@tangle-network/agent-eval 0.173.1 → 0.173.3

package/dist/benchmarks/index.js

@@ -1,7 +1,7 @@
 import { t as __exportAll } from "../rolldown-runtime-8H4AJuhK.js";
-import { at as runCampaign } from "../llm-judge-B2YxbAJb.js";
-import { S as fsCampaignStorage } from "../external-optimizer-subprocess-DgNebftP.js";
-import "../campaign-pxS0wmo4.js";
+import { at as runCampaign } from "../llm-judge-BfqMFo4h.js";
+import { S as fsCampaignStorage } from "../external-optimizer-subprocess-wBWeoG6A.js";
+import "../campaign-B3kPMU8S.js";
 import { join } from "node:path";
 //#region src/benchmarks/calibration.ts
 async function calibrateBenchmarkMetric(options) {

package/dist/campaign/index.js

@@ -1,10 +1,10 @@
-import { A as costFromLedgerSummary, B as assertSearchHistoryMatchesReplay, E as openAutoPr, F as renderSurfaceDiff, G as campaignMeanComposite, H as searchHistoryCoverageRow, I as surfaceContentHash, K as compareRankKeys, L as surfaceHash, M as assertCodeSurfaceIdentity, N as codeSurfaceIdentityMaterial, O as combineComparisonCosts, P as componentSurfaceIdentityMaterial, Q as defaultProductionGate, R as SearchHistoryRequiredError, S as recordCandidatePopulationSearch, U as verifySearchHistoryReceipt, V as createSearchHistoryReceipt, W as campaignBreakdown, Z as readGepaCandidatePopulationArtifact, _ as runImprovementLoop, _t as readCachedCell, a as transientDispatchFailure, at as runCampaign, b as labelTrustRank, c as buildLoopProvenanceRecord, ct as tangleTracesRoot, d as emitLoopProvenance, f as loopProvenanceArgsFromResult, ft as assertCampaignDesign, g as verifyLoopProvenanceRecord, gt as campaignSplitDigestFromIdentities, h as provenanceSpansPath, ht as campaignSplitDigest, i as quotaExhaustedUntil, it as runEval, j as optimizationTokenUsageFromSummary, k as compareOptimizationMethods, l as campaignMeasurementDigest, m as provenanceRecordPath, mt as campaignScenarioIdentity, ot as planCampaignRun, p as loopProvenanceSpans, pt as assertCampaignSplitIdentity, r as isTransientTransportFailure, st as resolveRunDir, t as llmJudge, u as canonicalDigest, v as runOptimization, vt as buildCellSchedule, x as SearchRecorder, y as isProposedCandidate, yt as cellCachePath, z as assertCompleteSearchHistory } from "../llm-judge-B2YxbAJb.js";
-import { C as inMemoryCampaignStorage, D as SearchLedgerIntegrityError, E as SearchLedgerError, S as fsCampaignStorage, T as SearchLedgerConflictError, l as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, u as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, v as resolveExternalOptimizerCallbackLimits, x as createRunCostLedger, y as resolveExternalOptimizerProcessLimits } from "../external-optimizer-subprocess-DgNebftP.js";
+import { A as costFromLedgerSummary, B as assertSearchHistoryMatchesReplay, E as openAutoPr, F as renderSurfaceDiff, G as campaignMeanComposite, H as searchHistoryCoverageRow, I as surfaceContentHash, K as compareRankKeys, L as surfaceHash, M as assertCodeSurfaceIdentity, N as codeSurfaceIdentityMaterial, O as combineComparisonCosts, P as componentSurfaceIdentityMaterial, Q as defaultProductionGate, R as SearchHistoryRequiredError, S as recordCandidatePopulationSearch, U as verifySearchHistoryReceipt, V as createSearchHistoryReceipt, W as campaignBreakdown, Z as readGepaCandidatePopulationArtifact, _ as runImprovementLoop, _t as