@tangle-network/agent-eval 0.215.1 → 0.216.0

package/dist/cli.js

@@ -5,10 +5,10 @@ import { t as renderSearchSummary } from "./search-summary-CBvz3u_S.js";
 import { c as hostedTenantFromEnv, g as SearchRunKindSchema, r as shipSearchLedger, t as SearchShipConflictError, u as SEARCH_LEDGER_BATCH_MAX_BYTES } from "./search-shipper-er4pK5_Y.js";
 import { i as runRolloutReleaseCli } from "./hf-dataset-BR2yhuNn.js";
 import { t as createChatClient } from "./chat-client-DthPOYGd.js";
-import { t as runAnalystBenchmarkCommand } from "./benchmark-command-CgpGi7vY.js";
+import { t as runAnalystBenchmarkCommand } from "./benchmark-command-CByl9dxG.js";
 import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-DbcC7xMl.js";
 import { C as editCredit, S as front, _ as surfaceTextEdits, c as metaSearch, d as renderMetaSearchText, m as landscape, n as taskMatrix, o as skillManifold, p as formatLandscape, r as formatSkillManifold, t as tree, w as editCreditText } from "./tree-rXLW3M_4.js";
-import { t as runSupervisorRunCommand } from "./report-command-CtIIsQKP.js";
+import { t as runSupervisorRunCommand } from "./report-command-y8H6RSX8.js";
 import { createHash } from "node:crypto";
 import { existsSync, readFileSync, writeFileSync } from "node:fs";
 import { dirname, join } from "node:path";

package/dist/cluster-bootstrap-B3JywQQn.js

@@ -0,0 +1,66 @@
+import { s as ValidationError } from "./errors-Dngq5h35.js";
+import { i as makeRng } from "./internal-CRHwS9iz.js";
+//#region src/meta-eval/cluster-bootstrap.ts
+/**
+* Percentile cluster bootstrap of an observation mean. Whole clusters are resampled with
+* replacement and each resample's statistic is Σ values / Σ observations over the drawn
+* clusters, so observations that share a source move together and a large cluster cannot
+* pose as many independent results. For a paired comparison pass the per-observation
+* difference (candidate − baseline) under the observation's cluster.
+*
+* With one cluster the interval collapses to the point estimate: there is no between-cluster
+* spread to measure, which callers must treat as no interval, not as certainty.
+*/
+function clusterBootstrapMean(observations, options = {}) {
+	const confidence = options.confidence ?? .95;
+	const resamples = options.resamples ?? 2e3;
+	if (!(confidence > 0 && confidence < 1)) throw new ValidationError(`clusterBootstrapMean: confidence must be in (0,1), got ${confidence}`);
+	if (!Number.isInteger(resamples) || resamples < 1) throw new ValidationError(`clusterBootstrapMean: resamples must be a positive integer, got ${resamples}`);
+	if (!observations.length) throw new ValidationError("clusterBootstrapMean: requires at least one observation");
+	const sums = /* @__PURE__ */ new Map();
+	const values = [];
+	for (const [index, observation] of observations.entries()) {
+		if (typeof observation?.cluster !== "string" || observation.cluster === "") throw new ValidationError(`clusterBootstrapMean: observation ${index} needs a cluster id`);
+		if (!Number.isFinite(observation.value)) throw new ValidationError(`clusterBootstrapMean: observation ${index} value must be finite`);
+		const cell = sums.get(observation.cluster) ?? {
+			sum: 0,
+			count: 0
+		};
+		cell.sum += observation.value;
+		cell.count += 1;
+		sums.set(observation.cluster, cell);
+		values.push(observation.value);
+	}
+	const cells = [...sums.values()];
+	const mean = cells.reduce((acc, cell) => acc + cell.sum, 0) / observations.length;
+	const k = cells.length;
+	const rng = makeRng(options.seed, values);
+	const samples = new Array(resamples);
+	for (let b = 0; b < resamples; b++) {
+		let sum = 0;
+		let count = 0;
+		for (let draw = 0; draw < k; draw++) {
+			const cell = cells[Math.floor(rng() * k)];
+			sum += cell.sum;
+			count += cell.count;
+		}
+		samples[b] = sum / count;
+	}
+	samples.sort((a, b) => a - b);
+	const alpha = 1 - confidence;
+	const lowIdx = Math.floor(alpha / 2 * resamples);
+	const highIdx = Math.min(resamples - 1, Math.ceil((1 - alpha / 2) * resamples) - 1);
+	return {
+		n: observations.length,
+		clusters: k,
+		mean,
+		low: samples[lowIdx],
+		high: samples[Math.max(highIdx, lowIdx)],
+		confidence,
+		resamples
+	};
+}
+//#endregion
+export { clusterBootstrapMean as t };
+
+//# sourceMappingURL=cluster-bootstrap-B3JywQQn.js.map

