@tangle-network/agent-eval 0.203.2 → 0.205.0

package/dist/benchmarks/index.js

@@ -1,7 +1,7 @@
 import { t as __exportAll } from "../rolldown-runtime-8H4AJuhK.js";
-import { rt as runCampaign } from "../llm-judge-DyZdrp6I.js";
+import { rt as runCampaign } from "../llm-judge-Bka5q-ds.js";
 import { S as fsCampaignStorage } from "../external-optimizer-subprocess-BmKtH3T6.js";
-import "../campaign-DYhg4SjN.js";
+import "../campaign-0lmkAMpE.js";
 import { join, resolve } from "node:path";
 import { readFile, readdir, stat } from "node:fs/promises";
 import { parse } from "smol-toml";

package/dist/builder-eval/index.js

@@ -1,4 +1,4 @@
-import { a as spearmanR, r as pearsonR } from "../descriptive-D9LUFk1-.js";
+import { a as spearmanR, r as pearsonR } from "../descriptive-CnU8Yq29.js";
 import { s as judgeSpans } from "../query-DlV0pdx4.js";
 import { t as newRecordId } from "../record-id-DUgsK5qp.js";
 import { t as TraceEmitter } from "../emitter-DbMXUIVP.js";

package/dist/campaign/index.js

@@ -1,15 +1,15 @@
-import { a as heldoutSignificance, i as dimensionRegressions, o as pairHoldout, r as detectScale } from "../statistical-heldout-CNLOvRq_.js";
-import { A as assertCompleteSearchHistory, B as searchReceiptAccounting, C as incumbentWithOperatorBandit, D as surfaceDiff, E as searchEdgeId, F as verifySearchHistoryArtifact, H as uniform, I as verifySearchHistoryReceipt, K as defaultProductionGate, L as campaignCellSearchResult, M as assertSearchHistoryMatchesState, N as createSearchHistoryReceipt, O as surfaceNode, P as searchHistoryCoverageRow, R as searchModelIdentity, S as incumbent, T as developmentClaim, V as asha, W as openAutoPr, _ as isProposedCandidate, a as transientDispatchFailure, at as resolveRunDir, b as beam, c as compareOptimizationMethods, ct as buildCellSchedule, d as combineComparisonCosts, f as costFromLedgerSummary, g as runOptimization, h as runImprovementLoop, i as quotaExhaustedUntil, it as planCampaignRun, k as SearchHistoryRequiredError, l as optimizationTokenUsageFromSummary, lt as cellCachePath, m as readGepaCandidatePopulationArtifact, nt as runEval, ot as tangleTracesRoot, r as isTransientTransportFailure, rt as runCampaign, st as readCachedCell, t as llmJudge, v as labelTrustRank, w as SearchRecorder, x as crowdedFrontierParent, y as aide, z as searchProposalExecution } from "../llm-judge-DyZdrp6I.js";
-import { D as powerPreflight, F as surfaceContentHash, G as campaignCoverage, H as assertCampaignDesign, I as surfaceDispatchRef, J as campaignSplitDigestFromIdentities, K as campaignScenarioIdentity, L as surfaceHash, M as codeSurfaceIdentityMaterial, N as componentSurfaceIdentityMaterial, P as renderSurfaceDiff, U as assertCampaignSplitIdentity, _ as provenanceRecordPath, b as campaignBreakdown, d as buildLoopProvenanceRecord, f as campaignMeasurementDigest, g as loopProvenanceSpans, h as loopProvenanceArgsFromResult, j as assertCodeSurfaceIdentity, m as emitLoopProvenance, p as canonicalDigest, q as campaignSplitDigest, v as provenanceSpansPath, x as campaignMeanComposite, y as verifyLoopProvenanceRecord } from "../campaign-evidence-Dt_rWPk2.js";
+import { a as heldoutSignificance, i as dimensionRegressions, o as pairHoldout, r as detectScale } from "../statistical-heldout-C26hOFWD.js";
