@tangle-network/agent-eval 0.135.1 → 0.135.3
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/dist/analyst/index.js +1 -1
- package/dist/{analyst-LsnNpSkm.js → analyst-j5je5J7c.js} +2 -2
- package/dist/{analyst-LsnNpSkm.js.map → analyst-j5je5J7c.js.map} +1 -1
- package/dist/{analyze-runs-DMo3Lb_y.d.ts → analyze-runs-Cda5Xkj1.d.ts} +3 -3
- package/dist/{analyze-runs-DMo3Lb_y.d.ts.map → analyze-runs-Cda5Xkj1.d.ts.map} +1 -1
- package/dist/{analyze-runs-qk8op0tN.js → analyze-runs-jjCmF8pU.js} +14 -7
- package/dist/analyze-runs-jjCmF8pU.js.map +1 -0
- package/dist/{baseline-BaPxoROc.js → baseline-BUeFcgrn.js} +2 -2
- package/dist/{baseline-BaPxoROc.js.map → baseline-BUeFcgrn.js.map} +1 -1
- package/dist/benchmarks/index.d.ts +1 -1
- package/dist/benchmarks/index.js +1 -1
- package/dist/{benchmarks-DwFrCrCS.js → benchmarks-Dfgm9ts5.js} +3 -3
- package/dist/{benchmarks-DwFrCrCS.js.map → benchmarks-Dfgm9ts5.js.map} +1 -1
- package/dist/builder-eval/index.js +1 -1
- package/dist/campaign/index.d.ts +2 -2
- package/dist/campaign/index.js +2 -2
- package/dist/{campaign-B0oyngTs.js → campaign-Dz8uQnhC.js} +6 -6
- package/dist/{campaign-B0oyngTs.js.map → campaign-Dz8uQnhC.js.map} +1 -1
- package/dist/{client-BIyh1RCr.d.ts → client-DcvgkaZi.d.ts} +13 -4
- package/dist/client-DcvgkaZi.d.ts.map +1 -0
- package/dist/contract/index.d.ts +3 -3
- package/dist/contract/index.d.ts.map +1 -1
- package/dist/contract/index.js +20 -8
- package/dist/contract/index.js.map +1 -1
- package/dist/{default-registry-BAhV-lbE.js → default-registry-CHmdy2An.js} +2 -2
- package/dist/{default-registry-BAhV-lbE.js.map → default-registry-CHmdy2An.js.map} +1 -1
- package/dist/{eval-campaign-aMG73hvX.js → eval-campaign-Cc8WZJ6b.js} +2 -2
- package/dist/{eval-campaign-aMG73hvX.js.map → eval-campaign-Cc8WZJ6b.js.map} +1 -1
- package/dist/{extract-usage-2j25whHw.js → extract-usage-DIQpN-ww.js} +2 -2
- package/dist/{extract-usage-2j25whHw.js.map → extract-usage-DIQpN-ww.js.map} +1 -1
- package/dist/hosted/index.d.ts +1 -1
- package/dist/{index-BoJNQR6n.d.ts → index-B4Fjfo5U.d.ts} +93 -15
- package/dist/index-B4Fjfo5U.d.ts.map +1 -0
- package/dist/{index-C21xKtxu.d.ts → index-CQsJcqch.d.ts} +3 -3
- package/dist/{index-C21xKtxu.d.ts.map → index-CQsJcqch.d.ts.map} +1 -1
- package/dist/{index-DSC51roc2.d.ts → index-DSC51roc.d.ts} +1 -1
- package/dist/index-DSC51roc.d.ts.map +1 -0
- package/dist/index.d.ts +42 -11
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +136 -58
- package/dist/index.js.map +1 -1
- package/dist/matrix/index.d.ts +1 -1
- package/dist/meta-eval/index.d.ts +1 -2
- package/dist/meta-eval/index.d.ts.map +1 -1
- package/dist/meta-eval/index.js +2 -2
- package/dist/multishot/index.d.ts +1 -1
- package/dist/openapi.json +1 -1
- package/dist/{paired-arms-CA_8pN01.js → paired-arms-BbFKrAU-.js} +2 -2
- package/dist/{paired-arms-CA_8pN01.js.map → paired-arms-BbFKrAU-.js.map} +1 -1
- package/dist/pipelines/index.js +2 -2
- package/dist/{release-report-BVZBmRZp.js → release-report-DooPguBc.js} +4 -3
- package/dist/{release-report-BVZBmRZp.js.map → release-report-DooPguBc.js.map} +1 -1
- package/dist/{release-report-CuULWKyk.d.ts → release-report-DpBxGGI1.d.ts} +2 -2
- package/dist/{release-report-CuULWKyk.d.ts.map → release-report-DpBxGGI1.d.ts.map} +1 -1
- package/dist/{replay-BI6CVKkp.d.ts → replay-BRfMIs81.d.ts} +39 -21
- package/dist/replay-BRfMIs81.d.ts.map +1 -0
- package/dist/{replay-CJfGLdx4.js → replay-C6wRg47C.js} +23 -41
- package/dist/replay-C6wRg47C.js.map +1 -0
- package/dist/reporting.d.ts +3 -3
- package/dist/reporting.js +4 -4
