@tangle-network/agent-eval 0.145.11 → 0.145.13

package/dist/benchmarks/index.js

@@ -1,7 +1,7 @@
 import { t as __exportAll } from "../rolldown-runtime-8H4AJuhK.js";
-import { K as runCampaign } from "../llm-judge-DhiJqSjB.js";
-import { x as fsCampaignStorage } from "../external-optimizer-subprocess-BhKYK0Jv.js";
-import "../campaign-B3GJpJ4h.js";
+import { K as runCampaign } from "../llm-judge-DqAkIQA1.js";
+import { x as fsCampaignStorage } from "../external-optimizer-subprocess-x5AbhAYX.js";
+import "../campaign-Ct6r9oNe.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 { $ as assertCampaignDesign, A as surfaceHash, C as optimizationTokenUsageFromSummary, D as componentSurfaceIdentityMaterial, E as codeSurfaceIdentityMaterial, G as runEval, J as resolveRunDir, K as runCampaign, M as campaignMeanComposite, N as compareRankKeys, O as renderSurfaceDiff, R as readGepaCandidatePopulationArtifact, S as costFromLedgerSummary, T as assertComponentSurface, Y as tangleTracesRoot, a as canonicalDigest, at as cellCachePath, b as combineComparisonCosts, c as loopProvenanceSpans, d as verifyLoopProvenanceRecord, et as assertCampaignSplitIdentity, f as runImprovementLoop, h as labelTrustRank, i as campaignMeasurementDigest, it as readCachedCell, j as campaignBreakdown, k as surfaceContentHash, l as provenanceRecordPath, m as isProposedCandidate, nt as campaignSplitDigest, o as emitLoopProvenance, p as runOptimization, q as planCampaignRun, r as buildLoopProvenanceRecord, rt as campaignSplitDigestFromIdentities, s as loopProvenanceArgsFromResult, t as llmJudge, tt as campaignScenarioIdentity, u as provenanceSpansPath, v as openAutoPr, w as assertCodeSurfaceIdentity, x as compareOptimizationMethods, z as defaultProductionGate } from "../llm-judge-DhiJqSjB.js";
-import { E as SearchLedgerIntegrityError, S as inMemoryCampaignStorage, T as SearchLedgerError, _ as resolveExternalOptimizerCallbackLimits, b as createRunCostLedger, c as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, l as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, v as resolveExternalOptimizerProcessLimits, w as SearchLedgerConflictError, x as fsCampaignStorage } from "../external-optimizer-subprocess-BhKYK0Jv.js";
+import { $ as assertCampaignDesign, A as surfaceHash, C as optimizationTokenUsageFromSummary, D as componentSurfaceIdentityMaterial, E as codeSurfaceIdentityMaterial, G as runEval, J as resolveRunDir, K as runCampaign, M as campaignMeanComposite, N as compareRankKeys, O as renderSurfaceDiff, R as readGepaCandidatePopulationArtifact, S as costFromLedgerSummary, T as assertComponentSurface, Y as tangleTracesRoot, a as canonicalDigest, at as cellCachePath, b as combineComparisonCosts, c as loopProvenanceSpans, d as verifyLoopProvenanceRecord, et as assertCampaignSplitIdentity, f as runImprovementLoop, h as labelTrustRank, i as campaignMeasurementDigest, it as readCachedCell, j as campaignBreakdown, k as surfaceContentHash, l as provenanceRecordPath, m as isProposedCandidate, nt as campaignSplitDigest, o as emitLoopProvenance, p as runOptimization, q as planCampaignRun, r as buildLoopProvenanceRecord, rt as campaignSplitDigestFromIdentities, s as loopProvenanceArgsFromResult, t as llmJudge, tt as campaignScenarioIdentity, u as provenanceSpansPath, v as openAutoPr, w as assertCodeSurfaceIdentity, x as compareOptimizationMethods, z as defaultProductionGate } from "../llm-judge-DqAkIQA1.js";
+import { E as SearchLedgerIntegrityError, S as inMemoryCampaignStorage, T as SearchLedgerError, _ as resolveExternalOptimizerCallbackLimits, b as createRunCostLedger, c as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, l as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, v as resolveExternalOptimizerProcessLimits, w as SearchLedgerConflictError, x as fsCampaignStorage } from "../external-optimizer-subprocess-x5AbhAYX.js";
 import { a as heldoutSignificance, i as dimensionRegressions, o as pairHoldout, r as detectScale, t as powerPreflight } from "../power-preflight-DEw-uC7q.js";
 import { r as makeProposalFinding } from "../types-BI4fT3HN.js";
-import { n as acquireSingleRunLock } from "../external-optimizer-process-BRE56woM.js";
