@tangle-network/agent-eval 0.150.0 → 0.150.1
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 +14 -0
- package/dist/analyst/index.d.ts +13 -13
- package/dist/analyst/index.js +4 -4
- package/dist/{backend-integrity-DaeroaEs.d.ts → backend-integrity-DOCa_QrR.d.ts} +2 -2
- package/dist/{backend-integrity-DaeroaEs.d.ts.map → backend-integrity-DOCa_QrR.d.ts.map} +1 -1
- package/dist/{benchmark-A0E-JZ1H.d.ts → benchmark-BtAWA8nT.d.ts} +3 -3
- package/dist/{benchmark-A0E-JZ1H.d.ts.map → benchmark-BtAWA8nT.d.ts.map} +1 -1
- package/dist/{benchmark-command-CqGA0SJQ.js → benchmark-command-CAFwbH0L.js} +7 -7
- package/dist/{benchmark-command-CqGA0SJQ.js.map → benchmark-command-CAFwbH0L.js.map} +1 -1
- package/dist/benchmarks/index.d.ts +4 -4
- package/dist/benchmarks/index.js +3 -3
- package/dist/campaign/index.d.ts +8 -8
- package/dist/campaign/index.js +5 -5
- package/dist/{campaign-CR6Yt5ZS.js → campaign-CN_7xJdV.js} +18 -9
- package/dist/campaign-CN_7xJdV.js.map +1 -0
- package/dist/{capture-fetch-Im0UN1i6.d.ts → capture-fetch-BBVFzhkk.d.ts} +2 -2
- package/dist/{capture-fetch-Im0UN1i6.d.ts.map → capture-fetch-BBVFzhkk.d.ts.map} +1 -1
- package/dist/{chat-client-sFiClniQ.js → chat-client-2bVfrzhN.js} +2 -2
- package/dist/{chat-client-sFiClniQ.js.map → chat-client-2bVfrzhN.js.map} +1 -1
- package/dist/cli.js +2 -2
- package/dist/{client-w6vVJhSp.d.ts → client-kPQYT_56.d.ts} +2 -2
- package/dist/{client-w6vVJhSp.d.ts.map → client-kPQYT_56.d.ts.map} +1 -1
- package/dist/contract/index.d.ts +10 -10
- package/dist/contract/index.js +5 -5
- package/dist/{default-registry-BhRNsP-D.d.ts → default-registry-Cw0Ohdoj.d.ts} +6 -6
- package/dist/{default-registry-BhRNsP-D.d.ts.map → default-registry-Cw0Ohdoj.d.ts.map} +1 -1
- package/dist/{define-agent-eval-Cw3mkB4-.d.ts → define-agent-eval-CEQWL9Hy.d.ts} +5 -5
- package/dist/{define-agent-eval-Cw3mkB4-.d.ts.map → define-agent-eval-CEQWL9Hy.d.ts.map} +1 -1
- package/dist/{define-agent-eval-ChrtErmg.js → define-agent-eval-rqNyVhVV.js} +4 -4
- package/dist/{define-agent-eval-ChrtErmg.js.map → define-agent-eval-rqNyVhVV.js.map} +1 -1
- package/dist/{dspy-rlm-engine-Csi3CUAT.js → dspy-rlm-engine-DHI0WrUU.js} +2 -2
- package/dist/{dspy-rlm-engine-Csi3CUAT.js.map → dspy-rlm-engine-DHI0WrUU.js.map} +1 -1
- package/dist/{engine-CgQdUh-r.d.ts → engine-BLzhNzoY.d.ts} +4 -4
- package/dist/{engine-CgQdUh-r.d.ts.map → engine-BLzhNzoY.d.ts.map} +1 -1
- package/dist/{eval-campaign-Czc2TcW5.js → eval-campaign-C4jmuM-b.js} +2 -2
- package/dist/{eval-campaign-Czc2TcW5.js.map → eval-campaign-C4jmuM-b.js.map} +1 -1
- package/dist/{exact-types-D9_T83Ol.d.ts → exact-types-ccQAyut1.d.ts} +2 -2
