@tangle-network/agent-eval 0.145.6 → 0.145.8
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 +19 -0
- package/README.md +4 -0
- package/dist/analyst/index.d.ts +10 -10
- package/dist/analyst/index.js +1 -1
- package/dist/{backend-integrity-CsVin_Wb.d.ts → backend-integrity-BffcHGdm.d.ts} +2 -2
- package/dist/{backend-integrity-CsVin_Wb.d.ts.map → backend-integrity-BffcHGdm.d.ts.map} +1 -1
- package/dist/{benchmark-Ceoan7vk.d.ts → benchmark-DbiZbPdH.d.ts} +3 -3
- package/dist/{benchmark-Ceoan7vk.d.ts.map → benchmark-DbiZbPdH.d.ts.map} +1 -1
- package/dist/{benchmark-command-DspwA7cv.js → benchmark-command-rWP4-dJ6.js} +2 -2
- package/dist/{benchmark-command-DspwA7cv.js.map → benchmark-command-rWP4-dJ6.js.map} +1 -1
- package/dist/benchmarks/index.d.ts +5 -5
- package/dist/benchmarks/index.js +2 -2
- package/dist/campaign/index.d.ts +8 -8
- package/dist/campaign/index.js +4 -4
- package/dist/{campaign-jTOvqnse.js → campaign-Dg2B4W6d.js} +35 -11
- package/dist/campaign-Dg2B4W6d.js.map +1 -0
- package/dist/{capture-fetch-DDvpjVRU.d.ts → capture-fetch-hykwHfDI.d.ts} +2 -2
- package/dist/{capture-fetch-DDvpjVRU.d.ts.map → capture-fetch-hykwHfDI.d.ts.map} +1 -1
- package/dist/cli.js +1 -1
- package/dist/{client-DCVe0CwG.d.ts → client-D_TIV9pJ.d.ts} +4 -4
- package/dist/{client-DCVe0CwG.d.ts.map → client-D_TIV9pJ.d.ts.map} +1 -1
- package/dist/contract/index.d.ts +11 -11
- package/dist/contract/index.js +4 -4
- package/dist/{default-registry-CgzxnJsj.d.ts → default-registry-cpHtP2Og.d.ts} +6 -6
- package/dist/{default-registry-CgzxnJsj.d.ts.map → default-registry-cpHtP2Og.d.ts.map} +1 -1
- package/dist/{define-agent-eval-DLhZ8IHi.d.ts → define-agent-eval-BgG5DIOS.d.ts} +6 -6
- package/dist/{define-agent-eval-DLhZ8IHi.d.ts.map → define-agent-eval-BgG5DIOS.d.ts.map} +1 -1
- package/dist/{define-agent-eval-CViDh2P9.js → define-agent-eval-cSMEFzVu.js} +4 -4
- package/dist/{define-agent-eval-CViDh2P9.js.map → define-agent-eval-cSMEFzVu.js.map} +1 -1
- package/dist/{engine-FRZ3RLvT.d.ts → engine-DJqRKbhs.d.ts} +7 -7
- package/dist/{engine-FRZ3RLvT.d.ts.map → engine-DJqRKbhs.d.ts.map} +1 -1
- package/dist/{eval-campaign-UB-usSQ2.js → eval-campaign-aNqpefCS.js} +2 -2
- package/dist/{eval-campaign-UB-usSQ2.js.map → eval-campaign-aNqpefCS.js.map} +1 -1
- package/dist/{exact-types-BH1twmAJ.d.ts → exact-types-CzbhVDr2.d.ts} +2 -2
- package/dist/{exact-types-BH1twmAJ.d.ts.map → exact-types-CzbhVDr2.d.ts.map} +1 -1
- package/dist/experiment/index.d.ts +3 -3
- package/dist/{feedback-trajectory-juOozjAc.d.ts → feedback-trajectory-D9vYSob_.d.ts} +3 -3
- package/dist/{feedback-trajectory-juOozjAc.d.ts.map → feedback-trajectory-D9vYSob_.d.ts.map} +1 -1
- package/dist/hosted/index.d.ts +3 -3
- package/dist/index-BKShPTwZ.d.ts +1 -0
- package/dist/{index-COQYtuRF.d.ts → index-BNPtkBPf.d.ts} +2 -2
- package/dist/{index-COQYtuRF.d.ts.map → index-BNPtkBPf.d.ts.map} +1 -1
- package/dist/{index-DF3ynqmJ.d.ts → index-CM-e2LiV.d.ts} +11 -11
