@tangle-network/agent-eval 0.191.0 → 0.192.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +38 -0
- package/dist/adapters/http.d.ts +2 -2
- package/dist/analyst/index.d.ts +6 -6
- package/dist/analyst/index.js +3 -3
- package/dist/{benchmark-command-3tZJnD82.js → benchmark-command-C458yEFK.js} +7 -7
- package/dist/{benchmark-command-3tZJnD82.js.map → benchmark-command-C458yEFK.js.map} +1 -1
- package/dist/benchmarks/index.d.ts +3 -3
- package/dist/benchmarks/index.js +3 -3
- package/dist/campaign/index.d.ts +4 -4
- package/dist/campaign/index.js +8 -9
- package/dist/{campaign-C8iWEF0M.js → campaign-BVQVBEfH.js} +12 -326
- package/dist/campaign-BVQVBEfH.js.map +1 -0
- package/dist/{campaign-evidence-BqjlTM3P.js → campaign-evidence-DcQiYhJI.js} +10 -45
- package/dist/campaign-evidence-DcQiYhJI.js.map +1 -0
- package/dist/cli.js +2 -2
- package/dist/contract/index.d.ts +6 -6
- package/dist/contract/index.js +6 -6
- package/dist/{default-registry-DK_9aVJf.d.ts → default-registry-DdqA_-_o.d.ts} +3 -3
- package/dist/{default-registry-DK_9aVJf.d.ts.map → default-registry-DdqA_-_o.d.ts.map} +1 -1
- package/dist/{define-agent-eval-BE43YTm4.js → define-agent-eval-DQuos3Aw.js} +8 -9
- package/dist/define-agent-eval-DQuos3Aw.js.map +1 -0
- package/dist/{define-agent-eval-P28y7mI1.d.ts → define-agent-eval-DcludNJS.d.ts} +13 -22
- package/dist/{define-agent-eval-P28y7mI1.d.ts.map → define-agent-eval-DcludNJS.d.ts.map} +1 -1
- package/dist/{dspy-rlm-engine-MfpwoEaj.js → dspy-rlm-engine-BX34vPVY.js} +2 -2
- package/dist/{dspy-rlm-engine-MfpwoEaj.js.map → dspy-rlm-engine-BX34vPVY.js.map} +1 -1
- package/dist/{engine-vNiM02Ub.d.ts → engine-C1DIAEpn.d.ts} +2 -2
- package/dist/{engine-vNiM02Ub.d.ts.map → engine-C1DIAEpn.d.ts.map} +1 -1
- package/dist/evaluation.d.ts +1 -1
- package/dist/experiment/index.d.ts +1 -1
- package/dist/experiment/index.js +1 -1
- package/dist/{external-optimizer-process-xFTliuiK.js → external-optimizer-process-D-Sab4v3.js} +3 -62
- package/dist/external-optimizer-process-D-Sab4v3.js.map +1 -0
- package/dist/{external-optimizer-subprocess-B234V9V0.js → external-optimizer-subprocess-kbqC-iWf.js} +64 -3
- package/dist/external-optimizer-subprocess-kbqC-iWf.js.map +1 -0
- package/dist/hosted/index.js +1 -1
- package/dist/{index-I4UJGRno.d.ts → index-Bj96W0jZ.d.ts} +28 -2
- package/dist/{index-I4UJGRno.d.ts.map → index-Bj96W0jZ.d.ts.map} +1 -1
- package/dist/{index-D-TN7ZRM.d.ts → index-DBoXFaz0.d.ts} +3 -3
- package/dist/{index-D-TN7ZRM.d.ts.map → index-DBoXFaz0.d.ts.map} +1 -1
- package/dist/{index-DH-MWH5f.d.ts → index-DNVzu9Vv.d.ts} +6 -6
- package/dist/{index-DH-MWH5f.d.ts.map → index-DNVzu9Vv.d.ts.map} +1 -1
- package/dist/{index-CzfnASj_.d.ts → index-DwQxdWve.d.ts} +338 -129
- package/dist/index-DwQxdWve.d.ts.map +1 -0
- package/dist/index.d.ts +9 -9
- package/dist/index.js +6 -6
- package/dist/jev.d.ts +1 -1
- package/dist/{journal-c3qY0SDP.js → journal-DvKCvjty.js} +94 -2
- package/dist/{journal-c3qY0SDP.js.map → journal-DvKCvjty.js.map} +1 -1
- package/dist/ledger-core/index.d.ts +2 -2
- package/dist/ledger-core/index.js +2 -2
- package/dist/{llm-judge-NJOHM0iA.js → llm-judge-Z6MkTkd0.js} +1819 -609
- package/dist/llm-judge-Z6MkTkd0.js.map +1 -0
- package/dist/{matrix-CgqdFm5N.d.ts → matrix-CQMMrEQ5.d.ts} +2 -2
- package/dist/{matrix-CgqdFm5N.d.ts.map → matrix-CQMMrEQ5.d.ts.map} +1 -1
- package/dist/multishot/golden/index.d.ts +1 -1
- package/dist/multishot/index.d.ts +2 -2
- package/dist/openapi.json +1 -1
- package/dist/{produced-state-CY-L7CqS.js → produced-state-CVmCPQDJ.js} +3 -3
- package/dist/{produced-state-CY-L7CqS.js.map → produced-state-CVmCPQDJ.js.map} +1 -1
- package/dist/rl.d.ts +1 -1
- package/dist/{search-ledger-CQ8Uy4FK.js → search-ledger-Ck7tok0h.js} +78 -12
- package/dist/search-ledger-Ck7tok0h.js.map +1 -0
- package/dist/{search-ledger-file-EvdDZRhv.js → search-ledger-file-DY0AXX3W.js} +2 -2
- package/dist/{search-ledger-file-EvdDZRhv.js.map → search-ledger-file-DY0AXX3W.js.map} +1 -1
- package/dist/{search-shipper-C55V7HE3.js → search-shipper-BG182p-U.js} +3 -3
- package/dist/{search-shipper-C55V7HE3.js.map → search-shipper-BG182p-U.js.map} +1 -1
- package/dist/{semantic-concept-judge-Bt7bXAAo.js → semantic-concept-judge-Bg-560H0.js} +3 -3
- package/dist/{semantic-concept-judge-Bt7bXAAo.js.map → semantic-concept-judge-Bg-560H0.js.map} +1 -1
- package/dist/{skillopt-optimization-method-DwO3TOAV.js → skillopt-optimization-method-DoayuM6D.js} +6 -6
- package/dist/{skillopt-optimization-method-DwO3TOAV.js.map → skillopt-optimization-method-DoayuM6D.js.map} +1 -1
- package/dist/{statistical-heldout-BED3wBtX.d.ts → statistical-heldout-C1Ej5Ce8.d.ts} +2 -2
- package/dist/{statistical-heldout-BED3wBtX.d.ts.map → statistical-heldout-C1Ej5Ce8.d.ts.map} +1 -1
- package/dist/{store-tool-spans-Cfh9XUCm.d.ts → store-tool-spans-ByLdU26j.d.ts} +3 -3
- package/dist/{store-tool-spans-Cfh9XUCm.d.ts.map → store-tool-spans-ByLdU26j.d.ts.map} +1 -1
- package/dist/{tool-groups-D-mkeN0N.d.ts → tool-groups-DiBi16ns.d.ts} +2 -2
- package/dist/tool-groups-DiBi16ns.d.ts.map +1 -0
- package/dist/trace-repair/index.d.ts +1 -1
- package/dist/traces.d.ts +2 -2
- package/dist/{types-ClK-sHs8.d.ts → types-ByZslIlr.d.ts} +2 -2
- package/dist/{types-ClK-sHs8.d.ts.map → types-ByZslIlr.d.ts.map} +1 -1
- package/dist/{types-BmQRs661.d.ts → types-C1CAD70_.d.ts} +14 -15
- package/dist/types-C1CAD70_.d.ts.map +1 -0
- package/docs/campaign-proposers.md +14 -16
- package/docs/search-ledger.md +62 -1
- package/package.json +1 -1
- package/dist/campaign-C8iWEF0M.js.map +0 -1
- package/dist/campaign-evidence-BqjlTM3P.js.map +0 -1
- package/dist/define-agent-eval-BE43YTm4.js.map +0 -1
- package/dist/external-optimizer-process-xFTliuiK.js.map +0 -1
- package/dist/external-optimizer-subprocess-B234V9V0.js.map +0 -1
- package/dist/index-CzfnASj_.d.ts.map +0 -1
- package/dist/llm-judge-NJOHM0iA.js.map +0 -1
- package/dist/search-ledger-CQ8Uy4FK.js.map +0 -1
- package/dist/tool-groups-D-mkeN0N.d.ts.map +0 -1
- package/dist/types-BmQRs661.d.ts.map +0 -1
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import { c as RunSplitTag } from "../run-record-DF-fuu31.js";
|
|
2
|
-
import { a as CampaignResult } from "../types-
|
|
3
|
-
import {
|
|
4
|
-
import { a as BenchmarkFamily, c as BenchmarkSource, i as BenchmarkEvaluation, l as BenchmarkTaskKind, n as BenchmarkAdapter, o as BenchmarkResponder, r as BenchmarkDatasetItem, s as BenchmarkScenario, t as BENCHMARK_SPLIT_SEED, u as deterministicSplit } from "../types-
|
|
2
|
+
import { a as CampaignResult } from "../types-C1CAD70_.js";
|
|
3
|
+
import { qo as CampaignStorage } from "../index-DwQxdWve.js";
|
|
4
|
+
import { a as BenchmarkFamily, c as BenchmarkSource, i as BenchmarkEvaluation, l as BenchmarkTaskKind, n as BenchmarkAdapter, o as BenchmarkResponder, r as BenchmarkDatasetItem, s as BenchmarkScenario, t as BENCHMARK_SPLIT_SEED, u as deterministicSplit } from "../types-ByZslIlr.js";
|
|
5
5
|
//#region src/benchmarks/calibration.d.ts
|
|
6
6
|
interface BenchmarkMetricCalibrationOptions<TPayload = unknown, TArtifact = string> {
|
|
7
7
|
adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, TArtifact>;
|
package/dist/benchmarks/index.js
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import { t as __exportAll } from "../rolldown-runtime-8H4AJuhK.js";
|
|
2
|
-
import {
|
|
3
|
-
import {
|
|
4
|
-
import "../campaign-
|
|
2
|
+
import { ot as runCampaign } from "../llm-judge-Z6MkTkd0.js";
|
|
3
|
+
import { C as fsCampaignStorage } from "../external-optimizer-subprocess-kbqC-iWf.js";
|
|
4
|
+
import "../campaign-BVQVBEfH.js";
|
|
5
5
|
import { join, resolve } from "node:path";
|
|
6
6
|
import { readFile, readdir, stat } from "node:fs/promises";
|
|
7
7
|
import { parse } from "smol-toml";
|
package/dist/campaign/index.d.ts
CHANGED
|
@@ -1,9 +1,9 @@
|
|
|
1
1
|
import { c as CostLedgerHandle, w as PendingCostCallView } from "../cost-ledger-DbQdN3nO.js";
|
|
2
2
|
import { _ as ProposalFinding, v as ProposalFindingOrigin, x as makeProposalFinding } from "../types-CeDfCRXp.js";
|
|
3
3
|
import { t as SeriesDistribution } from "../descriptive-B2iPaT9J.js";
|
|
4
|
-
import { $ as paretoSignificanceGate, B as PowerPreflight, G as BuildEvidenceVectorOptions, H as powerPreflight, J as ParetoSignificanceGateOptions, K as EvidenceVector, Q as paretoPolicy, U as AxisEvidence, V as PowerPreflightOptions, W as AxisVerdict, X as PromotionPolicy, Y as PromotionObjective, Z as buildEvidenceVector, _ as sequentialPairedGate, a as detectScale, c as pairHoldout, d as SequentialDecision, f as SequentialObservation, g as sequentialDecide, h as SequentialStreamState, i as PairedHoldout, l as SequentialDecideFn, m as SequentialPairedGateOptions, n as HeldoutSignificance, o as dimensionRegressions, p as SequentialPairedGate, q as ObjectiveSource, r as HeldoutSignificanceOptions, s as heldoutSignificance, t as DimensionRegression, u as SequentialDecideOptions } from "../statistical-heldout-