+import { A as assertCompleteSearchHistory, B as searchReceiptAccounting, C as incumbentWithOperatorBandit, D as surfaceDiff, E as searchEdgeId, F as verifySearchHistoryArtifact, H as uniform, I as verifySearchHistoryReceipt, K as defaultProductionGate, L as campaignCellSearchResult, M as assertSearchHistoryMatchesState, N as createSearchHistoryReceipt, O as surfaceNode, P as searchHistoryCoverageRow, R as searchModelIdentity, S as incumbent, T as developmentClaim, V as asha, W as openAutoPr, _ as isProposedCandidate, a as transientDispatchFailure, at as resolveRunDir, b as beam, c as compareOptimizationMethods, ct as buildCellSchedule, d as combineComparisonCosts, f as costFromLedgerSummary, g as runOptimization, h as runImprovementLoop, i as quotaExhaustedUntil, it as planCampaignRun, k as SearchHistoryRequiredError, l as optimizationTokenUsageFromSummary, lt as cellCachePath, m as readGepaCandidatePopulationArtifact, nt as runEval, ot as tangleTracesRoot, r as isTransientTransportFailure, rt as runCampaign, st as readCachedCell, t as llmJudge, v as labelTrustRank, w as SearchRecorder, x as crowdedFrontierParent, y as aide, z as searchProposalExecution } from "../llm-judge-Bka5q-ds.js";
+import { D as powerPreflight, F as surfaceContentHash, G as campaignCoverage, H as assertCampaignDesign, I as surfaceDispatchRef, J as campaignSplitDigestFromIdentities, K as campaignScenarioIdentity, L as surfaceHash, M as codeSurfaceIdentityMaterial, N as componentSurfaceIdentityMaterial, P as renderSurfaceDiff, U as assertCampaignSplitIdentity, _ as provenanceRecordPath, b as campaignBreakdown, d as buildLoopProvenanceRecord, f as campaignMeasurementDigest, g as loopProvenanceSpans, h as loopProvenanceArgsFromResult, j as assertCodeSurfaceIdentity, m as emitLoopProvenance, p as canonicalDigest, q as campaignSplitDigest, v as provenanceSpansPath, x as campaignMeanComposite, y as verifyLoopProvenanceRecord } from "../campaign-evidence-C__cUpCt.js";
 import { a as SearchLedgerError, i as SearchLedgerConflictError, o as SearchLedgerIntegrityError } from "../search-ledger-file-DY0AXX3W.js";
 import { C as inMemoryCampaignStorage, S as fsCampaignStorage, 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-BmKtH3T6.js";
-import { S as searchCellSetDigest, a as searchPolicyView, b as estimateNode, c as planSearchClaim, i as searchExpansionIndex, l as searchClaimReserveUsd, n as runSearch, o as SEARCH_CLAIM_RULE, r as searchDivergence, s as decideSearchClaim, t as SEARCH_KERNEL_SOURCE, u as verifySearchClaim, w as searchPosterior, x as estimateNodeFromCells, y as SEARCH_ESTIMATOR } from "../search-kernel-Dpniloks.js";
+import { S as searchCellSetDigest, a as searchPolicyView, b as estimateNode, c as planSearchClaim, i as searchExpansionIndex, l as searchClaimReserveUsd, n as runSearch, o as SEARCH_CLAIM_RULE, r as searchDivergence, s as decideSearchClaim, t as SEARCH_KERNEL_SOURCE, u as verifySearchClaim, w as searchPosterior, x as estimateNodeFromCells, y as SEARCH_ESTIMATOR } from "../search-kernel-CsISc5t8.js";
