@tangle-network/agent-eval 0.103.0 → 0.103.1

package/dist/{insight-report-k0sRTzKg.d.ts → insight-report-DumCfEur.d.ts}

@@ -1,5 +1,5 @@
-import { G as GainDistributionBin, P as ParetoFigureSpec } from './summary-report-C0nnxOD8.js';
-import { C as ContinuousAgreement } from './judge-calibration-0p2QcWNE.js';
+import { G as GainDistributionBin, P as ParetoFigureSpec } from './summary-report-Fc_YFJat.js';
+import { C as ContinuousAgreement } from './judge-calibration-7C-IDmKr.js';
 
 /**
  * # InsightReport — the rigorous decision packet for any set of agent runs.

package/dist/{judge-calibration-0p2QcWNE.d.ts → judge-calibration-7C-IDmKr.d.ts}

@@ -45,6 +45,9 @@ interface CalibrationResult {
         delta: number;
     }>;
 }
+/**
+ * Measure judge quality against human gold labels: computes Cohen's κ, Pearson correlation, and MAE over matched item ids.
+ */
 declare function calibrateJudge(golden: GoldenItem[], candidate: CandidateScore[]): CalibrationResult;
 interface PositionalBiasResult {
     /**

package/dist/{kind-factory-D0nk7AKV.d.ts → kind-factory-BLHxwX71.d.ts}

@@ -1,7 +1,7 @@
 import { AxAIService, AxFunction } from '@ax-llm/ax';
 import { T as TraceAnalysisStore } from './store-C1YxJDEK.js';
 import { z } from 'zod';
-import { g as AnalystCost, b as AnalystContext, a as Analyst } from './types-Dz9cKF0g.js';
+import { g as AnalystCost, b as AnalystContext, a as Analyst } from './types-DA9yj-Jd.js';
 
 /**
  * Typed Ax output for analyst findings.

package/dist/meta-eval/index.d.ts

@@ -1,12 +1,12 @@
 export { C as CalibrationBin, a as CalibrationOptions, b as CalibrationPair, c as CalibrationReport, d as CorrelationResult, e as CorrelationStudyOptions, f as CorrelationStudyResult, E as EvalMetricSpec, O as OutcomePair, g as calibrationCurve, h as calibrationFromPairs, i as correlationStudy } from '../calibration-BPmzuVPk.js';
 export { D as DeploymentOutcome, F as FileSystemOutcomeStore, a as FileSystemOutcomeStoreOptions, I as InMemoryOutcomeStore, O as OutcomeFilter, b as OutcomeStore } from '../outcome-store-rnXLEqSn.js';
-export { R as RubricOutcomePair, a as RubricPredictiveValidityInput, b as RubricPredictiveValidityReport, c as RubricRanking, r as rubricPredictiveValidity } from '../rubric-predictive-validity-0MdjTt8R.js';
-import { C as ContinuousAgreement, a as CalibrationResult, b as ContinuousCalibrationResult, c as CandidateScore, G as GoldenItem } from '../judge-calibration-0p2QcWNE.js';
+export { R as RubricOutcomePair, a as RubricPredictiveValidityInput, b as RubricPredictiveValidityReport, c as RubricRanking, r as rubricPredictiveValidity } from '../rubric-predictive-validity-ZGIlJNce.js';
+import { C as ContinuousAgreement, a as CalibrationResult, b as ContinuousCalibrationResult, c as CandidateScore, G as GoldenItem } from '../judge-calibration-7C-IDmKr.js';
 import { S as SeriesConvergenceOptions, a as SeriesConvergenceResult } from '../series-convergence-D5OWMBg6.js';
-import { C as CorpusAgreementReport } from '../statistics-xP-cWc5k.js';
+import { C as CorpusAgreementReport } from '../statistics-D88peojY.js';
 import '../store-BcFXE6LG.js';
 import '../schema-m0gsnbt3.js';
-import '../run-record-MRdJ-Kq2.js';
+import '../run-record-CPfd1ARZ.js';
 import '@tangle-network/agent-interface';
 import '../errors-CzMUYo7b.js';
 import '../types-C7DGg5ex.js';

package/dist/meta-eval/index.js

@@ -11,15 +11,15 @@ import {
 } from "../chunk-3RF76KTD.js";
 import {
   rubricPredictiveValidity
-} from "../chunk-NF7OZ4J7.js";
+} from "../chunk-QCBB6ZIU.js";
 import {
   pearsonR,
   spearmanR
-} from "../chunk-HRGUJTER.js";
+} from "../chunk-HYGRFL7C.js";
 import {
   aggregateLlm,
   llmSpans
-} from "../chunk-UFSG7ACU.js";
+} from "../chunk-JZXGWLK5.js";
 import "../chunk-5BKGXME7.js";
 import {
   ValidationError

package/dist/{multi-layer-verifier-DUZXrPDA.d.ts → multi-layer-verifier-CI4jdX-q.d.ts}