package/dist/cluster-bootstrap-B3JywQQn.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"cluster-bootstrap-B3JywQQn.js","names":[],"sources":["../src/meta-eval/cluster-bootstrap.ts"],"sourcesContent":["import { ValidationError } from '../errors'\nimport { makeRng } from '../statistics/internal'\n\n/** One observation and the source cluster it belongs to. */\nexport interface ClusteredObservation {\n  cluster: string\n  value: number\n}\n\nexport interface ClusterBootstrapOptions {\n  /** Confidence level. Default 0.95. */\n  confidence?: number\n  /** Bootstrap resample count. Default 2000. */\n  resamples?: number\n  /** Deterministic seed. If omitted, derived from the values so the interval is reproducible. */\n  seed?: number\n}\n\nexport interface ClusterBootstrapResult {\n  n: number\n  clusters: number\n  /** Mean over observations, each observation weighted equally. */\n  mean: number\n  low: number\n  high: number\n  confidence: number\n  resamples: number\n}\n\n/**\n * Percentile cluster bootstrap of an observation mean. Whole clusters are resampled with\n * replacement and each resample's statistic is Σ values / Σ observations over the drawn\n * clusters, so observations that share a source move together and a large cluster cannot\n * pose as many independent results. For a paired comparison pass the per-observation\n * difference (candidate − baseline) under the observation's cluster.\n *\n * With one cluster the interval collapses to the point estimate: there is no between-cluster\n * spread to measure, which callers must treat as no interval, not as certainty.\n */\nexport function clusterBootstrapMean(\n  observations: readonly ClusteredObservation[],\n  options: ClusterBootstrapOptions = {},\n): ClusterBootstrapResult {\n  const confidence = options.confidence ?? 0.95\n  const resamples = options.resamples ?? 2000\n  if (!(confidence > 0 && confidence < 1)) {\n    throw new ValidationError(\n      `clusterBootstrapMean: confidence must be in (0,1), got ${confidence}`,\n    )\n  }\n  if (!Number.isInteger(resamples) || resamples < 1) {\n    throw new ValidationError(\n      `clusterBootstrapMean: resamples must be a positive integer, got ${resamples}`,\n    )\n  }\n  if (!observations.length) {\n    throw new ValidationError('clusterBootstrapMean: requires at least one observation')\n  }\n\n  const sums = new Map<string, { sum: number; count: number }>()\n  const values: number[] = []\n  for (const [index, observation] of observations.entries()) {\n    if (typeof observation?.cluster !== 'string' || observation.cluster === '') {\n      throw new ValidationError(`clusterBootstrapMean: observation ${index} needs a cluster id`)\n    }\n    if (!Number.isFinite(observation.value)) {\n      throw new ValidationError(`clusterBootstrapMean: observation ${index} value must be finite`)\n    }\n    const cell = sums.get(observation.cluster) ?? { sum: 0, count: 0 }\n    cell.sum += observation.value\n    cell.count += 1\n    sums.set(observation.cluster, cell)\n    values.push(observation.value)\n  }\n\n  const cells = [...sums.values()]\n  const total = cells.reduce((acc, cell) => acc + cell.sum, 0)\n  const mean = total / observations.length\n  const k = cells.length\n  const rng = makeRng(options.seed, values)\n  const samples = new Array<number>(resamples)\n  for (let b = 0; b < resamples; b++) {\n    let sum = 0\n    let count = 0\n    for (let draw = 0; draw < k; draw++) {\n      const cell = cells[Math.floor(rng() * k)]!\n      sum += cell.sum\n      count += cell.count\n    }\n    samples[b] = sum / count\n  }\n  samples.sort((a, b) => a - b)\n  const alpha = 1 - confidence\n  const lowIdx = Math.floor((alpha / 2) * resamples)\n  const highIdx = Math.min(resamples - 1, Math.ceil((1 - alpha / 2) * resamples) - 1)\n  return {\n    n: observations.length,\n    clusters: k,\n    mean,\n    low: samples[lowIdx]!,\n    high: samples[Math.max(highIdx, lowIdx)]!,\n    confidence,\n    resamples,\n  }\n}\n"],"mappings":";;;;;;;;;;;;;AAuCA,SAAgB,qBACd,cACA,UAAmC,CAAC,GACZ;CACxB,MAAM,aAAa,QAAQ,cAAc;CACzC,MAAM,YAAY,QAAQ,aAAa;CACvC,IAAI,EAAE,aAAa,KAAK,aAAa,IACnC,MAAM,IAAI,gBACR,0DAA0D,YAC5D;CAEF,IAAI,CAAC,OAAO,UAAU,SAAS,KAAK,YAAY,GAC9C,MAAM,IAAI,gBACR,mEAAmE,WACrE;CAEF,IAAI,CAAC,aAAa,QAChB,MAAM,IAAI,gBAAgB,yDAAyD;CAGrF,MAAM,uBAAO,IAAI,IAA4C;CAC7D,MAAM,SAAmB,CAAC;CAC1B,KAAK,MAAM,CAAC,OAAO,gBAAgB,aAAa,QAAQ,GAAG;EACzD,IAAI,OAAO,aAAa,YAAY,YAAY,YAAY,YAAY,IACtE,MAAM,IAAI,gBAAgB,qCAAqC,MAAM,oBAAoB;EAE3F,IAAI,CAAC,OAAO,SAAS,YAAY,KAAK,GACpC,MAAM,IAAI,gBAAgB,qCAAqC,MAAM,sBAAsB;EAE7F,MAAM,OAAO,KAAK,IAAI,YAAY,OAAO,KAAK;GAAE,KAAK;GAAG,OAAO;EAAE;EACjE,KAAK,OAAO,YAAY;EACxB,KAAK,SAAS;EACd,KAAK,IAAI,YAAY,SAAS,IAAI;EAClC,OAAO,KAAK,YAAY,KAAK;CAC/B;CAEA,MAAM,QAAQ,CAAC,GAAG,KAAK,OAAO,CAAC;CAE/B,MAAM,OADQ,MAAM,QAAQ,KAAK,SAAS,MAAM,KAAK,KAAK,CACzC,IAAI,aAAa;CAClC,MAAM,IAAI,MAAM;CAChB,MAAM,MAAM,QAAQ,QAAQ,MAAM,MAAM;CACxC,MAAM,UAAU,IAAI,MAAc,SAAS;CAC3C,KAAK,IAAI,IAAI,GAAG,IAAI,WAAW,KAAK;EAClC,IAAI,MAAM;EACV,IAAI,QAAQ;EACZ,KAAK,IAAI,OAAO,GAAG,OAAO,GAAG,QAAQ;GACnC,MAAM,OAAO,MAAM,KAAK,MAAM,IAAI,IAAI,CAAC;GACvC,OAAO,KAAK;GACZ,SAAS,KAAK;EAChB;EACA,QAAQ,KAAK,MAAM;CACrB;CACA,QAAQ,MAAM,GAAG,MAAM,IAAI,CAAC;CAC5B,MAAM,QAAQ,IAAI;CAClB,MAAM,SAAS,KAAK,MAAO,QAAQ,IAAK,SAAS;CACjD,MAAM,UAAU,KAAK,IAAI,YAAY,GAAG,KAAK,MAAM,IAAI,QAAQ,KAAK,SAAS,IAAI,CAAC;CAClF,OAAO;EACL,GAAG,aAAa;EAChB,UAAU;EACV;EACA,KAAK,QAAQ;EACb,MAAM,QAAQ,KAAK,IAAI,SAAS,MAAM;EACtC;EACA;CACF;AACF"}