 import { a as validateSearchLedgerEvent, c as SearchStateView, d as searchTaskSetDigest, f as searchUnitScores, i as replaySearchLedgerText, l as searchCellId, n as openSearchLedger, o as SEARCH_LEDGER_SCHEMA, r as parseSearchLedgerLine, s as SearchState, t as FileSearchLedger, u as searchNodeId } from "../search-ledger-CXQPwdWz.js";
 import { t as acquireSingleRunLock } from "../single-run-lock-fVu3MZWH.js";
-import { n as searchProposerView, t as renderSearchSummary } from "../search-summary-C1nAHgkB.js";
+import { n as searchProposerView, t as renderSearchSummary } from "../search-summary-BZkoLwbB.js";
 import { i as makeProposalFinding } from "../types-CW5b90JH.js";
-import { a as importGepaPopulation, c as heldOutGate, d as decodeExternalTextCandidate, f as readExternalOptimizerObservationArtifact, h as createReferenceEquivalenceJudge, i as importExternalEvaluations, l as composeGate, n as gepaOptimizationMethod, o as recordGepaSearch, r as externalSurface, s as recordSkillOptSearch, t as skillOptOptimizationMethod, u as externalTextOptimizationMethod } from "../skillopt-optimization-method-Dz5CGZb8.js";
-import { A as loadEvalFixtureScenarios, C as FsLabeledScenarioStore, D as neutralizationGate, E as rolloutArgumentDiff, M as analyzeCrossSurfaceInteractions, N as buildTraceAnalystSurfaceDispatch, O as discoverEvalFixtures, P as traceAnalystQualityJudge, S as neutralizeText, T as classifyUngroundedLiterals, _ as scoreUserStory, a as autoevalsScorerJudge, b as scopedOptimizationMethod, c as selectDiscriminative, d as finalizeProfileMatrix, f as runProfileMatrixSegment, g as renderScoreboardMarkdown, h as makePlaybackDispatch, i as verifyCodeSurface, j as planEvalFixtureRun, k as loadEvalFixture, l as createRetainedDispatch, m as runProfileMatrix, n as gitWorktreeAdapter, o as phoenixEvaluatorJudge, p as ProfileMatrixError, r as resolveWorktreePath, s as scoreDiscrimination, t as WorktreeAdapterError, u as createProfileMatrixPlan, v as scoreboardSummary, w as LabeledScenarioStoreError, x as sequentialOptimizationMethod, y as userStoryScoreboard } from "../campaign-DYhg4SjN.js";
-import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-BI1RxzwD.js";
-import { n as sequentialPairedGate, t as sequentialDecide } from "../sequential-DWn_np6P.js";
+import { a as importGepaPopulation, c as heldOutGate, d as decodeExternalTextCandidate, f as readExternalOptimizerObservationArtifact, h as createReferenceEquivalenceJudge, i as importExternalEvaluations, l as composeGate, n as gepaOptimizationMethod, o as recordGepaSearch, r as externalSurface, s as recordSkillOptSearch, t as skillOptOptimizationMethod, u as externalTextOptimizationMethod } from "../skillopt-optimization-method-bEm4jYV4.js";
+import { A as loadEvalFixtureScenarios, C as FsLabeledScenarioStore, D as neutralizationGate, E as rolloutArgumentDiff, M as analyzeCrossSurfaceInteractions, N as buildTraceAnalystSurfaceDispatch, O as discoverEvalFixtures, P as traceAnalystQualityJudge, S as neutralizeText, T as classifyUngroundedLiterals, _ as scoreUserStory, a as autoevalsScorerJudge, b as scopedOptimizationMethod, c as selectDiscriminative, d as finalizeProfileMatrix, f as runProfileMatrixSegment, g as renderScoreboardMarkdown, h as makePlaybackDispatch, i as verifyCodeSurface, j as planEvalFixtureRun, k as loadEvalFixture, l as createRetainedDispatch, m as runProfileMatrix, n as gitWorktreeAdapter, o as phoenixEvaluatorJudge, p as ProfileMatrixError, r as resolveWorktreePath, s as scoreDiscrimination, t as WorktreeAdapterError, u as createProfileMatrixPlan, v as scoreboardSummary, w as LabeledScenarioStoreError, x as sequentialOptimizationMethod, y as userStoryScoreboard } from "../campaign-0lmkAMpE.js";