@@ -138,6 +138,9 @@ declare function gradeSemanticStatus(input: {
     available: boolean;
     threshold?: number;
 }): LayerStatus;
+/**
+ * Ordered DAG of verification layers with dependency-based skipping, per-layer findings, soft-fail semantics, and a blended composite score across all passed layers.
+ */
 declare class MultiLayerVerifier<Env = unknown> {
     private readonly layers;
     constructor(layers: Layer<Env>[]);

package/dist/multishot/index.d.ts

@@ -1,7 +1,7 @@
-import { J as JudgeScore } from '../types-DFI_Z-ZL.js';
+import { J as JudgeScore } from '../types-Bihq6-a3.js';
 import { AgentProfile } from '@tangle-network/agent-interface';
 import { M as MatrixResult } from '../types-BUxNaJ8c.js';
-import '../run-record-MRdJ-Kq2.js';
+import '../run-record-CPfd1ARZ.js';
 import '../errors-CzMUYo7b.js';
 import '../schema-m0gsnbt3.js';
 import '../verdict-C9MlYujm.js';

package/dist/openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "@tangle-network/agent-eval — wire protocol",
-    "version": "0.103.0",
+    "version": "0.103.1",
     "description": "HTTP and stdio RPC interface to agent-eval. The TypeScript runtime is the source of truth; this spec is the contract that cross-language clients (Python, Rust, Go) generate from.\n\nWire-protocol version: 1.0.0. Bumps on breaking changes to request/response schemas.",
     "contact": {
       "name": "Tangle Network",

package/dist/pipelines/index.d.ts

@@ -4,7 +4,7 @@ export { a as FailureCluster, F as FailureClusterReport, f as failureClusterView
 import { a as TrajectoryStep } from '../trajectory-2TkpSEVh.js';
 import { B as BaselineOptions, a as BaselineReport } from '../baseline-Bbid3WoO.js';
 export { c as computeToolUseMetrics } from '../baseline-Bbid3WoO.js';
-import { l as llmSpans } from '../query-B7GGjRox.js';
+import { l as llmSpans } from '../query-0aTmbmQe.js';
 
 /**
  * BudgetBreachView — aggregates breach events across the corpus.

package/dist/pipelines/index.js

@@ -3,21 +3,21 @@ import {
   classifyFailure,
   compareToBaseline,
   computeToolUseMetrics
-} from "../chunk-IH7LYRHL.js";
+} from "../chunk-BXZVBX3D.js";
 import {
   buildTrajectory
 } from "../chunk-RZTMDUO7.js";
 import {
   interRaterReliability,
   pearsonR
-} from "../chunk-HRGUJTER.js";
+} from "../chunk-HYGRFL7C.js";
 import {
   aggregateLlm,
   argHash,
   llmSpans,
   runFailureClass,
   toolSpans
-} from "../chunk-UFSG7ACU.js";
+} from "../chunk-JZXGWLK5.js";
 import "../chunk-5BKGXME7.js";
 import "../chunk-3BFEG2F6.js";
 import "../chunk-PZ5AY32C.js";

package/dist/{policy-edit-BDQzzsBU.d.ts → policy-edit-nUhuFtLF.d.ts}

@@ -1,6 +1,6 @@
-import { A as AgentProfileCell, a as AgentProfileJson } from './run-record-MRdJ-Kq2.js';
+import { A as AgentProfileCell, a as AgentProfileJson } from './run-record-CPfd1ARZ.js';
 import { V as ValidationError } from './errors-CzMUYo7b.js';
-import { A as AnalystFinding, E as EvidenceRef } from './types-Dz9cKF0g.js';
+import { A as AnalystFinding, E as EvidenceRef } from './types-DA9yj-Jd.js';
 
 type PolicyEditSchemaVersion = 'policy-edit/v1';
 declare const POLICY_EDIT_AXES: readonly ["carrier", "representation", "budget", "sampling", "output_contract", "tool_contract", "routing", "memory", "agent_profile", "deployment_target"];

package/dist/{pre-registration-mWG2w8d-.d.ts → pre-registration-CUOSGAZK.d.ts}

@@ -1,7 +1,7 @@
 import { A as AgentEvalError } from './errors-CzMUYo7b.js';
-import { R as RunRecord } from './run-record-MRdJ-Kq2.js';
-import { C as ChatClient } from './types-Dz9cKF0g.js';
-import { c as JudgeDimension, S as Scenario, a as JudgeConfig } from './types-DFI_Z-ZL.js';
+import { R as RunRecord } from './run-record-CPfd1ARZ.js';
+import { C as ChatClient } from './types-DA9yj-Jd.js';
+import { c as JudgeDimension, S as Scenario, a as JudgeConfig } from './types-Bihq6-a3.js';
 import { TCloud } from '@tangle-network/tcloud';
 import { R as RawProviderSink } from './raw-provider-sink-C46HDghv.js';
 import { D as DefaultVerdict } from './verdict-C9MlYujm.js';

package/dist/product-benchmark/index.d.ts

@@ -1,4 +1,4 @@
-import { R as RunRecord } from '../run-record-MRdJ-Kq2.js';
+import { R as RunRecord } from '../run-record-CPfd1ARZ.js';
 import '@tangle-network/agent-interface';
 import '../errors-CzMUYo7b.js';
 import '../schema-m0gsnbt3.js';

package/dist/{provenance-BhJm32vN.d.ts → provenance-B2VsP0jP.d.ts}

@@ -1,9 +1,9 @@
-import { S as Scenario, g as Gate, G as GateResult, h as GateContext, C as CampaignResult, i as Mutator, f as SurfaceProposer, M as MutableSurface, j as GateDecision } from './types-DFI_Z-ZL.js';
+import { S as Scenario, g as Gate, G as GateResult, h as GateContext, C as CampaignResult, i as Mutator, f as SurfaceProposer, M as MutableSurface, j as GateDecision } from './types-Bihq6-a3.js';
 import { R as RedTeamCase } from './red-team-BWdoyleI.js';
-import { R as RunRecord } from './run-record-MRdJ-Kq2.js';
+import { R as RunRecord } from './run-record-CPfd1ARZ.js';
 import { D as Direction } from './pareto-E-pembql.js';
-import { a as PairedBootstrapResult } from './statistics-xP-cWc5k.js';
-import { R as RunCampaignOptions, C as CampaignStorage } from './gepa-DeyPTlvx.js';
+import { a as PairedBootstrapResult } from './statistics-D88peojY.js';
+import { R as RunCampaignOptions, C as CampaignStorage } from './gepa-COlCAkHN.js';
 import { HostedClient, TraceSpanEvent } from './hosted/index.js';
 
