@tangle-network/agent-eval 0.149.0 → 0.150.0
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 +28 -0
- package/README.md +5 -1
- package/dist/analyst/index.d.ts +13 -13
- package/dist/analyst/index.d.ts.map +1 -1
- package/dist/analyst/index.js +4 -4
- package/dist/{backend-integrity-Bz8nSrRE.d.ts → backend-integrity-DaeroaEs.d.ts} +2 -2
- package/dist/{backend-integrity-Bz8nSrRE.d.ts.map → backend-integrity-DaeroaEs.d.ts.map} +1 -1
- package/dist/{benchmark-Cn7ZMFNW.d.ts → benchmark-A0E-JZ1H.d.ts} +3 -3
- package/dist/{benchmark-Cn7ZMFNW.d.ts.map → benchmark-A0E-JZ1H.d.ts.map} +1 -1
- package/dist/{benchmark-command-gufocQ-N.js → benchmark-command-CqGA0SJQ.js} +8 -7
- package/dist/benchmark-command-CqGA0SJQ.js.map +1 -0
- package/dist/benchmarks/index.d.ts +4 -4
- package/dist/benchmarks/index.js +3 -3
- package/dist/campaign/index.d.ts +9 -9
- package/dist/campaign/index.js +6 -6
- package/dist/{campaign-BcfXzmPM.js → campaign-CR6Yt5ZS.js} +80 -7
- package/dist/campaign-CR6Yt5ZS.js.map +1 -0
- package/dist/{capture-fetch-BAx3Gntt.d.ts → capture-fetch-Im0UN1i6.d.ts} +2 -2
- package/dist/{capture-fetch-BAx3Gntt.d.ts.map → capture-fetch-Im0UN1i6.d.ts.map} +1 -1
- package/dist/{chat-client-BX8Wh3fn.js → chat-client-sFiClniQ.js} +2 -2
- package/dist/{chat-client-BX8Wh3fn.js.map → chat-client-sFiClniQ.js.map} +1 -1
- package/dist/cli.js +2 -2
- package/dist/{client-BV7icIfy.d.ts → client-w6vVJhSp.d.ts} +2 -2
- package/dist/{client-BV7icIfy.d.ts.map → client-w6vVJhSp.d.ts.map} +1 -1
- package/dist/contract/index.d.ts +10 -10
- package/dist/contract/index.js +5 -5
- package/dist/{default-registry-BUtRMRKk.d.ts → default-registry-BhRNsP-D.d.ts} +6 -6
- package/dist/{default-registry-BUtRMRKk.d.ts.map → default-registry-BhRNsP-D.d.ts.map} +1 -1
- package/dist/{define-agent-eval-CvZQW4u9.js → define-agent-eval-ChrtErmg.js} +3 -3
- package/dist/{define-agent-eval-CvZQW4u9.js.map → define-agent-eval-ChrtErmg.js.map} +1 -1
- package/dist/{define-agent-eval-BWuk2O_N.d.ts → define-agent-eval-Cw3mkB4-.d.ts} +5 -5
- package/dist/{define-agent-eval-BWuk2O_N.d.ts.map → define-agent-eval-Cw3mkB4-.d.ts.map} +1 -1
- package/dist/{dspy-rlm-engine-BmtOl_kP.js → dspy-rlm-engine-Csi3CUAT.js} +2 -2
- package/dist/{dspy-rlm-engine-BmtOl_kP.js.map → dspy-rlm-engine-Csi3CUAT.js.map} +1 -1
- package/dist/{engine-CIy18RTX.d.ts → engine-CgQdUh-r.d.ts} +4 -4
- package/dist/{engine-CIy18RTX.d.ts.map → engine-CgQdUh-r.d.ts.map} +1 -1
- package/dist/{eval-campaign-bmZ6NIIP.js → eval-campaign-Czc2TcW5.js} +2 -2
- package/dist/{eval-campaign-bmZ6NIIP.js.map → eval-campaign-Czc2TcW5.js.map} +1 -1
- package/dist/{exact-types-rdKFzEnK.d.ts → exact-types-D9_T83Ol.d.ts} +2 -2
- package/dist/{exact-types-rdKFzEnK.d.ts.map → exact-types-D9_T83Ol.d.ts.map} +1 -1
- package/dist/experiment/index.d.ts +73 -4
- package/dist/experiment/index.d.ts.map +1 -1
- package/dist/experiment/index.js +175 -1
- package/dist/experiment/index.js.map +1 -1
- package/dist/{experiment-tracker-D9VHwRm-.d.ts → experiment-tracker-Cp9ougHs.d.ts} +2 -2
- package/dist/{experiment-tracker-D9VHwRm-.d.ts.map → experiment-tracker-Cp9ougHs.d.ts.map} +1 -1
- package/dist/{external-optimizer-contracts-CSjDLLmr.d.ts → external-optimizer-contracts-mbhR9Ldu.d.ts} +13 -6
- package/dist/{external-optimizer-contracts-CSjDLLmr.d.ts.map → external-optimizer-contracts-mbhR9Ldu.d.ts.map} +1 -1
- package/dist/{external-optimizer-process-C4AyCcQd.js → external-optimizer-process-tDe72VRO.js} +4 -4
- package/dist/{external-optimizer-process-C4AyCcQd.js.map → external-optimizer-process-tDe72VRO.js.map} +1 -1
- package/dist/{external-optimizer-subprocess-DU1KM8yV.js → external-optimizer-subprocess-BepyNwtj.js} +341 -25
