@tangle-network/agent-eval 0.221.8 → 0.222.0

package/dist/{analyze-runs-DfX6dHP0.d.ts → analyze-runs-C-pP1sK9.d.ts}

@@ -1,10 +1,9 @@
-import { o as FailureClass } from "./schema-D6EVKzTB.js";
-import { a as RunRecord, u as RunTerminalOutcome } from "./run-record-C3awiXPf.js";
-import { r as ContinuousAgreement } from "./judge-calibration-DFtEMlde.js";
-import { r as DatasetScenario } from "./dataset-DQqhOCPt.js";
-import { f as ParetoFigureSpec, l as GainDistributionBin } from "./summary-report-C8V49Jt3.js";
-import { t as AnalystRegistry } from "./registry-BcdJmyoU.js";
-//#region src/contract/insight-report.d.ts
+import { o as FailureClass } from "./schema-CkIvnmLU.js";
+import { a as RunRecord, u as RunTerminalOutcome } from "./run-record-D2BFMPGl.js";
+import { r as ContinuousAgreement } from "./judge-calibration-DnrbQaXI.js";
+import { r as DatasetScenario } from "./dataset-DThC1S6G.js";
+import { f as ParetoFigureSpec, l as GainDistributionBin } from "./summary-report-0La5vejY.js";
+import { t as AnalystRegistry } from "./registry-dhdDoAV8.js";
 interface InsightReport {
   /** Number of runs analyzed. */
   n: number;
@@ -414,8 +413,6 @@ interface Recommendation {
   /** Optional pointer back into the report for the evidence. */
   evidencePath?: string;
 }
-//#endregion
-//#region src/contract/execution-insight.d.ts
 interface SummarizeExecutionOptions {
   runs: RunRecord[];
   histogramBins?: number;
@@ -426,8 +423,6 @@ interface ExecutionReport {
 }
 /** Summarize runtime facts without interpreting task quality or promotion readiness. */
 declare function summarizeExecution(opts: SummarizeExecutionOptions): ExecutionReport;
-//#endregion
-//#region src/contract/analyze-runs.d.ts
 interface AnalyzeRunsOptions {
   /** The runs to analyze. */
   runs: RunRecord[];
@@ -484,6 +479,5 @@ interface AnalyzeRunsOptions {
   baselineLabel?: string;
 }
 declare function analyzeRuns(opts: AnalyzeRunsOptions): Promise<InsightReport>;
-//#endregion
 export { ReleaseSummary as _, summarizeExecution as a, ExecutionInsight as c, InsightReport as d, InterRaterInsight as f, Recommendation as g, OutcomeCorrelationInsight as h, SummarizeExecutionOptions as i, FailureClassTally as l, LiftInsight as m, analyzeRuns as n, CostProvenanceSummary as o, JudgeInsight as p, ExecutionReport as r, ExecutionErrorOutcomeCell as s, AnalyzeRunsOptions as t, FailureClusterInsight as u, ScalarDistribution as v, TokenUsageInsight as y };
-//# sourceMappingURL=analyze-runs-DfX6dHP0.d.ts.map
+//# sourceMappingURL=analyze-runs-C-pP1sK9.d.ts.map