|
|
5
|
-
import { A as LabeledScenarioWrite, B as ScoredSurfaceOutcome, C as JudgeDimension, D as LabeledScenarioSampleArgs, E as LabeledScenarioRecord, F as ProposeContext, G as labelTrustRank, H as SurfaceProposer, I as ProposedCandidate, L as RedactionStatus, M as OptimizerConfig, N as ParetoParent, O as LabeledScenarioSource, P as ProposalTrackContext, R as Scenario, S as JudgeConfig, T as LabelTrust, U as TraceSpan, V as SessionScript, W as isProposedCandidate, _ as GateDecision, a as CampaignResult, b as GenerationRecord, c as CampaignTraceWriter, d as DispatchContext, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, m as GateCheckStatus, n as CampaignArtifactWriter, o as CampaignScenarioIdentity, p as Gate, r as CampaignCellResult, s as CampaignTokenUsage, t as CampaignAggregates, u as ComponentSurface, v as GateResult, w as JudgeScore, x as JudgeAggregate, y as GenerationCandidate, z as ScenarioAggregate } from "../types-
|
|
6
|
-
import { $ as
|
|
4
|
+
import { $ as paretoSignificanceGate, B as PowerPreflight, G as BuildEvidenceVectorOptions, H as powerPreflight, J as ParetoSignificanceGateOptions, K as EvidenceVector, Q as paretoPolicy, U as AxisEvidence, V as PowerPreflightOptions, W as AxisVerdict, X as PromotionPolicy, Y as PromotionObjective, Z as buildEvidenceVector, _ as sequentialPairedGate, a as detectScale, c as pairHoldout, d as SequentialDecision, f as SequentialObservation, g as sequentialDecide, h as SequentialStreamState, i as PairedHoldout, l as SequentialDecideFn, m as SequentialPairedGateOptions, n as HeldoutSignificance, o as dimensionRegressions, p as SequentialPairedGate, q as ObjectiveSource, r as HeldoutSignificanceOptions, s as heldoutSignificance, t as DimensionRegression, u as SequentialDecideOptions } from "../statistical-heldout-C1Ej5Ce8.js";
|
|
5
|
+
import { A as LabeledScenarioWrite, B as ScoredSurfaceOutcome, C as JudgeDimension, D as LabeledScenarioSampleArgs, E as LabeledScenarioRecord, F as ProposeContext, G as labelTrustRank, H as SurfaceProposer, I as ProposedCandidate, L as RedactionStatus, M as OptimizerConfig, N as ParetoParent, O as LabeledScenarioSource, P as ProposalTrackContext, R as Scenario, S as JudgeConfig, T as LabelTrust, U as TraceSpan, V as SessionScript, W as isProposedCandidate, _ as GateDecision, a as CampaignResult, b as GenerationRecord, c as CampaignTraceWriter, d as DispatchContext, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, m as GateCheckStatus, n as CampaignArtifactWriter, o as CampaignScenarioIdentity, p as Gate, r as CampaignCellResult, s as CampaignTokenUsage, t as CampaignAggregates, u as ComponentSurface, v as GateResult, w as JudgeScore, x as JudgeAggregate, y as GenerationCandidate, z as ScenarioAggregate } from "../types-C1CAD70_.js";
|
|
6
|
+
import { $ as ProfileMatrixSegmentResult, $a as searchHistoryCoverageRow, $i as ExternalOptimizerObservationSummary, $n as CrossSurfaceInteractionPath, $o as assertCampaignDesign, $t as RolloutCall, A as SearchExecutor, Aa as RecordSearchEdgeInput, Ai as OptimizationMethod, An as searchCellSetDigest, Ao as openAutoPr, Ar as campaignMeasurementDigest, B as searchPolicyView, Ba as surfaceDiff, Bi as OptimizationTokenUsage, Bn as CrossSurfaceCandidateComparison, Bo as planCampaignRun, Br as campaignBreakdown, C as SingleRunLockOptions, Ca as runOptimization, Ci as ExternalTextOptimizationMethodConfig, Cn as planEvalFixtureRun, Co as searchTaskSetDigest, Cr as LoopProvenanceArgsFromResult, Ct as runProfileMatrix, D as SearchArtifactCodec, Da as crowdedFrontierParent, Di as ExternalTextEvaluationResponse, Dn as SearchPosterior, Do as SearchLedgerIntegrityError, Dr as LoopProvenanceOptimizationMethod, E as SEARCH_KERNEL_SOURCE, Ea as SearchPolicyView, Ei as ExternalOptimizationExample, En as SEARCH_ESTIMATOR, Eo as SearchLedgerError, Er as LoopProvenanceEvidence, F as SearchProposedChild, Fa as SearchRecorderOptions, Fi as OptimizationMethodProvenance, Fn as CrossSurfaceAdditionRejectionReason, Fo as RunCampaignOptions, Fr as provenanceRecordPath, G as resolveRunDir, Ga as SearchHistoryCoverage, Gi as costFromLedgerSummary, Gn as CrossSurfaceComponentEvidence, Go as buildCellSchedule, Gr as DefaultProductionGateOptions, Gt as scopedOptimizationMethod, H as ScenarioSignal, Ha as CreateSearchHistoryReceiptInput, Hi as optimizationTokenUsageFromSummary, Hn as CrossSurfaceCandidateOutcome, Ho as CacheRead, Hr as HeldOutGateOptions, I as SearchProposerPort, Ia as SearchRunIdentity, Ii as OptimizationMethodResult, In as CrossSurfaceAttemptCompleteness, Io as runCampaign, Ir as provenanceSpansPath, J as FinalizeProfileMatrixOptions, Ja as SearchHistoryReceipt, Ji as GepaCandidatePopulationSummary, Jn as CrossSurfaceEligibility, Jo as createRunCostLedger, Jt as FsLabeledScenarioStore, K as tangleTracesRoot, Ka as SearchHistoryCoverageRow, Ki as GepaCandidatePopulationArtifact, Kn as CrossSurfaceCompositionStep, Ko as cellCachePath, Kr as DefaultProductionRewardHackingOptions, Kt as sequentialOptimizationMethod, L as SearchRunResult, La as SettleSearchCellInput, Li as OptimizationMethodRunOptions, Ln as CrossSurfaceBestSingleSelection, Lo as CampaignRunPlan, Lr as verifyLoopProvenanceRecord, M as SearchProposalBlob, Ma as SearchLedgerBinding, Mi as OptimizationMethodComposition, Mn as analyzeCrossSurfaceInteractions, Mo as runEval, Mr as emitLoopProvenance, N as SearchProposalRequest, Na as SearchOpening, Ni as OptimizationMethodInput, Nn as AnalyzeCrossSurfaceInteractionsInput, No as CampaignCellFailureReceipt, Nr as loopProvenanceArgsFromResult, O as SearchCellResult, Oa as incumbent, Oi as decodeExternalTextCandidate, On as estimateNode, Oo as OpenAutoPrOptions, Or as LoopProvenanceRecord, P as SearchProposalResult, Pa as SearchRecorder, Pi as OptimizationMethodPairwise, Pn as CrossSurfaceAdditionDecision, Po as CampaignCellRetryPolicy, Pr as loopProvenanceSpans, Q as ProfileMatrixRow, Qa as createSearchHistoryReceipt, Qi as ExternalOptimizerObservationArtifact, Qn as CrossSurfaceInteractionEffect, Qo as FinalEvidenceUse, Qt as RolloutArgumentDiffOptions, R as runSearch, Ra as developmentClaim, Ri as OptimizationMethodScore, Rn as CrossSurfaceBootstrapPolicy, Ro as CampaignRunPlanCell, Rr as CampaignBreakdown, S as SingleRunLock, Sa as RunOptimizationResult, Si as externalTextOptimizationMethod, Sn as loadEvalFixtureScenarios, So as searchNodeId, Sr as EmitLoopProvenanceResult, St as ScenarioRollup, T as RunSearchOptions, Ta as SearchPolicy, Ti as ExternalTextOptimizerResult, Tn as NodePosterior, To as SearchLedgerConflictError, Tr as LoopProvenanceCandidate, U as scoreDiscrimination, Ua as SearchHistoryAdmissionOptions, Ui as ComparisonCost, Un as CrossSurfaceCandidateSummary, Uo as readCachedCell, Ur as heldOutGate, V as DiscriminationScore, Va as surfaceNode, Vi as compareOptimizationMethods, Vn as CrossSurfaceCandidateEvidence, Vo as CacheIssueReason, Vr as campaignMeanComposite, W as selectDiscriminative, Wa as SearchHistoryAuditSummary, Wi as combineComparisonCosts, Wn as CrossSurfaceComponent, Wo as CellScheduleSlot, Wr as DefaultProductionGateCheck, X as ProfileMatrixCoverage, Xa as assertCompleteSearchHistory, Xi as readGepaCandidatePopulationArtifact, Xn as CrossSurfaceIneligibilityReason, Xo as inMemoryCampaignStorage, Xt as LabeledScenarioStoreError, Y as FinalizedProfileMatrixResult, Ya as SearchHistoryRequiredError, Yi as GepaCandidateSelectionScore, Yn as CrossSurfaceEvidenceBreakdown, Yo as fsCampaignStorage, Yt as FsLabeledScenarioStoreOptions, Z as ProfileMatrixPlan, Za as assertSearchHistoryMatchesState, Zi as ExternalOptimizerExecutionSummary, Zn as CrossSurfaceInteractionAwareSelection, Zo as FinalEvidencePolicy, Zt as RolloutArgumentDiff, _ as componentSurfaceIdentityMaterial, _a as RunImprovementLoopOptions, _i as GepaOptimizationRecipe, _n as EvalFixtureValidationMode, _o as SearchSpend, _r as SearchAllocator, _t as ProfileDispatchFn, a as WorktreeAdapterError, aa as ReferenceEquivalenceJudgeOptions, an as GepaPopulationImport, ao as SearchLedgerAppendResult, ar as CrossSurfacePairIncompatibilityReason, at as PlaybackDriver, b as surfaceDispatchRef, ba as PremeasuredOptimizationBaseline, bi as OpenAICompatibleOptimizerModel, bn as discoverEvalFixtures, bo as SearchUnitScore, br as BuildLoopProvenanceArgs, bt as RunProfileMatrixOptions, c as verifyCodeSurface, ca as createReferenceEquivalenceJudge, ci as composeGate, cn as importGepaPopulation, co as parseSearchLedgerLine, cr as CrossSurfaceRelativeCost, ct as ScoreboardRow, d as PhoenixEvaluationResultLike, da as LlmJudgeOptions, di as SkillOptTrainerConfig, dn as neutralizationGate, do as SearchCell, dr as CrossSurfaceTaskRow, dt as UserStoryVerdict, ea as ExternalOptimizerSubmittedCandidate, en as ScoredRollout, eo as verifySearchHistoryArtifact, er as