- package/dist/external-optimizer-subprocess-BepyNwtj.js.map +1 -0
- package/dist/{feedback-trajectory-niGeR6uJ.d.ts → feedback-trajectory-Cc9M2tNF.d.ts} +3 -3
- package/dist/{feedback-trajectory-niGeR6uJ.d.ts.map → feedback-trajectory-Cc9M2tNF.d.ts.map} +1 -1
- package/dist/hosted/index.d.ts +2 -2
- package/dist/{index-aHDnhBC7.d.ts → index-DL4MJuJL.d.ts} +7 -7
- package/dist/{index-aHDnhBC7.d.ts.map → index-DL4MJuJL.d.ts.map} +1 -1
- package/dist/index-DjpxROrT.d.ts +1 -0
- package/dist/{index-BCDyf_kT.d.ts → index-dh0kp2GM.d.ts} +42 -11
- package/dist/index-dh0kp2GM.d.ts.map +1 -0
- package/dist/index.d.ts +20 -20
- package/dist/index.js +8 -8
- package/dist/{integrity-CGfpTE5-.d.ts → integrity-D2MAlC_K.d.ts} +2 -2
- package/dist/{integrity-CGfpTE5-.d.ts.map → integrity-D2MAlC_K.d.ts.map} +1 -1
- package/dist/{llm-client-BMuxYoZy.js → llm-client-CxyJ8Oft.js} +15 -6
- package/dist/llm-client-CxyJ8Oft.js.map +1 -0
- package/dist/{llm-judge-DWq1Ptco.js → llm-judge-D7UuMI2t.js} +4 -3
- package/dist/llm-judge-D7UuMI2t.js.map +1 -0
- package/dist/{matrix-Bc111FKv.d.ts → matrix-D3hL68uV.d.ts} +2 -2
- package/dist/{matrix-Bc111FKv.d.ts.map → matrix-D3hL68uV.d.ts.map} +1 -1
- package/dist/meta-eval/index.d.ts +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-C0oJ4vr-.js → produced-state-WJs_enbZ.js} +3 -3
- package/dist/{produced-state-C0oJ4vr-.js.map → produced-state-WJs_enbZ.js.map} +1 -1
- package/dist/{promotion-policy-CewdjfCJ.d.ts → promotion-policy-DYRzXdTV.d.ts} +2 -2
- package/dist/{promotion-policy-CewdjfCJ.d.ts.map → promotion-policy-DYRzXdTV.d.ts.map} +1 -1
- package/dist/{provenance-DA-Pmyfv.d.ts → provenance-Bh9k3sjl.d.ts} +17 -6
- package/dist/{provenance-DA-Pmyfv.d.ts.map → provenance-Bh9k3sjl.d.ts.map} +1 -1
- package/dist/{registry-CaqJIDdN.d.ts → registry-DhpX9Bsg.d.ts} +4 -4
- package/dist/{registry-CaqJIDdN.d.ts.map → registry-DhpX9Bsg.d.ts.map} +1 -1
- package/dist/{researcher-BySlpla_.d.ts → researcher-CbCK-ppF.d.ts} +3 -3
- package/dist/{researcher-BySlpla_.d.ts.map → researcher-CbCK-ppF.d.ts.map} +1 -1
- package/dist/rl.d.ts +2 -2
- package/dist/rl.js +1 -1
- package/dist/{semantic-concept-judge-BI7Rrl5-.js → semantic-concept-judge-B7KPD57W.js} +2 -2
- package/dist/{semantic-concept-judge-BI7Rrl5-.js.map → semantic-concept-judge-B7KPD57W.js.map} +1 -1
- package/dist/{series-convergence-D2fsoJ2w.d.ts → series-convergence-VhTad5TM.d.ts} +2 -2
- package/dist/{series-convergence-D2fsoJ2w.d.ts.map → series-convergence-VhTad5TM.d.ts.map} +1 -1
- package/dist/{server-D9wQclzG.js → server-DNiUVirV.js} +2 -2
- package/dist/{server-D9wQclzG.js.map → server-DNiUVirV.js.map} +1 -1
- package/dist/{skillopt-optimization-method-ChCTN4FD.d.ts → skillopt-optimization-method-BtgL6PMH.d.ts} +19 -13
- package/dist/{skillopt-optimization-method-ChCTN4FD.d.ts.map → skillopt-optimization-method-BtgL6PMH.d.ts.map} +1 -1
- package/dist/{skillopt-optimization-method-Donmq4sq.js → skillopt-optimization-method-L8vi4h4m.js} +92 -14
- package/dist/skillopt-optimization-method-L8vi4h4m.js.map +1 -0
- package/dist/{statistical-heldout-uxSpFEjm.d.ts → statistical-heldout-Caxd30ZF.d.ts} +2 -2
- package/dist/{statistical-heldout-uxSpFEjm.d.ts.map → statistical-heldout-Caxd30ZF.d.ts.map} +1 -1
- package/dist/{store-tool-spans-D5FhM0_A.d.ts → store-tool-spans-BGg2QFk1.d.ts} +4 -4
- package/dist/{store-tool-spans-D5FhM0_A.d.ts.map → store-tool-spans-BGg2QFk1.d.ts.map} +1 -1
- package/dist/{tool-groups-zpufabP8.d.ts → tool-groups-BuhFx38z.d.ts} +3 -3
- package/dist/tool-groups-BuhFx38z.d.ts.map +1 -0
- package/dist/trace-repair/index.d.ts +1 -1
- package/dist/traces.d.ts +7 -7
- package/dist/{types-ColZlZtc.d.ts → types-B61iI2ZG.d.ts} +2 -2