- package/dist/{index-DF3ynqmJ.d.ts.map → index-CM-e2LiV.d.ts.map} +1 -1
- package/dist/{index-PgqfwhsM.d.ts → index-YrUFx3FU.d.ts} +5 -5
- package/dist/{index-PgqfwhsM.d.ts.map → index-YrUFx3FU.d.ts.map} +1 -1
- package/dist/index.d.ts +22 -22
- package/dist/index.js +9 -9
- package/dist/{insight-report-CRi-Ufrj.d.ts → insight-report-BeT8KCgI.d.ts} +3 -3
- package/dist/{insight-report-CRi-Ufrj.d.ts.map → insight-report-BeT8KCgI.d.ts.map} +1 -1
- package/dist/{llm-judge-DuYa4SEA.js → llm-judge-DhiJqSjB.js} +3 -3
- package/dist/{llm-judge-DuYa4SEA.js.map → llm-judge-DhiJqSjB.js.map} +1 -1
- package/dist/meta-eval/index.d.ts +1 -1
- package/dist/{mint-Dj9Ww_3I.js → mint-BV6tLVWl.js} +2 -2
- package/dist/{mint-Dj9Ww_3I.js.map → mint-BV6tLVWl.js.map} +1 -1
- package/dist/multishot/index.d.ts +1 -1
- package/dist/openapi.json +1 -1
- package/dist/{pre-registration-DeRvl9sE.d.ts → pre-registration-zFSLEiFU.d.ts} +2 -2
- package/dist/{pre-registration-DeRvl9sE.d.ts.map → pre-registration-zFSLEiFU.d.ts.map} +1 -1
- package/dist/{produced-state-D6j7qy1Q.js → produced-state-BNyyud4g.js} +2 -2
- package/dist/{produced-state-D6j7qy1Q.js.map → produced-state-BNyyud4g.js.map} +1 -1
- package/dist/{promotion-policy-CsMZJOB-.d.ts → promotion-policy-u3wj6w3U.d.ts} +2 -2
- package/dist/{promotion-policy-CsMZJOB-.d.ts.map → promotion-policy-u3wj6w3U.d.ts.map} +1 -1
- package/dist/{provenance-8L-_4xiL.d.ts → provenance-AACfvbUR.d.ts} +124 -108
- package/dist/provenance-AACfvbUR.d.ts.map +1 -0
- package/dist/{registry-oJeeI4-a.d.ts → registry-B_1Frl8a.d.ts} +3 -3
- package/dist/{registry-oJeeI4-a.d.ts.map → registry-B_1Frl8a.d.ts.map} +1 -1
- package/dist/{release-confidence-BFRE5WSp.d.ts → release-confidence-4XrqlpFD.d.ts} +3 -3
- package/dist/{release-confidence-BFRE5WSp.d.ts.map → release-confidence-4XrqlpFD.d.ts.map} +1 -1
- package/dist/{release-confidence-CxDuiAev.js → release-confidence-BknrpBnO.js} +2 -2
- package/dist/{release-confidence-CxDuiAev.js.map → release-confidence-BknrpBnO.js.map} +1 -1
- package/dist/reporting.d.ts +3 -3
- package/dist/reporting.js +1 -1
- package/dist/{researcher-DaN4GST-.d.ts → researcher-Du-oniHp.d.ts} +3 -3
- package/dist/{researcher-DaN4GST-.d.ts.map → researcher-Du-oniHp.d.ts.map} +1 -1
- package/dist/{reward-hacking-DSSTuI9r.d.ts → reward-hacking-Bu8ev6PR.d.ts} +2 -2
- package/dist/{reward-hacking-DSSTuI9r.d.ts.map → reward-hacking-Bu8ev6PR.d.ts.map} +1 -1
- package/dist/{reward-hacking-DNgjilrV.js → reward-hacking-DFo2FU5J.js} +2 -2
- package/dist/{reward-hacking-DNgjilrV.js.map → reward-hacking-DFo2FU5J.js.map} +1 -1
- package/dist/rl.d.ts +5 -5
- package/dist/rl.js +4 -4
- package/dist/rollout/index.d.ts +1 -1
- package/dist/rollout/index.js +2 -2
- package/dist/{rollout-BWtw0I_6.js → rollout-ytVQ7WT8.js} +2 -2
- package/dist/{rollout-BWtw0I_6.js.map → rollout-ytVQ7WT8.js.map} +1 -1
- package/dist/{rubric-predictive-validity-BgxtKe4G.d.ts → rubric-predictive-validity-C7LnNvF2.d.ts} +2 -2