CrossSurfaceInteractionReport, es as assertCampaignSplitIdentity, et as RunProfileMatrixSegmentOptions, f as PhoenixEvaluatorLike, fa as llmJudge, fi as skillOptOptimizationMethod, fn as EvalFixture, fo as SearchCompletion, fr as BuildTraceAnalystSurfaceDispatchOptions, ft as makePlaybackDispatch, g as codeSurfaceIdentityMaterial, ga as transientDispatchFailure, gi as GepaOptimizationMethodConfig, gn as EvalFixtureScenario, go as SearchScoredCell, gr as traceAnalystQualityJudge, gt as userStoryScoreboard, h as assertCodeSurfaceIdentity, ha as quotaExhaustedUntil, hi as GepaEngineRun, hn as EvalFixtureRunPlan, ho as SearchOperation, hr as buildTraceAnalystSurfaceDispatch, ht as scoreboardSummary, i as WorktreeAdapter, in as GepaEvaluationImport, io as SearchLedger, ir as CrossSurfacePairEvidence, it as PlaybackContext, j as SearchLane, ja as RegisterSearchNodeInput, ji as OptimizationMethodComparison, jn as searchPosterior, jo as RunEvalOptions, jr as canonicalDigest, k as SearchCellWork, ka as AllocateSearchCellInput, ki as CompareOptimizationMethodsOptions, kn as estimateNodeFromCells, ko as OpenAutoPrResult, kr as buildLoopProvenanceRecord, l as AutoevalsScoreLike, li as SkillOptOptimizationMethodConfig, ln as recordGepaSearch, lo as replaySearchLedgerText, lr as CrossSurfaceSelectionPolicy, lt as ScoreboardSummary, m as phoenixEvaluatorJudge, ma as isTransientTransportFailure, mi as GepaEngineOptions, mn as EvalFixtureLoadOptions, mo as SearchNode, mr as TraceAnalystScenario, mt as scoreUserStory, n as GitWorktreeAdapterOptions, nn as classifyUngroundedLiterals, no as FileSearchLedger, nr as CrossSurfaceNaiveStackSelection, ns as campaignSplitDigest, nt as finalizeProfileMatrix, o as gitWorktreeAdapter, on as externalSurface, oo as SearchLedgerTrustedHeadMode, or as CrossSurfacePairwiseEntry, ot as PlaybackStep, p as autoevalsScorerJudge, pa as TransientFailureOptions, pi as GepaAdaptiveEngineRun, pn as EvalFixtureFile, po as SearchDecisionRecord, pr as TraceAnalystArtifact, pt as renderScoreboardMarkdown, q as CreateProfileMatrixPlanOptions, qa as SearchHistoryPolicy, qi as GepaCandidatePopulationCandidate, qn as CrossSurfaceDistribution, qo as CampaignStorage, qr as defaultProductionGate, qt as neutralizeText, r as Worktree, rn as rolloutArgumentDiff, ro as OpenSearchLedgerOptions, rr as CrossSurfacePairCompatibility, rs as campaignSplitDigestFromIdentities, rt as runProfileMatrixSegment, s as resolveWorktreePath, sa as ReferenceEquivalenceScenario, sn as importExternalEvaluations, so as openSearchLedger, sr as CrossSurfaceRankedSingle, st as ScoreboardRenderOptions, t as CodeSurfaceVerification, ta as readExternalOptimizerObservationArtifact, tn as UngroundedLiteralReport, to as verifySearchHistoryReceipt, tr as CrossSurfaceInteractionTask, ts as campaignScenarioIdentity, tt as createProfileMatrixPlan, u as AutoevalsScorerLike, ua as LlmJudgeDimension, ui as SkillOptRunnerCommand, un as NeutralizationGateOptions, uo as validateSearchLedgerEvent, ur as CrossSurfaceSelections, ut as UserStory, v as renderSurfaceDiff, va as RunImprovementLoopResult, vi as GepaRunnerCommand, vn as LoadEvalFixtureScenariosOptions, vo as SearchState, vr as SearchCellPlan, vt as ProfileMatrixError, w as acquireSingleRunLock, wa as SearchExpansion, wi as ExternalTextOptimizerContext, wn as EstimateNodeCellsInput, wo as searchUnitScores, wr as LoopProvenanceBackend, x as surfaceHash, xa as RunOptimizationOptions, xi as OptimizerModelBudget, xn as loadEvalFixture, xo as searchCellId, xr as EmitLoopProvenanceArgs, xt as RunProfileMatrixResult, y as surfaceContentHash, ya as runImprovementLoop, yi as gepaOptimizationMethod, yn as PlanEvalFixtureRunOptions, yo as SearchStateView, yr as uniform, yt as ProfileSummary, z as searchExpansionIndex, za as searchEdgeId, zi as OptimizationPackageSource, zn as CrossSurfaceCandidate, zo as PlanCampaignRunOptions } from "../index-DwQxdWve.js";
|
|
7
7
|
import { A as SearchNodeDecision, B as SearchReservation, C as SearchExecutionIdentity, D as SearchLedgerHash, E as SearchLedgerEvent, F as SearchOperationKind, G as SearchSurfaceEffect, H as SearchSplit, I as SearchOperationRecordedEvent, J as SearchTask, K as SearchSurfaceEvidence, L as SearchOperationStartedEvent, M as SearchNodeRegisteredEvent, N as SearchNodeStatus, O as SearchModelIdentity, P as SearchOpenedEvent, Q as SearchUnknown, R as SearchProposer, S as SearchEstimateMethod, T as SearchLedgerEntry, U as SearchSplitTasks, V as SearchSourceRef, W as SearchSplits, X as SearchTokenAccounting, Y as SearchTaskOutcome, Z as SearchTraceRef, _ as SearchClosedEvent, a as SearchAttemptAccounting, b as SearchEdgeOperator, c as SearchCancelReason, d as SearchCellCancelledEvent, f as SearchCellSettledEvent, g as SearchCloseReason, h as SearchClaimPower, i as SearchArtifactRef, j as SearchNodeRef, k as SearchNodeDecidedEvent, l as SearchCandidateSurface, m as SearchClaim, n as SEARCH_LEDGER_SCHEMA, o as SearchAudit, p as SearchCellStage, q as SearchSurfaceKind, r as SearchArtifactKind, s as SearchBudget, t as NodeEstimate, u as SearchCellAllocatedEvent, v as SearchCostAccounting, w as SearchFailureReason, x as SearchEdgeRecordedEvent, y as SearchEdgeAttribution, z as SearchProposerKind } from "../search-ledger-types-C9anU2DX.js";
|
|
8
8
|
import { B as ExternalOptimizerModelBudget, F as ExternalOptimizerCallbackLimits, G as ExternalOptimizerProcessLimits, H as ExternalOptimizerModelCallRequest, I as ExternalOptimizerChatRequest, L as ExternalOptimizerEndpointFormat, N as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, P as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, Q as resolveExternalOptimizerProcessLimits, R as ExternalOptimizerEvaluationObservation, U as ExternalOptimizerModelCallResult, V as ExternalOptimizerModelCall, W as ExternalOptimizerModelExecutionObservation, Y as ExternalTextCandidate, Z as resolveExternalOptimizerCallbackLimits, q as ExternalOptimizerRunnerCommand, z as ExternalOptimizerEvaluationRefusalReason } from "../agent-profile-DDfIjdyS.js";
|
|
9
|
-
export { type AllocateSearchCellInput, type AnalyzeCrossSurfaceInteractionsInput, type AutoevalsScoreLike, type AutoevalsScorerLike, type AxisEvidence, type AxisVerdict, type BuildEvidenceVectorOptions, type BuildLoopProvenanceArgs, type BuildTraceAnalystSurfaceDispatchOptions, type CacheIssueReason, type CacheRead, type CampaignAggregates, type CampaignArtifactWriter, type CampaignBreakdown, type CampaignCellFailureReceipt, type CampaignCellResult, type CampaignCellRetryPolicy, type CampaignCostMeter, type CampaignResult, type CampaignRunPlan, type CampaignRunPlanCell, type CampaignScenarioIdentity, type CampaignStorage, type CampaignTokenUsage, type CampaignTraceWriter, type CellScheduleSlot, type CodeSurface, type CodeSurfaceVerification, type CompareOptimizationMethodsOptions, type ComparisonCost, type ComponentSurface, type CostLedgerHandle, type CreateProfileMatrixPlanOptions, type CreateSearchHistoryReceiptInput, type CrossSurfaceAdditionDecision, type CrossSurfaceAdditionRejectionReason, type CrossSurfaceAttemptCompleteness, type CrossSurfaceBestSingleSelection, type CrossSurfaceBootstrapPolicy, type CrossSurfaceCandidate, type CrossSurfaceCandidateComparison, type CrossSurfaceCandidateEvidence, type CrossSurfaceCandidateOutcome, type CrossSurfaceCandidateSummary, type CrossSurfaceComponent, type CrossSurfaceComponentEvidence, type CrossSurfaceCompositionStep, type CrossSurfaceDistribution, type CrossSurfaceEligibility, type CrossSurfaceEvidenceBreakdown, type CrossSurfaceIneligibilityReason, type CrossSurfaceInteractionAwareSelection, type CrossSurfaceInteractionEffect, type CrossSurfaceInteractionPath, type CrossSurfaceInteractionReport, type CrossSurfaceInteractionTask, type CrossSurfaceNaiveStackSelection, type CrossSurfacePairCompatibility, type CrossSurfacePairEvidence, type CrossSurfacePairIncompatibilityReason, type CrossSurfacePairwiseEntry, type CrossSurfaceRankedSingle, type CrossSurfaceRelativeCost, type CrossSurfaceSelectionPolicy, type CrossSurfaceSelections, type CrossSurfaceTaskRow, type CrowdedFrontierParentOptions, DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, type DimensionRegression, type DiscriminationScore, type DispatchContext, type DispatchFn, type EmitLoopProvenanceArgs, type EmitLoopProvenanceResult, type EstimateNodeCellsInput, type EvalFixture, type EvalFixtureFile, type EvalFixtureLoadOptions, type EvalFixtureRunPlan, type EvalFixtureScenario, type EvalFixtureValidationMode, type EvidenceVector, type ExternalOptimizationExample, type ExternalOptimizerCallbackLimits, type ExternalOptimizerChatRequest, type ExternalOptimizerEndpointFormat, type ExternalOptimizerEvaluationObservation, type ExternalOptimizerEvaluationRefusalReason, type ExternalOptimizerExecutionSummary, type ExternalOptimizerModelBudget, type ExternalOptimizerModelCall, type