- package/dist/{exact-types-D9_T83Ol.d.ts.map → exact-types-ccQAyut1.d.ts.map} +1 -1
- package/dist/experiment/index.d.ts +3 -3
- package/dist/{experiment-tracker-Cp9ougHs.d.ts → experiment-tracker-0MhuPArU.d.ts} +2 -2
- package/dist/{experiment-tracker-Cp9ougHs.d.ts.map → experiment-tracker-0MhuPArU.d.ts.map} +1 -1
- package/dist/{external-optimizer-contracts-mbhR9Ldu.d.ts → external-optimizer-contracts-DbLsm4Po.d.ts} +2 -2
- package/dist/{external-optimizer-contracts-mbhR9Ldu.d.ts.map → external-optimizer-contracts-DbLsm4Po.d.ts.map} +1 -1
- package/dist/{external-optimizer-process-tDe72VRO.js → external-optimizer-process-Bhmzngf-.js} +2 -2
- package/dist/{external-optimizer-process-tDe72VRO.js.map → external-optimizer-process-Bhmzngf-.js.map} +1 -1
- package/dist/{external-optimizer-subprocess-BepyNwtj.js → external-optimizer-subprocess-Dn90UqN2.js} +273 -42
- package/dist/external-optimizer-subprocess-Dn90UqN2.js.map +1 -0
- package/dist/{feedback-trajectory-Cc9M2tNF.d.ts → feedback-trajectory-DpTTjo0q.d.ts} +3 -3
- package/dist/{feedback-trajectory-Cc9M2tNF.d.ts.map → feedback-trajectory-DpTTjo0q.d.ts.map} +1 -1
- package/dist/hosted/index.d.ts +2 -2
- package/dist/index-C1ravkGA.d.ts +1 -0
- package/dist/{index-DL4MJuJL.d.ts → index-CTKpu9ry.d.ts} +7 -7
- package/dist/{index-DL4MJuJL.d.ts.map → index-CTKpu9ry.d.ts.map} +1 -1
- package/dist/{index-dh0kp2GM.d.ts → index-IQccV3Ou.d.ts} +15 -10
- package/dist/index-IQccV3Ou.d.ts.map +1 -0
- package/dist/index.d.ts +19 -19
- package/dist/index.js +8 -8
- package/dist/{integrity-D2MAlC_K.d.ts → integrity-B0dZ96EO.d.ts} +2 -2
- package/dist/{integrity-D2MAlC_K.d.ts.map → integrity-B0dZ96EO.d.ts.map} +1 -1
- package/dist/{llm-client-CxyJ8Oft.js → llm-client-Bg32RW0j.js} +47 -2
- package/dist/llm-client-Bg32RW0j.js.map +1 -0
- package/dist/{llm-judge-D7UuMI2t.js → llm-judge-CVq33oz1.js} +4 -4
- package/dist/{llm-judge-D7UuMI2t.js.map → llm-judge-CVq33oz1.js.map} +1 -1
- package/dist/{matrix-D3hL68uV.d.ts → matrix-DrVnRp4G.d.ts} +2 -2
- package/dist/{matrix-D3hL68uV.d.ts.map → matrix-DrVnRp4G.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-WJs_enbZ.js → produced-state-jfk8Du3b.js} +3 -3
- package/dist/{produced-state-WJs_enbZ.js.map → produced-state-jfk8Du3b.js.map} +1 -1
- package/dist/{promotion-policy-DYRzXdTV.d.ts → promotion-policy-DLOUkYhI.d.ts} +2 -2
- package/dist/{promotion-policy-DYRzXdTV.d.ts.map → promotion-policy-DLOUkYhI.d.ts.map} +1 -1
- package/dist/{provenance-Bh9k3sjl.d.ts → provenance-oA4-zUqm.d.ts} +6 -6
- package/dist/{provenance-Bh9k3sjl.d.ts.map → provenance-oA4-zUqm.d.ts.map} +1 -1
- package/dist/{registry-DhpX9Bsg.d.ts → registry-BQwrSYpC.d.ts} +4 -4