- package/dist/{rubric-predictive-validity-BgxtKe4G.d.ts.map → rubric-predictive-validity-C7LnNvF2.d.ts.map} +1 -1
- package/dist/{run-record-BvHPVS-i.js → run-record-D2lDdSAz.js} +5 -3
- package/dist/run-record-D2lDdSAz.js.map +1 -0
- package/dist/{run-record-CKiihE6f.d.ts → run-record-DVV82Gwh.d.ts} +8 -3
- package/dist/run-record-DVV82Gwh.d.ts.map +1 -0
- package/dist/{skillopt-optimization-method-BTls2l-Z.js → skillopt-optimization-method-C6JSfYxT.js} +2 -2
- package/dist/{skillopt-optimization-method-BTls2l-Z.js.map → skillopt-optimization-method-C6JSfYxT.js.map} +1 -1
- package/dist/{skillopt-optimization-method-CFDSzD-7.d.ts → skillopt-optimization-method-DO6OFgcz.d.ts} +5 -5
- package/dist/{skillopt-optimization-method-CFDSzD-7.d.ts.map → skillopt-optimization-method-DO6OFgcz.d.ts.map} +1 -1
- package/dist/{statistical-heldout-C4De2tRI.d.ts → statistical-heldout-CBGPSZ5X.d.ts} +3 -3
- package/dist/{statistical-heldout-C4De2tRI.d.ts.map → statistical-heldout-CBGPSZ5X.d.ts.map} +1 -1
- package/dist/{store-tool-spans-CggeC1LB.d.ts → store-tool-spans-C9c4R7ca.d.ts} +4 -4
- package/dist/{store-tool-spans-CggeC1LB.d.ts.map → store-tool-spans-C9c4R7ca.d.ts.map} +1 -1
- package/dist/{summary-report-CaL-Hnxt.d.ts → summary-report-B__Y5ub3.d.ts} +2 -2
- package/dist/{summary-report-CaL-Hnxt.d.ts.map → summary-report-B__Y5ub3.d.ts.map} +1 -1
- package/dist/{tool-groups-BdcoEgtV.d.ts → tool-groups-CZotz-e_.d.ts} +3 -3
- package/dist/tool-groups-CZotz-e_.d.ts.map +1 -0
- package/dist/trace-repair/index.d.ts +2 -2
- package/dist/traces.d.ts +6 -6
- package/dist/traces.js +1 -1
- package/dist/{types-BxLccGMf.d.ts → types-BjsNDR49.d.ts} +3 -3
- package/dist/{types-BxLccGMf.d.ts.map → types-BjsNDR49.d.ts.map} +1 -1
- package/dist/{types-BRsxjg7z.d.ts → types-C4bSVIr7.d.ts} +3 -3
- package/dist/{types-BRsxjg7z.d.ts.map → types-C4bSVIr7.d.ts.map} +1 -1
- package/dist/{types-DLQx4mKU.d.ts → types-vXyshMwx.d.ts} +2 -2
- package/dist/{types-DLQx4mKU.d.ts.map → types-vXyshMwx.d.ts.map} +1 -1
- package/dist/wire/index.d.ts +1 -1
- package/docs/eval-surface-map.md +2 -1
- package/docs/wire-protocol.md +4 -0
- package/package.json +1 -1
- package/dist/campaign-jTOvqnse.js.map +0 -1
- package/dist/index-Ba3YrbAL.d.ts +0 -1
- package/dist/provenance-8L-_4xiL.d.ts.map +0 -1
- package/dist/run-record-BvHPVS-i.js.map +0 -1
- package/dist/run-record-CKiihE6f.d.ts.map +0 -1
- package/dist/tool-groups-BdcoEgtV.d.ts.map +0 -1
|
@@ -1,10 +1,10 @@
|
|
|
1
1
|
import { c as CostLedgerHandle } from "./cost-ledger-DbQdN3nO.js";
|
|
2
|
-
import { a as RunRecord } from "./run-record-
|
|
2
|
+
import { a as RunRecord } from "./run-record-DVV82Gwh.js";
|
|
3
3
|
import { E as LlmCallMetadata, p as ChatClient } from "./types-Cx3YUh2r.js";
|
|
4
4
|
import { _ as ExternalTextEvaluationRequest, c as ExternalOptimizerModelBudget, g as ExternalTextCandidate, h as ExternalOptimizerRunnerCommand, l as ExternalOptimizerModelCall, m as ExternalOptimizerResumeMode, r as ExternalOptimizerCallbackLimits } from "./external-optimizer-contracts-lixrOZdX.js";