- package/dist/{types-ColZlZtc.d.ts.map → types-B61iI2ZG.d.ts.map} +1 -1
- package/dist/{types-BohHewKK.d.ts → types-BNbhD_Vj.d.ts} +2 -2
- package/dist/{types-BohHewKK.d.ts.map → types-BNbhD_Vj.d.ts.map} +1 -1
- package/dist/{types-DOhGq4S1.d.ts → types-COAJy8Ku.d.ts} +15 -3
- package/dist/types-COAJy8Ku.d.ts.map +1 -0
- package/dist/{types-C1Bmeb8X.d.ts → types-COEjZMLY.d.ts} +3 -3
- package/dist/{types-C1Bmeb8X.d.ts.map → types-COEjZMLY.d.ts.map} +1 -1
- package/dist/wire/index.d.ts +2 -2
- package/dist/wire/index.js +1 -1
- package/docs/campaign-proposers.md +60 -3
- package/docs/experiment.md +2 -0
- package/docs/multishot-golden-records.md +3 -0
- package/package.json +4 -2
- package/dist/benchmark-command-gufocQ-N.js.map +0 -1
- package/dist/campaign-BcfXzmPM.js.map +0 -1
- package/dist/external-optimizer-subprocess-DU1KM8yV.js.map +0 -1
- package/dist/index-BCDyf_kT.d.ts.map +0 -1
- package/dist/index-CKblBxtr.d.ts +0 -1
- package/dist/llm-client-BMuxYoZy.js.map +0 -1
- package/dist/llm-judge-DWq1Ptco.js.map +0 -1
- package/dist/skillopt-optimization-method-Donmq4sq.js.map +0 -1
- package/dist/tool-groups-zpufabP8.d.ts.map +0 -1
- package/dist/types-DOhGq4S1.d.ts.map +0 -1
package/dist/cli.js
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
#!/usr/bin/env node
|
|
2
2
|
import { o as runRolloutReleaseCli } from "./hf-dataset-XggBupCr.js";
|
|
3
|
-
import { n as runAnalystBenchmarkCommand } from "./benchmark-command-
|
|
4
|
-
import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-
|
|
3
|
+
import { n as runAnalystBenchmarkCommand } from "./benchmark-command-CqGA0SJQ.js";
|
|
4
|
+
import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-DNiUVirV.js";
|
|
5
5
|
import { writeFileSync } from "node:fs";
|
|
6
6
|
//#region src/cli-config.ts
|
|
7
7
|
function resolveCliLlmConfig(env = process.env) {
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import { l as RunTerminalOutcome } from "./run-record-DVV82Gwh.js";
|
|
2
|
-
import { _ as GateDecision, j as MutableSurface } from "./types-
|
|
2
|
+
import { _ as GateDecision, j as MutableSurface } from "./types-COEjZMLY.js";
|
|
3
3
|
import { o as InsightReport } from "./insight-report-BeT8KCgI.js";
|
|
4
4
|
//#region src/hosted/types.d.ts
|
|
5
5
|
declare const HOSTED_WIRE_VERSION: '2026-07-24.v1';
|
|
@@ -199,4 +199,4 @@ declare function hostedClientFromEnv(overrides?: Partial<HostedTenant> & {
|
|
|
199
199
|
}): HostedClient | undefined;
|
|
200
200
|
//#endregion
|
|
201
201
|
export { UnixNanoTimestamp as _, hostedTenantFromEnv as a, EvalRunGenerationSnapshot as c, HostedIngestHeaders as d, HostedWireVersion as f, TraceSpanEvent as g, IngestTracesRequest as h, hostedClientFromEnv as i, EvalRunStatus as l, IngestResponse as m, HostedTenant as n, EvalRunCellScore as o, IngestEvalRunsRequest as p, createHostedClient as r, EvalRunEvent as s, HostedClient as t, HOSTED_WIRE_VERSION as u };
|
|
202
|
-
//# sourceMappingURL=client-
|
|
202
|
+
//# sourceMappingURL=client-w6vVJhSp.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"client-
|
|
1
|
+
{"version":3,"file":"client-w6vVJhSp.d.ts","names":[],"sources":["../src/hosted/types.ts","../src/hosted/client.ts"],"mappings":";;;;cAgCa;KACD,2BAA2B;;UAKtB;;EAEf;;EAEA;;EAEA,yBAAyB;;EAEzB;;;KAMU;UAQK;;EAEf;;EAEA;;EAEA;;EAEA,YAAY,eAAe;;EAE3B,iBAAiB;;EAEjB;;EAEA;;UAGe;;EAEf;;;EAGA;;;EAGA,UAAU;;EAEV,OAAO;;EAEP;;EAEA;;EAEA;;;;;;;UAQe;;EAEf;;EAEA;;EAEA;;EAEA,QAAQ;;EAER,QAAQ;;EAER,WAAW;;EAEX,aAAa;;EAEb,eAAe;;EAEf;;EAEA;;EAEA;;EAEA;;;;;;EAMA,gBAAgB;;;;;;KASN;;;;;;UAOK;EACf;EACA;EACA;EACA;EACA,mBAAmB;EACnB,iBAAiB;EACjB,YAAY;EACZ,SAAS;IACP,cAAc;IACd;IACA,aAAa;;EAEf;IAAW;IAAgC;;;EAE3C;;EAEA;;EAEA;;EAEA;;UAKe;EACf,aAAa;EACb,QAAQ;;UAGO;EACf,aAAa;EACb,OAAO;;UAGQ;;EAEf;;EAEA,UAAU;IAAQ;IAAe;;;;;UC1JlB;;EAEf;;EAEA;;EAEA;;EAEA,mBAAmB;;EAEnB;;EAEA;;UAGe;EACf,cAAc,OAAO,cAAc,0BAA0B,QAAQ;EACrE,eAAe,QAAQ,gBAAgB,0BAA0B,QAAQ;EACzE,aAAa,OAAO,kBAAkB,0BAA0B,QAAQ;WAC/D,QAAQ;WACR,aAAa;;iBAyGR,mBAAmB,QAAQ,eAAe;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAsE1C,oBACd,YAAW,QAAQ;EAAkB,MAAM;IAC1C;iBAiBa,oBACd,YAAW,QAAQ;EAAkB,MAAM;IAC1C"}