readCachedCell, a as transientDispatchFailure, at as runCampaign, b as labelTrustRank, c as buildLoopProvenanceRecord, ct as tangleTracesRoot, d as emitLoopProvenance, f as loopProvenanceArgsFromResult, ft as assertCampaignDesign, g as verifyLoopProvenanceRecord, gt as campaignSplitDigestFromIdentities, h as provenanceSpansPath, ht as campaignSplitDigest, i as quotaExhaustedUntil, it as runEval, j as optimizationTokenUsageFromSummary, k as compareOptimizationMethods, l as campaignMeasurementDigest, m as provenanceRecordPath, mt as campaignScenarioIdentity, ot as planCampaignRun, p as loopProvenanceSpans, pt as assertCampaignSplitIdentity, r as isTransientTransportFailure, st as resolveRunDir, t as llmJudge, u as canonicalDigest, v as runOptimization, vt as buildCellSchedule, x as SearchRecorder, y as isProposedCandidate, yt as cellCachePath, z as assertCompleteSearchHistory } from "../llm-judge-BfqMFo4h.js";
+import { C as inMemoryCampaignStorage, D as SearchLedgerIntegrityError, E as SearchLedgerError, S as fsCampaignStorage, T as SearchLedgerConflictError, l as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, u as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, v as resolveExternalOptimizerCallbackLimits, x as createRunCostLedger, y as resolveExternalOptimizerProcessLimits } from "../external-optimizer-subprocess-wBWeoG6A.js";
 import { a as heldoutSignificance, i as dimensionRegressions, o as pairHoldout, r as detectScale, t as powerPreflight } from "../power-preflight-CFXm0Vjo.js";
 import { i as makeProposalFinding } from "../types-CiWITkGo.js";
-import { n as acquireSingleRunLock } from "../external-optimizer-process-Cq_Pg15r.js";
-import { a as validateSearchLedgerEvent, c as externalTextOptimizationMethod, i as openSearchLedger, l as decodeExternalTextCandidate, n as gepaOptimizationMethod, o as heldOutGate, p as createReferenceEquivalenceJudge, r as FileSearchLedger, s as composeGate, t as skillOptOptimizationMethod, u as readExternalOptimizerObservationArtifact } from "../skillopt-optimization-method-BzdphODy.js";
-import { A as analyzeCrossSurfaceInteractions, C as classifyUngroundedLiterals, D as loadEvalFixture, E as discoverEvalFixtures, M as traceAnalystQualityJudge, O as loadEvalFixtureScenarios, S as LabeledScenarioStoreError, T as neutralizationGate, _ as scoreboardSummary, a as autoevalsScorerJudge, b as neutralizeText, c as selectDiscriminative, d as runProfileMatrixSegment, f as ProfileMatrixError, g as scoreUserStory, h as renderScoreboardMarkdown, i as verifyCodeSurface, j as buildTraceAnalystSurfaceDispatch, k as planEvalFixtureRun, l as createProfileMatrixPlan, m as makePlaybackDispatch, n as gitWorktreeAdapter, o as phoenixEvaluatorJudge, p as runProfileMatrix, r as resolveWorktreePath, s as scoreDiscrimination, t as WorktreeAdapterError, u as finalizeProfileMatrix, v as userStoryScoreboard, w as rolloutArgumentDiff, x as FsLabeledScenarioStore, y as crowdedFrontierParent } from "../campaign-pxS0wmo4.js";
+import { n as acquireSingleRunLock } from "../external-optimizer-process-BwITA9Jp.js";
+import { a as validateSearchLedgerEvent, c as externalTextOptimizationMethod, i as openSearchLedger, l as decodeExternalTextCandidate, n as gepaOptimizationMethod, o as heldOutGate, p as createReferenceEquivalenceJudge, r as FileSearchLedger, s as composeGate, t as skillOptOptimizationMethod, u as readExternalOptimizerObservationArtifact } from "../skillopt-optimization-method-DDw3v3gA.js";