|
|
5
|
-
import { R as Scenario, S as JudgeConfig, i as CampaignCostMeter, j as MutableSurface, o as CampaignScenarioIdentity, p as Gate } from "./types-
|
|
6
|
-
import { Q as OptimizationPackageSource, W as OptimizationMethod, k as RedTeamCase, z as CanaryOptions } from "./provenance-
|
|
7
|
-
import { t as DetectRewardHackingInput } from "./reward-hacking-
|
|
5
|
+
import { R as Scenario, S as JudgeConfig, i as CampaignCostMeter, j as MutableSurface, o as CampaignScenarioIdentity, p as Gate } from "./types-BjsNDR49.js";
|
|
6
|
+
import { Q as OptimizationPackageSource, W as OptimizationMethod, k as RedTeamCase, z as CanaryOptions } from "./provenance-AACfvbUR.js";
|
|
7
|
+
import { t as DetectRewardHackingInput } from "./reward-hacking-Bu8ev6PR.js";
|
|
8
8
|
//#region src/reference-equivalence-judge.d.ts
|
|
9
9
|
declare const REFERENCE_EQUIVALENCE_JUDGE_VERSION = "reference-equivalence-judge-v1-2026-07-13";
|
|
10
10
|
declare const REFERENCE_EQUIVALENCE_INPUT_LIMITS: {
|
|
@@ -431,4 +431,4 @@ interface SkillOptOptimizationMethodConfig<TScenario extends Scenario, TArtifact
|
|
|
431
431
|
declare function skillOptOptimizationMethod<TScenario extends Scenario, TArtifact>(config: SkillOptOptimizationMethodConfig<TScenario, TArtifact>): OptimizationMethod<TScenario, TArtifact>;
|
|
432
432
|
//#endregion
|
|
433
433
|
export { campaignScenarioIdentity as A, runReferenceEquivalenceJudge as B, ExternalTextOptimizerContext as C, decodeExternalTextCandidate as D, ExternalTextEvaluationResponse as E, ReferenceEquivalenceJudgeInput as F, ReferenceEquivalenceJudgeOptions as I, ReferenceEquivalenceJudgeResult as L, campaignSplitDigestFromIdentities as M, REFERENCE_EQUIVALENCE_INPUT_LIMITS as N, assertCampaignDesign as O, REFERENCE_EQUIVALENCE_JUDGE_VERSION as P, ReferenceEquivalenceScenario as R, ExternalTextOptimizationMethodConfig as S, ExternalOptimizationExample as T, DefaultProductionGateOptions as _, GepaAdaptiveEngineRun as a, composeGate as b, GepaOptimizationMethodConfig as c, gepaOptimizationMethod as d, OpenAICompatibleOptimizerModel as f, DefaultProductionGateCheck as g, heldOutGate as h, skillOptOptimizationMethod as i, campaignSplitDigest as j, assertCampaignSplitIdentity as k, GepaOptimizationRecipe as l, HeldOutGateOptions as m, SkillOptRunnerCommand as n, GepaEngineOptions as o, OptimizerModelBudget as p, SkillOptTrainerConfig as r, GepaEngineRun as s, SkillOptOptimizationMethodConfig as t, GepaRunnerCommand as u, DefaultProductionRewardHackingOptions as v, ExternalTextOptimizerResult as w, externalTextOptimizationMethod as x, defaultProductionGate as y, createReferenceEquivalenceJudge as z };
|
|
434
|
-
//# sourceMappingURL=skillopt-optimization-method-
|
|
434
|
+
//# sourceMappingURL=skillopt-optimization-method-DO6OFgcz.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"skillopt-optimization-method-
|
|
1
|
+
