;IAChC,MAAM,EAAE,MAAM,CAAA;IACd,KAAK,EAAE,MAAM,CAAA;IACb,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,kBAAkB,CAAC,EAAE,MAAM,CAAA;CAC5B;AASD,qBAAa,yBAAyB;IACpC,QAAQ,CACN,IAAI,EAAE,uBAAuB,EAAE,EAC/B,MAAM,GAAE,uBAA4B,GACnC,0BAA0B;CAU9B;AAED,qBAAa,uBAAuB;IACtB,OAAO,CAAC,QAAQ,CAAC,MAAM;gBAAN,MAAM,EAAE,yBAAyB;IAExD,QAAQ,CAAC,IAAI,EAAE,uBAAuB,EAAE,GAAG,OAAO,CAAC,0BAA0B,CAAC;CAuFrF"}

package/dist/steering.d.ts

@@ -0,0 +1,24 @@
+export interface SteeringRolePrompt {
+    system?: string;
+    append?: string;
+}
+export interface SteeringBundle {
+    id: string;
+    coderPrompt?: string;
+    continuePrompt?: string;
+    reviewerPrompts?: Record<string, string>;
+    skills?: string[];
+    rolePrompts?: Record<string, SteeringRolePrompt>;
+    metadata?: Record<string, unknown>;
+}
+export interface SteeringDelta {
+    coderPrompt?: string;
+    continuePrompt?: string;
+    reviewerPrompts?: Record<string, string>;
+    skills?: string[];
+    rolePrompts?: Record<string, SteeringRolePrompt>;
+    metadata?: Record<string, unknown>;
+}
+export declare function mergeSteeringBundle(base: SteeringBundle, delta: SteeringDelta): SteeringBundle;
+export declare function renderSteeringText(bundle: SteeringBundle): string;
+//# sourceMappingURL=steering.d.ts.map

package/dist/steering.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"steering.d.ts","sourceRoot":"","sources":["../src/steering.ts"],"names":[],"mappings":"AAAA,MAAM,WAAW,kBAAkB;IACjC,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,MAAM,CAAC,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,WAAW,cAAc;IAC7B,EAAE,EAAE,MAAM,CAAA;IACV,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,eAAe,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IACxC,MAAM,CAAC,EAAE,MAAM,EAAE,CAAA;IACjB,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,kBAAkB,CAAC,CAAA;IAChD,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,aAAa;IAC5B,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,eAAe,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IACxC,MAAM,CAAC,EAAE,MAAM,EAAE,CAAA;IACjB,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,kBAAkB,CAAC,CAAA;IAChD,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,wBAAgB,mBAAmB,CAAC,IAAI,EAAE,cAAc,EAAE,KAAK,EAAE,aAAa,GAAG,cAAc,CAmB9F;AAED,wBAAgB,kBAAkB,CAAC,MAAM,EAAE,cAAc,GAAG,MAAM,CAejE"}