 /**
@@ -72,9 +72,13 @@ interface DefaultProductionGateOptions {
      *  fires at the `gaming` severity. Default true. */
     blockOnRewardHackingGaming?: boolean;
 }
+/**
+ * Opinionated production gate composing held-out significance, red-team, reward-hacking, and canary checks into a single `Gate.decide` decision.
+ */
 declare function defaultProductionGate<TArtifact, TScenario extends Scenario>(options: DefaultProductionGateOptions): Gate<TArtifact, TScenario>;
 
 /**
+ * @module
  * Thin Gate adapter — exposes delta-threshold-on-holdout as a composable
  * `Gate`. Use when you want held-out as one of N composed gates instead of
  * the full `defaultProductionGate` stack.
@@ -84,6 +88,9 @@ interface HeldOutGateOptions<TScenario extends Scenario = Scenario> {
     scenarios: TScenario[];
     deltaThreshold?: number;
 }
+/**
+ * Composable held-out delta gate: ships only when the candidate's mean composite on `scenarios` beats the baseline by at least `deltaThreshold`.
+ */
 declare function heldOutGate<TArtifact, TScenario extends Scenario>(options: HeldOutGateOptions<TScenario>): Gate<TArtifact, TScenario>;
 
 /**
@@ -220,6 +227,9 @@ declare function paretoSignificanceGate<TArtifact = unknown, TScenario extends S
 interface RunEvalOptions<TScenario extends Scenario, TArtifact> extends Omit<RunCampaignOptions<TScenario, TArtifact>, 'runDir'> {
     runDir: string;
 }
+/**
+ * Simplest evaluation preset: run scenarios through dispatch, score with judges, and return a `CampaignResult` — no optimizer, no gate, no PR.
+ */
 declare function runEval<TScenario extends Scenario, TArtifact>(opts: RunEvalOptions<TScenario, TArtifact>): Promise<CampaignResult<TArtifact, TScenario>>;
 
 /**
@@ -240,6 +250,9 @@ interface EvolutionaryProposerOptions<TFindings = unknown> {
     /** External findings fed to the mutator each generation. Default: []. */
     findings?: TFindings[];
 }
+/**
+ * Wrap a stateless `Mutator` (GEPA, AxGEPA, reflective-mutation) as a `SurfaceProposer` that mutates the current best surface into N candidates each generation.
+ */
 declare function evolutionaryProposer<TFindings = unknown>(opts: EvolutionaryProposerOptions<TFindings>): SurfaceProposer<TFindings>;
 
 /**
@@ -396,6 +409,9 @@ declare function loopProvenanceSpans(record: LoopProvenanceRecord, opts?: {
 }): TraceSpanEvent[];
 /** Canonical durable paths under the run dir. */
 declare function provenanceRecordPath(runDir: string): string;
+/**
+ * Canonical path for the durable OTLP spans JSONL file under a loop run directory.
+ */
 declare function provenanceSpansPath(runDir: string): string;
 interface EmitLoopProvenanceResult {
     record: LoopProvenanceRecord;

package/dist/{query-B7GGjRox.d.ts → query-0aTmbmQe.d.ts}

@@ -13,6 +13,7 @@ import { T as TraceStore } from './store-BcFXE6LG.js';
 declare function runsForScenario(store: TraceStore, scenarioId: string): Promise<Run[]>;
 declare function llmSpans(store: TraceStore, runId?: string): Promise<LlmSpan[]>;
 declare function toolSpans(store: TraceStore, runId?: string, toolName?: string): Promise<ToolSpan[]>;
+/** Query judge-kind spans from the trace store, optionally scoped to a single run. */
 declare function judgeSpans(store: TraceStore, runId?: string): Promise<JudgeSpan[]>;
 /** Group spans by any key selector. */
 declare function groupBy<T, K extends string | number>(items: T[], key: (t: T) => K): Map<K, T[]>;

package/dist/{release-report-BQ1Ziyu-.d.ts → release-report-DkaCZ9k4.d.ts}

@@ -1,6 +1,6 @@
 import { D as DatasetSplit, c as DatasetManifest, a as DatasetScenario } from './dataset-BbGkaN2I.js';
-import { m as GateDecision } from './summary-report-C0nnxOD8.js';
-import { R as RunRecord, b as RunSplitTag } from './run-record-MRdJ-Kq2.js';
+import { m as GateDecision } from './summary-report-Fc_YFJat.js';
+import { R as RunRecord, b as RunSplitTag } from './run-record-CPfd1ARZ.js';
 