{"version":3,"file":"skillopt-optimization-method-DO6OFgcz.d.ts","names":[],"sources":["../src/reference-equivalence-judge.ts","../src/campaign/coverage.ts","../src/campaign/external-text-evaluation.ts","../src/campaign/external-text-optimization-contract.ts","../src/campaign/external-text-optimization.ts","../src/campaign/gates/compose.ts","../src/campaign/gates/default-production-gate.ts","../src/campaign/gates/heldout-gate.ts","../src/campaign/optimizer-model.ts","../src/campaign/gepa-optimization-method.ts","../src/campaign/skillopt-optimization-method.ts"],"mappings":";;;;;;;;cAOa;cAEA;WACX;WACA;WACA;;UAGe,qCAAqC;EACpD;EACA;;UAGe;EACf;EACA;EACA;;UAGe;;EAEf,MAAM;;EAEN;;EAEA,SAAS;;EAET,aAAa;;UAGE,wCAAwC;EACvD;EACA;EACA;EACA;;;iBA0Bc,gCACd,SAAS,mCACR,oBAAoB;;iBA2CD,6BACpB,OAAO,gCACP,SAAS,mCACR,QAAQ;;;;iBCxGK,qBAAqB,kBAAkB,UACrD,oBAAoB,aACpB;;iBA2Bc,yBAAyB,kBAAkB,UACzD,UAAU,YACT,2BAA2B,KAAK;;iBAUnB,kCACd,oBAAoB,4BACpB;;iBAgBc,oBAAoB,kBAAkB,UACpD,oBAAoB,aACpB;;iBAOc,4BACd,oBAAoB,4BACpB,cACA;;;UCpEe;EACf;EACA;;UAGe;EACf;EACA;IACE;IACA,YAAY;IACZ;IACA;;;IAGA;IACA;MACE;MACA;;;;iBAwHU,4BAA4B,WAAW,wBAAwB;;;UC3I9D;WACN;WACA;WACA;WACA;WACA;WACA,eAAe;WACf,mBAAmB;WACnB,uBAAuB;WACvB;WACA;;WAEA;WACA;WACA;WACA,QAAQ;;WAER,MAAM;WACN,WACP,SAFa,kCAGV,QAAQ;;UAGE;EACf,eAAe;EACf;EACA;IACM;;IACA;;IACA;IAAkB;;;;;;;;;UAqBT,qCACf,kBAAkB,UAClB;EAEA;EACA,QAAQ,KAAK;EACb;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA,SAAS;EACT;EACA;EACA,oBAAoB,UAAU;EAC9B,oBAAoB,UAAU,WAAW,UAAU;EACnD,MAAM,SAAS,iCAAiC,QAAQ;;;;;;;;;;iBCN1C,+BAA+B,kBAAkB,UAAU,WACzE,QAAQ,qCAAqC,WAAW,aACvD,mBAAmB,WAAW;;;;;;iBClEjB,YAAY,qBAAqB,kBAAkB,WAAW,aACzE,OAAO,MAAM,KAAK,WAAW,cAC/B,KAAK,WAAW;;;KCaP;KAOA,wCAAwC,KAClD;EAGA,SAAS,YAAY;;UAGN;;;EAGf,kBAAkB;;;;;EAKlB;;EAEA;;EAEA;;EAEA;;;;EAIA;;;;EAIA;;;;;EAKA;;;;EAIA;;;EAGA;;;;EAIA,iBAAiB;;;EAGjB,aAAa;;EAEb,gBAAgB;;EAEhB,SAAS;;;;;EAKT,iBAAiB;;;;;iBAMH,sBAAsB,WAAW,kBAAkB,UACjE,SAAS,+BACR,KAAK,WAAW;;;UC/EF,mBAAmB,kBAAkB,WAAW;EAC/D,WAAW;;;EAGX;;EAEA;;;;EAIA;;EAEA;;EAEA;;;;;;;;iBASc,YAAY,WAAW,kBAAkB,UACvD,SAAS,mBAAmB,aAC3B,KAAK,WAAW;;;KCvCP,uBAAuB;;UAGlB;EACf;EACA,QAAQ;;EAER,MAAM;;EAEN;;;;;UCmDe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;;;EAKA,eAAe;;;UAIA,sBAAsB;;EAErC;;;KAIU,wBAAwB;;;;;;;KAQxB;EAEN;EACA,KAAK;;EAGL;EACA,eAAe;;EAGf;EACA,eAAe;;EAEf;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;;EAGA;EACA,eAAe;EACf;;EAGA;EACA,eAAe;EACf;;EAGA;EACA,kBAAkB;EAClB,cAAc;EACd;;;KAIM,oBAAoB;UAEf,6BAA6B,kBAAkB,UAAU;;EAExE;;EAEA,QAAQ;;EAER;;EAEA;;EAEA;;;;;EAKA;;EAEA;;EAEA;;EAEA,2BAA2B,QAAQ;;EAEnC;;;;;;EAMA,YAAY;;;;;;EAMZ,oBAAoB,UAAU;;EAE9B,oBAAoB,UAAU,WAAW,UAAU;;EAEnD,SAAS;;;;;EAKT;EACA,SAAS;;;;;;;;;;iBAWK,uBAAuB,kBAAkB,UAAU,WACjE,QAAQ,6BAA6B,WAAW,aAC/C,mBAAmB,WAAW;;;UC5IhB;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;;;;EAKA,YAAY;;KAGF,wBAAwB;UAEnB,iCAAiC,kBAAkB,UAAU;EAC5E;;EAEA;EACA;;EAEA;EACA,SAAS;;;;;EAKT,WAAW;;EAEX;;EAEA;EACA;;EAEA;;EAEA,2BAA2B,QAAQ;EACnC;EACA,oBAAoB,UAAU;EAC9B,oBAAoB,UAAU,WAAW,UAAU;EACnD,SAAS;EACT,SAAS;;;iBAIK,2BAA2B,kBAAkB,UAAU,WACrE,QAAQ,iCAAiC,WAAW,aACnD,mBAAmB,WAAW"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { R as Scenario, b as GenerationRecord, p as Gate, w as JudgeScore } from "./types-