+import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-BSbTAMtE.js";
+import { n as sequentialPairedGate, t as sequentialDecide } from "../sequential-7uxpO0Vm.js";
 export { DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, FileSearchLedger, FsLabeledScenarioStore, LabeledScenarioStoreError, ProfileMatrixError, SEARCH_CLAIM_RULE, SEARCH_ESTIMATOR, SEARCH_KERNEL_SOURCE, SEARCH_LEDGER_SCHEMA, SearchHistoryRequiredError, SearchLedgerConflictError, SearchLedgerError, SearchLedgerIntegrityError, SearchRecorder, SearchState, SearchStateView, WorktreeAdapterError, acquireSingleRunLock, aide, analyzeCrossSurfaceInteractions, asha, assertCampaignDesign, assertCampaignSplitIdentity, assertCodeSurfaceIdentity, assertCompleteSearchHistory, assertSearchHistoryMatchesState, autoevalsScorerJudge, beam, buildCellSchedule, buildEvidenceVector, buildLoopProvenanceRecord, buildTraceAnalystSurfaceDispatch, campaignBreakdown, campaignCellSearchResult, campaignCoverage, campaignMeanComposite, campaignMeasurementDigest, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, combineComparisonCosts, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createReferenceEquivalenceJudge, createRetainedDispatch, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decideSearchClaim, decodeExternalTextCandidate, defaultProductionGate, detectScale, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, incumbent, incumbentWithOperatorBandit, isProposedCandidate, isTransientTransportFailure, labelTrustRank, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, loopProvenanceArgsFromResult, loopProvenanceSpans, makePlaybackDispatch, makeProposalFinding, neutralizationGate, neutralizeText, openAutoPr, openSearchLedger, optimizationTokenUsageFromSummary, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, phoenixEvaluatorJudge, planCampaignRun, planEvalFixtureRun, planSearchClaim, powerPreflight, provenanceRecordPath, provenanceSpansPath, quotaExhaustedUntil, readCachedCell, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, recordGepaSearch, recordSkillOptSearch, renderScoreboardMarkdown, renderSearchSummary, renderSurfaceDiff, replaySearchLedgerText, resolveExternalOptimizerCallbackLimits, resolveExternalOptimizerProcessLimits, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchClaimReserveUsd, searchDivergence, searchEdgeId, searchExpansionIndex, searchHistoryCoverageRow, searchModelIdentity, searchNodeId, searchPolicyView, searchPosterior, searchProposalExecution, searchProposerView, searchReceiptAccounting, searchTaskSetDigest, searchUnitScores, selectDiscriminative, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, tangleTracesRoot, traceAnalystQualityJudge, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord, verifySearchClaim, verifySearchHistoryArtifact, verifySearchHistoryReceipt };