 /**
  * Release confidence gate.
@@ -216,6 +216,9 @@ interface JudgeReplayGateArgs<TOutput> {
     /** Maximum concurrent judge calls. Default 4. */
     judgeConcurrency?: number;
 }
+/**
+ * Confirm a candidate's win with a stronger judge: score baseline and candidate outputs independently, then bootstrap a CI to verify the lift generalises beyond the inner loop.
+ */
 declare function judgeReplayGate<TOutput>(args: JudgeReplayGateArgs<TOutput>): Promise<BootstrapResult & {
     baselineSamples: number;
     candidateSamples: number;

package/dist/reporting.d.ts

@@ -1,15 +1,15 @@
-export { R as RubricOutcomePair, a as RubricPredictiveValidityInput, b as RubricPredictiveValidityReport, c as RubricRanking, r as rubricPredictiveValidity } from './rubric-predictive-validity-0MdjTt8R.js';
-export { B as BootstrapOptions, a as BootstrapResult, J as JudgeReplayGateArgs, R as ReleaseConfidenceAxis, b as ReleaseConfidenceAxisName, c as ReleaseConfidenceInput, d as ReleaseConfidenceIssue, e as ReleaseConfidenceMetrics, f as ReleaseConfidenceScorecard, g as ReleaseConfidenceStatus, h as ReleaseConfidenceThresholds, i as ReleaseTraceEvidence, j as RenderReleaseReportOptions, V as Verdict, k as assertReleaseConfidence, l as bootstrapCi, m as evaluateReleaseConfidence, n as judgeReplayGate, r as renderReleaseReport } from './release-report-BQ1Ziyu-.js';
+export { R as RubricOutcomePair, a as RubricPredictiveValidityInput, b as RubricPredictiveValidityReport, c as RubricRanking, r as rubricPredictiveValidity } from './rubric-predictive-validity-ZGIlJNce.js';
+export { B as BootstrapOptions, a as BootstrapResult, J as JudgeReplayGateArgs, R as ReleaseConfidenceAxis, b as ReleaseConfidenceAxisName, c as ReleaseConfidenceInput, d as ReleaseConfidenceIssue, e as ReleaseConfidenceMetrics, f as ReleaseConfidenceScorecard, g as ReleaseConfidenceStatus, h as ReleaseConfidenceThresholds, i as ReleaseTraceEvidence, j as RenderReleaseReportOptions, V as Verdict, k as assertReleaseConfidence, l as bootstrapCi, m as evaluateReleaseConfidence, n as judgeReplayGate, r as renderReleaseReport } from './release-report-DkaCZ9k4.js';
 export { I as InterimReleaseConfidence, a as InterimReleaseConfidenceInput, P as PairedEvalueOptions, b as PairedEvalueSequence, c as PairedEvalueStep, S as SequentialDecision, e as evaluateInterimReleaseConfidence, p as pairedEvalueSequence } from './sequential-5iSVfzl2.js';
-export { P as PairedBootstrapOptions, a as PairedBootstrapResult, b as benjaminiHochberg, p as pairedBootstrap, w as wilcoxonSignedRank } from './statistics-xP-cWc5k.js';
-export { G as GainDistributionBin, a as GainDistributionFigureSpec, b as GainDistributionOptions, P as ParetoFigureSpec, c as ParetoPoint, R as RESEARCH_REPORT_HARD_PAIR_FLOOR, d as ResearchReport, e as ResearchReportCandidate, f as ResearchReportDecision, g as ResearchReportMethodology, h as ResearchReportOptions, i as ResearchReportRecommendation, S as SummaryTable, j as SummaryTableOptions, k as SummaryTableRow, l as gainHistogram, p as paretoChart, r as researchReport, s as summaryTable } from './summary-report-C0nnxOD8.js';
-import './run-record-MRdJ-Kq2.js';
+export { P as PairedBootstrapOptions, a as PairedBootstrapResult, b as benjaminiHochberg, p as pairedBootstrap, w as wilcoxonSignedRank } from './statistics-D88peojY.js';
+export { G as GainDistributionBin, a as GainDistributionFigureSpec, b as GainDistributionOptions, P as ParetoFigureSpec, c as ParetoPoint, R as RESEARCH_REPORT_HARD_PAIR_FLOOR, d as ResearchReport, e as ResearchReportCandidate, f as ResearchReportDecision, g as ResearchReportMethodology, h as ResearchReportOptions, i as ResearchReportRecommendation, S as SummaryTable, j as SummaryTableOptions, k as SummaryTableRow, l as gainHistogram, p as paretoChart, r as researchReport, s as summaryTable } from './summary-report-Fc_YFJat.js';
+import './run-record-CPfd1ARZ.js';
 import '@tangle-network/agent-interface';
 import './errors-CzMUYo7b.js';
 import './schema-m0gsnbt3.js';
 import './outcome-store-rnXLEqSn.js';
 import './dataset-BbGkaN2I.js';
-import './judge-calibration-0p2QcWNE.js';
+import './judge-calibration-7C-IDmKr.js';
 import './types-C7DGg5ex.js';
 import '@tangle-network/tcloud';
 import './failure-cluster-DH9Flgcf.js';

package/dist/reporting.js

@@ -4,10 +4,10 @@ import {
   evaluateReleaseConfidence,
   judgeReplayGate,
   renderReleaseReport
-} from "./chunk-VDGPPGE3.js";
+} from "./chunk-UMEAR2FI.js";
 import {
   rubricPredictiveValidity
-} from "./chunk-NF7OZ4J7.js";
+} from "./chunk-QCBB6ZIU.js";
 import {
   evaluateInterimReleaseConfidence,
   pairedEvalueSequence
@@ -18,12 +18,12 @@ import {
   paretoChart,
   researchReport,
   summaryTable
-} from "./chunk-6GT4NI4V.js";
+} from "./chunk-CLS3374R.js";
 import {
   benjaminiHochberg,
   pairedBootstrap,
   wilcoxonSignedRank
-} from "./chunk-HRGUJTER.js";
+} from "./chunk-HYGRFL7C.js";
 import "./chunk-VSMTAMNK.js";
 import "./chunk-3BFEG2F6.js";
 import "./chunk-PZ5AY32C.js";