package/dist/storyboard/code-edit.d.ts

@@ -0,0 +1,64 @@
+/**
+ * Code-edit extraction + the "watch the agent write code" animation.
+ *
+ * The base storyboard (./index) compresses a run into scenes but keeps scenes
+ * payload-free (title + narration + evidence span ids). To actually SHOW the
+ * code an agent wrote, this module pulls the concrete edit (path, diff,
+ * before/after, language) out of the raw edit-tool spans and renders a
+ * self-contained HTML replay that types each file out, character by character,
+ * like an editor recording.
+ *
+ * Additive on purpose: it reuses the canonical `Span` (trace/schema) and the
+ * `Storyboard` IR (./index) without modifying the reducer/compiler — so it
+ * composes with the existing pipeline and a Remotion consumer can read the same
+ * `CodeEdit[]` to render an MP4.
+ */
+import type { Span } from '../trace/schema';
+import type { Scene, Storyboard } from './index';
+/** A concrete code change lifted from a tool span. */
+export interface CodeEdit {
+    /** Span the edit came from — back-reference into the trace. */
+    spanId: string;
+    path: string;
+    language?: string;
+    /** Unified diff, when the tool carried one. */
+    diff?: string;
+    /** Full file body before / after, when present — enables true keystroke
+     *  animation rather than only flashing a diff. */
+    before?: string;
+    after?: string;
+    additions: number;
+    deletions: number;
+    startedAt: number;
+}
+/** Extract a CodeEdit from a single span, or undefined if it is not an edit. */
+export declare function codeEditFromSpan(span: Span): CodeEdit | undefined;
+/** All code edits in a run, in chronological order. */
+export declare function extractCodeEdits(spans: readonly Span[]): CodeEdit[];
+/** Pair each code-edit (`diff`) scene with its concrete edit, matched via the
+ *  scene's evidence span ids. Scenes without a recoverable edit are dropped. */
+export declare function codeEditsForStoryboard(storyboard: Storyboard, spans: readonly Span[]): Array<{
+    scene: Scene;
+    edit: CodeEdit;
+}>;
+/** The text the animation types for an edit: prefer the full new file body,
+ *  then the added lines of a diff, then a minimal placeholder so the scene is
+ *  never blank. */
+export declare function editAnimationText(edit: CodeEdit): string;
+export interface CodeAnimationOptions {
+    /** Page title. Default "Agent writing code". */
+    title?: string;
+    /** Characters typed per animation tick. Higher = faster. Default 4. */
+    charsPerTick?: number;
+    /** Pause between files, ms. Default 900. */
+    pauseBetweenMs?: number;
+}
+/**
+ * Render a self-contained HTML page that animates the agent writing each file,
+ * character by character, with a filename tab, line numbers, and a blinking
+ * cursor — the shareable "watch it build" clip. No build step, no assets, no
+ * deps. The same `CodeEdit[]` feeds a Remotion MP4 renderer in a consumer
+ * package; this is the dependency-free baseline.
+ */
+export declare function renderCodeAnimationHtml(source: readonly Span[] | readonly CodeEdit[], opts?: CodeAnimationOptions): string;
+//# sourceMappingURL=code-edit.d.ts.map

package/dist/storyboard/code-edit.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"code-edit.d.ts","sourceRoot":"","sources":["../../src/storyboard/code-edit.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAEH,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,iBAAiB,CAAA;AAC3C,OAAO,KAAK,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,SAAS,CAAA;AAKhD,sDAAsD;AACtD,MAAM,WAAW,QAAQ;IACvB,+DAA+D;IAC/D,MAAM,EAAE,MAAM,CAAA;IACd,IAAI,EAAE,MAAM,CAAA;IACZ,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,+CAA+C;IAC/C,IAAI,CAAC,EAAE,MAAM,CAAA;IACb;sDACkD;IAClD,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,SAAS,EAAE,MAAM,CAAA;IACjB,SAAS,EAAE,MAAM,CAAA;IACjB,SAAS,EAAE,MAAM,CAAA;CAClB;AA2DD,gFAAgF;AAChF,wBAAgB,gBAAgB,CAAC,IAAI,EAAE,IAAI,GAAG,QAAQ,GAAG,SAAS,CAqBjE;AAED,uDAAuD;AACvD,wBAAgB,gBAAgB,CAAC,KAAK,EAAE,SAAS,IAAI,EAAE,GAAG,QAAQ,EAAE,CAKnE;AAED;gFACgF;AAChF,wBAAgB,sBAAsB,CACpC,UAAU,EAAE,UAAU,EACtB,KAAK,EAAE,SAAS,IAAI,EAAE,GACrB,KAAK,CAAC;IAAE,KAAK,EAAE,KAAK,CAAC;IAAC,IAAI,EAAE,QAAQ,CAAA;CAAE,CAAC,CAWzC;AAED;;mBAEmB;AACnB,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,QAAQ,GAAG,MAAM,CAWxD;AAMD,MAAM,WAAW,oBAAoB;IACnC,gDAAgD;IAChD,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,uEAAuE;IACvE,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,4CAA4C;IAC5C,cAAc,CAAC,EAAE,MAAM,CAAA;CACxB;AAED;;;;;;GAMG;AACH,wBAAgB,uBAAuB,CACrC,MAAM,EAAE,SAAS,IAAI,EAAE,GAAG,SAAS,QAAQ,EAAE,EAC7C,IAAI,GAAE,oBAAyB,GAC9B,MAAM,CA+FR"}

package/dist/storyboard/code-edit.test.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=code-edit.test.d.ts.map

package/dist/storyboard/code-edit.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"code-edit.test.d.ts","sourceRoot":"","sources":["../../src/storyboard/code-edit.test.ts"],"names":[],"mappings":""}