- package/dist/{registry-DhpX9Bsg.d.ts.map → registry-BQwrSYpC.d.ts.map} +1 -1
- package/dist/{researcher-CbCK-ppF.d.ts → researcher-DJnoUE8c.d.ts} +3 -3
- package/dist/{researcher-CbCK-ppF.d.ts.map → researcher-DJnoUE8c.d.ts.map} +1 -1
- package/dist/{reward-hacking-DFo2FU5J.js → reward-hacking-DKI9T52l.js} +19 -1
- package/dist/reward-hacking-DKI9T52l.js.map +1 -0
- package/dist/rl.d.ts +2 -2
- package/dist/rl.js +2 -2
- package/dist/{semantic-concept-judge-B7KPD57W.js → semantic-concept-judge-laMCnTLn.js} +2 -2
- package/dist/{semantic-concept-judge-B7KPD57W.js.map → semantic-concept-judge-laMCnTLn.js.map} +1 -1
- package/dist/{series-convergence-VhTad5TM.d.ts → series-convergence-BxKEgBwA.d.ts} +2 -2
- package/dist/{series-convergence-VhTad5TM.d.ts.map → series-convergence-BxKEgBwA.d.ts.map} +1 -1
- package/dist/{server-DNiUVirV.js → server-dIWwF3j_.js} +2 -2
- package/dist/{server-DNiUVirV.js.map → server-dIWwF3j_.js.map} +1 -1
- package/dist/{skillopt-optimization-method-BtgL6PMH.d.ts → skillopt-optimization-method-BDD_o1xE.d.ts} +5 -5
- package/dist/{skillopt-optimization-method-BtgL6PMH.d.ts.map → skillopt-optimization-method-BDD_o1xE.d.ts.map} +1 -1
- package/dist/{skillopt-optimization-method-L8vi4h4m.js → skillopt-optimization-method-DLeUcK-K.js} +5 -5
- package/dist/{skillopt-optimization-method-L8vi4h4m.js.map → skillopt-optimization-method-DLeUcK-K.js.map} +1 -1
- package/dist/{statistical-heldout-Caxd30ZF.d.ts → statistical-heldout-_woZ9q9j.d.ts} +2 -2
- package/dist/{statistical-heldout-Caxd30ZF.d.ts.map → statistical-heldout-_woZ9q9j.d.ts.map} +1 -1
- package/dist/{store-tool-spans-BGg2QFk1.d.ts → store-tool-spans-Br2_IUhm.d.ts} +4 -4
- package/dist/{store-tool-spans-BGg2QFk1.d.ts.map → store-tool-spans-Br2_IUhm.d.ts.map} +1 -1
- package/dist/{tool-groups-BuhFx38z.d.ts → tool-groups-B4tqh8jB.d.ts} +3 -3
- package/dist/tool-groups-B4tqh8jB.d.ts.map +1 -0
- package/dist/trace-repair/index.d.ts +1 -1
- package/dist/traces.d.ts +7 -7
- package/dist/{types-BNbhD_Vj.d.ts → types-B2NsbrNy.d.ts} +2 -2
- package/dist/{types-BNbhD_Vj.d.ts.map → types-B2NsbrNy.d.ts.map} +1 -1
- package/dist/{types-B61iI2ZG.d.ts → types-CLAwnY-L.d.ts} +2 -2
- package/dist/{types-B61iI2ZG.d.ts.map → types-CLAwnY-L.d.ts.map} +1 -1
- package/dist/{types-COEjZMLY.d.ts → types-DdFNuyxQ.d.ts} +3 -3
- package/dist/{types-COEjZMLY.d.ts.map → types-DdFNuyxQ.d.ts.map} +1 -1
- package/dist/{types-COAJy8Ku.d.ts → types-jUBXJ7Iz.d.ts} +36 -3
- package/dist/types-jUBXJ7Iz.d.ts.map +1 -0
- package/dist/wire/index.d.ts +2 -2
- package/dist/wire/index.js +1 -1
- package/docs/campaign-proposers.md +8 -1
- package/package.json +1 -1
- package/dist/campaign-CR6Yt5ZS.js.map +0 -1