package/dist/contract/index.d.ts

@@ -1,14 +1,14 @@
 import { c as CostLedgerHandle } from "../cost-ledger-CO7UyEcQ.js";
-import { _ as ProposalFinding, i as AnalystFinding, v as ProposalFindingOrigin, x as makeProposalFinding } from "../types-DHb5TI36.js";
+import { a as AnalystFinding, b as ProposalFinding, w as makeProposalFinding, x as ProposalFindingOrigin } from "../types-oKRhQAde.js";
 import { A as createChatClient, T as CreateChatClientOpts, x as ChatClient } from "../types-DBZ7keha.js";
-import { $ as paretoSignificanceGate, G as BuildEvidenceVectorOptions, J as ParetoSignificanceGateOptions, K as EvidenceVector, Q as paretoPolicy, U as AxisEvidence, W as AxisVerdict, X as PromotionPolicy, Y as PromotionObjective, Z as buildEvidenceVector, q as ObjectiveSource } from "../statistical-heldout-CJTj2EoS.js";
-import { C as JudgeDimension, H as SurfaceProposer, M as OptimizerConfig, R as Scenario, S as JudgeConfig, V as SessionScript, _ as GateDecision, a as CampaignResult, b as GenerationRecord, c as CampaignTraceWriter, d as DispatchContext, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, j as MutableSurface, l as CodeSurface, m as GateCheckStatus, n as CampaignArtifactWriter, p as Gate, r as CampaignCellResult, t as CampaignAggregates, v as GateResult, w as JudgeScore, y as GenerationCandidate } from "../types-CGEmVYOk.js";
-import { $i as RunImprovementLoopOptions, $r as GepaOptimizationRecipe, Bi as ReferenceEquivalenceJudgeInput, Ci as compareOptimizationMethods, Dr as DefaultProductionGateOptions, Er as DefaultProductionGateCheck, Fa as campaignSplitDigest, Gi as runReferenceEquivalenceJudge, Gr as SkillOptOptimizationMethodConfig, Hi as ReferenceEquivalenceJudgeResult, Ji as llmJudge, Jr as skillOptOptimizationMethod, Ki as LlmJudgeDimension, Kr as SkillOptRunnerCommand, Oa as FinalEvidencePolicy, Or as DefaultProductionRewardHackingOptions, Qi as transientDispatchFailure, Qr as GepaOptimizationMethodConfig, Ri as REFERENCE_EQUIVALENCE_INPUT_LIMITS, Si as OptimizationTokenUsage, Ti as ComparisonCost, Tr as heldOutGate, Ui as ReferenceEquivalenceScenario, Vi as ReferenceEquivalenceJudgeOptions, Wi as createReferenceEquivalenceJudge, Wr as composeGate, Xr as GepaEngineOptions, Yr as GepaAdaptiveEngineRun, Zr as GepaEngineRun, _a as runCampaign, _i as OptimizationMethodProvenance, ai as ExternalTextOptimizationMethodConfig, ci as ExternalOptimizationExample, di as CompareOptimizationMethodsOptions, ea as RunImprovementLoopResult, ei as GepaRunnerCommand, fa as RunEvalOptions, fi as OptimizationMethod, ga as RunCampaignOptions, ha as CampaignCellRetryPolicy, hi as OptimizationMethodInput, ii as externalTextOptimizationMethod, ka as FinalEvidenceUse, kr as defaultProductionGate, li as ExternalTextEvaluationResponse, ma as CampaignCellFailureReceipt, ni as OpenAICompatibleOptimizerModel, oi as ExternalTextOptimizerContext, pa as runEval, pi as OptimizationMethodComparison, qi as LlmJudgeOptions, qr as SkillOptTrainerConfig, ri as OptimizerModelBudget, si as ExternalTextOptimizerResult, ta as runImprovementLoop, ti as gepaOptimizationMethod, vi as OptimizationMethodResult, wr as HeldOutGateOptions, xi as OptimizationPackageSource, zi as REFERENCE_EQUIVALENCE_JUDGE_VERSION } from "../index-B3WBm953.js";
-import { $t as fsCampaignStorage, Zt as CampaignStorage, en as inMemoryCampaignStorage } from "../search-kernel-BqyBZ9Wm.js";
+import { $ as paretoSignificanceGate, G as BuildEvidenceVectorOptions, J as ParetoSignificanceGateOptions, K as EvidenceVector, Q as paretoPolicy, U as AxisEvidence, W as AxisVerdict, X as PromotionPolicy, Y as PromotionObjective, Z as buildEvidenceVector, q as ObjectiveSource } from "../statistical-heldout-BeR4bOdg.js";
+import { C as JudgeDimension, H as SurfaceProposer, M as OptimizerConfig, R as Scenario, S as JudgeConfig, V as SessionScript, _ as GateDecision, a as CampaignResult, b as GenerationRecord, c as CampaignTraceWriter, d as DispatchContext, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, j as MutableSurface, l as CodeSurface, m as GateCheckStatus, n as CampaignArtifactWriter, p as Gate, r as CampaignCellResult, t as CampaignAggregates, v as GateResult, w as JudgeScore, y as GenerationCandidate } from "../types-B4C5nFw6.js";
