@tangle-network/agent-eval 0.193.0 → 0.193.2

package/dist/benchmarks/index.js

@@ -1,7 +1,7 @@
 import { t as __exportAll } from "../rolldown-runtime-8H4AJuhK.js";
-import { ct as runCampaign } from "../llm-judge-ChbWfeWl.js";
-import { C as fsCampaignStorage } from "../external-optimizer-subprocess-kbqC-iWf.js";
-import "../campaign-DdE6S80e.js";
+import { ct as runCampaign } from "../llm-judge-Cififl_V.js";
+import { C as fsCampaignStorage } from "../external-optimizer-subprocess-BrghTvEp.js";
+import "../campaign-DnL0S4XL.js";
 import { join, resolve } from "node:path";
 import { readFile, readdir, stat } from "node:fs/promises";
 import { parse } from "smol-toml";

package/dist/campaign/index.js

@@ -1,13 +1,13 @@
 import { a as heldoutSignificance, i as dimensionRegressions, o as pairHoldout, r as detectScale } from "../statistical-heldout-DANVmyug.js";
-import { A as createSearchHistoryReceipt, B as decideSearchClaim, C as searchEdgeId, D as assertCompleteSearchHistory, E as SearchHistoryRequiredError, F as runSearch, G as uniform, H as searchClaimReserveUsd, I as searchDivergence, L as searchExpansionIndex, M as verifySearchHistoryArtifact, N as verifySearchHistoryReceipt, P as SEARCH_KERNEL_SOURCE, R as searchPolicyView, S as developmentClaim, T as surfaceNode, V as planSearchClaim, Y as openAutoPr, Z as defaultProductionGate, _ as isProposedCandidate, a as transientDispatchFailure, b as incumbent, c as compareOptimizationMethods, ct as runCampaign, d as combineComparisonCosts, dt as tangleTracesRoot, f as costFromLedgerSummary, ft as readCachedCell, g as runOptimization, h as runImprovementLoop, i as quotaExhaustedUntil, j as searchHistoryCoverageRow, k as assertSearchHistoryMatchesState, l as optimizationTokenUsageFromSummary, lt as planCampaignRun, m as readGepaCandidatePopulationArtifact, mt as cellCachePath, pt as buildCellSchedule, r as isTransientTransportFailure, st as runEval, t as llmJudge, ut as resolveRunDir, v as labelTrustRank, w as surfaceDiff, x as SearchRecorder, y as crowdedFrontierParent, z as SEARCH_CLAIM_RULE } from "../llm-judge-ChbWfeWl.js";
+import { A as createSearchHistoryReceipt, B as decideSearchClaim, C as searchEdgeId, D as assertCompleteSearchHistory, E as SearchHistoryRequiredError, F as runSearch, G as uniform, H as searchClaimReserveUsd, I as searchDivergence, L as searchExpansionIndex, M as verifySearchHistoryArtifact, N as verifySearchHistoryReceipt, P as SEARCH_KERNEL_SOURCE, R as searchPolicyView, S as developmentClaim, T as surfaceNode, V as planSearchClaim, Y as openAutoPr, Z as defaultProductionGate, _ as isProposedCandidate, a as transientDispatchFailure, b as incumbent, c as compareOptimizationMethods, ct as runCampaign, d as combineComparisonCosts, dt as tangleTracesRoot, f as costFromLedgerSummary, ft as readCachedCell, g as runOptimization, h as runImprovementLoop, i as quotaExhaustedUntil, j as searchHistoryCoverageRow, k as assertSearchHistoryMatchesState, l as optimizationTokenUsageFromSummary, lt as planCampaignRun, m as readGepaCandidatePopulationArtifact, mt as cellCachePath, pt as buildCellSchedule, r as isTransientTransportFailure, st as runEval, t as llmJudge, ut as resolveRunDir, v as labelTrustRank, w as surfaceDiff, x as SearchRecorder, y as crowdedFrontierParent, z as SEARCH_CLAIM_RULE } from "../llm-judge-Cififl_V.js";
 import { D as powerPreflight, F as surfaceContentHash, 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-DJc60F9C.js";
 import { a as SearchLedgerError, i as SearchLedgerConflictError, o as SearchLedgerIntegrityError } from "../search-ledger-file-DY0AXX3W.js";
-import { C as fsCampaignStorage, S as createRunCostLedger, l as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, u as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, v as resolveExternalOptimizerCallbackLimits, w as inMemoryCampaignStorage, x as acquireSingleRunLock, y as resolveExternalOptimizerProcessLimits } from "../external-optimizer-subprocess-kbqC-iWf.js";