- package/dist/{researcher-DVtruQ9U.d.ts → researcher-Doo95b50.d.ts} +2 -2
- package/dist/{researcher-DVtruQ9U.d.ts.map → researcher-Doo95b50.d.ts.map} +1 -1
- package/dist/{reward-hacking-DCdRK9TY.js → reward-hacking-a-kYs0-i.js} +2 -2
- package/dist/{reward-hacking-DCdRK9TY.js.map → reward-hacking-a-kYs0-i.js.map} +1 -1
- package/dist/rl.d.ts +45 -3
- package/dist/rl.d.ts.map +1 -1
- package/dist/rl.js +108 -22
- package/dist/rl.js.map +1 -1
- package/dist/{rubric-predictive-validity-D6Q6n9oq.js → rubric-predictive-validity-BJf-8ejY.js} +2 -2
- package/dist/{rubric-predictive-validity-D6Q6n9oq.js.map → rubric-predictive-validity-BJf-8ejY.js.map} +1 -1
- package/dist/{skillopt-optimization-method-C9yGg-4C.js → skillopt-optimization-method-0UmPD6aP.js} +340 -54
- package/dist/skillopt-optimization-method-0UmPD6aP.js.map +1 -0
- package/dist/{skillopt-optimization-method-DJ3l4w8W.d.ts → skillopt-optimization-method-CwSYkv35.d.ts} +39 -9
- package/dist/skillopt-optimization-method-CwSYkv35.d.ts.map +1 -0
- package/dist/{statistics-D_4Snl-5.d.ts → statistics-CKOqre5S.d.ts} +329 -3
- package/dist/statistics-CKOqre5S.d.ts.map +1 -0
- package/dist/{statistics-RwRNu2__.js → statistics-CnGCLLqc.js} +315 -2
- package/dist/statistics-CnGCLLqc.js.map +1 -0
- package/dist/{summary-report-BxtossFi.js → summary-report-BEk8OFLs.js} +11 -6
- package/dist/summary-report-BEk8OFLs.js.map +1 -0
- package/dist/{summary-report-DGp0-_XO.d.ts → summary-report-CPMINBqs.d.ts} +182 -7
- package/dist/summary-report-CPMINBqs.d.ts.map +1 -0
- package/dist/{tools-BmuN627J.js → tools-D8yTtNSN.js} +132 -27
- package/dist/tools-D8yTtNSN.js.map +1 -0
- package/dist/traces.d.ts +2 -2
- package/dist/traces.js +5 -5
- package/docs/trace-analysis.md +24 -0
- package/package.json +1 -1
- package/dist/analyze-runs-qk8op0tN.js.map +0 -1
- package/dist/client-BIyh1RCr.d.ts.map +0 -1
- package/dist/index-BoJNQR6n.d.ts.map +0 -1
- package/dist/index-DSC51roc2.d.ts.map +0 -1
- package/dist/judge-calibration-DFtEMlde.d.ts +0 -146
- package/dist/judge-calibration-DFtEMlde.d.ts.map +0 -1
- package/dist/replay-BI6CVKkp.d.ts.map +0 -1
- package/dist/replay-CJfGLdx4.js.map +0 -1
- package/dist/skillopt-optimization-method-C9yGg-4C.js.map +0 -1
- package/dist/skillopt-optimization-method-DJ3l4w8W.d.ts.map +0 -1
- package/dist/statistics-D_4Snl-5.d.ts.map +0 -1
- package/dist/statistics-RwRNu2__.js.map +0 -1
- package/dist/summary-report-BxtossFi.js.map +0 -1
- package/dist/summary-report-DGp0-_XO.d.ts.map +0 -1
- package/dist/tools-BmuN627J.js.map +0 -1
|
@@ -1,8 +1,8 @@
|
|
|
1
1
|
import { o as FailureClass } from "./schema-BtVldJ3T.js";
|
|
2
2
|
import { l as RunTerminalOutcome } from "./run-record-DcObtIGh.js";
|
|
3
3
|
import { _ as GateDecision, j as MutableSurface } from "./types-DiWLru6Z.js";
|
|
4
|
-
import {
|
|
5
|
-
import { i as ParetoFigureSpec, t as GainDistributionBin } from "./summary-report-
|
|
4
|
+
import { bt as ContinuousAgreement } from "./statistics-CKOqre5S.js";
|
|
5
|
+
import { i as ParetoFigureSpec, t as GainDistributionBin } from "./summary-report-CPMINBqs.js";
|
|
6
6
|
//#region src/contract/insight-report.d.ts
|
|
7
7
|
interface InsightReport {
|
|
8
8
|
/** Number of runs analyzed. */
|
|
@@ -256,7 +256,16 @@ interface LiftInsight {
|
|
|
256
256
|
n: number;
|
|
257
257
|
/** Minimum paired observations required before the interval can drive a decision. */
|
|
258
258
|
minimumRequired: number;
|
|
259
|
-
/**
|
|
259
|
+
/**
|
|
260
|
+
* Whether the bootstrap interval has enough observations to drive a decision.