+import { $i as RunImprovementLoopOptions, $r as GepaOptimizationRecipe, Bi as ReferenceEquivalenceJudgeInput, Ci as compareOptimizationMethods, Dr as DefaultProductionGateOptions, Er as DefaultProductionGateCheck, Fa as campaignSplitDigest, Gi as runReferenceEquivalenceJudge, Gr as SkillOptOptimizationMethodConfig, Hi as ReferenceEquivalenceJudgeResult, Ji as llmJudge, Jr as skillOptOptimizationMethod, Ki as LlmJudgeDimension, Kr as SkillOptRunnerCommand, Oa as FinalEvidencePolicy, Or as DefaultProductionRewardHackingOptions, Qi as transientDispatchFailure, Qr as GepaOptimizationMethodConfig, Ri as REFERENCE_EQUIVALENCE_INPUT_LIMITS, Si as OptimizationTokenUsage, Ti as ComparisonCost, Tr as heldOutGate, Ui as ReferenceEquivalenceScenario, Vi as ReferenceEquivalenceJudgeOptions, Wi as createReferenceEquivalenceJudge, Wr as composeGate, Xr as GepaEngineOptions, Yr as GepaAdaptiveEngineRun, Zr as GepaEngineRun, _a as runCampaign, _i as OptimizationMethodProvenance, ai as ExternalTextOptimizationMethodConfig, ci as ExternalOptimizationExample, di as CompareOptimizationMethodsOptions, ea as RunImprovementLoopResult, ei as GepaRunnerCommand, fa as RunEvalOptions, fi as OptimizationMethod, ga as RunCampaignOptions, ha as CampaignCellRetryPolicy, hi as OptimizationMethodInput, ii as externalTextOptimizationMethod, ka as FinalEvidenceUse, kr as defaultProductionGate, li as ExternalTextEvaluationResponse, ma as CampaignCellFailureReceipt, ni as OpenAICompatibleOptimizerModel, oi as ExternalTextOptimizerContext, pa as runEval, pi as OptimizationMethodComparison, qi as LlmJudgeOptions, qr as SkillOptTrainerConfig, ri as OptimizerModelBudget, si as ExternalTextOptimizerResult, ta as runImprovementLoop, ti as gepaOptimizationMethod, vi as OptimizationMethodResult, wr as HeldOutGateOptions, xi as OptimizationPackageSource, zi as REFERENCE_EQUIVALENCE_JUDGE_VERSION } from "../index-NvGBiO8H.js";
+import { $t as fsCampaignStorage, Zt as CampaignStorage, en as inMemoryCampaignStorage } from "../search-kernel-CpcKH5Ih.js";
 import { i as InMemoryOutcomeStore, n as FileSystemOutcomeStore, o as OutcomeStore, r as FileSystemOutcomeStoreOptions, t as DeploymentOutcome } from "../outcome-store-CNt4iZ67.js";
 import { r as HostedTenant } from "../client-BzL9niqi.js";
-import { _ as ReleaseSummary, a as summarizeExecution, c as ExecutionInsight, d as InsightReport, f as InterRaterInsight, g as Recommendation, h as OutcomeCorrelationInsight, i as analyzeRuns, l as FailureClassTally, m as LiftInsight, n as ExecutionReport, o as CostProvenanceSummary, p as JudgeInsight, r as SummarizeExecutionOptions, s as ExecutionErrorOutcomeCell, t as AnalyzeRunsOptions, u as FailureClusterInsight, v as ScalarDistribution, y as TokenUsageInsight } from "../analyze-runs-BhFLOztS.js";
-import { _ as SelfImproveMethodResult, a as DefinedAgentEval, c as SelfImproveMethodOptions, d as SelfImproveProposerOptions, f as SelfImproveProposerResult, g as SelfImproveMethodProvenance, h as selfImprove, i as DefineAgentEvalOptions, l as SelfImproveOptions, m as SelfImproveRunError, n as AgentEvalEvaluateOptions, o as defineAgentEval, p as SelfImproveResult, r as AgentEvalImproveOptions, s as SelfImproveBudget, t as AgentEvalAgent, u as SelfImproveProgressEvent } from "../define-agent-eval-DfXvNho5.js";
-import { $ as FromRunRecordDirOptions, A as CodeAgentSessionActionKind, At as verifyCandidateExperimentComparison, B as AgentTraceConversation, C as fromOpenCodeSession, Ct as evaluatePairedMeasurements, D as parseCodeAgentJsonlFile, Dt as sealCandidateBenchmarkTask, E as parseCodeAgentJsonl, Et as sealCandidateBenchmarkSuite, F as CodeAgentSessionSource, G as AuthoringProvenance, H as AgentTraceIndex, I as CodeAgentSessionTerminalStatus, J as partitionRunsByAuthoringModel, K as PartitionByAuthoringModelResult, L as observeCodeAgentSession, M as CodeAgentSessionActionSurface, N as CodeAgentSessionExecutionReceipt, O as streamCodeAgentJsonlFile, Ot as sealCandidateExperiment, P as CodeAgentSessionObservation, Q as evalReportingSuite, R as AgentTraceContributor, S as fromKimiCodeSession, St as SealCandidateBenchmarkSuiteOptions, T as fromPiSession, Tt as runCandidateExperiment, U as AgentTraceRange, V as AgentTraceFile, W as AgentTraceRecord, X as EvalReportingSuiteOptions, Y as EvalReportingSuiteInput, Z as EvalReportingSuiteResult, _ as CodeAgentSessionIntakeResult, _t as EvaluatePairedMeasurementsOptions, a as FromFeedbackTableOptions, at as