ExternalOptimizerModelCallRequest, type ExternalOptimizerModelCallResult, type ExternalOptimizerModelExecutionObservation, type ExternalOptimizerObservationArtifact, type ExternalOptimizerObservationSummary, type ExternalOptimizerProcessLimits, type ExternalOptimizerRunnerCommand, type ExternalOptimizerSubmittedCandidate, type ExternalTextCandidate, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, FileSearchLedger, type FinalEvidencePolicy, type FinalEvidenceUse, type FinalizeProfileMatrixOptions, type FinalizedProfileMatrixResult, FsLabeledScenarioStore, type FsLabeledScenarioStoreOptions, type Gate, type GateCheckStatus, type GateContext, type GateContribution, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaAdaptiveEngineRun, type GepaCandidatePopulationArtifact, type GepaCandidatePopulationCandidate, type GepaCandidatePopulationSummary, type GepaCandidateSelectionScore, type GepaEngineOptions, type GepaEngineRun, type GepaEvaluationImport, type GepaOptimizationMethodConfig, type GepaOptimizationRecipe, type GepaPopulationImport, type GepaRunnerCommand, type GitWorktreeAdapterOptions, type HeldOutGateOptions, type HeldoutSignificance, type HeldoutSignificanceOptions, type JudgeAggregate, type JudgeConfig, type JudgeDimension, type JudgeScore, type LabelTrust, type LabeledScenarioRecord, type LabeledScenarioSampleArgs, type LabeledScenarioSource, type LabeledScenarioStore, LabeledScenarioStoreError, type LabeledScenarioWrite, type LlmJudgeDimension, type LlmJudgeOptions, type LoadEvalFixtureScenariosOptions, type LoopProvenanceArgsFromResult, type LoopProvenanceBackend, type LoopProvenanceCandidate, type LoopProvenanceEvidence, type LoopProvenanceOptimizationMethod, type LoopProvenanceRecord, type MutableSurface, type NeutralizationGateOptions, type NodeEstimate, type NodePosterior, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OpenAutoPrOptions, type OpenAutoPrResult, type OpenSearchLedgerOptions, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodComposition, type OptimizationMethodInput, type OptimizationMethodPairwise, type OptimizationMethodProvenance, type OptimizationMethodResult, type OptimizationMethodRunOptions, type OptimizationMethodScore, type OptimizationPackageSource, type OptimizationTokenUsage, type OptimizerConfig, type OptimizerModelBudget, type PairedHoldout, type ParentSelectionContext, type ParentSelector, type ParetoParent, type ParetoSignificanceGateOptions, type PendingCostCallView, type PhoenixEvaluationResultLike, type PhoenixEvaluatorLike, type PlanCampaignRunOptions, type PlanEvalFixtureRunOptions, type PlaybackContext, type PlaybackDriver, type PlaybackStep, type PowerPreflight, type PowerPreflightOptions, type PremeasuredOptimizationBaseline, type ProfileDispatchFn, type ProfileMatrixCoverage, ProfileMatrixError, type ProfileMatrixPlan, type ProfileMatrixRow, type ProfileMatrixSegmentResult, type ProfileSummary, type PromotionObjective, type PromotionPolicy, type ProposalFinding, type ProposalFindingOrigin, type ProposalTrackContext, type ProposeContext, type ProposedCandidate, type RecordSearchEdgeInput, type RedactionStatus, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceScenario, type RegisterSearchNodeInput, type RolloutArgumentDiff, type RolloutArgumentDiffOptions, type RolloutCall, type RunCampaignOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunOptimizationOptions, type RunOptimizationResult, type RunProfileMatrixOptions, type RunProfileMatrixResult, type RunProfileMatrixSegmentOptions, SEARCH_ESTIMATOR, SEARCH_LEDGER_SCHEMA, type Scenario, type ScenarioAggregate, type ScenarioRollup, type ScenarioSignal, type ScoreboardRenderOptions, type ScoreboardRow, type ScoreboardSummary, type ScoredRollout, type ScoredSurfaceOutcome, type SearchArtifactKind, type SearchArtifactRef, type SearchAttemptAccounting, type SearchAudit, type SearchBudget, type SearchCancelReason, type SearchCandidateSurface, type SearchCell, type SearchCellAllocatedEvent, type SearchCellCancelledEvent, type SearchCellSettledEvent, type SearchCellStage, type SearchClaim, type SearchClaimPower, type SearchCloseReason, type SearchClosedEvent, type SearchCompletion, type SearchCostAccounting, type SearchDecisionRecord, type SearchEdgeAttribution, type SearchEdgeOperator, type SearchEdgeRecordedEvent, type SearchEstimateMethod, type SearchExecutionIdentity, type SearchFailureReason, type SearchHistoryAdmissionOptions, type SearchHistoryAuditSummary, type SearchHistoryCoverage, type SearchHistoryCoverageRow, type SearchHistoryPolicy, type SearchHistoryReceipt, SearchHistoryRequiredError, type SearchLedger, type SearchLedgerAppendResult, type SearchLedgerBinding, SearchLedgerConflictError, type SearchLedgerEntry, SearchLedgerError, type SearchLedgerEvent, type SearchLedgerHash, SearchLedgerIntegrityError, type SearchLedgerTrustedHeadMode, type SearchModelIdentity, type SearchNode, type SearchNodeDecidedEvent, type SearchNodeDecision, type SearchNodeRef, type SearchNodeRegisteredEvent, type SearchNodeStatus, type SearchOpenedEvent, type SearchOpening, type SearchOperation, type SearchOperationKind, type SearchOperationRecordedEvent, type SearchOperationStartedEvent, type SearchPosterior, type SearchProposer, type SearchProposerKind, SearchRecorder, type SearchRecorderOptions, type SearchReservation, type SearchRunIdentity, type SearchScoredCell, type SearchSourceRef, type SearchSpend, type SearchSplit, type SearchSplitTasks, type SearchSplits, SearchState, SearchStateView, type SearchSurfaceEffect, type SearchSurfaceEvidence, type SearchSurfaceKind, type SearchTask, type SearchTaskOutcome, type SearchTokenAccounting, type SearchTraceRef, type SearchUnitScore, type SearchUnknown, type SequentialDecideFn, type SequentialDecideOptions, type SequentialDecision, type SequentialObservation, type SequentialPairedGate, type SequentialPairedGateOptions, type SequentialStreamState, type SeriesDistribution, type SessionScript, type SettleSearchCellInput, type SingleRunLock, type SingleRunLockOptions, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SurfaceProposer, type TraceAnalystArtifact, type TraceAnalystScenario, type TraceSpan, type TransientFailureOptions, type UngroundedLiteralReport, type UserStory, type UserStoryVerdict, type Worktree, type WorktreeAdapter, 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, compareRankKeys, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createReferenceEquivalenceJudge, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decodeExternalTextCandidate, defaultProductionGate, detectScale, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, isProposedCandidate, isTransientTransportFailure, labelTrustRank, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, loopProvenanceArgsFromResult, loopProvenanceSpans, makePlaybackDispatch, makeProposalFinding, neutralizationGate, neutralizeText, openAutoPr, openSearchLedger, optimizationTokenUsageFromSummary, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, phoenixEvaluatorJudge, planCampaignRun, planEvalFixtureRun, powerPreflight, provenanceRecordPath, provenanceSpansPath, quotaExhaustedUntil, readCachedCell, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, recordGepaSearch, renderScoreboardMarkdown, renderSurfaceDiff, replaySearchLedgerText, resolveExternalOptimizerCallbackLimits, resolveExternalOptimizerProcessLimits, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchEdgeId, searchHistoryCoverageRow, searchNodeId, searchPosterior, searchTaskSetDigest, searchUnitScores, selectDiscriminative, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, tangleTracesRoot, traceAnalystQualityJudge, transientDispatchFailure, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord, verifySearchHistoryArtifact, verifySearchHistoryReceipt };
|
|
9
|
+