|
|
2
|
-
import { r as SignedManifest, w as EProcessState } from "./pre-registration-
|
|
1
|
+
import { R as Scenario, b as GenerationRecord, p as Gate, w as JudgeScore } from "./types-BjsNDR49.js";
|
|
2
|
+
import { r as SignedManifest, w as EProcessState } from "./pre-registration-zFSLEiFU.js";
|
|
3
3
|
import { a as PairedPromotionDecision, d as PairedBootstrapResult, i as PairedMcNemarEvidence, r as PairedDecisionStatistic, t as PairedDecisionMethod } from "./paired-promotion-decision-CGzg0cI_.js";
|
|
4
4
|
//#region src/campaign/gates/sequential.d.ts
|
|
5
5
|
type SequentialDecision = 'promote' | 'continue' | 'undecided-at-maxN';
|
|
@@ -279,4 +279,4 @@ declare function dimensionRegressions(candidate: Map<string, Record<string, Judg
|
|
|
279
279
|
}): DimensionRegression[];
|
|
280
280
|
//#endregion
|
|
281
281
|
export { detectScale as a, pairHoldout as c, SequentialDecision as d, SequentialObservation as f, sequentialPairedGate as g, sequentialDecide as h, PairedHoldout as i, SequentialDecideFn as l, SequentialPairedGateOptions as m, HeldoutSignificance as n, dimensionRegressions as o, SequentialPairedGate as p, HeldoutSignificanceOptions as r, heldoutSignificance as s, DimensionRegression as t, SequentialDecideOptions as u };
|
|
282
|
-
//# sourceMappingURL=statistical-heldout-
|
|
282
|
+
//# sourceMappingURL=statistical-heldout-CBGPSZ5X.d.ts.map
|
package/dist/{statistical-heldout-C4De2tRI.d.ts.map → statistical-heldout-CBGPSZ5X.d.ts.map}
RENAMED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"statistical-heldout-
|
|
1
|
+
{"version":3,"file":"statistical-heldout-CBGPSZ5X.d.ts","names":[],"sources":["../src/campaign/gates/sequential.ts","../src/campaign/gates/statistical-heldout.ts"],"mappings":";;;;KAuCY;UAEK;EACf,UAAU;;EAEV;;EAEA;;;EAGA;;UAGe;;;EAGf;;;;EAIA;;;EAGA;;EAEA;;;;EAIA;;;EAGA;;;EAGA,kBAAkB;;EAElB;;UAGe,qBAAqB,qBAAqB,kBAAkB,WAAW,kBAC9E,KAAK,WAAW;;;;;;;;EAQxB,QAAQ,gBAAgB;;EAExB,SAAS;IAAkB,UAAU;;;;;;;;;;;;;;;;;iBA0MvB,qBAAqB,qBAAqB,kBAAkB,WAAW,UACrF,SAAS,8BACR,qBAAqB,WAAW;UAsFlB;;EAEf;;EAEA;;EAEA;;EAEA;;UAGe;GACd;IAAQ,SAAS;;IAAyB;IAAe;;;EAE1D,SAAS;;;;;;;;;;;;;;;;;;;;;;;;;iBA0BK,iBAAiB,UAAS,0BAA+B;;;UCtXxD;;EAEf;;EAEA;;EAEA;;;;;;;;;;;;;iBAcc,YACd,WAAW,YAAY,eAAe,cACtC,UAAU,YAAY,eAAe,cACrC,aAAa,aACb,SAAS,GAAG,oCACX;UAsDc;EACf,QAAQ;;;;;;;;;;;;EAYR,WAAW;;;;EAIX,iBAAiB;;;;;;;EAOjB,UAAU;;EAEV,mBAAmB;;EAEnB,SAAS;;;;EAIT;;EAEA;;EAEA;;EAEA,gBAAgB;;EAEhB;;;;EAIA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;;EAEA;EACA;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA6Bc,oBACd,QAAQ,eACR,OAAM,6BACL;UAkEc;EACf;;;EAGA,WAAW;;;;EAIX;;EAEA;IAAM;IAAa;;;EAEnB,mBAAmB;;EAEnB,SAAS;;EAET;;;;;EAKA;EACA;EACA;;;;;iBAMc,YAAY;;;;;;;;;;;;;;;;;;;iBAsBZ,qBACd,WAAW,YAAY,eAAe,cACtC,UAAU,YAAY,eAAe,cACrC,aAAa,aACb,8BACA;EACE;EACA;EACA;EACA;;;EAGA;IAED"}