|
package/dist/contract/index.d.ts
CHANGED
|
@@ -1,17 +1,17 @@
|
|
|
1
1
|
import { c as CostLedgerHandle } from "../cost-ledger-DbQdN3nO.js";
|
|
2
2
|
import { a as RunRecord, n as RunCostProvenance, s as RunSplitTag } from "../run-record-DVV82Gwh.js";
|
|
3
|
-
import { p as ChatClient, v as CreateChatClientOpts, w as createChatClient } from "../types-
|
|
4
|
-
import { _ as ProposalFinding, i as AnalystFinding, v as ProposalFindingOrigin, x as makeProposalFinding } from "../types-
|
|
5
|
-
import { n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-
|
|
6
|
-
import { C as JudgeDimension, H as SurfaceProposer, M as OptimizerConfig, R as Scenario, S as JudgeConfig, V as SessionScript, _ 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, l as CodeSurface, m as GateCheckStatus, n as CampaignArtifactWriter, p as Gate, r as CampaignCellResult, t as CampaignAggregates, v as GateResult, w as JudgeScore, y as GenerationCandidate } from "../types-
|
|
3
|
+
import { p as ChatClient, v as CreateChatClientOpts, w as createChatClient } from "../types-COAJy8Ku.js";
|
|
4
|
+
import { _ as ProposalFinding, i as AnalystFinding, v as ProposalFindingOrigin, x as makeProposalFinding } from "../types-BNbhD_Vj.js";
|
|
5
|
+
import { n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-BhRNsP-D.js";
|
|
6
|
+
import { C as JudgeDimension, H as SurfaceProposer, M as OptimizerConfig, R as Scenario, S as JudgeConfig, V as SessionScript, _ 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, l as CodeSurface, m as GateCheckStatus, n as CampaignArtifactWriter, p as Gate, r as CampaignCellResult, t as CampaignAggregates, v as GateResult, w as JudgeScore, y as GenerationCandidate } from "../types-COEjZMLY.js";
|
|
7
7
|
import { a as FailureClusterInsight, c as JudgeInsight, d as Recommendation, f as ReleaseSummary, i as FailureClassTally, l as LiftInsight, m as TokenUsageInsight, n as ExecutionErrorOutcomeCell, o as InsightReport, p as ScalarDistribution, r as ExecutionInsight, s as InterRaterInsight, t as CostProvenanceSummary, u as OutcomeCorrelationInsight } from "../insight-report-BeT8KCgI.js";
|
|
8
|
-
import { C as analyzeRuns, S as SummarizeExecutionOptions, b as AnalyzeRunsOptions, w as summarizeExecution, x as ExecutionReport } from "../engine-
|
|
9
|
-
import { $ as OptimizationTokenUsage, An as LlmJudgeDimension, Cn as CampaignStorage, D as runEval, E as RunEvalOptions, En as inMemoryCampaignStorage, G as OptimizationMethodComparison, H as CompareOptimizationMethodsOptions, J as OptimizationMethodProvenance, K as OptimizationMethodInput, Mn as llmJudge, Q as OptimizationPackageSource, Tn as fsCampaignStorage, U as ComparisonCost, W as OptimizationMethod, Y as OptimizationMethodResult, _n as RunCampaignOptions, b as RunImprovementLoopResult, gn as CampaignCellFailureReceipt, jn as LlmJudgeOptions, tt as compareOptimizationMethods, vn as runCampaign, x as runImprovementLoop, y as RunImprovementLoopOptions } from "../provenance-
|
|
10
|
-
