@tangle-network/agent-eval 0.150.1 → 0.150.2

package/dist/benchmarks/index.js

@@ -1,7 +1,7 @@
 import { t as __exportAll } from "../rolldown-runtime-8H4AJuhK.js";
-import { et as runCampaign } from "../llm-judge-CVq33oz1.js";
-import { x as fsCampaignStorage } from "../external-optimizer-subprocess-Dn90UqN2.js";
-import "../campaign-CN_7xJdV.js";
+import { et as runCampaign } from "../llm-judge-BqqMS8t7.js";
+import { x as fsCampaignStorage } from "../external-optimizer-subprocess-CKNb42oM.js";
+import "../campaign-la-gEYNz.js";
 import { join } from "node:path";
 //#region src/benchmarks/calibration.ts
 async function calibrateBenchmarkMetric(options) {

package/dist/builder-eval/index.js

@@ -1,4 +1,4 @@
-import { a as spearmanR, r as pearsonR } from "../descriptive-B5MwKfbf.js";
+import { a as spearmanR, r as pearsonR } from "../descriptive-jDOuI6mz.js";
 import { a as judgeSpans } from "../query-Di7eEQ79.js";
 import { t as TraceEmitter } from "../emitter-CPBAhxum.js";
 //#region src/sandbox-harness.ts

package/dist/campaign/index.js

@@ -1,10 +1,10 @@
-import { $ as runEval, A as surfaceHash, B as campaignMeanComposite, C as optimizationTokenUsageFromSummary, D as componentSurfaceIdentityMaterial, E as codeSurfaceIdentityMaterial, F as assertSearchHistoryMatchesReplay, I as createSearchHistoryReceipt, K as readGepaCandidatePopulationArtifact, L as searchHistoryCoverageRow, M as SEARCH_HISTORY_RECEIPT_SCHEMA_VERSION, N as SearchHistoryRequiredError, O as renderSurfaceDiff, P as assertCompleteSearchHistory, R as verifySearchHistoryReceipt, S as costFromLedgerSummary, T as assertComponentSurface, V as compareRankKeys, a as canonicalDigest, b as combineComparisonCosts, c as loopProvenanceSpans, ct as assertCampaignSplitIdentity, d as verifyLoopProvenanceRecord, dt as campaignSplitDigestFromIdentities, et as runCampaign, f as runImprovementLoop, ft as readCachedCell, h as labelTrustRank, i as campaignMeasurementDigest, j as SEARCH_HISTORY_RECEIPT_DIGEST_ALGORITHM, k as surfaceContentHash, l as provenanceRecordPath, lt as campaignScenarioIdentity, m as isProposedCandidate, nt as resolveRunDir, o as emitLoopProvenance, p as runOptimization, pt as cellCachePath, q as defaultProductionGate, r as buildLoopProvenanceRecord, rt as tangleTracesRoot, s as loopProvenanceArgsFromResult, st as assertCampaignDesign, t as llmJudge, tt as planCampaignRun, u as provenanceSpansPath, ut as campaignSplitDigest, v as openAutoPr, w as assertCodeSurfaceIdentity, x as compareOptimizationMethods, z as campaignBreakdown } from "../llm-judge-CVq33oz1.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-Dn90UqN2.js";
+import { $ as runEval, A as surfaceHash, B as campaignMeanComposite, C as optimizationTokenUsageFromSummary, D as componentSurfaceIdentityMaterial, E as codeSurfaceIdentityMaterial, F as assertSearchHistoryMatchesReplay, I as createSearchHistoryReceipt, K as readGepaCandidatePopulationArtifact, L as searchHistoryCoverageRow, M as SEARCH_HISTORY_RECEIPT_SCHEMA_VERSION, N as SearchHistoryRequiredError, O as renderSurfaceDiff, P as assertCompleteSearchHistory, R as verifySearchHistoryReceipt, S as costFromLedgerSummary, T as assertComponentSurface, V as compareRankKeys, a as canonicalDigest, b as combineComparisonCosts, c as loopProvenanceSpans, ct as assertCampaignSplitIdentity, d as verifyLoopProvenanceRecord, dt as campaignSplitDigestFromIdentities, et as runCampaign, f as runImprovementLoop, ft as readCachedCell, h as labelTrustRank, i as campaignMeasurementDigest, j as SEARCH_HISTORY_RECEIPT_DIGEST_ALGORITHM, k as surfaceContentHash, l as provenanceRecordPath, lt as campaignScenarioIdentity, m as isProposedCandidate, nt as resolveRunDir, o as emitLoopProvenance, p as runOptimization, pt as cellCachePath, q as defaultProductionGate, r as buildLoopProvenanceRecord, rt as tangleTracesRoot, s as loopProvenanceArgsFromResult, st as assertCampaignDesign, t as llmJudge, tt as planCampaignRun, u as provenanceSpansPath, ut as campaignSplitDigest, v as openAutoPr, w as assertCodeSurfaceIdentity, x as compareOptimizationMethods, z as campaignBreakdown } from "../llm-judge-BqqMS8t7.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-CKNb42oM.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-Bhmzngf-.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-DLeUcK-K.js";
-import { A as rolloutArgumentDiff, C as scoreboardSummary, D as FsLabeledScenarioStore, E as neutralizeText, F as planEvalFixtureRun, I as analyzeCrossSurfaceInteractions, L as buildTraceAnalystSurfaceDispatch, M as discoverEvalFixtures, N as loadEvalFixture, O as LabeledScenarioStoreError, P as loadEvalFixtureScenarios, R as traceAnalystQualityJudge, S as scoreUserStory, T as createOpenAiCompatibleExecutionOwner, _ 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 neutralizationGate, k as classifyUngroundedLiterals, 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-CN_7xJdV.js";
+import { n as acquireSingleRunLock } from "../external-optimizer-process-x9oEXKsU.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-CPBlTcj5.js";
+import { A as rolloutArgumentDiff, C as scoreboardSummary, D as FsLabeledScenarioStore, E as neutralizeText, F as planEvalFixtureRun, I as analyzeCrossSurfaceInteractions, L as buildTraceAnalystSurfaceDispatch, M as discoverEvalFixtures, N as loadEvalFixture, O as LabeledScenarioStoreError, P as loadEvalFixtureScenarios, R as traceAnalystQualityJudge, S as scoreUserStory, T as createOpenAiCompatibleExecutionOwner, _ 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 neutralizationGate, k as classifyUngroundedLiterals, 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-la-gEYNz.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_HISTORY_RECEIPT_DIGEST_ALGORITHM, SEARCH_HISTORY_RECEIPT_SCHEMA_VERSION, SEARCH_LEDGER_SCHEMA, SearchHistoryRequiredError, SearchLedgerConflictError, SearchLedgerError, SearchLedgerIntegrityError, SegmentedProfileMatrixError, WorktreeAdapterError, acquireSingleRunLock, analyzeCrossSurfaceInteractions, assertCampaignDesign, assertCampaignSplitIdentity, assertCodeSurfaceIdentity, assertCompleteSearchHistory, assertComponentSurface, assertSearchHistoryMatchesReplay, autoevalsScorerJudge, buildEvidenceVector, buildLoopProvenanceRecord, buildTraceAnalystSurfaceDispatch, campaignBreakdown, campaignMeanComposite, campaignMeasurementDigest, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, combineComparisonCosts, compareOptimizationMethods, compareRankKeys, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createOpenAiCompatibleExecutionOwner, createProfileMatrixPlan, createReferenceEquivalenceJudge, createRunCostLedger, createSearchHistoryReceipt, 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, searchHistoryCoverageRow, selectDiscriminative, sequentialDecide, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceHash, tangleTracesRoot, traceAnalystQualityJudge, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord, verifySearchHistoryReceipt };