|
|
261
|
+
*
|
|
262
|
+
* Sample size only. A ZERO-WIDTH interval clears this and is still refused at
|
|
263
|
+
* the ship decision itself: n identical paired deltas make every resample
|
|
264
|
+
* identical, so `[g, g]` clears every threshold below g and `[0, 0]` clears
|
|
265
|
+
* every negative one on no spread at all. That is an absence of evidence, not
|
|
266
|
+
* a certainty, and it is a different finding from "too few runs" — so it is
|
|
267
|
+
* reported where the decision is made, not folded in here.
|
|
268
|
+
*/
|
|
260
269
|
decisionEligible: boolean;
|
|
261
270
|
/** Scored baseline observations without a candidate match. */
|
|
262
271
|
unpairedBaseline: number;
|
|
@@ -592,4 +601,4 @@ declare function hostedClientFromEnv(overrides?: Partial<HostedTenant> & {
|
|
|
592
601
|
}): HostedClient | undefined;
|
|
593
602
|
//#endregion
|
|
594
603
|
export { ScalarDistribution as A, InsightReport as C, OutcomeCorrelationInsight as D, LiftInsight as E, Recommendation as O, FailureClusterInsight as S, JudgeInsight as T, UnixNanoTimestamp as _, hostedTenantFromEnv as a, ExecutionInsight as b, EvalRunGenerationSnapshot as c, HostedIngestHeaders as d, HostedWireVersion as f, TraceSpanEvent as g, IngestTracesRequest as h, hostedClientFromEnv as i, TokenUsageInsight as j, ReleaseSummary as k, 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, CostProvenanceSummary as v, InterRaterInsight as w, FailureClassTally as x, ExecutionErrorOutcomeCell as y };
|
|
595
|
-
//# sourceMappingURL=client-
|
|
604
|
+
//# sourceMappingURL=client-DcvgkaZi.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"client-DcvgkaZi.d.ts","names":[],"sources":["../src/contract/insight-report.ts","../src/hosted/types.ts","../src/hosted/client.ts"],"mappings":";;;;;;UAqCiB;;EAEf;;;;EAKA,WAAW;;EAGX,WAAW;;;EAIX,cAAc,eAAe;;EAG7B;IACE,MAAM;IACN,QAAQ;;;;IAIR,aAAa;;;;;IAKb;MAAa;MAAe;;;;;;EAM9B,QAAQ,eAAe;;;;EAKvB,aAAa;;;;EAKb,OAAO;;;EAIP,kBAAkB;;;EAIlB,gBAAgB;;;;;EAMhB,qBAAqB;;;;;EAMrB,SAAS;;;;;;EAOT,wBAAwB;;;;EAKxB,iBAAiB;;;EAIjB,iBAAiB;;UAGF;EACf;IAAY;IAAW;;EACvB;IAAa;IAAW;;EACxB;IAAc;;EACd;;UAGe;;EAEf,YAAY;;EAEZ,SAAS;;;EAGT,YAAY;;;EAGZ;IACE;IACA,YAAY;IACZ,SAAS;IACT;;;EAGF,QAAQ;IAAQ;IAAe;;;;;EAI/B;IACE;IACA;IACA;;;;;EAKF;IACE;;;IAGA;;IAEA;;IAEA;;IAEA;;IAEA;;;;;IAKA,mBAAmB,OAAO,oBAAoB;;;;EAIhD;IACE;IACA;IACA;IACA;IACA;;;UAIa;;EAEf;;EAEA;;EAEA;;UAGe;EACf,OAAO;EACP,QAAQ;EACR,WAAW;EACX,QAAQ;EACR,YAAY;EACZ;IACE;IACA;IACA;IACA;IACA;;;;UAOa;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,WAAW;;;;;EAKX,WAAW;IAAQ;IAAe;;;UAGnB;;EAEf;;EAEA;;;EAGA,cAAc;;;EAGd;;;EAGA;;;EAGA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,SAAS;;EAET,mBAAmB;IACjB;IACA,SAAS;MAAQ;MAAe;;IAChC;;;UAIa;EACf;EACA;;EAEA;;EAEA;;;EAGA;;EAEA;;EAEA;;;;;;;;;;;EAWA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;;EAEf,UAAU;IACR;IACA;;IAEA;;IAEA;;IAEA;;EAEF;;;;;UAMe;;EAEf,cAAc;;EAEd;;EAEA;;UAGe;;EAEf;;;EAGA;EACA,UAAU;IAAQ;IAAe;IAAgB;;;UAGlC;;;EAGf;;EAEA;;EAEA;;EAEA;;EAEA;IACE;IACA;IACA;;;UAIa;;;EAGf;EACA,MAAM;IACJ;IACA;IACA;;;;EAIF;;UAGQ;;EAER;;EAEA;;;;EAIA;;EAEA;EACA;;KAGU,eACP;EACC;;EAEA;;EAEA;;EAEA;;EAEA;MAED;EACC;;EAEA;EACA;EACA;EACA;;UAGW;;EAEf;EACA;;EAEA;;;;EAIA,SAAS,eAAe;;;EAGxB;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;;EAEA;;;;cC5ZW;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
|
@@ -4,10 +4,10 @@ import { A as ChatClient, F as CreateChatClientOpts, V as createChatClient } fro
|
|
|
4
4
|