CompareAgentProfileImprovementExperimentOptions, b as fromClaudeCodeSession, bt as PairedMeasurementEvaluation, c as CodeAgentStoreObservation, ct as measuredComparisonFromAgentProfileImprovementExperiment, d as CodeAgentStoreScope, dt as sealAgentProfileImprovementSuite, et as FromRunRecordDirResult, f as CodeAgentStoreSession, ft as sealAgentProfileImprovementTask, g as CodeAgentSessionIntakeOptions, gt as CompareCandidateExperimentOptions, h as CodeAgentSessionDiagnostic, ht as CandidateExperimentRun, i as FeedbackTableRow, it as AgentProfileImprovementExperimentRun, j as CodeAgentSessionActionStatus, k as CodeAgentSessionAction, kt as verifyCandidateExperiment, l as CodeAgentStoreRejection, lt as runAgentProfileImprovementExperiment, m as CodeAgentJsonlLine, mt as CandidateExperimentExecutionInput, n as fromOtelSpans, nt as fromRunRecordDir, o as FromFeedbackTableResult, ot as RunAgentProfileImprovementExperimentOptions, p as observeCodeAgentStore, pt as verifyAgentProfileImprovementExperimentComparison, q as parseAgentTrace, r as FeedbackTableMeta, rt as AgentProfileImprovementExperimentExecutionInput, s as fromFeedbackTable, st as SealAgentProfileImprovementSuiteOptions, t as FromOtelSpansOptions, tt as RunRecordRejection, u as CodeAgentStoreRejections, ut as sealAgentProfileImprovementExperiment, v as CodeAgentSessionMetrics, vt as PairedMeasurement, w as fromPiGraphSession, wt as measuredComparisonFromCandidateExperiment, x as fromCodexSession, xt as RunCandidateExperimentOptions, y as ParsedCodeAgentJsonl, yt as PairedMeasurementAdapter, z as AgentTraceContributorType } from "../index-B7m2uzJL.js";
-import { n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-B4B7GMvI.js";
+import { _ as ReleaseSummary, a as summarizeExecution, c as ExecutionInsight, d as InsightReport, f as InterRaterInsight, g as Recommendation, h as OutcomeCorrelationInsight, i as analyzeRuns, l as FailureClassTally, m as LiftInsight, n as ExecutionReport, o as CostProvenanceSummary, p as JudgeInsight, r as SummarizeExecutionOptions, s as ExecutionErrorOutcomeCell, t as AnalyzeRunsOptions, u as FailureClusterInsight, v as ScalarDistribution, y as TokenUsageInsight } from "../analyze-runs-CSDAh8A7.js";
+import { _ as SelfImproveMethodResult, a as DefinedAgentEval, c as SelfImproveMethodOptions, d as SelfImproveProposerOptions, f as SelfImproveProposerResult, g as SelfImproveMethodProvenance, h as selfImprove, i as DefineAgentEvalOptions, l as SelfImproveOptions, m as SelfImproveRunError, n as AgentEvalEvaluateOptions, o as defineAgentEval, p as SelfImproveResult, r as AgentEvalImproveOptions, s as SelfImproveBudget, t as AgentEvalAgent, u as SelfImproveProgressEvent } from "../define-agent-eval-CEWkX5wZ.js";
+import { $ as FromRunRecordDirOptions, A as CodeAgentSessionActionKind, At as verifyCandidateExperimentComparison, B as AgentTraceConversation, C as fromOpenCodeSession, Ct as evaluatePairedMeasurements, D as parseCodeAgentJsonlFile, Dt as sealCandidateBenchmarkTask, E as parseCodeAgentJsonl, Et as sealCandidateBenchmarkSuite, F as CodeAgentSessionSource, G as AuthoringProvenance, H as AgentTraceIndex, I as CodeAgentSessionTerminalStatus, J as partitionRunsByAuthoringModel, K as PartitionByAuthoringModelResult, L as observeCodeAgentSession, M as CodeAgentSessionActionSurface, N as CodeAgentSessionExecutionReceipt, O as streamCodeAgentJsonlFile, Ot as sealCandidateExperiment, P as CodeAgentSessionObservation, Q as evalReportingSuite, R as AgentTraceContributor, S as fromKimiCodeSession, St as SealCandidateBenchmarkSuiteOptions, T as fromPiSession, Tt as runCandidateExperiment, U as AgentTraceRange, V as AgentTraceFile, W as AgentTraceRecord, X as EvalReportingSuiteOptions, Y as EvalReportingSuiteInput, Z as EvalReportingSuiteResult, _ as CodeAgentSessionIntakeResult, _t as EvaluatePairedMeasurementsOptions, a as FromFeedbackTableOptions, at as CompareAgentProfileImprovementExperimentOptions, b as fromClaudeCodeSession, bt as PairedMeasurementEvaluation, c as CodeAgentStoreObservation, ct as measuredComparisonFromAgentProfileImprovementExperiment, d as CodeAgentStoreScope, dt as sealAgentProfileImprovementSuite, et as FromRunRecordDirResult, f as CodeAgentStoreSession, ft as sealAgentProfileImprovementTask, g as CodeAgentSessionIntakeOptions, gt as CompareCandidateExperimentOptions, h as CodeAgentSessionDiagnostic, ht as CandidateExperimentRun, i as FeedbackTableRow, it as AgentProfileImprovementExperimentRun, j as CodeAgentSessionActionStatus, k as CodeAgentSessionAction, kt as verifyCandidateExperiment, l as CodeAgentStoreRejection, lt as