|
|
@@ -2,9 +2,9 @@ import { r as CaptureIntegrityError } from "./errors-DEE6u6ot.js";
|
|
|
2
2
|
import { c as CostLedgerHandle } from "./cost-ledger-DbQdN3nO.js";
|
|
3
3
|
import { S as ToolSpan } from "./schema-BtVldJ3T.js";
|
|
4
4
|
import { s as TraceStore } from "./store-CT9YIIve.js";
|
|
5
|
-
import { A as SearchSpanResult, B as ViewSpansResult, D as DatasetOverview, F as TraceAnalystFilters, H as ViewTraceResult, S as BoundedTraceAnalysisStoreOptions, T as TraceAnalysisStoreContext, f as AnalystUsageReceipt, i as AnalystFinding, j as SearchTraceResult, k as QueryTracesPage, w as TraceAnalysisStore } from "./types-
|
|
6
|
-
import { a as TraceAnalystLimits, i as TraceAnalysisEngineResult, n as TraceAnalysisEngine } from "./engine-
|
|
7
|
-
import { t as TraceToolGroupName } from "./tool-groups-
|
|
5
|
+
import { A as SearchSpanResult, B as ViewSpansResult, D as DatasetOverview, F as TraceAnalystFilters, H as ViewTraceResult, S as BoundedTraceAnalysisStoreOptions, T as TraceAnalysisStoreContext, f as AnalystUsageReceipt, i as AnalystFinding, j as SearchTraceResult, k as QueryTracesPage, w as TraceAnalysisStore } from "./types-vXyshMwx.js";
|
|
6
|
+
import { a as TraceAnalystLimits, i as TraceAnalysisEngineResult, n as TraceAnalysisEngine } from "./engine-DJqRKbhs.js";
|
|
7
|
+
import { t as TraceToolGroupName } from "./tool-groups-CZotz-e_.js";
|
|
8
8
|
//#region src/trace/otel.d.ts
|
|
9
9
|
declare const OTEL_AGENT_EVAL_SCOPE: {
|
|
10
10
|
name: string;
|
|
@@ -339,4 +339,4 @@ declare class ToolTraceMissingError extends CaptureIntegrityError {
|
|
|
339
339
|
declare function toolSpansToTraceAnalysisStore(spans: readonly ToolSpan[] | null | undefined, options?: ToolSpansToTraceAnalysisStoreOptions): TraceAnalysisStore;
|
|
340
340
|
//#endregion
|
|
341
341
|
export { createBoundedTraceAnalysisStore as A, OtlpSpan as B, scoreTraceInsightReadiness as C, AnalyzeTracesResult as D, AnalyzeTracesOptions as E, redactString as F, redactValue as I, OTEL_AGENT_EVAL_SCOPE as L, REDACTION_VERSION as M, RedactionReport as N, analyzeTraces as O, RedactionRule as P, OtlpExport as R, planTraceInsightQuestions as S, AnalyzeTracesInput as T, exportRunAsOtlp as V, buildTraceInsightPrompt as _, ToolSpansToTraceAnalysisStoreOptions as a, domainEvidencePattern as b, TraceInsightFinding as c, TraceInsightQualityGate as d, TraceInsightQuestion as f, buildTraceInsightContext as g, TraceInsightTask as h, OtlpFileTraceStoreOptions as i, DEFAULT_REDACTION_RULES as j, OtlpFlatLine as k, TraceInsightPanelRole as l, TraceInsightSuite as m, toolSpansToTraceAnalysisStore as n, otlpTextToTraceAnalysisStore as o, TraceInsightReadiness as p, OtlpFileTraceStore as r, TraceInsightContext as s, ToolTraceMissingError as t, TraceInsightPromptInput as u, defaultTraceInsightPanel as v, tokenizeDomainWords as w, inferDomainKeywords as x, describeTraceInsightScope as y, OtlpResourceSpans as z };
|
|
342
|
-
//# sourceMappingURL=store-tool-spans-