package/dist/analyze-runs-C-pP1sK9.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"analyze-runs-C-pP1sK9.d.ts","names":[],"sources":["../src/contract/insight-report.ts","../src/contract/execution-insight.ts","../src/contract/analyze-runs.ts"],"mappings":";;;;;;UAqCiB;;EAEf;;;;EAKA,WAAW;;EAGX,WAAW;;;EAIX,cAAc,eAAe;;EAG7B;IACE,MAAM;IACN,QAAQ;;;;;IAKR,aAAa;;;;;IAKb;MAAa;MAAe;;;;;;EAM9B,QAAQ,eAAe;;;;EAKvB,aAAa;;;;EAKb,OAAO;;;EAIP,kBAAkB;;;EAIlB,gBAAgB;;;;;EAMhB,qBAAqB;;;;;EAMrB,SAAS;;;;;;EAOT,wBAAwB;;;;EAKxB,iBAAiB;;;EAIjB,iBAAiB;;UAGF;EACf;IAAY;IAAW;;EACvB;IAAa;IAAW;;;;EAGxB;IAAc;IAAW;;EACzB;IAAc;;EACd;;UAGe;;EAEf,YAAY;;EAEZ,SAAS;;;EAGT,YAAY;;;EAGZ;IACE;IACA,YAAY;IACZ,SAAS;IACT;;;EAGF,QAAQ;IAAQ;IAAe;;;;;EAI/B;IACE;IACA;IACA;;;;;EAKF;IACE;;;IAGA;;IAEA;;IAEA;;IAEA;;IAEA;;;;;IAKA,mBAAmB,OAAO,oBAAoB;;;;EAIhD;IACE;IACA;IACA;IACA;IACA;;;UAIa;;EAEf;;EAEA;;EAEA;;UAGe;EACf,OAAO;EACP,QAAQ;EACR,WAAW;EACX,QAAQ;EACR,YAAY;EACZ;IACE;IACA;IACA;IACA;IACA;;;;UAOa;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,WAAW;;;;;EAKX,WAAW;IAAQ;IAAe;;;UAGnB;;EAEf;;EAEA;;;EAGA,cAAc;;;EAGd;;;EAGA;;;EAGA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,SAAS;;EAET,mBAAmB;IACjB;IACA,SAAS;MAAQ;MAAe;;IAChC;;;UAIa;EACf;EACA;;EAEA;;EAEA;;;EAGA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;EAWA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;;EAEf,UAAU;IACR;IACA;;IAEA;;IAEA;;IAEA;;EAEF;;;;;UAMe;;EAEf,cAAc;;EAEd;;EAEA;;UAGe;;EAEf;;;EAGA;EACA,UAAU;IAAQ;IAAe;IAAgB;;;UAGlC;;;EAGf;;EAEA;;EAEA;;EAEA;;EAEA;IACE;IACA;IACA;;;UAIa;;;EAGf;EACA,MAAM;IACJ;IACA;IACA;;;;EAIF;;UAGQ;;EAER;;EAEA;;;;EAIA;;EAEA;EACA;;KAGU,eACP;EACC;;EAEA;;EAEA;;EAEA;;EAEA;MAED;EACC;;EAEA;EACA;EACA;EACA;;UAGW;;EAEf;EACA;;EAEA;;;;EAIA,SAAS,eAAe;;;EAGxB;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;;EAEA;;UCzbe;EACf,MAAM;EACN;;UAGe;EACf,WAAW;EACX,gBAAgB;;;iBAIF,mBAAmB,MAAM,4BAA4B;UC4BpD;;EAEf,MAAM;;;EAGN;;;;;EAKA;EACA;;;EAGA,8BAA8B;;;EAG9B,kBAAkB;;;EAGlB,UAAU;;;;EAIV;IACE;IACA,cAAc;;;;;EAKhB,cAAc;IAAQ;IAAe;IAAe;;;EAEpD;;;;EAIA;;;;;;;;;EASA,eAAe;;;EAGf;;iBAGoB,YAAY,MAAM,qBAAqB,QAAQ"}