export { type AllocateSearchCellInput, type AnalyzeCrossSurfaceInteractionsInput, type AutoevalsScoreLike, type AutoevalsScorerLike, type AxisEvidence, type AxisVerdict, type BuildEvidenceVectorOptions, type BuildLoopProvenanceArgs, type BuildTraceAnalystSurfaceDispatchOptions, type CacheIssueReason, type CacheRead, type CampaignAggregates, type CampaignArtifactWriter, type CampaignBreakdown, type CampaignCellFailureReceipt, type CampaignCellResult, type CampaignCellRetryPolicy, type CampaignCostMeter, type CampaignResult, type CampaignRunPlan, type CampaignRunPlanCell, type CampaignScenarioIdentity, type CampaignStorage, type CampaignTokenUsage, type CampaignTraceWriter, type CellScheduleSlot, type CodeSurface, type CodeSurfaceVerification, type CompareOptimizationMethodsOptions, type ComparisonCost, type ComponentSurface, type CostLedgerHandle, type CreateProfileMatrixPlanOptions, type CreateSearchHistoryReceiptInput, type CrossSurfaceAdditionDecision, type CrossSurfaceAdditionRejectionReason, type CrossSurfaceAttemptCompleteness, type CrossSurfaceBestSingleSelection, type CrossSurfaceBootstrapPolicy, type CrossSurfaceCandidate, type CrossSurfaceCandidateComparison, type CrossSurfaceCandidateEvidence, type CrossSurfaceCandidateOutcome, type CrossSurfaceCandidateSummary, type CrossSurfaceComponent, type CrossSurfaceComponentEvidence, type CrossSurfaceCompositionStep, type CrossSurfaceDistribution, type CrossSurfaceEligibility, type CrossSurfaceEvidenceBreakdown, type CrossSurfaceIneligibilityReason, type CrossSurfaceInteractionAwareSelection, type CrossSurfaceInteractionEffect, type CrossSurfaceInteractionPath, type CrossSurfaceInteractionReport, type CrossSurfaceInteractionTask, type CrossSurfaceNaiveStackSelection, type CrossSurfacePairCompatibility, type CrossSurfacePairEvidence, type CrossSurfacePairIncompatibilityReason, type CrossSurfacePairwiseEntry, type CrossSurfaceRankedSingle, type CrossSurfaceRelativeCost, type CrossSurfaceSelectionPolicy, type CrossSurfaceSelections, type CrossSurfaceTaskRow, DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, type DimensionRegression, type DiscriminationScore, type DispatchContext, type DispatchFn, type EmitLoopProvenanceArgs, type EmitLoopProvenanceResult, type EstimateNodeCellsInput, type EvalFixture, type EvalFixtureFile, type EvalFixtureLoadOptions, type EvalFixtureRunPlan, type EvalFixtureScenario, type EvalFixtureValidationMode, type EvidenceVector, type ExternalOptimizationExample, type ExternalOptimizerCallbackLimits, type ExternalOptimizerChatRequest, type ExternalOptimizerEndpointFormat, type ExternalOptimizerEvaluationObservation, type ExternalOptimizerEvaluationRefusalReason, type ExternalOptimizerExecutionSummary, type ExternalOptimizerModelBudget, type ExternalOptimizerModelCall, type ExternalOptimizerModelCallRequest, type ExternalOptimizerModelCallResult, type ExternalOptimizerModelExecutionObservation, type ExternalOptimizerObservationArtifact, type ExternalOptimizerObservationSummary, type ExternalOptimizerProcessLimits, type ExternalOptimizerRunnerCommand, type ExternalOptimizerSubmittedCandidate, type ExternalTextCandidate, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, FileSearchLedger, type FinalEvidencePolicy, type FinalEvidenceUse, type FinalizeProfileMatrixOptions, type FinalizedProfileMatrixResult, FsLabeledScenarioStore, type FsLabeledScenarioStoreOptions, type Gate, type GateCheckStatus, type GateContext, type GateContribution, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaAdaptiveEngineRun, type GepaCandidatePopulationArtifact, type GepaCandidatePopulationCandidate, type GepaCandidatePopulationSummary, type GepaCandidateSelectionScore, type GepaEngineOptions, type GepaEngineRun, type GepaEvaluationImport, type GepaOptimizationMethodConfig, type GepaOptimizationRecipe, type GepaPopulationImport, type GepaRunnerCommand, type GitWorktreeAdapterOptions, type HeldOutGateOptions, type HeldoutSignificance, type HeldoutSignificanceOptions, type JudgeAggregate, type JudgeConfig, type JudgeDimension, type JudgeScore, type LabelTrust, type LabeledScenarioRecord, type LabeledScenarioSampleArgs, type LabeledScenarioSource, type LabeledScenarioStore, LabeledScenarioStoreError, type LabeledScenarioWrite, type LlmJudgeDimension, type LlmJudgeOptions, type LoadEvalFixtureScenariosOptions, type LoopProvenanceArgsFromResult, type LoopProvenanceBackend, type LoopProvenanceCandidate, type LoopProvenanceEvidence, type LoopProvenanceOptimizationMethod, type LoopProvenanceRecord, type MutableSurface, type NeutralizationGateOptions, type NodeEstimate, type NodePosterior, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OpenAutoPrOptions, type OpenAutoPrResult, type OpenSearchLedgerOptions, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodComposition, type OptimizationMethodInput, type OptimizationMethodPairwise, type OptimizationMethodProvenance, type OptimizationMethodResult, type OptimizationMethodRunOptions, type OptimizationMethodScore, type OptimizationPackageSource, type OptimizationTokenUsage, type OptimizerConfig, type OptimizerModelBudget, type PairedHoldout, type ParetoParent, type ParetoSignificanceGateOptions, type PendingCostCallView, type PhoenixEvaluationResultLike, type PhoenixEvaluatorLike, type PlanCampaignRunOptions, type PlanEvalFixtureRunOptions, type PlaybackContext, type PlaybackDriver, type PlaybackStep, type PowerPreflight, type PowerPreflightOptions, type PremeasuredOptimizationBaseline, type ProfileDispatchFn, type ProfileMatrixCoverage, ProfileMatrixError, type ProfileMatrixPlan, type ProfileMatrixRow, type ProfileMatrixSegmentResult, type ProfileSummary, type PromotionObjective, type PromotionPolicy, type ProposalFinding, type ProposalFindingOrigin, type ProposalTrackContext, type ProposeContext, type ProposedCandidate, type RecordSearchEdgeInput, type RedactionStatus, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceScenario, type RegisterSearchNodeInput, type RolloutArgumentDiff, type RolloutArgumentDiffOptions, type RolloutCall, type RunCampaignOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunOptimizationOptions, type RunOptimizationResult, type RunProfileMatrixOptions, type RunProfileMatrixResult, type RunProfileMatrixSegmentOptions, type RunSearchOptions, SEARCH_ESTIMATOR, SEARCH_KERNEL_SOURCE, SEARCH_LEDGER_SCHEMA, type Scenario, type ScenarioAggregate, type ScenarioRollup, type ScenarioSignal, type ScoreboardRenderOptions, type ScoreboardRow, type ScoreboardSummary, type ScoredRollout, type ScoredSurfaceOutcome, type SearchAllocator, type SearchArtifactCodec, type SearchArtifactKind, type SearchArtifactRef, type SearchAttemptAccounting, type SearchAudit, type SearchBudget, type SearchCancelReason, type SearchCandidateSurface, type SearchCell, type SearchCellAllocatedEvent, type SearchCellCancelledEvent, type SearchCellPlan, type SearchCellResult, type SearchCellSettledEvent, type SearchCellStage, type SearchCellWork, type SearchClaim, type SearchClaimPower, type SearchCloseReason, type SearchClosedEvent, type SearchCompletion, type SearchCostAccounting, type SearchDecisionRecord, type SearchEdgeAttribution, type SearchEdgeOperator, type SearchEdgeRecordedEvent, type SearchEstimateMethod, type SearchExecutionIdentity, type SearchExecutor, type SearchExpansion, type SearchFailureReason, type SearchHistoryAdmissionOptions, type SearchHistoryAuditSummary, type SearchHistoryCoverage, type SearchHistoryCoverageRow, type SearchHistoryPolicy, type SearchHistoryReceipt, SearchHistoryRequiredError, type SearchLane, type SearchLedger, type SearchLedgerAppendResult, type SearchLedgerBinding, SearchLedgerConflictError, type SearchLedgerEntry, SearchLedgerError, type SearchLedgerEvent, type SearchLedgerHash, SearchLedgerIntegrityError, type SearchLedgerTrustedHeadMode, type SearchModelIdentity, type SearchNode, type SearchNodeDecidedEvent, type SearchNodeDecision, type SearchNodeRef, type SearchNodeRegisteredEvent, type SearchNodeStatus, type SearchOpenedEvent, type SearchOpening, type SearchOperation, type SearchOperationKind, type SearchOperationRecordedEvent, type SearchOperationStartedEvent, type SearchPolicy, type SearchPolicyView, type SearchPosterior, type SearchProposalBlob, type SearchProposalRequest, type SearchProposalResult, type SearchProposedChild, type SearchProposer, type SearchProposerKind, type SearchProposerPort, SearchRecorder, type SearchRecorderOptions, type SearchReservation, type SearchRunIdentity, type SearchRunResult, type SearchScoredCell, type SearchSourceRef, type SearchSpend, type SearchSplit, type SearchSplitTasks, type SearchSplits, SearchState, SearchStateView, type SearchSurfaceEffect, type SearchSurfaceEvidence, type SearchSurfaceKind, type SearchTask, type SearchTaskOutcome, type SearchTokenAccounting, type SearchTraceRef, type SearchUnitScore, type SearchUnknown, type SequentialDecideFn, type SequentialDecideOptions, type SequentialDecision, type SequentialObservation, type SequentialPairedGate, type SequentialPairedGateOptions, type SequentialStreamState, type SeriesDistribution, type SessionScript, type SettleSearchCellInput, type SingleRunLock, type SingleRunLockOptions, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SurfaceProposer, type TraceAnalystArtifact, type TraceAnalystScenario, type TraceSpan, type TransientFailureOptions, type UngroundedLiteralReport, type UserStory, type UserStoryVerdict, type Worktree, type WorktreeAdapter, 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, 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, powerPreflight, provenanceRecordPath, provenanceSpansPath, quotaExhaustedUntil, readCachedCell, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, recordGepaSearch, renderScoreboardMarkdown, renderSurfaceDiff, replaySearchLedgerText, resolveExternalOptimizerCallbackLimits, resolveExternalOptimizerProcessLimits, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchEdgeId, searchExpansionIndex, searchHistoryCoverageRow, searchNodeId, searchPolicyView, searchPosterior, 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/index.js