+import { A as analyzeCrossSurfaceInteractions, C as classifyUngroundedLiterals, D as loadEvalFixture, E as discoverEvalFixtures, M as traceAnalystQualityJudge, O as loadEvalFixtureScenarios, S as LabeledScenarioStoreError, T as neutralizationGate, _ as scoreboardSummary, a as autoevalsScorerJudge, b as neutralizeText, c as selectDiscriminative, d as runProfileMatrixSegment, f as ProfileMatrixError, g as scoreUserStory, h as renderScoreboardMarkdown, i as verifyCodeSurface, j as buildTraceAnalystSurfaceDispatch, k as planEvalFixtureRun, l as createProfileMatrixPlan, m as makePlaybackDispatch, n as gitWorktreeAdapter, o as phoenixEvaluatorJudge, p as runProfileMatrix, r as resolveWorktreePath, s as scoreDiscrimination, t as WorktreeAdapterError, u as finalizeProfileMatrix, v as userStoryScoreboard, w as rolloutArgumentDiff, x as FsLabeledScenarioStore, y as crowdedFrontierParent } from "../campaign-B3kPMU8S.js";
 import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-LY9mVQ7W.js";
 import { n as sequentialPairedGate, t as sequentialDecide } from "../sequential-B51qAYE4.js";
 export { DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, FileSearchLedger, FsLabeledScenarioStore, LabeledScenarioStoreError, ProfileMatrixError, SearchHistoryRequiredError, SearchLedgerConflictError, SearchLedgerError, SearchLedgerIntegrityError, SearchRecorder, WorktreeAdapterError, acquireSingleRunLock, analyzeCrossSurfaceInteractions, assertCampaignDesign, assertCampaignSplitIdentity, assertCodeSurfaceIdentity, assertCompleteSearchHistory, assertSearchHistoryMatchesReplay, autoevalsScorerJudge, buildCellSchedule, buildEvidenceVector, buildLoopProvenanceRecord, buildTraceAnalystSurfaceDispatch, campaignBreakdown, campaignMeanComposite, campaignMeasurementDigest, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, combineComparisonCosts, compareOptimizationMethods, compareRankKeys, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createReferenceEquivalenceJudge, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decodeExternalTextCandidate, defaultProductionGate, detectScale, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, inMemoryCampaignStorage, isProposedCandidate, isTransientTransportFailure, labelTrustRank, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, loopProvenanceArgsFromResult, loopProvenanceSpans, makePlaybackDispatch, makeProposalFinding, neutralizationGate, neutralizeText, openAutoPr, openSearchLedger, optimizationTokenUsageFromSummary, pairHoldout, paretoPolicy, paretoSignificanceGate, phoenixEvaluatorJudge, planCampaignRun, planEvalFixtureRun, powerPreflight, provenanceRecordPath, provenanceSpansPath, quotaExhaustedUntil, readCachedCell, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, recordCandidatePopulationSearch, renderScoreboardMarkdown, renderSurfaceDiff, resolveExternalOptimizerCallbackLimits, resolveExternalOptimizerProcessLimits, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchHistoryCoverageRow, selectDiscriminative, sequentialDecide, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceHash, tangleTracesRoot, traceAnalystQualityJudge, transientDispatchFailure, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord, verifySearchHistoryReceipt };