package/dist/{researcher-B_ODTAJs.d.ts → researcher-SDAezfML.d.ts}

@@ -1,6 +1,6 @@
-import { b as RunSplitTag, c as RunTokenUsage, d as RunJudgeMetadata, J as JudgeScoresRecord, A as AgentProfileCell, e as AgentProfileCellInput, R as RunRecord } from './run-record-MRdJ-Kq2.js';
+import { b as RunSplitTag, c as RunTokenUsage, d as RunJudgeMetadata, J as JudgeScoresRecord, A as AgentProfileCell, e as AgentProfileCellInput, R as RunRecord } from './run-record-CPfd1ARZ.js';
 import { L as LlmClientOptions, a as LlmRouteRequirements } from './llm-client-Bj7g0rqu.js';
-import { h as ResearchReportOptions, d as ResearchReport, m as GateDecision } from './summary-report-C0nnxOD8.js';
+import { h as ResearchReportOptions, d as ResearchReport, m as GateDecision } from './summary-report-Fc_YFJat.js';
 import { T as TraceEmitter, R as RunCompleteHook } from './emitter-C2rqGH_l.js';
 import { R as RunIntegrityExpectations, a as RunIntegrityReport } from './integrity-D2t12mMw.js';
 import { R as RawProviderSink } from './raw-provider-sink-C46HDghv.js';

package/dist/rl.d.ts

@@ -1,23 +1,23 @@
-import { R as RunRecord, b as RunSplitTag } from './run-record-MRdJ-Kq2.js';
+import { R as RunRecord, b as RunSplitTag } from './run-record-CPfd1ARZ.js';
 export { A as AdversarialMutation } from './adversarial-B7loGVVX.js';
-import { P as PreferenceExtractionReport, D as DpoExportRow, G as GrpoExportRow, S as SftExportRow, E as ExtractPreferencesOptions, a as DpoLookups, b as GrpoLookups, c as SftLookups } from './corpus-CiSzzLa5.js';
-export { d as CorpusAppendResult, C as CorpusRecord, e as DatasetFormat, f as ExtractStepRewardsOptions, H as HarvestOptions, g as PreferenceStrategy, h as PreferenceTriple, i as PrmExportRow, j as PrmLookups, k as PrmTrainingTriple, R as RewardKind, l as RewardStats, m as RlDatasetBundle, n as RlDatasetConfig, o as RlDatasetManifest, p as RlDatasetStats, q as RunwiseStepSummary, r as StepReward, s as StepRewardJsonlRow, t as StepScorer, u as appendToCorpus, v as buildDatasetFromCorpus, w as buildRlDataset, x as datasheetToMarkdown, y as extractPreferences, z as extractStepRewards, A as prmTrainingPairs, B as readCorpus, F as runwiseStepRewardSummary, I as stepRewardsToJsonl, J as toAnthropicFormat, K as toDpoJsonl, L as toDpoRows, M as toGrpoJsonl, N as toGrpoRows, O as toPrmJsonl, Q as toPrmRows, T as toSftJsonl, U as toSftRows, V as toTRLFormat } from './corpus-CiSzzLa5.js';
+import { P as PreferenceExtractionReport, D as DpoExportRow, G as GrpoExportRow, S as SftExportRow, E as ExtractPreferencesOptions, a as DpoLookups, b as GrpoLookups, c as SftLookups } from './corpus-CysLCiK4.js';
+export { d as CorpusAppendResult, C as CorpusRecord, e as DatasetFormat, f as ExtractStepRewardsOptions, H as HarvestOptions, g as PreferenceStrategy, h as PreferenceTriple, i as PrmExportRow, j as PrmLookups, k as PrmTrainingTriple, R as RewardKind, l as RewardStats, m as RlDatasetBundle, n as RlDatasetConfig, o as RlDatasetManifest, p as RlDatasetStats, q as RunwiseStepSummary, r as StepReward, s as StepRewardJsonlRow, t as StepScorer, u as appendToCorpus, v as buildDatasetFromCorpus, w as buildRlDataset, x as datasheetToMarkdown, y as extractPreferences, z as extractStepRewards, A as prmTrainingPairs, B as readCorpus, F as runwiseStepRewardSummary, I as stepRewardsToJsonl, J as toAnthropicFormat, K as toDpoJsonl, L as toDpoRows, M as toGrpoJsonl, N as toGrpoRows, O as toPrmJsonl, Q as toPrmRows, T as toSftJsonl, U as toSftRows, V as toTRLFormat } from './corpus-CysLCiK4.js';
 export { O as OffPolicyEstimate, a as OffPolicyOptions, b as OffPolicyTrajectory, d as doublyRobust, i as inverseProbabilityWeighting, o as offPolicyEstimateAll, s as selfNormalizedImportanceWeighting } from './off-policy-DiwuKKg7.js';
 import { b as OutcomeStore } from './outcome-store-rnXLEqSn.js';
 export { D as DeploymentOutcome, F as FileSystemOutcomeStore, a as FileSystemOutcomeStoreOptions, I as InMemoryOutcomeStore } from './outcome-store-rnXLEqSn.js';
-import { b as RubricPredictiveValidityReport } from './rubric-predictive-validity-0MdjTt8R.js';
-import { R as Researcher, F as FailureMode, S as SteeringChange, E as ExperimentPlan, a as ExperimentResult, b as EvalCampaignResult, c as EvalCampaignOptions } from './researcher-B_ODTAJs.js';
-export { r as runEvalCampaign } from './researcher-B_ODTAJs.js';
-import { a as VerificationReport } from './multi-layer-verifier-DUZXrPDA.js';
+import { b as RubricPredictiveValidityReport } from './rubric-predictive-validity-ZGIlJNce.js';
+import { R as Researcher, F as FailureMode, S as SteeringChange, E as ExperimentPlan, a as ExperimentResult, b as EvalCampaignResult, c as EvalCampaignOptions } from './researcher-SDAezfML.js';
+export { r as runEvalCampaign } from './researcher-SDAezfML.js';
+import { a as VerificationReport } from './multi-layer-verifier-CI4jdX-q.js';
 import { I as InterimReleaseConfidence } from './sequential-5iSVfzl2.js';
-import { C as CampaignResult } from './types-DFI_Z-ZL.js';
+import { C as CampaignResult } from './types-Bihq6-a3.js';
 import '@tangle-network/agent-interface';
 import './errors-CzMUYo7b.js';
 import './schema-m0gsnbt3.js';
 import './store-BcFXE6LG.js';
 import './llm-client-Bj7g0rqu.js';
 import './raw-provider-sink-C46HDghv.js';
-import './summary-report-C0nnxOD8.js';
+import './summary-report-Fc_YFJat.js';
 import './failure-cluster-DH9Flgcf.js';
 import './emitter-C2rqGH_l.js';
 import './integrity-D2t12mMw.js';