package/dist/{campaign-CN_7xJdV.js → campaign-la-gEYNz.js}

@@ -1,21 +1,21 @@
 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-jfk8Du3b.js";
+import { c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, s as agentProfileHash, t as extractProducedState } from "./produced-state-Bnq4FaDO.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 { at as assertRealBackend, et as runCampaign, h as labelTrustRank, k as surfaceContentHash, lt as campaignScenarioIdentity, mt as computeManifestHash, nt as resolveRunDir, ot as summarizeBackendIntegrity, pt as cellCachePath, st as assertCampaignDesign, tt as planCampaignRun, w as assertCodeSurfaceIdentity } from "./llm-judge-CVq33oz1.js";
-import { a as campaignCellCostProvenance, l as campaignCellToRunRecord, u as projectCampaignCellQuality } from "./reward-hacking-DKI9T52l.js";
+import { at as assertRealBackend, et as runCampaign, h as labelTrustRank, k as surfaceContentHash, lt as campaignScenarioIdentity, mt as computeManifestHash, nt as resolveRunDir, ot as summarizeBackendIntegrity, pt as cellCachePath, st as assertCampaignDesign, tt as planCampaignRun, w as assertCodeSurfaceIdentity } from "./llm-judge-BqqMS8t7.js";
+import { a as campaignCellCostProvenance, l as campaignCellToRunRecord, u as projectCampaignCellQuality } from "./reward-hacking-C0x0xihA.js";
 import { t as CostAccountingIncompleteError } from "./cost-ledger-B1qx30B4.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-Dn90UqN2.js";
+import { C as SEARCH_LEDGER_FILE_CONTEXT, E as SearchLedgerIntegrityError, T as SearchLedgerError, w as SearchLedgerConflictError, x as fsCampaignStorage } from "./external-optimizer-subprocess-CKNb42oM.js";
 import { o as pairHoldout } from "./power-preflight-DEw-uC7q.js";
 import { l as costReceiptFromLlm, s as callLlm, u as costReceiptFromLlmError } from "./llm-client-Bg32RW0j.js";
 import { a as scoreAnalystFindings } from "./benchmark-BhT16ep9.js";
-import "./external-optimizer-process-Bhmzngf-.js";
-import "./skillopt-optimization-method-DLeUcK-K.js";
+import "./external-optimizer-process-x9oEXKsU.js";
+import "./skillopt-optimization-method-CPBlTcj5.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";
@@ -3767,4 +3767,4 @@ function resolveWorktreePath(surface, worktreeDir) {
 //#endregion
 export { rolloutArgumentDiff as A, scoreboardSummary as C, FsLabeledScenarioStore as D, neutralizeText as E, planEvalFixtureRun as F, analyzeCrossSurfaceInteractions as I, buildTraceAnalystSurfaceDispatch as L, discoverEvalFixtures as M, loadEvalFixture as N, LabeledScenarioStoreError as O, loadEvalFixtureScenarios as P, traceAnalystQualityJudge as R, scoreUserStory as S, createOpenAiCompatibleExecutionOwner 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, neutralizationGate as j, classifyUngroundedLiterals 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-CN_7xJdV.js.map
+//# sourceMappingURL=campaign-la-gEYNz.js.map