CHANGED
|
@@ -1,13 +1,12 @@
|
|
|
1
1
|
import { a as heldoutSignificance, i as dimensionRegressions, o as pairHoldout, r as detectScale } from "../statistical-heldout-GcLIPxk4.js";
|
|
2
|
-
import {
|
|
3
|
-
import {
|
|
4
|
-
import { a as SearchLedgerError, i as SearchLedgerConflictError, o as SearchLedgerIntegrityError } from "../search-ledger-file-
|
|
5
|
-
import { C as
|
|
6
|
-
import { a as validateSearchLedgerEvent, c as
|
|
2
|
+
import { A as createSearchHistoryReceipt, B as estimateNodeFromCells, C as searchEdgeId, D as assertCompleteSearchHistory, E as SearchHistoryRequiredError, F as runSearch, H as searchPosterior, I as searchExpansionIndex, L as searchPolicyView, M as verifySearchHistoryArtifact, N as verifySearchHistoryReceipt, P as SEARCH_KERNEL_SOURCE, R as SEARCH_ESTIMATOR, S as developmentClaim, T as surfaceNode, U as uniform, V as searchCellSetDigest, Y as defaultProductionGate, _ as isProposedCandidate, a as transientDispatchFailure, at as runEval, b as incumbent, c as compareOptimizationMethods, ct as resolveRunDir, d as combineComparisonCosts, dt as buildCellSchedule, f as costFromLedgerSummary, ft as cellCachePath, g as runOptimization, h as runImprovementLoop, i as quotaExhaustedUntil, j as searchHistoryCoverageRow, k as assertSearchHistoryMatchesState, l as optimizationTokenUsageFromSummary, lt as tangleTracesRoot, m as readGepaCandidatePopulationArtifact, ot as runCampaign, q as openAutoPr, r as isTransientTransportFailure, st as planCampaignRun, t as llmJudge, ut as readCachedCell, v as labelTrustRank, w as surfaceDiff, x as SearchRecorder, y as crowdedFrontierParent, z as estimateNode } from "../llm-judge-Z6MkTkd0.js";
|
|
3
|
+
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-DcQiYhJI.js";
|
|
4
|
+
import { a as SearchLedgerError, i as SearchLedgerConflictError, o as SearchLedgerIntegrityError } from "../search-ledger-file-DY0AXX3W.js";
|
|
5
|
+
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";
|
|
6
|
+
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-Ck7tok0h.js";
|
|
7
7
|
import { i as makeProposalFinding } from "../types-CW5b90JH.js";
|
|
8
|
-
import { n as
|
|
9
|
-
import { a as
|
|
10
|
-
import { A as loadEvalFixtureScenarios, C as FsLabeledScenarioStore, D as neutralizationGate, E as rolloutArgumentDiff, F as searchCellSetDigest, I as searchPosterior, L as analyzeCrossSurfaceInteractions, M as SEARCH_ESTIMATOR, N as estimateNode, O as discoverEvalFixtures, P as estimateNodeFromCells, R as buildTraceAnalystSurfaceDispatch, S as neutralizeText, T as classifyUngroundedLiterals, _ as scoreboardSummary, a as autoevalsScorerJudge, b as scopedOptimizationMethod, c as selectDiscriminative, d as runProfileMatrixSegment, f as ProfileMatrixError, g as scoreUserStory, h as renderScoreboardMarkdown, i as verifyCodeSurface, j as planEvalFixtureRun, k as loadEvalFixture, 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 LabeledScenarioStoreError, x as sequentialOptimizationMethod, y as crowdedFrontierParent, z as traceAnalystQualityJudge } from "../campaign-C8iWEF0M.js";
|
|
8
|
+
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-DoayuM6D.js";
|
|
9
|
+
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-BVQVBEfH.js";
|
|
11
10
|
import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-CfMNUro2.js";
|
|
12
11
|
import { n as sequentialPairedGate, t as sequentialDecide } from "../sequential-lwJOTRON.js";
|
|
13
|
-
export { DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, FileSearchLedger, FsLabeledScenarioStore, LabeledScenarioStoreError, ProfileMatrixError, SEARCH_ESTIMATOR, 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,
|
|
12
|
+
export { DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, FileSearchLedger, FsLabeledScenarioStore, LabeledScenarioStoreError, ProfileMatrixError, 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, 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, powerPreflight, provenanceRecordPath, provenanceSpansPath, quotaExhaustedUntil, readCachedCell, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, recordGepaSearch, renderScoreboardMarkdown, renderSurfaceDiff, replaySearchLedgerText, resolveExternalOptimizerCallbackLimits, resolveExternalOptimizerProcessLimits, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchEdgeId, searchExpansionIndex, searchHistoryCoverageRow, searchNodeId, searchPolicyView, searchPosterior, searchTaskSetDigest, searchUnitScores, selectDiscriminative, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, tangleTracesRoot, traceAnalystQualityJudge, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord, verifySearchHistoryArtifact, verifySearchHistoryReceipt };
|
|
@@ -1,25 +1,22 @@
|
|
|
1
1
|
import { s as ValidationError, t as AgentEvalError } from "./errors-Dngq5h35.js";
|
|
2
|
-
import {
|
|
3
|
-
import { c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, s as agentProfileHash, t as extractProducedState } from "./produced-state-
|
|
2
|
+
import { i as compareCodeUnits } from "./canonical-BsZVgBhW.js";
|
|
3
|
+
import { c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, s as agentProfileHash, t as extractProducedState } from "./produced-state-CVmCPQDJ.js";
|
|
4
4
|
import { s as buildAgentProfileCell } from "./agent-profile-cell-DcFW4vOc.js";
|
|
5
|
-
import { t as mulberry32 } from "./random-Dn5fPWkt.js";
|
|
6
5
|
import { t as comparePairedArms } from "./paired-arms-JSDcs2zx.js";
|
|
7
|
-
import {
|
|
8
|
-
import { i as pairedDeltaTest, r as minimumPairsForPairedDeltaTest } from "./paired-promotion-decision-6cn3bLc3.js";
|
|
6
|
+
import { r as pairedBootstrap } from "./paired-tests-C8iCsioC.js";
|
|
9
7
|
import { o as pairHoldout } from "./statistical-heldout-GcLIPxk4.js";
|
|
10
8
|
import { c as runTaskScore, i as modelHasSnapshot, n as UNKNOWN_MODEL, u as validateRunRecord } from "./run-record-BRlM99rX.js";
|
|
11
9
|
import { n as contentHash, t as canonicalJson } from "./verdict-cache-BQEsVy9n.js";
|
|
12
|
-
import {
|
|
10
|
+
import { ct as resolveRunDir, d as combineComparisonCosts, ft as cellCachePath, ot as runCampaign, pt as computeManifestHash, st as planCampaignRun, u as executeOptimizationMethod, v as labelTrustRank } from "./llm-judge-Z6MkTkd0.js";
|
|
13
11
|
import { n as campaignCellToRunRecord, r as projectCampaignCellQuality, t as campaignCellCostProvenance } from "./run-record-tNnlQIrD.js";
|
|
14
12
|
import { t as CostAccountingIncompleteError } from "./cost-ledger-_IbkejCF.js";
|
|
15
|
-
import {
|
|
16
|
-
import {
|
|
17
|
-
import {
|
|
18
|
-
import
|
|
19
|
-
import { d as searchUnitScores } from "./search-ledger-CQ8Uy4FK.js";
|
|
13
|
+
import { B as assertRealBackend, F as surfaceContentHash, H as assertCampaignDesign, K as campaignScenarioIdentity, V as summarizeBackendIntegrity, j as assertCodeSurfaceIdentity } from "./campaign-evidence-DcQiYhJI.js";
|
|
14
|
+
import { m as mapConcurrent } from "./journal-DvKCvjty.js";
|
|
15
|
+
import { C as fsCampaignStorage } from "./external-optimizer-subprocess-kbqC-iWf.js";
|
|
16
|
+
import "./search-ledger-Ck7tok0h.js";
|
|
20
17
|
import { a as scoreAnalystFindings } from "./benchmark-BeQJyrRC.js";
|
|
21
|
-
import "./external-optimizer-process-
|
|
22
|
-
import "./skillopt-optimization-method-
|
|
18
|
+
import "./external-optimizer-process-D-Sab4v3.js";
|
|
19
|
+
import "./skillopt-optimization-method-DoayuM6D.js";
|
|
23
20
|
import "./sequential-lwJOTRON.js";
|
|
24
21
|
import { createHash } from "node:crypto";
|
|
25
22
|
import { closeSync, constants, existsSync, fstatSync, lstatSync, mkdirSync, mkdtempSync, openSync, readFileSync, readSync, readdirSync, readlinkSync, realpathSync, rmSync, statSync, writeFileSync } from "node:fs";
|
|
@@ -838,250 +835,6 @@ function unique(values) {
|
|
|
838
835
|
return [...new Set(values)];
|
|
839
836
|
}
|
|
840
837
|
//#endregion
|
|
841
|
-
//#region src/campaign/estimate-node.ts
|
|
842
|
-
/**
|
|
843
|
-
* Per-node statistics of a search.
|
|
844
|
-
*
|
|
845
|
-
* `estimateNode` is the one paired contrast of two nodes: it averages each
|
|
846
|
-
* node's cells inside their units, pairs the units both nodes scored, and runs
|
|
847
|
-
* `pairedDeltaTest` on the per-unit deltas. `estimateNodeFromCells` is the same
|
|
848
|
-
* computation over cells a caller already holds, such as rows a server read
|
|
849
|
-
* from its database. The bootstrap seed comes from `cellSetDigest`, the digest
|
|
850
|
-
* of exactly the cells used, so a producer and a verifier that hold the same
|
|
851
|
-
* cells report the same bits.
|
|
852
|
-
*
|
|
853
|
-
* `searchPosterior` gives every node a normal posterior on its improvement over
|
|
854
|
-
* the root, which expansion policies sample parents from. Its numbers steer
|
|
855
|
-
* spend and claim nothing.