import { B as runReferenceEquivalenceJudge, C as ExternalTextOptimizerContext, E as ExternalTextEvaluationResponse, F as ReferenceEquivalenceJudgeInput, I as ReferenceEquivalenceJudgeOptions, L as ReferenceEquivalenceJudgeResult, N as REFERENCE_EQUIVALENCE_INPUT_LIMITS, P as REFERENCE_EQUIVALENCE_JUDGE_VERSION, R as ReferenceEquivalenceScenario, S as ExternalTextOptimizationMethodConfig, T as ExternalOptimizationExample, _ as DefaultProductionGateOptions, a as GepaAdaptiveEngineRun, b as composeGate, c as GepaOptimizationMethodConfig, d as gepaOptimizationMethod, f as OpenAICompatibleOptimizerModel, g as DefaultProductionGateCheck, h as heldOutGate, i as skillOptOptimizationMethod, j as campaignSplitDigest, l as GepaOptimizationRecipe, m as HeldOutGateOptions, n as SkillOptRunnerCommand, o as GepaEngineOptions, p as OptimizerModelBudget, r as SkillOptTrainerConfig, s as GepaEngineRun, t as SkillOptOptimizationMethodConfig, u as GepaRunnerCommand, v as DefaultProductionRewardHackingOptions, w as ExternalTextOptimizerResult, x as externalTextOptimizationMethod, y as defaultProductionGate, z as createReferenceEquivalenceJudge } from "../skillopt-optimization-method-
|
|
11
|
-
import { a as ObjectiveSource, c as PromotionPolicy, d as paretoSignificanceGate, i as EvidenceVector, l as buildEvidenceVector, n as AxisVerdict, o as ParetoSignificanceGateOptions, r as BuildEvidenceVectorOptions, s as PromotionObjective, t as AxisEvidence, u as paretoPolicy } from "../promotion-policy-
|
|
12
|
-
import { c as EvalRunGenerationSnapshot, g as TraceSpanEvent, n as HostedTenant, o as EvalRunCellScore, s as EvalRunEvent } from "../client-
|
|
8
|
+
import { C as analyzeRuns, S as SummarizeExecutionOptions, b as AnalyzeRunsOptions, w as summarizeExecution, x as ExecutionReport } from "../engine-CgQdUh-r.js";
|
|
9
|
+
import { $ as OptimizationTokenUsage, An as LlmJudgeDimension, Cn as CampaignStorage, D as runEval, E as RunEvalOptions, En as inMemoryCampaignStorage, G as OptimizationMethodComparison, H as CompareOptimizationMethodsOptions, J as OptimizationMethodProvenance, K as OptimizationMethodInput, Mn as llmJudge, Q as OptimizationPackageSource, Tn as fsCampaignStorage, U as ComparisonCost, W as OptimizationMethod, Y as OptimizationMethodResult, _n as RunCampaignOptions, b as RunImprovementLoopResult, gn as CampaignCellFailureReceipt, jn as LlmJudgeOptions, tt as compareOptimizationMethods, vn as runCampaign, x as runImprovementLoop, y as RunImprovementLoopOptions } from "../provenance-Bh9k3sjl.js";
|
|
10
|
+
import { B as runReferenceEquivalenceJudge, C as ExternalTextOptimizerContext, E as ExternalTextEvaluationResponse, F as ReferenceEquivalenceJudgeInput, I as ReferenceEquivalenceJudgeOptions, L as ReferenceEquivalenceJudgeResult, N as REFERENCE_EQUIVALENCE_INPUT_LIMITS, P as REFERENCE_EQUIVALENCE_JUDGE_VERSION, R as ReferenceEquivalenceScenario, S as ExternalTextOptimizationMethodConfig, T as ExternalOptimizationExample, _ as DefaultProductionGateOptions, a as GepaAdaptiveEngineRun, b as composeGate, c as GepaOptimizationMethodConfig, d as gepaOptimizationMethod, f as OpenAICompatibleOptimizerModel, g as DefaultProductionGateCheck, h as heldOutGate, i as skillOptOptimizationMethod, j as campaignSplitDigest, l as GepaOptimizationRecipe, m as HeldOutGateOptions, n as SkillOptRunnerCommand, o as GepaEngineOptions, p as OptimizerModelBudget, r as SkillOptTrainerConfig, s as GepaEngineRun, t as SkillOptOptimizationMethodConfig, u as GepaRunnerCommand, v as DefaultProductionRewardHackingOptions, w as ExternalTextOptimizerResult, x as externalTextOptimizationMethod, y as defaultProductionGate, z as createReferenceEquivalenceJudge } from "../skillopt-optimization-method-BtgL6PMH.js";
|
|
11
|
+
import { a as ObjectiveSource, c as PromotionPolicy, d as paretoSignificanceGate, i as EvidenceVector, l as buildEvidenceVector, n as AxisVerdict, o as ParetoSignificanceGateOptions, r as BuildEvidenceVectorOptions, s as PromotionObjective, t as AxisEvidence, u as paretoPolicy } from "../promotion-policy-DYRzXdTV.js";