- package/dist/external-optimizer-subprocess-BepyNwtj.js.map +0 -1
- package/dist/index-DjpxROrT.d.ts +0 -1
- package/dist/index-dh0kp2GM.d.ts.map +0 -1
- package/dist/llm-client-CxyJ8Oft.js.map +0 -1
- package/dist/reward-hacking-DFo2FU5J.js.map +0 -1
- package/dist/tool-groups-BuhFx38z.d.ts.map +0 -1
- package/dist/types-COAJy8Ku.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-CAFwbH0L.js";
|
|
4
|
+
import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-dIWwF3j_.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-DdFNuyxQ.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-kPQYT_56.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"client-
|
|
1
|
+
{"version":3,"file":"client-kPQYT_56.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-jUBXJ7Iz.js";
|
|
4
|
+
import { _ as ProposalFinding, i as AnalystFinding, v as ProposalFindingOrigin, x as makeProposalFinding } from "../types-B2NsbrNy.js";
|
|
5
|
+
import { n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-Cw0Ohdoj.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-DdFNuyxQ.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-BLzhNzoY.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-oA4-zUqm.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-BDD_o1xE.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-DLOUkYhI.js";
|
|
12
|
+
import { c as EvalRunGenerationSnapshot, g as TraceSpanEvent, n as HostedTenant, o as EvalRunCellScore, s as EvalRunEvent } from "../client-kPQYT_56.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-CEQWL9Hy.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-rqNyVhVV.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-CVq33oz1.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-Dn90UqN2.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-2bVfrzhN.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-DLeUcK-K.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-B2NsbrNy.js";
|
|
2
|
+
import { l as ExactCapableAnalyst } from "./exact-types-ccQAyut1.js";
|
|
3
|
+
import { n as AnalystRegistry, r as AnalystRegistryOptions } from "./registry-BQwrSYpC.js";
|
|
4
|
+
import { a as TraceAnalystLimits, d as RawAnalystFinding, i as TraceAnalysisEngineResult, n as TraceAnalysisEngine } from "./engine-BLzhNzoY.js";
|
|
5
|
+
import { t as TraceToolGroupName } from "./tool-groups-B4tqh8jB.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-Cw0Ohdoj.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"default-registry-
|
|
1
|
+
{"version":3,"file":"default-registry-Cw0Ohdoj.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"}
|
|
@@ -1,10 +1,10 @@
|
|
|
1
1
|