|
|
856
|
-
*/
|
|
857
|
-
const CONFIDENCE = .95;
|
|
858
|
-
const RESAMPLES = 2e3;
|
|
859
|
-
/** 6 at 95 %: the smallest sample an exact one-sided sign test can resolve. */
|
|
860
|
-
const DESCRIPTIVE_FROM = minimumPairsForPairedDeltaTest(CONFIDENCE);
|
|
861
|
-
/** The estimator every `NodeEstimate` names. */
|
|
862
|
-
const SEARCH_ESTIMATOR = {
|
|
863
|
-
uri: "npm:@tangle-network/agent-eval#estimateNode",
|
|
864
|
-
revision: hashCanonical({
|
|
865
|
-
name: "tangle.estimate-node.2026-09",
|
|
866
|
-
unit: "mean of the unit’s scored cells, summed in cellId order",
|
|
867
|
-
pairing: "units both nodes scored, in unitId order",
|
|
868
|
-
statistic: DECISION_PAIRED_DELTA_STATISTIC,
|
|
869
|
-
interval: "percentile paired bootstrap",
|
|
870
|
-
confidence: CONFIDENCE,
|
|
871
|
-
resamples: RESAMPLES,
|
|
872
|
-
rng: "mulberry32",
|
|
873
|
-
seed: "the first 32 bits of cellSetDigest",
|
|
874
|
-
signTest: "exact, one-sided toward improvement in the objective direction",
|
|
875
|
-
methods: {
|
|
876
|
-
none: "below 2 pairs",
|
|
877
|
-
insufficient: `below ${DESCRIPTIVE_FROM} pairs`,
|
|
878
|
-
descriptive: `below 20 pairs`,
|
|
879
|
-
bootstrap: `from 20 pairs`
|
|
880
|
-
},
|
|
881
|
-
indeterminate: "every paired delta equal: no interval and no sign p"
|
|
882
|
-
})
|
|
883
|
-
};
|
|
884
|
-
/** The method a paired sample of `pairs` units supports. */
|
|
885
|
-
function searchEstimateMethod(pairs) {
|
|
886
|
-
if (pairs < 2) return "none";
|
|
887
|
-
if (pairs < DESCRIPTIVE_FROM) return "insufficient";
|
|
888
|
-
if (pairs < 20) return "descriptive";
|
|
889
|
-
return "bootstrap";
|
|
890
|
-
}
|
|
891
|
-
/**
|
|
892
|
-
* Digest of the cells an estimate reads, with the contrast they serve. Cells
|
|
893
|
-
* are listed in cellId order, so the digest names the set, not the order it
|
|
894
|
-
* was read in.
|
|
895
|
-
*/
|
|
896
|
-
function searchCellSetDigest(input) {
|
|
897
|
-
const list = (cells) => [...cells].sort((left, right) => compareCodeUnits(left.cellId, right.cellId)).map((cell) => [
|
|
898
|
-
cell.cellId,
|
|
899
|
-
cell.unitId,
|
|
900
|
-
cell.attempt,
|
|
901
|
-
cell.score
|
|
902
|
-
]);
|
|
903
|
-
return hashCanonical({
|
|
904
|
-
contrast: [input.nodeId, input.against],
|
|
905
|
-
split: input.split,
|
|
906
|
-
nodeCells: list(input.nodeCells),
|
|
907
|
-
againstCells: list(input.againstCells)
|
|
908
|
-
});
|
|
909
|
-
}
|
|
910
|
-
/**
|
|
911
|
-
* The paired contrast of `nodeId` against `against` on the units both scored.
|
|
912
|
-
*
|
|
913
|
-
* `delta` is the mean per-unit difference, node minus `against`, in the
|
|
914
|
-
* metric's own units; for a `minimize` objective an improvement is negative.
|
|
915
|
-
* The exact sign test is one-sided toward improvement in the objective's
|
|
916
|
-
* direction. Unscored cells (errored, cancelled or in flight) are absent from
|
|
917
|
-
* the inputs, never zero.
|
|
918
|
-
*
|
|
919
|
-
* - `none` (0 or 1 pair): no delta.
|
|
920
|
-
* - `insufficient` (2 to 5): the delta only.
|
|
921
|
-
* - `descriptive` (6 to 19): the bootstrap interval as spread, and the sign-test p.
|
|
922
|
-
* - `bootstrap` (20 or more): the decision-grade bootstrap interval.
|
|
923
|
-
*
|
|
924
|
-
* When every paired delta is equal the interval would have zero width, which
|
|
925
|
-
* is an absence of evidence; the estimate is then `indeterminate` and carries
|
|
926
|
-
* neither an interval nor a p-value.
|
|
927
|
-
*/
|
|
928
|
-
function estimateNodeFromCells(input) {
|
|
929
|
-
const { nodeId, against, split, direction } = input;
|
|
930
|
-
if (nodeId === against) throw new SearchLedgerError(`node ${nodeId} cannot be estimated against itself`);
|
|
931
|
-
assertDistinctCells([...input.nodeCells, ...input.againstCells], nodeId, against);
|
|
932
|
-
const cellSetDigest = searchCellSetDigest(input);
|
|
933
|
-
const nodeUnits = searchUnitScores(input.nodeCells);
|
|
934
|
-
const { node, other } = pairUnits(nodeUnits, unitMeans(searchUnitScores(input.againstCells)));
|
|
935
|
-
const pairs = node.length;
|
|
936
|
-
const method = searchEstimateMethod(pairs);
|
|
937
|
-
const base = {
|
|
938
|
-
against,
|
|
939
|
-
split,
|
|
940
|
-
units: nodeUnits.length,
|
|
941
|
-
pairs,
|
|
942
|
-
method,
|
|
943
|
-
cellSetDigest,
|
|
944
|
-
estimator: SEARCH_ESTIMATOR
|
|
945
|
-
};
|
|
946
|
-
if (method === "none") return {
|
|
947
|
-
...base,
|
|
948
|
-
delta: null,
|
|
949
|
-
interval: null,
|
|
950
|
-
exactSignP: null,
|
|
951
|
-
indeterminate: false
|
|
952
|
-
};
|
|
953
|
-
const maximize = direction === "maximize";
|
|
954
|
-
const test = pairedDeltaTest(maximize ? other : node, maximize ? node : other, {
|
|
955
|
-
statistic: DECISION_PAIRED_DELTA_STATISTIC,
|
|
956
|
-
confidence: CONFIDENCE,
|
|
957
|
-
resamples: RESAMPLES,
|
|
958
|
-
seed: seedFromDigest(cellSetDigest)
|
|
959
|
-
});
|
|
960
|
-
const { mean, low, high } = test.bootstrap;
|
|
961
|
-
const indeterminate = test.indeterminate;
|
|
962
|
-
const spread = (method === "descriptive" || method === "bootstrap") && !indeterminate;
|
|
963
|
-
return {
|
|
964
|
-
...base,
|
|
965
|
-
delta: plain(maximize ? mean : -mean),
|
|
966
|
-
interval: spread ? maximize ? [plain(low), plain(high)] : [plain(-high), plain(-low)] : null,
|
|
967
|
-
exactSignP: method === "descriptive" && !indeterminate ? test.pValue : null,
|
|
968
|
-
indeterminate
|
|
969
|
-
};
|
|
970
|
-
}
|
|
971
|
-
/**
|
|
972
|
-
* `estimateNodeFromCells` over the cells a search state holds: `nodeId`
|
|
973
|
-
* against `against` on one split, in the direction the search's objective
|
|
974
|
-
* declares.
|
|
975
|
-
*/
|
|
976
|
-
function estimateNode(state, nodeId, options) {
|
|
977
|
-
const header = state.header;
|
|
978
|
-
if (!header) throw new SearchLedgerError(`search ${state.searchId} has not been opened`);
|
|
979
|
-
for (const id of [nodeId, options.against]) if (!state.hasNode(id)) throw new SearchLedgerError(`node ${id} is not in search ${state.searchId}`);
|
|
980
|
-
return estimateNodeFromCells({
|
|
981
|
-
nodeId,
|
|
982
|
-
against: options.against,
|
|
983
|
-
split: options.split,
|
|
984
|
-
direction: header.objective.direction,
|
|
985
|
-
nodeCells: state.scoredCells(nodeId, options.split),
|
|
986
|
-
againstCells: state.scoredCells(options.against, options.split)
|
|
987
|
-
});
|
|
988
|
-
}
|
|
989
|
-
/**
|
|
990
|
-
* The posterior every node's parent-selection draw uses. Each node's mean is
|
|
991
|
-
* its improvement over the root on shared units; its variance is the search's
|
|
992
|
-
* pooled between-unit variance divided by its unit count, so a node measured
|
|
993
|
-
* on one unit has a wide posterior rather than none. The root is the reference
|
|
994
|
-
* and sits at exactly 0.