import { h as ProposalFindingOrigin, i as AnalystFinding, m as ProposalFinding, v as makeProposalFinding } from "../types-DVjczBM9.js";
|
|
5
5
|
import { n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-Brxr728w.js";
|
|
6
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, k as LabeledScenarioStore, 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-DiWLru6Z.js";
|
|
7
|
-
import { A as RunEvalOptions, B as OptimizerModelBudget, Bt as ComparisonCost, C as RunImprovementLoopOptions, Cn as ReferenceEquivalenceJudgeResult, D as RunOptimizationOptions, Dn as LlmJudgeDimension, E as PremeasuredOptimizationBaseline, En as runReferenceEquivalenceJudge, F as GepaOptimizationMethodConfig, Ft as ExternalTextOptimizerContext, G as ObjectiveSource, Gt as OptimizationMethodProvenance, H as AxisVerdict, Ht as OptimizationMethodComparison, I as GepaOptimizationRecipe, It as ExternalTextOptimizerResult, J as PromotionPolicy, K as ParetoSignificanceGateOptions, Kt as OptimizationMethodResult, L as GepaRunnerCommand, Lt as ExternalOptimizationExample, M as GepaAdaptiveEngineRun, Mt as composeGate, N as GepaEngineOptions, Nt as externalTextOptimizationMethod, On as LlmJudgeOptions, P as GepaEngineRun, Pt as ExternalTextOptimizationMethodConfig, R as gepaOptimizationMethod, Rt as ExternalTextEvaluationResponse, Sn as ReferenceEquivalenceJudgeOptions, T as runImprovementLoop, Tn as createReferenceEquivalenceJudge, U as BuildEvidenceVectorOptions, Ut as OptimizationMethodInput, V as AxisEvidence, Vt as OptimizationMethod, W as EvidenceVector, X as paretoPolicy, Xt as OptimizationTokenUsage, Y as buildEvidenceVector, Yt as OptimizationPackageSource, Z as paretoSignificanceGate, Zt as compareOptimizationMethods, bn as REFERENCE_EQUIVALENCE_JUDGE_VERSION, dt as DefaultProductionGateCheck, en as CampaignCellFailureReceipt, ft as DefaultProductionGateOptions, i as skillOptOptimizationMethod, in as RunCampaignOptions, j as runEval, kn as llmJudge, ln as fsCampaignStorage, lt as HeldOutGateOptions, mn as campaignSplitDigest, mt as defaultProductionGate, n as SkillOptRunnerCommand, on as runCampaign, ot as PowerPreflight, p as LoopProvenanceRecord, pt as DefaultProductionRewardHackingOptions, q as PromotionObjective, r as SkillOptTrainerConfig, sn as CampaignStorage, t as SkillOptOptimizationMethodConfig, un as inMemoryCampaignStorage, ut as heldOutGate, w as RunImprovementLoopResult, wn as ReferenceEquivalenceScenario, xn as ReferenceEquivalenceJudgeInput, yn as REFERENCE_EQUIVALENCE_INPUT_LIMITS, z as OpenAICompatibleOptimizerModel, zt as CompareOptimizationMethodsOptions } from "../skillopt-optimization-method-
|
|
8
|
-
import { A as ScalarDistribution, C as InsightReport, D as OutcomeCorrelationInsight, E as LiftInsight, O as Recommendation, S as FailureClusterInsight, T as JudgeInsight, b as ExecutionInsight, c as EvalRunGenerationSnapshot, g as TraceSpanEvent, j as TokenUsageInsight, k as ReleaseSummary, n as HostedTenant, o as EvalRunCellScore, s as EvalRunEvent, v as CostProvenanceSummary, w as InterRaterInsight, x as FailureClassTally, y as ExecutionErrorOutcomeCell } from "../client-
|
|
7
|
+