package/dist/rl.js

@@ -9,26 +9,26 @@ import {
   extractVerifiableReward,
   extractVerifiableRewardsFromRecords,
   filterDeterministicallyRewarded
-} from "./chunk-AIGWQEME.js";
+} from "./chunk-LOZOZYHU.js";
 import {
   FileSystemOutcomeStore,
   InMemoryOutcomeStore
 } from "./chunk-3RF76KTD.js";
 import {
   runEvalCampaign
-} from "./chunk-6FIAJHCU.js";
+} from "./chunk-6YVAN6R3.js";
 import "./chunk-CWNP4DV4.js";
 import {
   rubricPredictiveValidity
-} from "./chunk-NF7OZ4J7.js";
+} from "./chunk-QCBB6ZIU.js";
 import {
   evaluateInterimReleaseConfidence
 } from "./chunk-MAZ26DC7.js";
-import "./chunk-6GT4NI4V.js";
+import "./chunk-CLS3374R.js";
 import {
   benjaminiHochberg,
   wilcoxonSignedRank
-} from "./chunk-HRGUJTER.js";
+} from "./chunk-HYGRFL7C.js";
 import {
   observationsFromRunRecords,
   thompsonCurriculum,
@@ -36,9 +36,9 @@ import {
 } from "./chunk-VZSRQ272.js";
 import "./chunk-SBCB6VZY.js";
 import "./chunk-PC4UYEBM.js";
-import "./chunk-2NSLDY4B.js";
+import "./chunk-J5MUFXEY.js";
 import "./chunk-TVVP3ZZQ.js";
-import "./chunk-ABOIVNXL.js";
+import "./chunk-AGYDMORK.js";
 import "./chunk-VSMTAMNK.js";
 import {
   ValidationError

package/dist/{rubric-predictive-validity-0MdjTt8R.d.ts → rubric-predictive-validity-ZGIlJNce.d.ts}

@@ -1,4 +1,4 @@
-import { R as RunRecord } from './run-record-MRdJ-Kq2.js';
+import { R as RunRecord } from './run-record-CPfd1ARZ.js';
 import { b as OutcomeStore } from './outcome-store-rnXLEqSn.js';
 
 /**

package/dist/{run-campaign-2L4WCJHR.js → run-campaign-DAHKO5CT.js}

@@ -1,12 +1,12 @@
 import {
   planCampaignRun,
   runCampaign
-} from "./chunk-XKA6ZGEY.js";
-import "./chunk-HRGUJTER.js";
+} from "./chunk-VNM52AGA.js";
+import "./chunk-HYGRFL7C.js";
 import "./chunk-3BFEG2F6.js";
 import "./chunk-PZ5AY32C.js";
 export {
   planCampaignRun,
   runCampaign
 };
-//# sourceMappingURL=run-campaign-2L4WCJHR.js.map
+//# sourceMappingURL=run-campaign-DAHKO5CT.js.map

package/dist/{run-record-MRdJ-Kq2.d.ts → run-record-CPfd1ARZ.d.ts}

@@ -49,6 +49,9 @@ declare class AgentProfileCellValidationError extends ValidationError {
 }
 declare function buildAgentProfileCell(input: AgentProfileCellInput): Promise<AgentProfileCell>;
 declare function agentProfileCellHashMaterial(cell: AgentProfileCell): Omit<AgentProfileCell, 'cellId'>;
+/**
+ * Verify an `AgentProfileCell`'s `cellId` matches the sha256 of its hash-material fields, confirming the record has not been tampered with.
+ */
 declare function verifyAgentProfileCell(cell: AgentProfileCell): Promise<boolean>;
 declare function validateAgentProfileCell(input: unknown): AgentProfileCell;
 declare function requireAgentProfileCell(record: {

package/dist/{runtime-trajectory-8w0_jmtR.d.ts → runtime-trajectory-DZ8ei-Jo.d.ts}

@@ -1,4 +1,4 @@
-import { b as RunSplitTag } from './run-record-MRdJ-Kq2.js';
+import { b as RunSplitTag } from './run-record-CPfd1ARZ.js';
 
 interface RuntimeTrajectoryHookEvent {
     id: string;

package/dist/{semantic-concept-judge-D-IlH5v1.d.ts → semantic-concept-judge-C6mDEBIo.d.ts}