import { c as CostLedgerHandle, f as CostLedgerSummary, m as CostReceipt } from "./cost-ledger-DbQdN3nO.js";
|
|
2
|
-
import { _ as ProposalFinding } from "./types-
|
|
3
|
-
import { H as SurfaceProposer, R as Scenario, S as JudgeConfig, a as CampaignResult, d as DispatchContext, j as MutableSurface, k as LabeledScenarioStore, p as Gate } from "./types-
|
|
2
|
+
import { _ as ProposalFinding } from "./types-B2NsbrNy.js";
|
|
3
|
+
import { H as SurfaceProposer, R as Scenario, S as JudgeConfig, a as CampaignResult, d as DispatchContext, j as MutableSurface, k as LabeledScenarioStore, p as Gate } from "./types-DdFNuyxQ.js";
|
|
4
4
|
import { o as InsightReport } from "./insight-report-BeT8KCgI.js";
|
|
5
|
-
import { C as RunOptimizationOptions, Cn as CampaignStorage, E as RunEvalOptions, J as OptimizationMethodProvenance, S as PremeasuredOptimizationBaseline, W as OptimizationMethod, Y as OptimizationMethodResult, b as RunImprovementLoopResult, l as LoopProvenanceRecord } from "./provenance-
|
|
5
|
+
import { C as RunOptimizationOptions, Cn as CampaignStorage, E as RunEvalOptions, J as OptimizationMethodProvenance, S as PremeasuredOptimizationBaseline, W as OptimizationMethod, Y as OptimizationMethodResult, b as RunImprovementLoopResult, l as LoopProvenanceRecord } from "./provenance-oA4-zUqm.js";
|
|
6
6
|
import { o as PowerPreflight } from "./pareto-BqNW3LJR.js";
|
|
7
|
-
import { n as HostedTenant } from "./client-
|
|
7
|
+
import { n as HostedTenant } from "./client-kPQYT_56.js";
|
|
8
8
|
//#region src/contract/self-improve.d.ts
|
|
9
9
|
interface SelfImproveBudget {
|
|
10
10
|
/** Hard spend cap across the full run. Each paid call reserves its enforced
|
|
@@ -384,4 +384,4 @@ interface DefinedAgentEval<TScenario extends Scenario, TArtifact> {
|
|
|
384
384
|
declare function defineAgentEval<TScenario extends Scenario, TArtifact>(defaults: DefineAgentEvalOptions<TScenario, TArtifact>): DefinedAgentEval<TScenario, TArtifact>;
|
|
385
385
|
//#endregion
|
|
386
386
|
export { DefinedAgentEval as a, SelfImproveOptions as c, SelfImproveRunError as d, selfImprove as f, DefineAgentEvalOptions as i, SelfImproveProgressEvent as l, AgentEvalEvaluateOptions as n, defineAgentEval as o, AgentEvalImproveOptions as r, SelfImproveBudget as s, AgentEvalAgent as t, SelfImproveResult as u };
|
|
387
|
-
//# sourceMappingURL=define-agent-eval-
|
|
387
|
+
//# sourceMappingURL=define-agent-eval-CEQWL9Hy.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"define-agent-eval-
|
|
1
|
+