-import { a as externalTextOptimizationMethod, i as composeGate, n as gepaOptimizationMethod, o as decodeExternalTextCandidate, r as heldOutGate, s as readExternalOptimizerObservationArtifact, t as skillOptOptimizationMethod, u as createReferenceEquivalenceJudge } from "../skillopt-optimization-method-C6JSfYxT.js";
-import { A as neutralizationGate, C as scoreboardSummary, D as LabeledScenarioStoreError, E as FsLabeledScenarioStore, F as analyzeCrossSurfaceInteractions, I as buildTraceAnalystSurfaceDispatch, L as traceAnalystQualityJudge, M as loadEvalFixture, N as loadEvalFixtureScenarios, O as classifyUngroundedLiterals, P as planEvalFixtureRun, S as scoreUserStory, T as neutralizeText, _ as runProfileMatrixSegment, a as autoevalsScorerJudge, b as makePlaybackDispatch, c as FileSearchLedger, d as validateSearchLedgerEvent, f as scoreDiscrimination, g as finalizeProfileMatrix, h as createProfileMatrixPlan, i as verifyCodeSurface, j as discoverEvalFixtures, k as rolloutArgumentDiff, l as SEARCH_LEDGER_SCHEMA, m as SegmentedProfileMatrixError, n as gitWorktreeAdapter, o as phoenixEvaluatorJudge, p as selectDiscriminative, r as resolveWorktreePath, s as isTransientTransportFailure, t as WorktreeAdapterError, u as openSearchLedger, v as ProfileMatrixError, w as userStoryScoreboard, x as renderScoreboardMarkdown, y as runProfileMatrix } from "../campaign-B3GJpJ4h.js";
+import { n as acquireSingleRunLock } from "../external-optimizer-process-D-9iX64j.js";
+import { a as externalTextOptimizationMethod, i as composeGate, n as gepaOptimizationMethod, o as decodeExternalTextCandidate, r as heldOutGate, s as readExternalOptimizerObservationArtifact, t as skillOptOptimizationMethod, u as createReferenceEquivalenceJudge } from "../skillopt-optimization-method-CcWZBaSz.js";
+import { A as neutralizationGate, C as scoreboardSummary, D as LabeledScenarioStoreError, E as FsLabeledScenarioStore, F as analyzeCrossSurfaceInteractions, I as buildTraceAnalystSurfaceDispatch, L as traceAnalystQualityJudge, M as loadEvalFixture, N as loadEvalFixtureScenarios, O as classifyUngroundedLiterals, P as planEvalFixtureRun, S as scoreUserStory, T as neutralizeText, _ as runProfileMatrixSegment, a as autoevalsScorerJudge, b as makePlaybackDispatch, c as FileSearchLedger, d as validateSearchLedgerEvent, f as scoreDiscrimination, g as finalizeProfileMatrix, h as createProfileMatrixPlan, i as verifyCodeSurface, j as discoverEvalFixtures, k as rolloutArgumentDiff, l as SEARCH_LEDGER_SCHEMA, m as SegmentedProfileMatrixError, n as gitWorktreeAdapter, o as phoenixEvaluatorJudge, p as selectDiscriminative, r as resolveWorktreePath, s as isTransientTransportFailure, t as WorktreeAdapterError, u as openSearchLedger, v as ProfileMatrixError, w as userStoryScoreboard, x as renderScoreboardMarkdown, y as runProfileMatrix } from "../campaign-Ct6r9oNe.js";
 import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-xzA40Evo.js";
 import { n as sequentialPairedGate, t as sequentialDecide } from "../sequential-C458DXNf.js";
 export { DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, FileSearchLedger, FsLabeledScenarioStore, LabeledScenarioStoreError, ProfileMatrixError, SEARCH_LEDGER_SCHEMA, SearchLedgerConflictError, SearchLedgerError, SearchLedgerIntegrityError, SegmentedProfileMatrixError, WorktreeAdapterError, acquireSingleRunLock, analyzeCrossSurfaceInteractions, assertCampaignDesign, assertCampaignSplitIdentity, assertCodeSurfaceIdentity, assertComponentSurface, autoevalsScorerJudge, buildEvidenceVector, buildLoopProvenanceRecord, buildTraceAnalystSurfaceDispatch, campaignBreakdown, campaignMeanComposite, campaignMeasurementDigest, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, combineComparisonCosts, compareOptimizationMethods, compareRankKeys, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createReferenceEquivalenceJudge, createRunCostLedger, 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, readCachedCell, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, renderScoreboardMarkdown, renderSurfaceDiff, resolveExternalOptimizerCallbackLimits, resolveExternalOptimizerProcessLimits, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, scoreDiscrimination, scoreUserStory, scoreboardSummary, selectDiscriminative, sequentialDecide, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceHash, tangleTracesRoot, traceAnalystQualityJudge, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord };

package/dist/{campaign-B3GJpJ4h.js → campaign-Ct6r9oNe.js}

@@ -1,20 +1,20 @@
 import { s as ValidationError, t as AgentEvalError } from "./errors-Dngq5h35.js";
-import { c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, s as agentProfileHash, t as extractProducedState } from "./produced-state-BNyyud4g.js";
+import { c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, s as agentProfileHash, t as extractProducedState } from "./produced-state-B9skdU8u.js";
 import { buildAgentProfileCell } from "./profile-cell.js";
 import { t as comparePairedArms } from "./paired-arms-D-XRF_fy.js";
 import { r as pairedBootstrap } from "./paired-tests-BHIhYVdu.js";
 import { c as validateRunRecord, i as modelHasSnapshot, n as UNKNOWN_MODEL, s as runTaskScore } from "./run-record-D2lDdSAz.js";
 import { n as contentHash, t as canonicalJson } from "./verdict-cache-mZf5FEiY.js";
-import { $ as assertCampaignDesign, J as resolveRunDir, K as runCampaign, Q as summarizeBackendIntegrity, Z as assertRealBackend, at as cellCachePath, h as labelTrustRank, k as surfaceContentHash, ot as computeManifestHash, q as planCampaignRun, tt as campaignScenarioIdentity, w as assertCodeSurfaceIdentity } from "./llm-judge-DhiJqSjB.js";
+import { $ as assertCampaignDesign, J as resolveRunDir, K as runCampaign, Q as summarizeBackendIntegrity, Z as assertRealBackend, at as cellCachePath, h as labelTrustRank, k as surfaceContentHash, ot as computeManifestHash, q as planCampaignRun, tt as campaignScenarioIdentity, w as assertCodeSurfaceIdentity } from "./llm-judge-DqAkIQA1.js";
 import { a as campaignCellCostProvenance, l as campaignCellToRunRecord, u as projectCampaignCellQuality } from "./reward-hacking-DFo2FU5J.js";
 import { t as CostAccountingIncompleteError } from "./cost-ledger-BSe92yAV.js";
-import { t as FileLedgerJournal, u as mapConcurrent } from "./ledger-core-BmZt19oQ.js";
+import { t as FileLedgerJournal, u as mapConcurrent } from "./ledger-core-DTae9rv_.js";
 import { r as canonicalString } from "./canonical-D-XsTQ6_.js";
-import { C as SEARCH_LEDGER_FILE_CONTEXT, E as SearchLedgerIntegrityError, T as SearchLedgerError, w as SearchLedgerConflictError, x as fsCampaignStorage } from "./external-optimizer-subprocess-BhKYK0Jv.js";
+import { C as SEARCH_LEDGER_FILE_CONTEXT, E as SearchLedgerIntegrityError, T as SearchLedgerError, w as SearchLedgerConflictError, x as fsCampaignStorage } from "./external-optimizer-subprocess-x5AbhAYX.js";
 import { o as pairHoldout } from "./power-preflight-DEw-uC7q.js";
 import { a as scoreAnalystFindings } from "./benchmark-BhT16ep9.js";
-import "./external-optimizer-process-BRE56woM.js";
-import "./skillopt-optimization-method-C6JSfYxT.js";
+import "./external-optimizer-process-D-9iX64j.js";
+import "./skillopt-optimization-method-CcWZBaSz.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";
@@ -3685,4 +3685,4 @@ function resolveWorktreePath(surface, worktreeDir) {
 //#endregion
 export { neutralizationGate as A, scoreboardSummary as C, LabeledScenarioStoreError as D, FsLabeledScenarioStore as E, analyzeCrossSurfaceInteractions as F, buildTraceAnalystSurfaceDispatch as I, traceAnalystQualityJudge as L, loadEvalFixture as M, loadEvalFixtureScenarios as N, classifyUngroundedLiterals as O, planEvalFixtureRun as P, scoreUserStory as S, neutralizeText as T, runProfileMatrixSegment as _, autoevalsScorerJudge as a, makePlaybackDispatch as b, FileSearchLedger as c, validateSearchLedgerEvent as d, scoreDiscrimination as f, finalizeProfileMatrix as g, createProfileMatrixPlan as h, verifyCodeSurface as i, discoverEvalFixtures as j, rolloutArgumentDiff as k, SEARCH_LEDGER_SCHEMA as l, SegmentedProfileMatrixError as m, gitWorktreeAdapter as n, phoenixEvaluatorJudge as o, selectDiscriminative as p, resolveWorktreePath as r, isTransientTransportFailure as s, WorktreeAdapterError as t, openSearchLedger as u, ProfileMatrixError as v, userStoryScoreboard as w, renderScoreboardMarkdown as x, runProfileMatrix as y };
 
-//# sourceMappingURL=campaign-B3GJpJ4h.js.map
+//# sourceMappingURL=campaign-Ct6r9oNe.js.map