package/dist/{campaign-DYhg4SjN.js → campaign-0lmkAMpE.js}

@@ -1,25 +1,25 @@
 import { s as ValidationError, t as AgentEvalError } from "./errors-Dngq5h35.js";
 import { a as hashCanonical, i as compareCodeUnits, r as canonicalString } from "./canonical-BsZVgBhW.js";
-import { c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, s as agentProfileHash, t as extractProducedState } from "./produced-state-DLToUXrD.js";
+import { c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, s as agentProfileHash, t as extractProducedState } from "./produced-state-FdbE-88t.js";
 import { s as buildAgentProfileCell } from "./agent-profile-cell-DcFW4vOc.js";
-import { r as pairedBootstrap } from "./paired-tests-Ccr16gcW.js";
-import { t as comparePairedArms } from "./paired-arms-DpS_h8ON.js";
-import { o as pairHoldout } from "./statistical-heldout-CNLOvRq_.js";
+import { r as pairedBootstrap } from "./paired-tests-jbwPdSz5.js";
+import { t as comparePairedArms } from "./paired-arms-Dsu-r6xd.js";
+import { o as pairHoldout } from "./statistical-heldout-C26hOFWD.js";
 import { c as runTaskScore, i as modelHasSnapshot, n as UNKNOWN_MODEL, u as validateRunRecord } from "./run-record-BRlM99rX.js";
 import { n as contentHash, t as canonicalJson } from "./verdict-cache-BQEsVy9n.js";
-import { at as resolveRunDir, d as combineComparisonCosts, it as planCampaignRun, lt as cellCachePath, rt as runCampaign, u as executeOptimizationMethod, ut as computeManifestHash, v as labelTrustRank } from "./llm-judge-DyZdrp6I.js";
+import { at as resolveRunDir, d as combineComparisonCosts, it as planCampaignRun, lt as cellCachePath, rt as runCampaign, u as executeOptimizationMethod, ut as computeManifestHash, v as labelTrustRank } from "./llm-judge-Bka5q-ds.js";
 import { n as campaignCellToRunRecord, r as projectCampaignCellQuality, t as campaignCellCostProvenance } from "./run-record-BB290Y9A.js";
 import { t as CostAccountingIncompleteError } from "./cost-ledger-_IbkejCF.js";
-import { B as assertRealBackend, F as surfaceContentHash, H as assertCampaignDesign, K as campaignScenarioIdentity, V as summarizeBackendIntegrity, j as assertCodeSurfaceIdentity } from "./campaign-evidence-Dt_rWPk2.js";
+import { B as assertRealBackend, F as surfaceContentHash, H as assertCampaignDesign, K as campaignScenarioIdentity, V as summarizeBackendIntegrity, j as assertCodeSurfaceIdentity } from "./campaign-evidence-C__cUpCt.js";
 import { m as mapConcurrent } from "./journal-DvKCvjty.js";
 import { S as fsCampaignStorage } from "./external-optimizer-subprocess-BmKtH3T6.js";
-import "./search-kernel-Dpniloks.js";
+import "./search-kernel-CsISc5t8.js";
 import "./search-ledger-CXQPwdWz.js";
-import "./search-summary-C1nAHgkB.js";
+import "./search-summary-BZkoLwbB.js";
 import { a as scoreAnalystFindings } from "./benchmark-BywPeAUo.js";
 import "./external-optimizer-process-CIV4EqiT.js";
-import "./skillopt-optimization-method-Dz5CGZb8.js";
-import "./sequential-DWn_np6P.js";
+import "./skillopt-optimization-method-bEm4jYV4.js";
+import "./sequential-7uxpO0Vm.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";
 import { z } from "zod";
@@ -3436,4 +3436,4 @@ function resolveWorktreePath(surface, worktreeDir) {
 //#endregion
 export { loadEvalFixtureScenarios as A, FsLabeledScenarioStore as C, neutralizationGate as D, rolloutArgumentDiff as E, analyzeCrossSurfaceInteractions as M, buildTraceAnalystSurfaceDispatch as N, discoverEvalFixtures as O, traceAnalystQualityJudge as P, neutralizeText as S, classifyUngroundedLiterals as T, scoreUserStory as _, autoevalsScorerJudge as a, scopedOptimizationMethod as b, selectDiscriminative as c, finalizeProfileMatrix as d, runProfileMatrixSegment as f, renderScoreboardMarkdown as g, makePlaybackDispatch as h, verifyCodeSurface as i, planEvalFixtureRun as j, loadEvalFixture as k, createRetainedDispatch as l, runProfileMatrix as m, gitWorktreeAdapter as n, phoenixEvaluatorJudge as o, ProfileMatrixError as p, resolveWorktreePath as r, scoreDiscrimination as s, WorktreeAdapterError as t, createProfileMatrixPlan as u, scoreboardSummary as v, LabeledScenarioStoreError as w, sequentialOptimizationMethod as x, userStoryScoreboard as y };
 
-//# sourceMappingURL=campaign-DYhg4SjN.js.map
+//# sourceMappingURL=campaign-0lmkAMpE.js.map