@@ -1,10 +1,10 @@
 import { AxAIService } from '@ax-llm/ax';
 import { z } from 'zod';
-import { A as AnalystFinding, a as Analyst, b as AnalystContext } from './types-Dz9cKF0g.js';
-import { T as TraceAnalystKindSpec } from './kind-factory-D0nk7AKV.js';
+import { A as AnalystFinding, a as Analyst, b as AnalystContext } from './types-DA9yj-Jd.js';
+import { T as TraceAnalystKindSpec } from './kind-factory-BLHxwX71.js';
 import { a as TraceAnalystSpan } from './store-C1YxJDEK.js';
 import { L as LlmClientOptions } from './llm-client-Bj7g0rqu.js';
-import { S as Severity } from './multi-layer-verifier-DUZXrPDA.js';
+import { S as Severity } from './multi-layer-verifier-CI4jdX-q.js';
 
 interface CreateAnalystAiConfig {
     /** OpenAI-compatible API key forwarded as `Authorization: Bearer`.

package/dist/{statistics-xP-cWc5k.d.ts → statistics-D88peojY.d.ts}

@@ -1,4 +1,4 @@
-import { d as ContinuousAgreementOptions, C as ContinuousAgreement } from './judge-calibration-0p2QcWNE.js';
+import { d as ContinuousAgreementOptions, C as ContinuousAgreement } from './judge-calibration-7C-IDmKr.js';
 import { J as JudgeScore } from './types-C7DGg5ex.js';
 
 /** Identity: dimensions already follow "higher = better" by prompt convention

package/dist/{summary-report-C0nnxOD8.d.ts → summary-report-Fc_YFJat.d.ts}

@@ -1,4 +1,4 @@
-import { R as RunRecord } from './run-record-MRdJ-Kq2.js';
+import { R as RunRecord } from './run-record-CPfd1ARZ.js';
 import { F as FailureClusterReport } from './failure-cluster-DH9Flgcf.js';
 
 /**

package/dist/traces.d.ts

@@ -6,7 +6,7 @@ export { S as SpanHandle, T as TraceEmitter, b as TraceEmitterOptions, l as llmS
 export { b as RunIntegrityError, R as RunIntegrityExpectations, c as RunIntegrityIssue, d as RunIntegrityIssueCode, a as RunIntegrityReport, e as assertRunCaptured, t as throwIfRunIncomplete } from './integrity-D2t12mMw.js';
 import { T as TraceStore } from './store-BcFXE6LG.js';
 export { E as EventFilter, F as FileSystemTraceStore, a as FileSystemTraceStoreOptions, I as InMemoryTraceStore, R as RunFilter, S as SpanFilter } from './store-BcFXE6LG.js';
-export { a as aggregateLlm, b as argHash, g as groupBy, j as judgeSpans, l as llmSpans, r as runFailureClass, c as runsForScenario, t as toolSpans } from './query-B7GGjRox.js';
+export { a as aggregateLlm, b as argHash, g as groupBy, j as judgeSpans, l as llmSpans, r as runFailureClass, c as runsForScenario, t as toolSpans } from './query-0aTmbmQe.js';
 export { D as DEFAULT_REDACTION_RULES, b as REDACTION_VERSION, a as RedactionReport, R as RedactionRule, r as redactString, c as redactValue } from './redact-B40YG2M_.js';
 import { R as Run } from './schema-m0gsnbt3.js';
 export { A as Artifact, B as BudgetLedgerEntry, h as BudgetSpec, E as EventKind, i as FAILURE_CLASSES, F as FailureClass, G as GenericSpan, J as JudgeSpan, L as LlmSpan, M as Message, d as RetrievalSpan, g as RunLayer, b as RunOutcome, f as RunStatus, e as SandboxSpan, S as Span, j as SpanBase, c as SpanKind, k as SpanStatus, l as TRACE_SCHEMA_VERSION, T as ToolSpan, a as TraceEvent, m as isJudgeSpan, n as isLlmSpan, o as isRetrievalSpan, p as isSandboxSpan, q as isToolSpan } from './schema-m0gsnbt3.js';
@@ -14,7 +14,7 @@ import { A as AnalyzeTracesOptions, b as AnalyzeTracesResult } from './analyst-C
 export { a as AnalyzeTracesInput, c as AnalyzeTracesTurnSnapshot, d as analyzeTraces } from './analyst-C8HHvfJp.js';
 import { h as TraceAnalystSpanKind, i as TraceAnalystSpanStatus, T as TraceAnalysisStore, g as TraceAnalystFilters, b as DatasetOverview, Q as QueryTracesPage, l as ViewTraceResult, V as ViewSpansResult, c as SearchTraceResult, S as SearchSpanResult } from './store-C1YxJDEK.js';
 export { D as DEFAULT_TRACE_ANALYST_BUDGETS, E as ErrorCluster, d as SpanMatchRecord, e as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, f as TraceAnalystByteBudgets, a as TraceAnalystSpan, j as TraceAnalystTraceSummary, k as ViewTraceOversized } from './store-C1YxJDEK.js';
-import { b as RunSplitTag, c as RunTokenUsage, R as RunRecord } from './run-record-MRdJ-Kq2.js';
+import { b as RunSplitTag, c as RunTokenUsage, R as RunRecord } from './run-record-CPfd1ARZ.js';
 import { AxFunction } from '@ax-llm/ax';
 import '@tangle-network/agent-interface';
 