import { A as RunEvalOptions, B as OptimizerModelBudget, Bt as ComparisonCost, C as RunImprovementLoopOptions, Cn as ReferenceEquivalenceJudgeResult, D as RunOptimizationOptions, Dn as LlmJudgeDimension, E as PremeasuredOptimizationBaseline, En as runReferenceEquivalenceJudge, F as GepaOptimizationMethodConfig, Ft as ExternalTextOptimizerContext, G as ObjectiveSource, Gt as OptimizationMethodProvenance, H as AxisVerdict, Ht as OptimizationMethodComparison, I as GepaOptimizationRecipe, It as ExternalTextOptimizerResult, J as PromotionPolicy, K as ParetoSignificanceGateOptions, Kt as OptimizationMethodResult, L as GepaRunnerCommand, Lt as ExternalOptimizationExample, M as GepaAdaptiveEngineRun, Mt as composeGate, N as GepaEngineOptions, Nt as externalTextOptimizationMethod, On as LlmJudgeOptions, P as GepaEngineRun, Pt as ExternalTextOptimizationMethodConfig, R as gepaOptimizationMethod, Rt as ExternalTextEvaluationResponse, Sn as ReferenceEquivalenceJudgeOptions, T as runImprovementLoop, Tn as createReferenceEquivalenceJudge, U as BuildEvidenceVectorOptions, Ut as OptimizationMethodInput, V as AxisEvidence, Vt as OptimizationMethod, W as EvidenceVector, X as paretoPolicy, Xt as OptimizationTokenUsage, Y as buildEvidenceVector, Yt as OptimizationPackageSource, Z as paretoSignificanceGate, Zt as compareOptimizationMethods, bn as REFERENCE_EQUIVALENCE_JUDGE_VERSION, dt as DefaultProductionGateCheck, en as CampaignCellFailureReceipt, ft as DefaultProductionGateOptions, i as skillOptOptimizationMethod, in as RunCampaignOptions, j as runEval, kn as llmJudge, ln as fsCampaignStorage, lt as HeldOutGateOptions, mn as campaignSplitDigest, mt as defaultProductionGate, n as SkillOptRunnerCommand, on as runCampaign, ot as PowerPreflight, p as LoopProvenanceRecord, pt as DefaultProductionRewardHackingOptions, q as PromotionObjective, r as SkillOptTrainerConfig, sn as CampaignStorage, t as SkillOptOptimizationMethodConfig, un as inMemoryCampaignStorage, ut as heldOutGate, w as RunImprovementLoopResult, wn as ReferenceEquivalenceScenario, xn as ReferenceEquivalenceJudgeInput, yn as REFERENCE_EQUIVALENCE_INPUT_LIMITS, z as OpenAICompatibleOptimizerModel, zt as CompareOptimizationMethodsOptions } from "../skillopt-optimization-method-CwSYkv35.js";
|
|
8
|
+
import { A as ScalarDistribution, C as InsightReport, D as OutcomeCorrelationInsight, E as LiftInsight, O as Recommendation, S as FailureClusterInsight, T as JudgeInsight, b as ExecutionInsight, c as EvalRunGenerationSnapshot, g as TraceSpanEvent, j as TokenUsageInsight, k as ReleaseSummary, n as HostedTenant, o as EvalRunCellScore, s as EvalRunEvent, v as CostProvenanceSummary, w as InterRaterInsight, x as FailureClassTally, y as ExecutionErrorOutcomeCell } from "../client-DcvgkaZi.js";
|
|
9
9
|
import { i as InMemoryOutcomeStore, n as FileSystemOutcomeStore, o as OutcomeStore, r as FileSystemOutcomeStoreOptions, t as DeploymentOutcome } from "../outcome-store-BYHIuO0e.js";
|
|
10
|
-
import { a as summarizeExecution, i as analyzeRuns, n as ExecutionReport, r as SummarizeExecutionOptions, t as AnalyzeRunsOptions } from "../analyze-runs-
|
|
10
|
+
import { a as summarizeExecution, i as analyzeRuns, n as ExecutionReport, r as SummarizeExecutionOptions, t as AnalyzeRunsOptions } from "../analyze-runs-Cda5Xkj1.js";
|
|
11
11
|
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";
|
|
12
12
|
//#region src/contract/self-improve.d.ts
|
|
13
13
|
interface SelfImproveBudget {
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/contract/self-improve.ts","../../src/contract/define-agent-eval.ts","../../src/contract/measured-comparison.ts","../../src/contract/profile-measured-comparison.ts","../../src/contract/intake/run-record-dir.ts","../../src/contract/eval-reporting-suite.ts","../../src/contract/diff.ts","../../src/contract/intake/agent-trace.ts","../../src/contract/intake/code-agent-observation.ts","../../src/contract/intake/code-agent-session.ts","../../src/contract/intake/feedback-table.ts","../../src/contract/intake/otel-spans.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;;;EAIX,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;;;KCviB5B,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;;;