|
|
12
|
+
import { c as EvalRunGenerationSnapshot, g as TraceSpanEvent, n as HostedTenant, o as EvalRunCellScore, s as EvalRunEvent } from "../client-w6vVJhSp.js";
|
|
13
13
|
import { i as InMemoryOutcomeStore, n as FileSystemOutcomeStore, o as OutcomeStore, r as FileSystemOutcomeStoreOptions, t as DeploymentOutcome } from "../outcome-store-BYHIuO0e.js";
|
|
14
|
-
import { a as DefinedAgentEval, c as SelfImproveOptions, d as SelfImproveRunError, f as selfImprove, i as DefineAgentEvalOptions, l as SelfImproveProgressEvent, n as AgentEvalEvaluateOptions, o as defineAgentEval, r as AgentEvalImproveOptions, s as SelfImproveBudget, t as AgentEvalAgent, u as SelfImproveResult } from "../define-agent-eval-
|
|
14
|
+
import { a as DefinedAgentEval, c as SelfImproveOptions, d as SelfImproveRunError, f as selfImprove, i as DefineAgentEvalOptions, l as SelfImproveProgressEvent, n as AgentEvalEvaluateOptions, o as defineAgentEval, r as AgentEvalImproveOptions, s as SelfImproveBudget, t as AgentEvalAgent, u as SelfImproveResult } from "../define-agent-eval-Cw3mkB4-.js";
|
|
15
15
|
import { AgentCandidateBenchmarkCellRef, AgentCandidateBenchmarkSuiteInputs, AgentCandidateBenchmarkTask, AgentCandidateBenchmarkTaskMaterial, AgentCandidateBundle, AgentCandidateEvaluationPolicy, AgentCandidateExperiment, AgentCandidateExperimentMaterial, AgentCandidateExperimentMeasurement, AgentImprovementCost, AgentImprovementMeasuredComparison, AgentProfileImprovementExperiment, AgentProfileImprovementExperimentMaterial, AgentProfileImprovementMeasuredComparison, AgentProfileImprovementMeasurement, AgentProfileImprovementRunCell, AgentProfileImprovementRunReceipt, AgentProfileImprovementSuiteInputs, AgentProfileImprovementTask, AgentProfileImprovementTaskMaterial, CandidateExecutionEvidence, Sha256Digest } from "@tangle-network/agent-interface";
|
|
16
16
|
//#region src/contract/measured-comparison.d.ts
|
|
17
17
|
interface SealCandidateBenchmarkSuiteOptions {
|
package/dist/contract/index.js
CHANGED
|
@@ -1,21 +1,21 @@
|
|
|
1
1
|
import { s as ValidationError } from "../errors-Dngq5h35.js";
|
|
2
2
|
import { r as pairedBootstrap } from "../paired-tests-BHIhYVdu.js";
|
|
3
|
-
import { a as summarizeExecution, i as analyzeRuns, n as SelfImproveRunError, r as selfImprove, t as defineAgentEval } from "../define-agent-eval-
|
|
3
|
+
import { a as summarizeExecution, i as analyzeRuns, n as SelfImproveRunError, r as selfImprove, t as defineAgentEval } from "../define-agent-eval-ChrtErmg.js";
|
|
4
4
|
import { a as parseRunRecordSafe } from "../run-record-D2lDdSAz.js";
|
|
5
|
-
import { $ as runEval, et as runCampaign, f as runImprovementLoop, q as defaultProductionGate, t as llmJudge, ut as campaignSplitDigest, x as compareOptimizationMethods } from "../llm-judge-
|
|
5
|
+
import { $ as runEval, et as runCampaign, f as runImprovementLoop, q as defaultProductionGate, t as llmJudge, ut as campaignSplitDigest, x as compareOptimizationMethods } from "../llm-judge-D7UuMI2t.js";
|
|
6
6
|
import { i as isModelPriced, n as estimateCost } from "../metrics-Qv-cpptD.js";
|
|
7
7
|
import { i as CostLedger } from "../cost-ledger-B1qx30B4.js";
|
|
8
8
|
import { d as mapConcurrentRange } from "../ledger-core-DTae9rv_.js";
|
|
9
|
-
import { S as inMemoryCampaignStorage, x as fsCampaignStorage } from "../external-optimizer-subprocess-
|
|
9
|
+
import { S as inMemoryCampaignStorage, x as fsCampaignStorage } from "../external-optimizer-subprocess-BepyNwtj.js";
|
|
10
10
|
import { a as heldoutSignificance, s as decidePairedPromotion } from "../power-preflight-DEw-uC7q.js";