+import { C as fsCampaignStorage, S as createRunCostLedger, l as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, u as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, v as resolveExternalOptimizerCallbackLimits, w as inMemoryCampaignStorage, x as acquireSingleRunLock, y as resolveExternalOptimizerProcessLimits } from "../external-optimizer-subprocess-BrghTvEp.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-CJiAEU3j.js";
 import { a as estimateNodeFromCells, i as estimateNode, n as searchProposerView, o as searchCellSetDigest, r as SEARCH_ESTIMATOR, s as searchPosterior, t as renderSearchSummary } from "../search-summary-BhY8p0Av.js";
 import { i as makeProposalFinding } from "../types-CW5b90JH.js";
-import { a as importGepaPopulation, c as composeGate, d as readExternalOptimizerObservationArtifact, i as importExternalEvaluations, l as externalTextOptimizationMethod, m as createReferenceEquivalenceJudge, n as gepaOptimizationMethod, o as recordGepaSearch, r as externalSurface, s as heldOutGate, t as skillOptOptimizationMethod, u as decodeExternalTextCandidate } from "../skillopt-optimization-method-DXQdaBjx.js";
-import { A as planEvalFixtureRun, C as LabeledScenarioStoreError, D as discoverEvalFixtures, E as neutralizationGate, M as buildTraceAnalystSurfaceDispatch, N as traceAnalystQualityJudge, O as loadEvalFixture, S as FsLabeledScenarioStore, T as rolloutArgumentDiff, _ as scoreboardSummary, a as autoevalsScorerJudge, b as sequentialOptimizationMethod, c as selectDiscriminative, d as runProfileMatrixSegment, f as ProfileMatrixError, g as scoreUserStory, h as renderScoreboardMarkdown, i as verifyCodeSurface, j as analyzeCrossSurfaceInteractions, k as loadEvalFixtureScenarios, 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 classifyUngroundedLiterals, x as neutralizeText, y as scopedOptimizationMethod } from "../campaign-DdE6S80e.js";
+import { a as importGepaPopulation, c as composeGate, d as readExternalOptimizerObservationArtifact, i as importExternalEvaluations, l as externalTextOptimizationMethod, m as createReferenceEquivalenceJudge, n as gepaOptimizationMethod, o as recordGepaSearch, r as externalSurface, s as heldOutGate, t as skillOptOptimizationMethod, u as decodeExternalTextCandidate } from "../skillopt-optimization-method-C6DbFKnw.js";
+import { A as planEvalFixtureRun, C as LabeledScenarioStoreError, D as discoverEvalFixtures, E as neutralizationGate, M as buildTraceAnalystSurfaceDispatch, N as traceAnalystQualityJudge, O as loadEvalFixture, S as FsLabeledScenarioStore, T as rolloutArgumentDiff, _ as scoreboardSummary, a as autoevalsScorerJudge, b as sequentialOptimizationMethod, c as selectDiscriminative, d as runProfileMatrixSegment, f as ProfileMatrixError, g as scoreUserStory, h as renderScoreboardMarkdown, i as verifyCodeSurface, j as analyzeCrossSurfaceInteractions, k as loadEvalFixtureScenarios, 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 classifyUngroundedLiterals, x as neutralizeText, y as scopedOptimizationMethod } from "../campaign-DnL0S4XL.js";
 import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-Cy5OV8BV.js";
 import { n as sequentialPairedGate, t as sequentialDecide } from "../sequential-C21_PP7F.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, analyzeCrossSurfaceInteractions, assertCampaignDesign, assertCampaignSplitIdentity, assertCodeSurfaceIdentity, assertCompleteSearchHistory, assertSearchHistoryMatchesState, autoevalsScorerJudge, buildCellSchedule, buildEvidenceVector, buildLoopProvenanceRecord, buildTraceAnalystSurfaceDispatch, campaignBreakdown, campaignMeanComposite, campaignMeasurementDigest, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, combineComparisonCosts, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createReferenceEquivalenceJudge, 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, 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, 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, searchNodeId, searchPolicyView, searchPosterior, searchProposerView, searchTaskSetDigest, searchUnitScores, selectDiscriminative, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, tangleTracesRoot, traceAnalystQualityJudge, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord, verifySearchHistoryArtifact, verifySearchHistoryReceipt };