|
|
1
|
+
{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/contract/self-improve.ts","../../src/contract/define-agent-eval.ts","../../src/contract/measured-comparison.ts","../../src/contract/profile-measured-comparison.ts","../../src/contract/intake/run-record-dir.ts","../../src/contract/eval-reporting-suite.ts","../../src/contract/diff.ts","../../src/contract/intake/agent-trace.ts","../../src/contract/intake/code-agent-observation.ts","../../src/contract/intake/code-agent-session.ts","../../src/contract/intake/feedback-table.ts","../../src/contract/intake/otel-spans.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;;;EAIX,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;;;KCviB5B,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;;;UCvDd;EACf,QAAQ,gCAAgC;EACxC;EACA;;UAGe;EACf,YAAY;EACZ;EACA,QAAQ;EACR,MAAM;EACN,eAAe;EACf;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,QAAQ,OAAO,oCAAoC,QAAQ;;EAE3D;;EAEA,aAAa;EACb,SAAS;;UAGM;EACf,cAAc;EACd;IACE;IACA,MAAM;;;UAIO;EACf,YAAY;EACZ,cAAc;EACd;IACE;IACA,MAAM;;EAER,aAAa;EACb;EACA,YAAY;EACZ;EACA,WAAW;;;UAII,kBAAkB;EACjC;EACA,UAAU;EACV,WAAW;;;UAII,yBAAyB;EACxC,MAAM,KAAK;EACX,WAAW,KAAK;IAAkB;IAAc;;EAChD,QAAQ,KAAK;EACb,eAAe,KAAK,OAAO;EAC3B,UAAU,KAAK;EACf,UAAU,KAAK;EACf,OAAO,KAAK;;UAGG,kCAAkC;EACjD,uBAAuB,kBAAkB;EACzC,QAAQ;EACR,SAAS,yBAAyB;;EAElC;;EAEA,kBAAkB;;EAElB,kBAAkB;;;KAIR,8BAA8B,KACxC;EAGA,iBAAiB;EACjB,WAAW;EACX;;;iBAIc,2BACd,UAAU,sCACT;;iBAQa,4BACd,SAAS,qCACR;;iBAiBa,wBACd,UAAU,mCACT;iBAQa,0BAA0B,iBAAiB;;iBAarC,uBACpB,SAAS,gCACR,QAAQ;;;;;;;;iBAkEK,2BAA2B,MACzC,SAAS,kCAAkC,QAC1C;;iBAwRa,0CACd,SAAS,oCACR;;iBAuEa,oCACd,iBACC;iBAkHa,6BAA6B,iBAAiB;iBAM9C,oCACd,iBACC;iBAkBa,8BAA8B;;;;;;;;;;UCvsB7B;EACf,aAAa;EACb,QAAQ,gCAAgC;EACxC;EACA;;UAGe;EACf,YAAY;EACZ;EACA,aAAa;EACb,MAAM;EACN,SAAS;EACT;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,QACE,OAAO,kDACN,QAAQ;;EAEX;;EAEA,aAAa;EACb,SAAS;;UAGM;EACf,cAAc;EACd;IACE;IACA,MAAM;;;UAIO;EACf,YAAY;EACZ,cAAc;EACd;IACE;IACA,MAAM;;EAER,aAAa;EACb;EACA,YAAY;EACZ;EACA,WAAW;;;iBAIG,gCACd,UAAU,sCACT;;iBAQa,iCACd,SAAS,0CACR;;iBAqBa,sCACd,UAAU,4CACT;iBAOa,kCAAkC,iBAAiB;iBAInD,yCACd,iBACC;iBAIa,wCACd,iBACC;;;;;;iBASmB,qCACpB,SAAS,8CACR,QAAQ;;iBA0CK,wDACd,SAAS,kDACR;;iBAuEa,kDACd,iBACC;;;;UCnPc;;EAEf;;EAEA;;EAEA;;UAGe;;;;;;EAMf;;;;;;;EAOA,WAAW;;;;;;EAMX;;UAGe;;EAEf,MAAM;;EAEN,UAAU;;EAEV;;;;;;;;;;iBAkBoB,iBACpB,cACA,UAAS,0BACR,QAAQ;;;;;KC7CC,0BAA0B;UAErB;;;;;EAKf,UAAU,KAAK;;EAEf,OAAO;;;;;;;;;;EAUP;;;;UAKe;;;EAGf,QAAQ;;EAER;;IAEE;;IAEA;;;IAGA;;IAEA;;;IAGA,UAAU;;;EAGZ;;;;;;;iBAUoB,mBACpB,OAAO,yBACP,UAAS,4BACR,QAAQ;;;;;;UC5DM;EACf;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;;;EAGA,YAAY,eAAe,eAAe;;;;UAK3B;EACf;EACA;EACA;EACA;EACA;;EAEA,SAAS;;EAET,SAAS;;EAET,OAAO;;EAEP;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;;;;UAMe;EACf;EACA;EACA;EACA;EACA,oBAAoB;EACpB,mBAAmB;EACnB;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA,cAAc;;;;EAId,aAAa;;;;;;;;;iBAoDC,gBACd,QAAQ,2BACR,OAAO,4BACN;;;;;;iBAqEa,SAAS,QAAQ,cAAc,OAAO,eAAe;;;;;;;iBAsCrD,wBAAwB,KAAK,eAAe;;;KC1OhD;UAEK;EACf,MAAM;;EAEN;;UAGe;EACf;EACA;EACA;;;EAGA,cAAc;;UAGC;EACf;EACA,cAAc;EACd,QAAQ;;UAGO;EACf;EACA,eAAe;;UAGA;EACf;EACA;EACA;EACA;IAAQ;IAAc;;EACtB;IAAS;IAAe;;EACxB,OAAO;;;;UAOQ;EACf;;EAEA;;EAEA;EACA;EACA;;EAEA;;EAEA;;KAGU,kBAAkB,YAAY;;;;;;iBAW1B,gBAAgB,SAAS,qBAAqB;UAmE7C;;;;EAIf,SAAS,YAAY;;;EAGrB,cAAc;;;;;;;;iBASA,8BACd,MAAM,aACN,OAAO,kBACN;;;KCjLS;KAEA;KAEA;KAEA;KASA;UAEK;EACf;EACA;EACA;;UAGe;EACf;EACA;EACA,MAAM;EACN,SAAS;EACT;EACA,QAAQ;EACR;EACA;EACA,UAAU;;UAGK;EACf,QAAQ;EACR;EACA;EACA;IACE,QAAQ;IACR;;EAEF,SAAS;;UAGM;EACf,QAAQ;EACR;EACA;EACA,YAAY;;;;;;;iBAeE,wBACd,SAAS,iCACR;;;UCrCc;EACf;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf,QAAQ;EACR;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;;UAGe;EACf,MAAM;EACN,aAAa;EACb,SAAS;EACT,cAAc;;UAGC;EACf;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;EACA;EACA;EACA;EACA;EACA;;;EAGA,iBAAiB;;;EAGjB,YAAY;;iBAGE,oBAAoB,gBAAgB;iBAepC,iBACd,SAAS,gCACR;iBAIa,sBACd,SAAS,gCACR;iBAIa,oBACd,SAAS,gCACR;iBAIa,oBACd,SAAS,gCACR;iBAIa,cACd,SAAS,gCACR;cAIU,2BAAkB;;;UCpJd;;;EAGf;;EAEA;;;EAGA;;;EAGA,WAAW;;UAGI;EACf;;;EAGA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA,WAAW;;;EAGX,SAAS;;UAGM;;EAEf,SAAS;;;EAGT,OAAO;;;;EAIP;IAAU;IAAa;;;;;EAIvB;;UAGe;EACf,MAAM;;;EAGN,aAAa;IAAQ;IAAe;IAAe;;;iBAGrC,kBAAkB,MAAM,2BAA2B;;;UCvBlD;EACf,OAAO;;EAEP,eAAe;;EAEf;;;;;;EAMA,eAAe,eAAe,gBAAgB;;iBAGhC,cAAc,MAAM,uBAAuB"}
|
package/dist/contract/index.js
CHANGED
|
@@ -1,17 +1,17 @@
|
|
|
1
1
|
import { s as ValidationError } from "../errors-8YnH8WlF.js";
|
|
2
|
-
import { F as createChatClient, t as buildDefaultAnalystRegistry } from "../default-registry-
|
|
2
|
+
import { F as createChatClient, t as buildDefaultAnalystRegistry } from "../default-registry-CHmdy2An.js";
|
|
3
3
|
import { i as estimateCost, o as isModelPriced } from "../metrics-C9YY1OcL.js";
|
|
4
4
|
import { i as CostLedger } from "../cost-ledger-DHAjwNj7.js";
|
|
5
5
|
import { LLM_MODEL_ATTR_KEYS, SPAN_KIND_ATTR_KEYS } from "../trace-attributes.js";
|
|
6
|
-
import {
|
|
6
|
+
import { E as isOtlpModelCall, T as classifyOtlpSpanRole } from "../tools-D8yTtNSN.js";
|
|
7
7
|
import { l as makeProposalFinding } from "../proposal-findings-DCawte-y.js";
|
|
8
8
|
import { r as mapConcurrentRange } from "../concurrency-MUjT7VjM.js";
|
|
9
|
-
import { B as surfaceContentHash, Ct as llmJudge, M as compareOptimizationMethods, O as composeGate, Q as inMemoryCampaignStorage, S as defaultProductionGate, T as heldoutSignificance, V as surfaceHash, X as createRunCostLedger, Y as runCampaign, Z as fsCampaignStorage, _ as buildEvidenceVector, a as emitLoopProvenance, b as powerPreflight, ct as campaignSplitDigest, d as runImprovementLoop, dt as REFERENCE_EQUIVALENCE_INPUT_LIMITS, ft as REFERENCE_EQUIVALENCE_JUDGE_VERSION, g as gepaOptimizationMethod, j as assertOptimizationResult, k as externalTextOptimizationMethod, mt as runReferenceEquivalenceJudge, o as loopProvenanceArgsFromResult, p as runEval, pt as createReferenceEquivalenceJudge, rt as resolveRunDir, t as skillOptOptimizationMethod, v as paretoPolicy, x as heldOutGate, y as paretoSignificanceGate } from "../skillopt-optimization-method-
|
|
10
|
-
import {
|
|
9
|
+
import { B as surfaceContentHash, Ct as llmJudge, Ft as decidePairedPromotion, M as compareOptimizationMethods, O as composeGate, Q as inMemoryCampaignStorage, S as defaultProductionGate, T as heldoutSignificance, V as surfaceHash, X as createRunCostLedger, Y as runCampaign, Z as fsCampaignStorage, _ as buildEvidenceVector, a as emitLoopProvenance, b as powerPreflight, ct as campaignSplitDigest, d as runImprovementLoop, dt as REFERENCE_EQUIVALENCE_INPUT_LIMITS, ft as REFERENCE_EQUIVALENCE_JUDGE_VERSION, g as gepaOptimizationMethod, j as assertOptimizationResult, k as externalTextOptimizationMethod, mt as runReferenceEquivalenceJudge, o as loopProvenanceArgsFromResult, p as runEval, pt as createReferenceEquivalenceJudge, rt as resolveRunDir, t as skillOptOptimizationMethod, v as paretoPolicy, x as heldOutGate, y as paretoSignificanceGate } from "../skillopt-optimization-method-0UmPD6aP.js";