|
|
995
|
-
*/
|
|
996
|
-
function searchPosterior(state, options = {}) {
|
|
997
|
-
const split = options.split ?? "selection";
|
|
998
|
-
const header = state.header;
|
|
999
|
-
if (!header) throw new SearchLedgerError(`search ${state.searchId} has not been opened`);
|
|
1000
|
-
const rootNodeId = state.rootNodeId;
|
|
1001
|
-
if (rootNodeId === null) throw new SearchLedgerError(`search ${state.searchId} has no root node yet`);
|
|
1002
|
-
const sign = header.objective.direction === "maximize" ? 1 : -1;
|
|
1003
|
-
const root = searchUnitScores(state.scoredCells(rootNodeId, split));
|
|
1004
|
-
const rootMeans = unitMeans(root);
|
|
1005
|
-
const nodes = [];
|
|
1006
|
-
let squares = 0;
|
|
1007
|
-
let degreesOfFreedom = 0;
|
|
1008
|
-
for (const nodeId of state.nodeIds()) {
|
|
1009
|
-
if (nodeId === rootNodeId) {
|
|
1010
|
-
const measured = root.length > 0;
|
|
1011
|
-
nodes.push({
|
|
1012
|
-
nodeId,
|
|
1013
|
-
pairs: root.length,
|
|
1014
|
-
mean: measured ? 0 : null,
|
|
1015
|
-
variance: measured ? 0 : null
|
|
1016
|
-
});
|
|
1017
|
-
continue;
|
|
1018
|
-
}
|
|
1019
|
-
const { node, other } = pairUnits(searchUnitScores(state.scoredCells(nodeId, split)), rootMeans);
|
|
1020
|
-
const gains = node.map((value, index) => sign * (value - other[index]));
|
|
1021
|
-
const mean = gains.length > 0 ? sum(gains) / gains.length : null;
|
|
1022
|
-
if (mean !== null && gains.length >= 2) {
|
|
1023
|
-
squares += sum(gains.map((gain) => (gain - mean) ** 2));
|
|
1024
|
-
degreesOfFreedom += gains.length - 1;
|
|
1025
|
-
}
|
|
1026
|
-
nodes.push({
|
|
1027
|
-
nodeId,
|
|
1028
|
-
pairs: gains.length,
|
|
1029
|
-
mean: mean === null ? null : plain(mean),
|
|
1030
|
-
variance: null
|
|
1031
|
-
});
|
|
1032
|
-
}
|
|
1033
|
-
const pooledVariance = degreesOfFreedom > 0 ? squares / degreesOfFreedom : null;
|
|
1034
|
-
if (pooledVariance !== null) {
|
|
1035
|
-
for (const posterior of nodes) if (posterior.nodeId !== rootNodeId && posterior.pairs > 0) posterior.variance = pooledVariance / posterior.pairs;
|
|
1036
|
-
}
|
|
1037
|
-
return {
|
|
1038
|
-
split,
|
|
1039
|
-
rootNodeId,
|
|
1040
|
-
pooledVariance,
|
|
1041
|
-
degreesOfFreedom,
|
|
1042
|
-
nodes
|
|
1043
|
-
};
|
|
1044
|
-
}
|
|
1045
|
-
function unitMeans(units) {
|
|
1046
|
-
return new Map(units.map((unit) => [unit.unitId, unit.mean]));
|
|
1047
|
-
}
|
|
1048
|
-
/** Unit means of two nodes on the units both scored, aligned in the node's
|
|
1049
|
-
* unitId order. */
|
|
1050
|
-
function pairUnits(node, otherMeans) {
|
|
1051
|
-
const paired = {
|
|
1052
|
-
node: [],
|
|
1053
|
-
other: []
|
|
1054
|
-
};
|
|
1055
|
-
for (const unit of node) {
|
|
1056
|
-
const mean = otherMeans.get(unit.unitId);
|
|
1057
|
-
if (mean === void 0) continue;
|
|
1058
|
-
paired.node.push(unit.mean);
|
|
1059
|
-
paired.other.push(mean);
|
|
1060
|
-
}
|
|
1061
|
-
return paired;
|
|
1062
|
-
}
|
|
1063
|
-
function assertDistinctCells(cells, nodeId, against) {
|
|
1064
|
-
const seen = /* @__PURE__ */ new Set();
|
|
1065
|
-
for (const cell of cells) {
|
|
1066
|
-
if (seen.has(cell.cellId)) throw new SearchLedgerError(`cell ${cell.cellId} appears twice in the estimate of ${nodeId} against ${against}`);
|
|
1067
|
-
if (!Number.isFinite(cell.score)) throw new SearchLedgerError(`cell ${cell.cellId} has a non-finite score ${cell.score}`);
|
|
1068
|
-
seen.add(cell.cellId);
|
|
1069
|
-
}
|
|
1070
|
-
}
|
|
1071
|
-
/** The first 32 bits of a `sha256:` digest, as the bootstrap's seed. */
|
|
1072
|
-
function seedFromDigest(digest) {
|
|
1073
|
-
return Number.parseInt(digest.slice(7, 15), 16) | 0;
|
|
1074
|
-
}
|
|
1075
|
-
function sum(values) {
|
|
1076
|
-
let total = 0;
|
|
1077
|
-
for (const value of values) total += value;
|
|
1078
|
-
return total;
|
|
1079
|
-
}
|
|
1080
|
-
/** Canonical JSON has one zero; never hand it a negative one. */
|
|
1081
|
-
function plain(value) {
|
|
1082
|
-
return value === 0 ? 0 : value;
|
|
1083
|
-
}
|
|
1084
|
-
//#endregion
|
|
1085
838
|
//#region src/campaign/fixtures.ts
|
|
1086
839
|
/** Walk `evalsDir` and return the relative name of every fixture directory (one containing an exact-case `PROMPT.md`). */
|
|
1087
840
|
function discoverEvalFixtures(evalsDir) {
|
|
@@ -1684,73 +1437,6 @@ function sequentialOptimizationMethod(options) {
|
|
|
1684
1437
|
};
|
|
1685
1438
|
}
|
|
1686
1439
|
//#endregion
|
|
1687
|
-
//#region src/campaign/parent-selection.ts
|
|
1688
|
-
/**
|
|
1689
|
-
* Parent selection for `runOptimization`.
|
|
1690
|
-
*
|
|
1691
|
-
* Each generation mutates one parent surface. The default parent is the
|
|
1692
|
-
* global incumbent, so the search is a hill-climb and the lineage recorded in
|
|
1693
|
-
* `GenerationCandidate.parentSurfaceHash` is a chain. A `ParentSelector` draws
|
|
1694
|
-
* the parent from the Pareto frontier instead, so the recorded lineage becomes
|
|
1695
|
-
* a tree. Promotion is unchanged: a candidate still has to beat the incumbent.
|
|
1696
|
-
*/
|
|
1697
|
-
/**
|
|
1698
|
-
* NSGA-II crowded tournament selection over the frontier. Each generation
|
|
1699
|
-
* draws two distinct frontier members with a PRNG seeded from `seed` and the
|
|
1700
|
-
* generation index, and keeps the one with the larger crowding distance (more
|
|
1701
|
-
* isolated on the frontier). Boundary parents carry infinite distance, so a
|
|
1702
|
-
* boundary parent always beats an interior one. A tie on distance falls back
|
|
1703
|
-
* to the higher mean composite, then to the smaller surface hash. A frontier
|
|
1704
|
-
* of one member returns that member.
|
|
1705
|
-
*/
|
|
1706
|
-
function crowdedFrontierParent(options) {
|
|
1707
|
-
const { seed } = options;
|
|
1708
|
-
if (!Number.isInteger(seed)) throw new TypeError(`crowdedFrontierParent: seed must be an integer, got ${String(seed)}`);
|
|
1709
|
-
return ({ frontier, generation }) => {
|
|
1710
|
-
if (frontier.length === 0) throw new Error("crowdedFrontierParent: frontier is empty");
|
|
1711
|
-
if (frontier.length === 1) return frontier[0];
|
|
1712
|
-
const distances = /* @__PURE__ */ new Map();
|
|
1713
|
-
for (const { candidate, distance } of paretoFrontierWithCrowding([...frontier], frontierObjectives(frontier))) distances.set(candidate, distance);
|
|
1714
|
-
const rng = mulberry32(tournamentSeed(seed, generation));
|
|
1715
|
-
const first = Math.floor(rng() * frontier.length);
|
|
1716
|
-
const offset = Math.floor(rng() * (frontier.length - 1));
|
|
1717
|
-
const second = offset >= first ? offset + 1 : offset;
|
|
1718
|
-
const a = frontier[first];
|
|
1719
|
-
const b = frontier[second];
|
|
1720
|
-
return compareCrowded(a, distanceOf(distances, a), b, distanceOf(distances, b)) <= 0 ? a : b;
|
|
1721
|
-
};
|
|
1722
|
-
}
|
|
1723
|
-
function distanceOf(distances, parent) {
|
|
1724
|
-
const distance = distances.get(parent);
|
|
1725
|
-
if (distance === void 0) throw new Error(`crowdedFrontierParent: frontier member "${parent.surfaceHash}" is dominated or has a non-finite objective`);
|
|
1726
|
-
return distance;
|
|
1727
|
-
}
|
|
1728
|
-
/** Negative when `a` wins the tournament, positive when `b` wins. */
|
|
1729
|
-
function compareCrowded(a, da, b, db) {
|
|
1730
|
-
if (da !== db) return da > db ? -1 : 1;
|
|
1731
|
-
if (a.composite !== b.composite) return a.composite > b.composite ? -1 : 1;
|
|
1732
|
-
if (a.surfaceHash === b.surfaceHash) return 0;
|
|
1733
|
-
return a.surfaceHash < b.surfaceHash ? -1 : 1;
|
|
1734
|
-
}
|
|
1735
|
-
/** One `maximize` objective per scenario id present on the frontier. */
|
|
1736
|
-
function frontierObjectives(frontier) {
|
|
1737
|
-
const ids = /* @__PURE__ */ new Set();
|
|
1738
|
-
for (const parent of frontier) for (const id of Object.keys(parent.objectives)) ids.add(id);
|
|
1739
|
-
if (ids.size === 0) throw new Error("crowdedFrontierParent: frontier parents carry no objectives");
|
|
1740
|
-
return [...ids].sort().map((id) => ({
|
|
1741
|
-
name: id,
|
|
1742
|
-
direction: "maximize",
|
|
1743
|
-
value: (parent) => {
|
|
1744
|
-
const value = parent.objectives[id];
|
|
1745
|
-
if (typeof value !== "number" || !Number.isFinite(value)) throw new Error(`crowdedFrontierParent: frontier member "${parent.surfaceHash}" has no finite objective "${id}"`);
|
|
1746
|
-
return value;
|
|
1747
|
-
}
|
|
1748
|
-
}));
|
|
1749
|
-
}
|
|
1750
|
-
function tournamentSeed(seed, generation) {
|
|
1751
|
-
return seed ^ Math.imul(generation + 1, 2654435761) | 0;
|
|
1752
|
-
}
|
|
1753
|
-
//#endregion
|
|
1754
1440
|
//#region src/campaign/presets/playback.ts
|
|
1755
1441
|
/**
|
|
1756
1442
|
* Adapt a `PlaybackDriver` into a `runProfileMatrix` dispatch. The artifact the
|
|
@@ -3398,6 +3084,6 @@ function resolveWorktreePath(surface, worktreeDir) {
|
|
|
3398
3084
|
return verifyCodeSurface(surface, worktreeDir).path;
|
|
3399
3085
|
}
|
|
3400
3086
|
//#endregion
|
|
3401
|
-
export {
|
|
3087
|
+
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 };
|
|
3402
3088
|
|
|
3403
|
-
//# sourceMappingURL=campaign-
|
|
3089
|
+
//# sourceMappingURL=campaign-BVQVBEfH.js.map
|