runAgentProfileImprovementExperiment, m as CodeAgentJsonlLine, mt as CandidateExperimentExecutionInput, n as fromOtelSpans, nt as fromRunRecordDir, o as FromFeedbackTableResult, ot as RunAgentProfileImprovementExperimentOptions, p as observeCodeAgentStore, pt as verifyAgentProfileImprovementExperimentComparison, q as parseAgentTrace, r as FeedbackTableMeta, rt as AgentProfileImprovementExperimentExecutionInput, s as fromFeedbackTable, st as SealAgentProfileImprovementSuiteOptions, t as FromOtelSpansOptions, tt as RunRecordRejection, u as CodeAgentStoreRejections, ut as sealAgentProfileImprovementExperiment, v as CodeAgentSessionMetrics, vt as PairedMeasurement, w as fromPiGraphSession, wt as measuredComparisonFromCandidateExperiment, x as fromCodexSession, xt as RunCandidateExperimentOptions, y as ParsedCodeAgentJsonl, yt as PairedMeasurementAdapter, z as AgentTraceContributorType } from "../index-DlH0Y9gI.js";
+import { n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-BrWh62zf.js";
 export { type AgentEvalAgent, type AgentEvalEvaluateOptions, type AgentEvalImproveOptions, type AgentProfileImprovementExperimentExecutionInput, type AgentProfileImprovementExperimentRun, type AgentTraceContributor, type AgentTraceContributorType, type AgentTraceConversation, type AgentTraceFile, type AgentTraceIndex, type AgentTraceRange, type AgentTraceRecord, type AnalystFinding, type AnalyzeRunsOptions, type AuthoringProvenance, type AxisEvidence, type AxisVerdict, type BuildEvidenceVectorOptions, type CampaignAggregates, type CampaignArtifactWriter, type CampaignCellFailureReceipt, type CampaignCellResult, type CampaignCellRetryPolicy, type CampaignCostMeter, type CampaignResult, type CampaignStorage, type CampaignTraceWriter, type CandidateExperimentExecutionInput, type CandidateExperimentRun, type ChatClient, type CodeAgentJsonlLine, type CodeAgentSessionAction, type CodeAgentSessionActionKind, type CodeAgentSessionActionStatus, type CodeAgentSessionActionSurface, type CodeAgentSessionDiagnostic, type CodeAgentSessionExecutionReceipt, type CodeAgentSessionIntakeOptions, type CodeAgentSessionIntakeResult, type CodeAgentSessionMetrics, type CodeAgentSessionObservation, type CodeAgentSessionSource, type CodeAgentSessionTerminalStatus, type CodeAgentStoreObservation, type CodeAgentStoreRejection, type CodeAgentStoreRejections, type CodeAgentStoreScope, type CodeAgentStoreSession, type CodeSurface, type CompareAgentProfileImprovementExperimentOptions, type CompareCandidateExperimentOptions, type CompareOptimizationMethodsOptions, type ComparisonCost, type CostLedgerHandle, type CostProvenanceSummary, type CreateChatClientOpts, type DefaultAnalystRegistryOptions, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, type DefineAgentEvalOptions, type DefinedAgentEval, type DeploymentOutcome, type DispatchFn as Dispatch, type DispatchContext, type EvalReportingSuiteInput, type EvalReportingSuiteOptions, type EvalReportingSuiteResult, type EvaluatePairedMeasurementsOptions, type EvidenceVector, type ExecutionErrorOutcomeCell, type ExecutionInsight, type ExecutionReport, type ExternalOptimizationExample, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, type FailureClassTally, type FailureClusterInsight, type FeedbackTableMeta, type FeedbackTableRow, FileSystemOutcomeStore, type FileSystemOutcomeStoreOptions, type FinalEvidencePolicy, type FinalEvidenceUse, type FromFeedbackTableOptions, type FromFeedbackTableResult, type FromOtelSpansOptions, type FromRunRecordDirOptions, type FromRunRecordDirResult, type Gate, type GateCheckStatus, type GateContext, type GateContribution, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaAdaptiveEngineRun, type GepaEngineOptions, type GepaEngineRun, type GepaOptimizationMethodConfig, type GepaOptimizationRecipe, type GepaRunnerCommand, type HeldOutGateOptions, type HostedTenant, InMemoryOutcomeStore, type InsightReport, type InterRaterInsight, type JudgeConfig, type JudgeDimension, type JudgeInsight, type JudgeScore, type LiftInsight, type LlmJudgeDimension, type LlmJudgeOptions, type MutableSurface, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodInput, type OptimizationMethodProvenance, type OptimizationMethodResult, type OptimizationPackageSource, type OptimizationTokenUsage, type OptimizerConfig, type OptimizerModelBudget, type OutcomeCorrelationInsight, type OutcomeStore, type PairedMeasurement, type PairedMeasurementAdapter, type PairedMeasurementEvaluation, type ParetoSignificanceGateOptions, type ParsedCodeAgentJsonl, type PartitionByAuthoringModelResult, type PromotionObjective, type PromotionPolicy, type