{"version":3,"file":"define-agent-eval-CEQWL9Hy.d.ts","names":[],"sources":["../src/contract/self-improve.ts","../src/contract/define-agent-eval.ts"],"mappings":";;;;;;;;UA+DiB;;;EAGf;;;;EAIA;;EAEA;;EAEA;;;EAGA;;;EAGA;;;;EAIA;;EAEA,mBAAmB;;;;;;;;;EASnB;;EAEA;;;;;EAKA;;KAGU;EACN;EAA0B;;EAC1B;EAA4B;EAAuB;;EACnD;EAA4B;EAAe;;EAC3C;EAA8B;EAAe;EAAuB;;EAGpE;EAAsB;EAAkB;;EACxC;EAAyB;EAAW;EAAY;EAAa;;UAElD,mBAAmB,kBAAkB,UAAU;;;;;;;;;;;;;;EAc9D,QAAQ,SAAS,gBAAgB,UAAU,WAAW,KAAK,oBAAoB,QAAQ;;;;;;;EAQvF;;;EAIA,WAAW;;EAGX,OAAO,YAAY,WAAW;;;EAI9B,iBAAiB;;EAGjB,SAAS;;;;;;;;;;;;EAaT,sBAAsB,gCAAgC,WAAW;;;;;EAMjE,WAAW,gBAAgB;;;;;;EAO3B,SAAS,mBAAmB,WAAW;;;EAIvC,qBAAqB;;;EAIrB,OAAO,KAAK,WAAW;;;;;;;;EASvB,cAAc,eAAe,gBAAgB,iBAAiB,mBAAmB;;;;EAKjF,UAAU;;;;EAKV;;;EAIA,gBAAgB,QAAQ;;;;;EAMxB,iBAAiB;IACf,UAAU;IACV;IACA;;;;EAKF;;;EAIA,cAAc,OAAO;;;;EAKrB;EACA;EACA;;;;;;;;;;;;;;EAeA,eAAe;;;EAIf,eAAe;;;;EAKf,eAAe;;EAGf;;;;;;;;EASA;;;;;;;;;EAUA,oBAAoB,uBAAuB,WAAW;;;EAItD,WAAW;;;;;;;EAQX,mBAAmB,uBAAuB,WAAW;;UAGtC,kBAAkB,kBAAkB,UAAU;;;;EAI7D;IACE;IACA,aAAa;;;;;EAKf;IACE;IACA,aAAa;IACb,SAAS;;;IAGT;;;;IAIA;;;;;;EAMF;;;EAGA;;;;EAIA,YAAY;;EAEZ;;;EAGA;;EAEA;;EAEA;;EAEA,MAAM;;;EAGN,UAAU;;EAEV;IACE;IACA,MAAM;IACN;IACA,aAAa;;;;;;;;EAQf,SAAS;;;;;EAKT,QAAQ;;;;;;EAMR,KAAK,yBAAyB,WAAW;;;cAI9B,4BAA4B;WAC9B,MAAM;WACN,UAAU;EAEnB,YAAY,gBAAgB,QAAQ;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA+LhB,YAAY,kBAAkB,UAAU,WAC5D,MAAM,mBAAmB,WAAW,aACnC,QAAQ,kBAAkB,WAAW;;;KCtiB5B,eAAe,kBAAkB,UAAU,cACrD,SAAS,gBACT,UAAU,WACV,KAAK,oBACF,QAAQ;KAED,uBAAuB,kBAAkB,UAAU,aAAa,mBAC1E,WACA;UAGe,yBAAyB,kBAAkB,UAAU,mBAC5D,KACN,eAAe,WAAW;;EAI5B,YAAY;;EAEZ,UAAU;;EAEV,QAAQ,eAAe,WAAW;;EAElC,QAAQ,YAAY,WAAW;;EAE/B,SAAS,YAAY,WAAW;;EAEhC;;KAGU,wBAAwB,kBAAkB,UAAU,aAAa,KAC3E,QAAQ,mBAAmB,WAAW;EAGtC,SAAS,QAAQ;EACjB,eAAe,QAAQ;;UAGR,iBAAiB,kBAAkB,UAAU;;WAEnD,oBAAoB;;WAEpB,iBAAiB;;;;;EAK1B,SACE,OAAO,yBAAyB,WAAW,aAC1C,QAAQ,eAAe,WAAW;;;;;;;EAOrC,QACE,OAAO,wBAAwB,WAAW,aACzC,QAAQ,kBAAkB,WAAW;;;;;;;;;iBAU1B,gBAAgB,kBAAkB,UAAU,WAC1D,UAAU,uBAAuB,WAAW,aAC3C,iBAAiB,WAAW"}
|
|
@@ -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-
|
|
13
|
-
import { c as campaignCellTaskScore, l as campaignCellToRunRecord, o as campaignCellExecutionEvidence, s as campaignCellJudgeDimensions } from "./reward-hacking-
|
|
14
|
-
import { S as inMemoryCampaignStorage, b as createRunCostLedger, x as fsCampaignStorage } from "./external-optimizer-subprocess-
|
|
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-CVq33oz1.js";
|
|
13
|
+
import { c as campaignCellTaskScore, l as campaignCellToRunRecord, o as campaignCellExecutionEvidence, s as campaignCellJudgeDimensions } from "./reward-hacking-DKI9T52l.js";
|
|
14
|
+
import { S as inMemoryCampaignStorage, b as createRunCostLedger, x as fsCampaignStorage } from "./external-optimizer-subprocess-Dn90UqN2.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-rqNyVhVV.js.map
|