|
|
11
11
|
import { r as makeProposalFinding } from "../types-BI4fT3HN.js";
|
|
12
12
|
import { $ as classifyOtlpSpanRole, et as isOtlpModelCall } from "../kind-factory-DmAa0h3K.js";
|
|
13
|
-
import { n as buildDefaultAnalystRegistry, t as createChatClient } from "../chat-client-
|
|
13
|
+
import { n as buildDefaultAnalystRegistry, t as createChatClient } from "../chat-client-sFiClniQ.js";
|
|
14
14
|
import { LLM_MODEL_ATTR_KEYS, SPAN_KIND_ATTR_KEYS } from "../trace-attributes.js";
|
|
15
15
|
import { t as extractUsage } from "../extract-usage-BrQ8mCLX.js";
|
|
16
16
|
import { n as InMemoryOutcomeStore, t as FileSystemOutcomeStore } from "../outcome-store-ChBKlTd_.js";
|
|
17
17
|
import { i as summarizeTraceErrors, n as recordAggregateMeasurements, r as summarizeExecutionMeasurements, t as readTaskFailureLabels } from "../task-failure-attributes--ZTP3tYO.js";
|
|
18
|
-
import { a as externalTextOptimizationMethod, c as REFERENCE_EQUIVALENCE_INPUT_LIMITS, d as runReferenceEquivalenceJudge, i as composeGate, l as REFERENCE_EQUIVALENCE_JUDGE_VERSION, n as gepaOptimizationMethod, r as heldOutGate, t as skillOptOptimizationMethod, u as createReferenceEquivalenceJudge } from "../skillopt-optimization-method-
|
|
18
|
+
import { a as externalTextOptimizationMethod, c as REFERENCE_EQUIVALENCE_INPUT_LIMITS, d as runReferenceEquivalenceJudge, i as composeGate, l as REFERENCE_EQUIVALENCE_JUDGE_VERSION, n as gepaOptimizationMethod, r as heldOutGate, t as skillOptOptimizationMethod, u as createReferenceEquivalenceJudge } from "../skillopt-optimization-method-L8vi4h4m.js";
|
|
19
19
|
import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-xzA40Evo.js";
|
|
20
20
|
import { createHash } from "node:crypto";
|
|
21
21
|
import { agentCandidateBenchmarkSuiteSchema, agentCandidateBenchmarkTaskSchema, agentCandidateBundleSchema, agentCandidateEvaluationPolicySchema, agentCandidateExperimentSchema, agentImprovementMeasuredComparisonSchema, agentProfileImprovementExperimentSchema, agentProfileImprovementMeasuredComparisonSchema, agentProfileImprovementRunCellSchema, agentProfileImprovementRunReceiptSchema, agentProfileImprovementSuiteInputsSchema, agentProfileImprovementSuiteSchema, agentProfileImprovementTaskSchema, candidateExecutionEvidenceSchema, canonicalCandidateDigest, canonicalCandidateJson, numbersApproximatelyEqual, omitTopLevelDigest } from "@tangle-network/agent-interface";
|
|
@@ -1,8 +1,8 @@
|
|
|
1
|
-
import { i as AnalystFinding, n as AnalystContext, w as TraceAnalysisStore } from "./types-
|
|
2
|
-
import { l as ExactCapableAnalyst } from "./exact-types-
|
|
3
|
-
import { n as AnalystRegistry, r as AnalystRegistryOptions } from "./registry-
|
|
4
|
-
import { a as TraceAnalystLimits, d as RawAnalystFinding, i as TraceAnalysisEngineResult, n as TraceAnalysisEngine } from "./engine-
|
|
5
|
-
import { t as TraceToolGroupName } from "./tool-groups-
|
|
1
|
+
import { i as AnalystFinding, n as AnalystContext, w as TraceAnalysisStore } from "./types-BNbhD_Vj.js";
|
|
2
|
+
import { l as ExactCapableAnalyst } from "./exact-types-D9_T83Ol.js";
|
|
3
|
+
import { n as AnalystRegistry, r as AnalystRegistryOptions } from "./registry-DhpX9Bsg.js";
|
|
4
|
+
import { a as TraceAnalystLimits, d as RawAnalystFinding, i as TraceAnalysisEngineResult, n as TraceAnalysisEngine } from "./engine-CgQdUh-r.js";
|
|
5
|
+
import { t as TraceToolGroupName } from "./tool-groups-BuhFx38z.js";
|
|
6
6
|
//#region src/trace-analyst/behavioral-metrics.d.ts
|
|
7
7
|
type SuboptimalCode = 'monotonic-input-growth' | 'output-length-decay' | 'single-tool-dependency' | 'no-self-verification';