package/dist/{campaign-pxS0wmo4.js → campaign-B3kPMU8S.js}

@@ -1,21 +1,21 @@
 import { s as ValidationError, t as AgentEvalError } from "./errors-Dngq5h35.js";
 import { i as compareCodeUnits } from "./canonical-DPyQ_rpt.js";
-import { c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, s as agentProfileHash, t as extractProducedState } from "./produced-state-7VYDwtkk.js";
+import { c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, s as agentProfileHash, t as extractProducedState } from "./produced-state-D91uDvQw.js";
 import { buildAgentProfileCell } from "./profile-cell.js";
 import { t as mulberry32 } from "./random-Dn5fPWkt.js";
 import { t as comparePairedArms } from "./paired-arms-D4aeIHUy.js";
 import { r as pairedBootstrap } from "./paired-tests-C8iCsioC.js";
 import { c as validateRunRecord, i as modelHasSnapshot, n as UNKNOWN_MODEL, s as runTaskScore } from "./run-record-ZIsR9Fif.js";
 import { n as contentHash, t as canonicalJson } from "./verdict-cache-B3eCVQtY.js";
-import { I as surfaceContentHash, M as assertCodeSurfaceIdentity, T as paretoFrontierWithCrowding, at as runCampaign, b as labelTrustRank, bt as computeManifestHash, dt as summarizeBackendIntegrity, ft as assertCampaignDesign, mt as campaignScenarioIdentity, ot as planCampaignRun, st as resolveRunDir, ut as assertRealBackend, yt as cellCachePath } from "./llm-judge-B2YxbAJb.js";
+import { I as surfaceContentHash, M as assertCodeSurfaceIdentity, T as paretoFrontierWithCrowding, at as runCampaign, b as labelTrustRank, bt as computeManifestHash, dt as summarizeBackendIntegrity, ft as assertCampaignDesign, mt as campaignScenarioIdentity, ot as planCampaignRun, st as resolveRunDir, ut as assertRealBackend, yt as cellCachePath } from "./llm-judge-BfqMFo4h.js";
 import { a as campaignCellCostProvenance, l as campaignCellToRunRecord, u as projectCampaignCellQuality } from "./reward-hacking-CKW4teig.js";
 import { t as CostAccountingIncompleteError } from "./cost-ledger-B1qx30B4.js";
 import { u as mapConcurrent } from "./ledger-core-PIfjCbKn.js";
-import { S as fsCampaignStorage } from "./external-optimizer-subprocess-DgNebftP.js";
+import { S as fsCampaignStorage } from "./external-optimizer-subprocess-wBWeoG6A.js";
 import { o as pairHoldout } from "./power-preflight-CFXm0Vjo.js";
 import { a as scoreAnalystFindings } from "./benchmark-C4wk_Sjr.js";
-import "./external-optimizer-process-Cq_Pg15r.js";
-import "./skillopt-optimization-method-BzdphODy.js";
+import "./external-optimizer-process-BwITA9Jp.js";
+import "./skillopt-optimization-method-DDw3v3gA.js";
 import "./sequential-B51qAYE4.js";
 import { createHash } from "node:crypto";
 import { closeSync, constants, existsSync, fstatSync, lstatSync, mkdirSync, mkdtempSync, openSync, readFileSync, readSync, readdirSync, readlinkSync, realpathSync, rmSync, statSync, writeFileSync } from "node:fs";
@@ -3050,4 +3050,4 @@ function resolveWorktreePath(surface, worktreeDir) {
 //#endregion
 export { analyzeCrossSurfaceInteractions as A, classifyUngroundedLiterals as C, loadEvalFixture as D, discoverEvalFixtures as E, traceAnalystQualityJudge as M, loadEvalFixtureScenarios as O, LabeledScenarioStoreError as S, neutralizationGate as T, scoreboardSummary as _, autoevalsScorerJudge as a, neutralizeText as b, selectDiscriminative as c, runProfileMatrixSegment as d, ProfileMatrixError as f, scoreUserStory as g, renderScoreboardMarkdown as h, verifyCodeSurface as i, buildTraceAnalystSurfaceDispatch as j, planEvalFixtureRun as k, createProfileMatrixPlan as l, makePlaybackDispatch as m, gitWorktreeAdapter as n, phoenixEvaluatorJudge as o, runProfileMatrix as p, resolveWorktreePath as r, scoreDiscrimination as s, WorktreeAdapterError as t, finalizeProfileMatrix as u, userStoryScoreboard as v, rolloutArgumentDiff as w, FsLabeledScenarioStore as x, crowdedFrontierParent as y };
 
-//# sourceMappingURL=campaign-pxS0wmo4.js.map
+//# sourceMappingURL=campaign-B3kPMU8S.js.map