package/dist/{analyze-runs-CHmsykNJ.js → analyze-runs-CTheASYV.js}

@@ -1,17 +1,17 @@
-import { s as ValidationError } from "./errors-BcOs5MKW.js";
-import { r as observedSplitScore } from "./reward-nw2xZGZG.js";
-import { c as validateRunRecord } from "./run-record-DhEJgoPz.js";
-import { c as pairedCohensDz, o as cohensD, r as paretoChart } from "./summary-report-DHDWxe6p.js";
-import { o as normalCdf, r as pairRunRecords } from "./paired-arms-B2ttLc0_.js";
-import { t as interpolatedQuantile } from "./quantile--0yb07pD.js";
-import { c as lnGamma, l as regularizedIncompleteBeta, t as assertFiniteSample$1 } from "./internal-BTK91Bqt.js";
-import { a as spearmanR, r as pearsonR } from "./descriptive-CVNy48rR.js";
-import { r as continuousAgreement } from "./judge-calibration-BkXMIWA4.js";
-import { i as pairedBootstrap, l as studentTQuantile, s as pairedTTest, t as mean$1 } from "./mean-BbA_JCzW.js";
-import { i as requiredPairedSampleSize, r as pairedMde } from "./power-and-mde-CQX-0weG.js";
-import { n as aggregatePairedHoldout } from "./statistical-heldout-Bksv92Th.js";
-import "./query-B_RZ36an.js";
-//#region src/statistics/inference.ts
+import { s as ValidationError } from "./errors-DeEEV9Vx.js";
+import { r as observedSplitScore } from "./reward-P0VD_Aft.js";
+import { c as validateRunRecord } from "./run-record-DEjx43cI.js";
+import { c as pairedCohensDz, o as cohensD, r as paretoChart } from "./summary-report-CPfsrLac.js";
+import { t as interpolatedQuantile } from "./quantile-CncXP-Ss.js";
+import { o as normalCdf, r as pairRunRecords } from "./paired-arms-D-54hO-k.js";
+import { c as lnGamma, l as regularizedIncompleteBeta, t as assertFiniteSample$1 } from "./internal-v6XfOeJW.js";
+import { a as spearmanR, r as pearsonR } from "./descriptive-cPwITVfy.js";
+import { r as continuousAgreement } from "./judge-calibration-DGqlfWW8.js";
+import { i as pairedBootstrap, l as studentTQuantile, s as pairedTTest, t as mean$1 } from "./mean-CdDYG-us.js";
+import { i as requiredPairedSampleSize, r as pairedMde } from "./power-and-mde-ClI6RWw4.js";
+import { n as aggregatePairedHoldout } from "./statistical-heldout-TP6ndL0B.js";
+import "./query-CGXzO9d4.js";
+// src/statistics/inference.ts
 /** Two-sided normal tail, evaluated without subtracting a near-one CDF. */
 function normalTwoSidedPValue(z) {
 	if (Number.isNaN(z)) throw new ValidationError("normalTwoSidedPValue: z must not be NaN");
@@ -110,8 +110,7 @@ function linearRegressionSlope(points) {
 		slopePValue: n < 3 ? null : residual === 0 ? slope === 0 ? 1 : 0 : studentTTwoSidedPValue(slope / Math.sqrt(residual / (n - 2) / sxx), n - 2)
 	};
 }
-//#endregion
-//#region src/baseline.ts
+// src/baseline.ts
 /**
 * Baseline regression detection.
 *
@@ -214,8 +213,7 @@ function assertFiniteSample(name, sample) {
 	const invalid = sample.find((value) => !Number.isFinite(value));
 	if (invalid !== void 0) throw new RangeError(`welchsTTest: sample ${name} must contain only finite values, got ${invalid}`);
 }
-//#endregion
-//#region src/contract/quality-insight.ts
+// src/contract/quality-insight.ts
 /** Observed score distributions, judge agreement, and downstream quality correlation. */
 function resolveSplit(runs, pref) {
 	if (pref !== "auto") return pref;
@@ -452,8 +450,7 @@ function isFiniteNumber(value) {
 function meanOrZero(arr) {
 	return arr.length === 0 ? 0 : mean$1(arr);
 }
-//#endregion
-//#region src/contract/comparison-insight.ts
+// src/contract/comparison-insight.ts
 /** Paired candidate lift and independent prior-period metric comparisons. */
 function computePriorPeriodComparison(current, baseline, split, windowLabel) {
 	if (current.length === 0 || baseline.length === 0) return void 0;
@@ -619,8 +616,7 @@ function computeLift(runs, baselineId, candidateId, split, independentUnitByScen
 		requiredN
 	};
 }
-//#endregion
-//#region src/contract/execution-insight.ts
+// src/contract/execution-insight.ts
 /** Runtime usage, terminal outcomes, execution errors, and cost capture facts. */
 /** Summarize runtime facts without interpreting task quality or promotion readiness. */
 function summarizeExecution(opts) {
@@ -823,8 +819,7 @@ function diagnoseCostCoverage(runs, provenance) {
 	if (unknown === runs.length) return `USD cost total unknown for all ${runs.length} runs${floor} — no observed or estimated USD values; token and wall-time metrics remain available.`;
 	return `USD cost total unknown for ${unknown}/${runs.length} runs${floor}; excluded those rows from cost statistics (${known}/${runs.length} retained: ${provenance.observed.n} observed, ${provenance.estimated.n} estimated).`;
 }
-//#endregion
-//#region src/contamination-guard.ts
+// src/contamination-guard.ts
 function checkCanaries(output, scenarios) {
 	const leaks = [];
 	for (const s of scenarios) {
@@ -844,8 +839,7 @@ function excerpt(source, needle) {
 	const end = Math.min(source.length, at + needle.length + 30);
 	return (start > 0 ? "…" : "") + source.slice(start, end) + (end < source.length ? "…" : "");
 }
-//#endregion
-//#region src/contract/failure-insight.ts
+// src/contract/failure-insight.ts
 /**
 * Model-free task-failure tally.
 *
@@ -934,8 +928,7 @@ function stringifyOutput(run) {
 	if (typeof metadata?.output === "string") return metadata.output;
 	if (typeof metadata?.text === "string") return metadata.text;
 }
-//#endregion
-//#region src/contract/insight-recommendations.ts
+// src/contract/insight-recommendations.ts
 /** A bootstrap interval with no spread: every resample landed on the same
 *  value, so the interval carries no information about how far the point
 *  estimate could be wrong and cannot support a directional claim. */
@@ -1121,8 +1114,7 @@ function buildRecommendations(ctx) {
 	});
 	return out;
 }
-//#endregion
-//#region src/contract/analyze-runs.ts
+// src/contract/analyze-runs.ts
 async function analyzeRuns(opts) {
 	const runs = opts.runs.map(validateRunRecord);
 	const bins = opts.histogramBins ?? 12;
@@ -1192,7 +1184,6 @@ async function analyzeRuns(opts) {
 		recommendations
 	};
 }
-//#endregion
 export { welchsTTest as a, fisherExactTwoSided as c, studentTTwoSidedPValue as d, twoByTwoPValue as f, iqr as i, linearRegressionSlope as l, checkCanaries as n, chiSquared as o, summarizeExecution as r, chiSquaredPValue as s, analyzeRuns as t, normalTwoSidedPValue as u };
 
-//# sourceMappingURL=analyze-runs-CHmsykNJ.js.map
+//# sourceMappingURL=analyze-runs-CTheASYV.js.map