package/dist/traces.js

@@ -28,7 +28,7 @@ import {
   scoreTraceInsightReadiness,
   tokenizeDomainWords,
   traceAnalystOnRunComplete
-} from "./chunk-7RBJANJD.js";
+} from "./chunk-3722PKPB.js";
 import {
   TRACE_ANALYST_ACTOR_DESCRIPTION,
   TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION,
@@ -50,7 +50,7 @@ import {
   runFailureClass,
   runsForScenario,
   toolSpans
-} from "./chunk-UFSG7ACU.js";
+} from "./chunk-JZXGWLK5.js";
 import {
   FAILURE_CLASSES,
   TRACE_SCHEMA_VERSION,
@@ -106,12 +106,12 @@ import {
   defaultProviderRedactor,
   providerFromBaseUrl
 } from "./chunk-PC4UYEBM.js";
-import "./chunk-2NSLDY4B.js";
+import "./chunk-J5MUFXEY.js";
 import {
   TraceEmitter,
   llmSpanFromProvider
 } from "./chunk-TVVP3ZZQ.js";
-import "./chunk-ABOIVNXL.js";
+import "./chunk-AGYDMORK.js";
 import "./chunk-VSMTAMNK.js";
 import "./chunk-3BFEG2F6.js";
 import "./chunk-PZ5AY32C.js";

package/dist/{types-DFI_Z-ZL.d.ts → types-Bihq6-a3.d.ts}

@@ -1,4 +1,4 @@
-import { c as RunTokenUsage } from './run-record-MRdJ-Kq2.js';
+import { c as RunTokenUsage } from './run-record-CPfd1ARZ.js';
 
 /**
  * Pass A substrate types — `runCampaign` is the one primitive every

package/dist/{types-Dz9cKF0g.d.ts → types-DA9yj-Jd.d.ts}

@@ -1,4 +1,4 @@
-import { R as RunRecord } from './run-record-MRdJ-Kq2.js';
+import { R as RunRecord } from './run-record-CPfd1ARZ.js';
 import { T as TraceAnalysisStore } from './store-C1YxJDEK.js';
 import { a as JudgeInput } from './types-C7DGg5ex.js';
 import { b as LlmCallRequest, c as LlmCallResult } from './llm-client-Bj7g0rqu.js';

package/dist/workflow/index.d.ts

@@ -1,13 +1,13 @@
 import { W as WorkflowTopology } from '../harness-optimizer-mOl9XX_O.js';
-import { b as RunSplitTag, c as RunTokenUsage, R as RunRecord } from '../run-record-MRdJ-Kq2.js';
-import { A as AnalystFinding, h as AnalystSeverity, E as EvidenceRef } from '../types-Dz9cKF0g.js';
-import { F as FailureClusterInsight } from '../insight-report-k0sRTzKg.js';
-import { a as VerificationReport, L as LayerResult } from '../multi-layer-verifier-DUZXrPDA.js';
+import { b as RunSplitTag, c as RunTokenUsage, R as RunRecord } from '../run-record-CPfd1ARZ.js';
+import { A as AnalystFinding, h as AnalystSeverity, E as EvidenceRef } from '../types-DA9yj-Jd.js';
+import { F as FailureClusterInsight } from '../insight-report-DumCfEur.js';
+import { a as VerificationReport, L as LayerResult } from '../multi-layer-verifier-CI4jdX-q.js';
 import { F as FailureClusterReport } from '../failure-cluster-DH9Flgcf.js';
 import { R as RedactionRule, a as RedactionReport } from '../redact-B40YG2M_.js';
 import { D as DatasetSplit } from '../dataset-BbGkaN2I.js';
 import { a as FeedbackTrajectory } from '../feedback-trajectory-BxY0cKfs.js';
-import { a as PairedBootstrapResult } from '../statistics-xP-cWc5k.js';
+import { a as PairedBootstrapResult } from '../statistics-D88peojY.js';
 import '../pareto-E-pembql.js';
 import '../run-critic-CmMf05uV.js';
 import '../schema-m0gsnbt3.js';
@@ -19,8 +19,8 @@ import '../types-C7DGg5ex.js';
 import '@tangle-network/tcloud';
 import '../llm-client-Bj7g0rqu.js';
 import '../raw-provider-sink-C46HDghv.js';
-import '../summary-report-C0nnxOD8.js';
-import '../judge-calibration-0p2QcWNE.js';
+import '../summary-report-Fc_YFJat.js';
+import '../judge-calibration-7C-IDmKr.js';
 import '../verdict-C9MlYujm.js';
 import '../control-runtime-Acf9CGhw.js';
 import '../emitter-C2rqGH_l.js';

package/dist/workflow/index.js

@@ -1,14 +1,14 @@
 import {
   pairedBootstrap
-} from "../chunk-HRGUJTER.js";
+} from "../chunk-HYGRFL7C.js";
 import {
   DEFAULT_REDACTION_RULES,
   redactString
 } from "../chunk-GGE4NNQT.js";
 import {
   validateRunRecord
-} from "../chunk-2NSLDY4B.js";
-import "../chunk-ABOIVNXL.js";
+} from "../chunk-J5MUFXEY.js";
+import "../chunk-AGYDMORK.js";
 import "../chunk-VSMTAMNK.js";
 import {
   ValidationError

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-eval",
-  "version": "0.103.0",
+  "version": "0.103.1",
   "description": "Evaluate and improve AI agents from runs, traces, judges, and feedback. Compare candidates, cluster failures, measure lift, and gate releases.",
   "homepage": "https://github.com/tangle-network/agent-eval#readme",
   "repository": {