|
|
10
|
+
import { E as pairedBootstrap } from "../statistics-CnGCLLqc.js";
|
|
11
11
|
import { i as parseRunRecordSafe, r as modelHasSnapshot } from "../run-record-BIwU2wdV.js";
|
|
12
|
-
import { a as recordAggregateMeasurements, i as readTaskFailureLabels, o as summarizeExecutionMeasurements, s as summarizeTraceErrors, t as extractUsage } from "../extract-usage-
|
|
13
|
-
import { n as summarizeExecution, t as analyzeRuns } from "../analyze-runs-
|
|
14
|
-
import { a as campaignCellExecutionEvidence, c as campaignCellToRunRecord, o as campaignCellJudgeDimensions, s as campaignCellTaskScore } from "../reward-hacking-
|
|
12
|
+
import { a as recordAggregateMeasurements, i as readTaskFailureLabels, o as summarizeExecutionMeasurements, s as summarizeTraceErrors, t as extractUsage } from "../extract-usage-DIQpN-ww.js";
|
|
13
|
+
import { n as summarizeExecution, t as analyzeRuns } from "../analyze-runs-jjCmF8pU.js";
|
|
14
|
+
import { a as campaignCellExecutionEvidence, c as campaignCellToRunRecord, o as campaignCellJudgeDimensions, s as campaignCellTaskScore } from "../reward-hacking-a-kYs0-i.js";
|
|
15
15
|
import { n as InMemoryOutcomeStore, t as FileSystemOutcomeStore } from "../outcome-store-ChBKlTd_.js";
|
|
16
16
|
import { t as createHostedClient } from "../client-LIuo-KPv.js";
|
|
17
17
|
import { dirname, join } from "node:path";
|
|
@@ -828,7 +828,19 @@ function evaluatePairedMeasurements(options) {
|
|
|
828
828
|
const powerSufficient = power.sufficient;
|
|
829
829
|
const guardedDimensions = new Set(criticalDimensions);
|
|
830
830
|
const missingCriticalDimensions = criticalDimensions.filter((dimension) => !dimensions.includes(dimension));
|
|
831
|
-
const
|
|
831
|
+
const regressionProven = new Set(dimensions.filter((name) => {
|
|
832
|
+
if (!guardedDimensions.has(name)) return false;
|
|
833
|
+
const baselineDim = measurements.map((m) => dimensionScore(m.baseline, name));
|
|
834
|
+
return decidePairedPromotion(measurements.map((m) => dimensionScore(m.candidate, name)), baselineDim, {
|
|
835
|
+
confidence,
|
|
836
|
+
resamples,
|
|
837
|
+
statistic: "mean",
|
|
838
|
+
seed: bootstrapSeed,
|
|
839
|
+
threshold: regressionTolerance,
|
|
840
|
+
minPairs: minProductiveRuns
|
|
841
|
+
}).promote;
|
|
842
|
+
}));
|
|
843
|
+
const regressions = objectives.filter((objective) => objective.kind === "dimension" && guardedDimensions.has(objective.name) && objective.availability === "measured" && (objective.confidenceInterval.lower < -regressionTolerance || regressionProven.has(objective.name)));
|
|
832
844
|
const measurementCostUsd = measurements.reduce((sum, measurement) => sum + measurement.baseline.costUsd + measurement.candidate.costUsd, 0);
|
|
833
845
|
const measurementWorkDurationMs = measurements.reduce((sum, measurement) => sum + measurement.baseline.latencyMs + measurement.candidate.latencyMs, 0);
|
|
834
846
|
const incompleteRuns = measurements.flatMap((measurement) => [measurement.baseline, measurement.candidate]).filter((run) => !run.completed);
|