ProposalFinding, type ProposalFindingOrigin, REFERENCE_EQUIVALENCE_INPUT_LIMITS, REFERENCE_EQUIVALENCE_JUDGE_VERSION, type Recommendation, type ReferenceEquivalenceJudgeInput, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceJudgeResult, type ReferenceEquivalenceScenario, type ReleaseSummary, type RunAgentProfileImprovementExperimentOptions, type RunCampaignOptions, type RunCandidateExperimentOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunRecordRejection, type ScalarDistribution, type Scenario, type SealAgentProfileImprovementSuiteOptions, type SealCandidateBenchmarkSuiteOptions, type SelfImproveBudget, type SelfImproveMethodOptions, type SelfImproveMethodProvenance, type SelfImproveMethodResult, type SelfImproveOptions, type SelfImproveProgressEvent, type SelfImproveProposerOptions, type SelfImproveProposerResult, type SelfImproveResult, SelfImproveRunError, type SessionScript, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SummarizeExecutionOptions, type SurfaceProposer, type TokenUsageInsight, analyzeRuns, buildDefaultAnalystRegistry, buildEvidenceVector, campaignSplitDigest, compareOptimizationMethods, composeGate, createChatClient, createReferenceEquivalenceJudge, defaultProductionGate, defineAgentEval, evalReportingSuite, evaluatePairedMeasurements, externalTextOptimizationMethod, fromClaudeCodeSession, fromCodexSession, fromFeedbackTable, fromKimiCodeSession, fromOpenCodeSession, fromOtelSpans, fromPiGraphSession, fromPiSession, fromRunRecordDir, fsCampaignStorage, gepaOptimizationMethod, heldOutGate, inMemoryCampaignStorage, llmJudge, makeProposalFinding, measuredComparisonFromAgentProfileImprovementExperiment, measuredComparisonFromCandidateExperiment, observeCodeAgentSession, observeCodeAgentStore, paretoPolicy, paretoSignificanceGate, parseAgentTrace, parseCodeAgentJsonl, parseCodeAgentJsonlFile, partitionRunsByAuthoringModel, runAgentProfileImprovementExperiment, runCampaign, runCandidateExperiment, runEval, runImprovementLoop, runReferenceEquivalenceJudge, sealAgentProfileImprovementExperiment, sealAgentProfileImprovementSuite, sealAgentProfileImprovementTask, sealCandidateBenchmarkSuite, sealCandidateBenchmarkTask, sealCandidateExperiment, selfImprove, skillOptOptimizationMethod, streamCodeAgentJsonlFile, summarizeExecution, transientDispatchFailure, verifyAgentProfileImprovementExperimentComparison, verifyCandidateExperiment, verifyCandidateExperimentComparison };

package/dist/contract/index.js

@@ -4,19 +4,19 @@ import { t as decidePairedPromotion } from "../paired-promotion-decision-BN6IQTi
 import { a as heldoutSignificance } from "../statistical-heldout-1Gbmex7v.js";
 import { n as summarizeExecution, t as analyzeRuns } from "../analyze-runs-3xpby3p2.js";
 import { a as parseRunRecordSafe } from "../run-record-DlTZ_lqe.js";
-import { K as defaultProductionGate, a as transientDispatchFailure, c as compareOptimizationMethods, h as runImprovementLoop, nt as runEval, rt as runCampaign, t as llmJudge } from "../llm-judge-Cd5s5Ei6.js";
+import { K as defaultProductionGate, a as transientDispatchFailure, c as compareOptimizationMethods, h as runImprovementLoop, nt as runEval, rt as runCampaign, t as llmJudge } from "../llm-judge-PXV7Xljl.js";
 import { i as isModelPriced, n as estimateCost } from "../metrics-BKaPBtIG.js";
 import { i as CostLedger } from "../cost-ledger-D2FnJPBF.js";
 import { G as campaignSplitDigest } from "../campaign-evidence-OcUciD-N.js";
 import { h as mapConcurrentRange } from "../journal-DvKCvjty.js";
-import { C as inMemoryCampaignStorage, S as fsCampaignStorage } from "../external-optimizer-subprocess-BUxifb8g.js";
-import { n as SelfImproveRunError, r as selfImprove, t as defineAgentEval } from "../define-agent-eval-BdfnqXeP.js";
-import { t as buildDefaultAnalystRegistry } from "../default-registry-KP4-Tolz.js";
+import { F as inMemoryCampaignStorage, P as fsCampaignStorage } from "../external-optimizer-subprocess-DcerlZSG.js";
+import { o as makeProposalFinding } from "../types-IXD2HJzu.js";
+import { n as SelfImproveRunError, r as selfImprove, t as defineAgentEval } from "../define-agent-eval-Ch3MWX6k.js";
+import { t as buildDefaultAnalystRegistry } from "../default-registry-Cn1LQ2Ve.js";
 import { t as extractUsage } from "../extract-usage-BrnzRJ6H.js";
-import { i as makeProposalFinding } from "../types-CW5b90JH.js";
 import { t as createChatClient } from "../chat-client-DthPOYGd.js";
 import { n as InMemoryOutcomeStore, t as FileSystemOutcomeStore } from "../outcome-store-DUcMj-3D.js";