package/dist/{campaign-DdE6S80e.js → campaign-DnL0S4XL.js}

@@ -1,23 +1,23 @@
 import { s as ValidationError, t as AgentEvalError } from "./errors-Dngq5h35.js";
 import { i as compareCodeUnits } 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-BqeIaIp2.js";
+import { c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, s as agentProfileHash, t as extractProducedState } from "./produced-state-B2AhDRnq.js";
 import { s as buildAgentProfileCell } from "./agent-profile-cell-DcFW4vOc.js";
 import { t as comparePairedArms } from "./paired-arms-JSDcs2zx.js";
 import { r as pairedBootstrap } from "./paired-tests-C8iCsioC.js";
 import { o as pairHoldout } from "./statistical-heldout-DANVmyug.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 { ct as runCampaign, d as combineComparisonCosts, ht as computeManifestHash, lt as planCampaignRun, mt as cellCachePath, u as executeOptimizationMethod, ut as resolveRunDir, v as labelTrustRank } from "./llm-judge-ChbWfeWl.js";
+import { ct as runCampaign, d as combineComparisonCosts, ht as computeManifestHash, lt as planCampaignRun, mt as cellCachePath, u as executeOptimizationMethod, ut as resolveRunDir, v as labelTrustRank } from "./llm-judge-Cififl_V.js";
 import { n as campaignCellToRunRecord, r as projectCampaignCellQuality, t as campaignCellCostProvenance } from "./run-record-tNnlQIrD.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-DJc60F9C.js";
 import { m as mapConcurrent } from "./journal-DvKCvjty.js";
-import { C as fsCampaignStorage } from "./external-optimizer-subprocess-kbqC-iWf.js";
+import { C as fsCampaignStorage } from "./external-optimizer-subprocess-BrghTvEp.js";
 import "./search-ledger-CJiAEU3j.js";
 import "./search-summary-BhY8p0Av.js";
 import { a as scoreAnalystFindings } from "./benchmark-BeQJyrRC.js";
-import "./external-optimizer-process-D-Sab4v3.js";
-import "./skillopt-optimization-method-DXQdaBjx.js";
+import "./external-optimizer-process-_YIECYJ7.js";
+import "./skillopt-optimization-method-C6DbFKnw.js";
 import "./sequential-C21_PP7F.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";
@@ -3087,4 +3087,4 @@ function resolveWorktreePath(surface, worktreeDir) {
 //#endregion
 export { planEvalFixtureRun as A, LabeledScenarioStoreError as C, discoverEvalFixtures as D, neutralizationGate as E, buildTraceAnalystSurfaceDispatch as M, traceAnalystQualityJudge as N, loadEvalFixture as O, FsLabeledScenarioStore as S, rolloutArgumentDiff as T, scoreboardSummary as _, autoevalsScorerJudge as a, sequentialOptimizationMethod as b, selectDiscriminative as c, runProfileMatrixSegment as d, ProfileMatrixError as f, scoreUserStory as g, renderScoreboardMarkdown as h, verifyCodeSurface as i, analyzeCrossSurfaceInteractions as j, loadEvalFixtureScenarios 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, classifyUngroundedLiterals as w, neutralizeText as x, scopedOptimizationMethod as y };
 
-//# sourceMappingURL=campaign-DdE6S80e.js.map
+//# sourceMappingURL=campaign-DnL0S4XL.js.map