|
|
8
8
|
interface SuboptimalSignal {
|
|
@@ -111,4 +111,4 @@ interface DefaultAnalystRegistryOptions {
|
|
|
111
111
|
declare function buildDefaultAnalystRegistry(options?: DefaultAnalystRegistryOptions): AnalystRegistry;
|
|
112
112
|
//#endregion
|
|
113
113
|
export { createTraceAnalyst as a, runTraceAnalyst as c, deriveEfficiencyFindings as d, TraceAnalystDefinition as i, BehavioralAnalystOptions as l, buildDefaultAnalystRegistry as n, renderPriorFindings as o, CreateTraceAnalystOptions as r, renderUpstreamFindings as s, DefaultAnalystRegistryOptions as t, behavioralAnalyst as u };
|
|
114
|
-
//# sourceMappingURL=default-registry-
|
|
114
|
+
//# sourceMappingURL=default-registry-BhRNsP-D.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"default-registry-
|
|
1
|
+
{"version":3,"file":"default-registry-BhRNsP-D.d.ts","names":[],"sources":["../src/trace-analyst/behavioral-metrics.ts","../src/analyst/behavioral-analyst.ts","../src/analyst/kind-factory.ts","../src/analyst/default-registry.ts"],"mappings":";;;;;;KAoBY;UAMK;EACf,MAAM;EACN;;EAEA;;EAEA,UAAU;;UAGK;;EAEf;EACA;;EAEA,gBAAgB;;EAEhB;EACA;EACA,eAAe;EACf;EACA;;EAEA;EACA;EACA,SAAS;;UAGM;EACf;EACA;EACA,sBAAsB;EACtB,uBAAuB;;;;UC1BR;;EAEf;;EAEA;;;;;;iBA+Dc,yBACd,SAAS,mBACT;EAAQ;EAAoB;IAC3B;;iBAkCa,kBACd,UAAS,2BACR,oBAAoB;;;;UCrHN;EACf;EACA;EACA;EACA;;EAEA,sBAAsB,SAAS;;EAE/B;;EAEA,WAAW;;EAEX,kBACE,OAAO,oBACP,SAAS,wCACe;EAC1B,SAAS,QAAQ;EACjB,eAAe,KAAK,mBAAmB,SAAS,mBAAmB;EACnE;EACA;;UAGe;EACf,QAAQ;;EAER;EACA;;;iBAIoB,gBAAgB;EACpC,YAAY;EACZ,QAAQ;EACR,OAAO;EACP,SAAS;EACT;IACE,QAAQ;;iBAgFI,mBACd,YAAY,wBACZ,SAAS,4BACR,oBAAoB;iBA+QP,oBAAoB,OAAO;iBAe3B,uBAAuB,UAAU;;;UCnahC;;;;;;EAMf,SAAS;EACT,uBAAuB;;EAEvB;EACA,aAAa;EACb,WAAW;;iBAGG,4BACd,UAAS,gCACR"}
|
|
@@ -9,9 +9,9 @@ import { r as welchsTTest } from "./baseline-BhPRQBVn.js";
|
|
|
9
9
|
import { r as observedSplitScore } from "./reward-nw2xZGZG.js";
|
|
10
10
|
import { c as validateRunRecord, i as modelHasSnapshot } from "./run-record-D2lDdSAz.js";
|
|
11
11
|
import { r as paretoChart } from "./summary-report-Blysd6Z2.js";
|
|
12
|
-
import { $ as runEval, A as surfaceHash, f as runImprovementLoop, k as surfaceContentHash, nt as resolveRunDir, o as emitLoopProvenance, q as defaultProductionGate, s as loopProvenanceArgsFromResult, y as assertOptimizationResult } from "./llm-judge-
|
|
12
|
+
import { $ as runEval, A as surfaceHash, f as runImprovementLoop, k as surfaceContentHash, nt as resolveRunDir, o as emitLoopProvenance, q as defaultProductionGate, s as loopProvenanceArgsFromResult, y as assertOptimizationResult } from "./llm-judge-D7UuMI2t.js";
|
|
13
13
|
import { c as campaignCellTaskScore, l as campaignCellToRunRecord, o as campaignCellExecutionEvidence, s as campaignCellJudgeDimensions } from "./reward-hacking-DFo2FU5J.js";
|
|
14
|
-
import { S as inMemoryCampaignStorage, b as createRunCostLedger, x as fsCampaignStorage } from "./external-optimizer-subprocess-
|
|
14
|
+
import { S as inMemoryCampaignStorage, b as createRunCostLedger, x as fsCampaignStorage } from "./external-optimizer-subprocess-BepyNwtj.js";
|
|
15
15
|
import { t as powerPreflight } from "./power-preflight-DEw-uC7q.js";
|
|
16
16
|
import { t as createHostedClient } from "./client-LIuo-KPv.js";
|
|
17
17
|
//#region src/contamination-guard.ts
|
|
@@ -1491,4 +1491,4 @@ function requirePositiveInteger(value, field) {
|
|
|
1491
1491
|
//#endregion
|
|
1492
1492
|
export { summarizeExecution as a, analyzeRuns as i, SelfImproveRunError as n, checkCanaries as o, selfImprove as r, defineAgentEval as t };
|
|
1493
1493
|
|
|
1494
|
-
//# sourceMappingURL=define-agent-eval-
|
|
1494
|
+
//# sourceMappingURL=define-agent-eval-ChrtErmg.js.map
|