-import { c as heldOutGate, g as runReferenceEquivalenceJudge, h as createReferenceEquivalenceJudge, l as composeGate, m as REFERENCE_EQUIVALENCE_JUDGE_VERSION, n as gepaOptimizationMethod, p as REFERENCE_EQUIVALENCE_INPUT_LIMITS, t as skillOptOptimizationMethod, u as externalTextOptimizationMethod } from "../skillopt-optimization-method-Dcepr6QL.js";
+import { c as heldOutGate, g as runReferenceEquivalenceJudge, h as createReferenceEquivalenceJudge, l as composeGate, m as REFERENCE_EQUIVALENCE_JUDGE_VERSION, n as gepaOptimizationMethod, p as REFERENCE_EQUIVALENCE_INPUT_LIMITS, t as skillOptOptimizationMethod, u as externalTextOptimizationMethod } from "../skillopt-optimization-method-CCtzralq.js";
 import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-CWW35UWq.js";
 import { t as fromOtelSpans } from "../otel-spans-C8YIJWy1.js";
 import { createHash } from "node:crypto";

package/dist/{default-registry-B4B7GMvI.d.ts → default-registry-BrWh62zf.d.ts}

@@ -1,8 +1,8 @@
-import { O as TraceAnalysisStore, i as AnalystFinding, n as AnalystContext } from "./types-DHb5TI36.js";
-import { l as ExactCapableAnalyst } from "./exact-types-Dlb73wN4.js";
-import { n as AnalystRegistry, r as AnalystRegistryOptions } from "./registry-CMYIaDgS.js";
-import { a as TraceAnalystLimits, i as TraceAnalysisEngineResult, l as RawAnalystFinding, n as TraceAnalysisEngine } from "./engine-C37HS2kr.js";
-import { t as TraceToolGroupName } from "./tool-groups-BKRrtqy4.js";
+import { X as TraceAnalysisStore, a as AnalystFinding, r as AnalystContext } from "./types-oKRhQAde.js";
+import { l as ExactCapableAnalyst } from "./exact-types-CD0lf5Qd.js";
+import { i as AnalystRegistryOptions, t as AnalystRegistry } from "./registry-JYfPxcVz.js";
+import { a as TraceAnalystLimits, i as TraceAnalysisEngineResult, l as RawAnalystFinding, n as TraceAnalysisEngine } from "./engine-C_zNWfM8.js";
+import { t as TraceToolGroupName } from "./tool-groups-mcxOXVlb.js";
 //#region src/trace-analyst/behavioral-metrics.d.ts
 type SuboptimalCode = 'monotonic-input-growth' | 'output-length-decay' | 'single-tool-dependency' | 'no-self-verification';
 interface SuboptimalSignal {
@@ -110,4 +110,4 @@ interface DefaultAnalystRegistryOptions {
 declare function buildDefaultAnalystRegistry(options?: DefaultAnalystRegistryOptions): AnalystRegistry;
 //#endregion
 export { createTraceAnalyst as a, BehavioralAnalystOptions as c, TraceAnalystDefinition as i, behavioralAnalyst as l, buildDefaultAnalystRegistry as n, renderPriorFindings as o, CreateTraceAnalystOptions as r, runTraceAnalyst as s, DefaultAnalystRegistryOptions as t, deriveEfficiencyFindings as u };
-//# sourceMappingURL=default-registry-B4B7GMvI.d.ts.map
+//# sourceMappingURL=default-registry-BrWh62zf.d.ts.map

package/dist/{default-registry-B4B7GMvI.d.ts.map → default-registry-BrWh62zf.d.ts.map}

@@ -1 +1 @@
-{"version":3,"file":"default-registry-B4B7GMvI.d.ts","names":[],"sources":["../src/trace-analyst/behavioral-metrics.ts","../src/analyst/behavioral-analyst.ts","../src/analyst/kind-factory.ts","../src/analyst/default-registry.ts"],"mappings":";;;;;;KAoBY;UAMK;EACf,MAAM;EACN;;EAEA;;EAEA,UAAU;;UAGK;;EAEf;EACA;;EAEA,gBAAgB;;EAEhB;EACA;EACA,eAAe;EACf;EACA;;EAEA;EACA;EACA,SAAS;;UAGM;EACf;EACA;EACA,sBAAsB;EACtB,uBAAuB;;;;UC1BR;;EAEf;;EAEA;;;;;;iBA+Dc,yBACd,SAAS,mBACT;EAAQ;EAAoB;IAC3B;;iBAkCa,kBACd,UAAS,2BACR,oBAAoB;;;;UCrHN;EACf;EACA;EACA;EACA;;EAEA,sBAAsB,SAAS;;EAE/B;;EAEA,WAAW;;EAEX,kBACE,OAAO,oBACP,SAAS,wCACe;EAC1B,SAAS,QAAQ;EACjB,eAAe,KAAK,mBAAmB,SAAS,mBAAmB;EACnE;EACA;;UAGe;EACf,QAAQ;;EAER;EACA;;;iBAIoB,gBAAgB;EACpC,YAAY;EACZ,QAAQ;EACR,OAAO;EACP,SAAS;EACT;IACE,QAAQ;;iBAgHI,mBACd,YAAY,wBACZ,SAAS,4BACR,oBAAoB;iBA2RP,oBAAoB,OAAO;;;UChc1B;;;;;;EAMf,SAAS;EACT,uBAAuB;;EAEvB;EACA,aAAa;EACb,WAAW;;iBAGG,4BACd,UAAS,gCACR"}
+{"version":3,"file":"default-registry-BrWh62zf.d.ts","names":[],"sources":["../src/trace-analyst/behavioral-metrics.ts","../src/analyst/behavioral-analyst.ts","../src/analyst/kind-factory.ts","../src/analyst/default-registry.ts"],"mappings":";;;;;;KAoBY;UAMK;EACf,MAAM;EACN;;EAEA;;EAEA,UAAU;;UAGK;;EAEf;EACA;;EAEA,gBAAgB;;EAEhB;EACA;EACA,eAAe;EACf;EACA;;EAEA;EACA;EACA,SAAS;;UAGM;EACf;EACA;EACA,sBAAsB;EACtB,uBAAuB;;;;UC1BR;;EAEf;;EAEA;;;;;;iBA+Dc,yBACd,SAAS,mBACT;EAAQ;EAAoB;IAC3B;;iBAkCa,kBACd,UAAS,2BACR,oBAAoB;;;;UCrHN;EACf;EACA;EACA;EACA;;EAEA,sBAAsB,SAAS;;EAE/B;;EAEA,WAAW;;EAEX,kBACE,OAAO,oBACP,SAAS,wCACe;EAC1B,SAAS,QAAQ;EACjB,eAAe,KAAK,mBAAmB,SAAS,mBAAmB;EACnE;EACA;;UAGe;EACf,QAAQ;;EAER;EACA;;;iBAIoB,gBAAgB;EACpC,YAAY;EACZ,QAAQ;EACR,OAAO;EACP,SAAS;EACT;IACE,QAAQ;;iBAgHI,mBACd,YAAY,wBACZ,SAAS,4BACR,oBAAoB;iBA2RP,oBAAoB,OAAO;;;UChc1B;;;;;;EAMf,SAAS;EACT,uBAAuB;;EAEvB;EACA,aAAa;EACb,WAAW;;iBAGG,4BACd,UAAS,gCACR"}