|
|
342
|
+
//# sourceMappingURL=store-tool-spans-C9c4R7ca.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"store-tool-spans-
|
|
1
|
+
{"version":3,"file":"store-tool-spans-C9c4R7ca.d.ts","names":[],"sources":["../src/trace/otel.ts","../src/trace/redact.ts","../src/trace-analyst/store-boundary.ts","../src/trace/otlp-flat.ts","../src/trace-analyst/analyst.ts","../src/trace-analyst/insights.ts","../src/trace-analyst/store-otlp.ts","../src/trace-analyst/store-tool-spans.ts"],"mappings":";;;;;;;;cAqBa;EAA0B;EAAoC;;UAE1D;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA,YAAY;IACV;IACA;MAAS;MAAsB;MAAmB;MAAsB;;;EAE1E,SAAS;IAAQ;IAAsB;IAAc,aAAa;;EAClE;IAAW;IAAc;;;UAGV;EACf;IAAY,YAAY;;EACxB,YAAY;IAAQ,cAAc;IAAuB,OAAO;;;UAGjD;EACf,eAAe;;;iBAIK,gBACpB,OAAO,YACP,eACA,gBAAe,4CACd,QAAQ;;;;;;;;;;;;;;;;UCvCM;EACf;EACA,SAAS;;EAET;;UAGe;EACf;EACA,QAAQ;;;cAIG,yBAAyB;cAezB;;;;;iBAMG,aACd,eACA,QAAO;EACJ;EAAgB,QAAQ;;;;;;;iBAuBb,YACd,gBACA,QAAO,iBACP,SAAQ;EACL;EAAgB,QAAQ;;;;;iBCjDb,gCACd,QAAQ,oBACR,UAAS,mCACR;;;KC5BS;UAEK;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;IACE,MAAM;IACN;;EAEF;IAAY,YAAY;;EACxB,YAAY;EACZ,SAAS;IAAQ;IAAc;IAAuB,aAAa;;;;;UCNpD;EACf;EACA;EACA;EACA;;KAGU,sBAAsB;UAEjB;;EAEf,iBAAiB;;EAEjB,QAAQ;;EAER;;EAEA,YAAY;EACZ,SAAS,QAAQ;EACjB;EACA;EACA,aAAa;EACb;EACA,yBAAyB;EACzB,4BAA4B;EAC5B,eAAe,SAAS;EACxB,OAAO;EACP,OAAO,iBAAiB,SAAS;EACjC,SAAS;;;;;;;;iBASW,cACpB,OAAO,oBACP,SAAS,uBACR,QAAQ;;;UCtDM;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,OAAO;;UAGQ;EACf;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;;UAGe;EACf;EACA;EACA;;UAGe;EACf,OAAO;EACP,WAAW;EACX,QAAQ;EACR,SAAS;EACT;;UAGe;EACf,OAAO;EACP;EACA;EACA,WAAW;EACX,OAAO;EACP,UAAU;EACV,OAAO;EACP,QAAQ;;UAGO;EACf;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,OAAO;;iBA0BO,oBAAoB;iBAMpB,oBAAoB,OAAO;iBAuB3B,sBAAsB,qBAAqB;iBAS3C,0BAA0B,OAAO;iBAsBjC,0BAA0B,OAAO,0BAA0B;iBAsD3D,yBAAyB,OAAO,0BAA0B;iBAa1D,2BAA2B,SAAS,sBAAsB;iBA2D1D,4BAA4B;iBA6B5B,wBAAwB,OAAO;;;UC1K9B;;EAEf;;EAEA;;EAEA;;EAEA;;KAGG,gCAAgC;UAEpB,kCAAkC;;EAEjD;;;;;;;;EAQA;;uBAMa,kCAAkC;mBAC9B;mBACA;mBACA;mBACA;UACT;UACA;UACA;UACA;EAER,YAAY,MAAM;EAyBZ,SAAS,kBAAkB,UAAU,4BAA4B;EAOjE,SACJ;IAAQ;IAAkB;KAC1B,UAAU,4BACT;EAeG,YACJ,UAAU,qBACV,UAAU,4BACT,QAAQ;EAoDL,YACJ;IAAQ,UAAU;IAAqB;IAAe;KACtD,UAAU,4BACT,QAAQ;EAmBL,YACJ,UAAU,qBACV,UAAU,4BACT;EAMG,UACJ;IAAQ;IAAkB;KAC1B,UAAU,4BACT,QAAQ;EAuCL,UACJ;IAAQ;IAAkB;IAA6B;KACvD,UAAU,4BACT,QAAQ;EAsFL,YACJ;IAAQ;IAAkB;IAAuB;KACjD,UAAU,4BACT,QAAQ;EA4CL,WACJ;IAAQ;IAAkB;IAAiB;IAAuB;KAClE,UAAU,4BACT,QAAQ;;;EA6CL,cAAc,UAAU,4BAA4B;UAI5C;qBAmBK,WAAW,SAAS,cAAc,QAAQ;qBAE1C;UAEL;UAmBA;UAsHA;UA+DN;UAkBA;UAwDA;UAmBM;;cA+CH,2BAA2B;mBACrB;mBACA;EAEjB,YAAY,MAAM;;;YAYF,WAAW,SAAS,cAAc,QAAQ;YAkBhD;;;iBAwBI,6BACd,cACA,OAAM,uCACL;;;;cC/7BU,8BAA8B;EACzC;;;;;;iBASc,8BACd,gBAAgB,+BAChB,UAAS,uCACR"}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { a as RunRecord } from "./run-record-
|
|
1
|
+
import { a as RunRecord } from "./run-record-DVV82Gwh.js";
|
|
2
2
|
import { r as PairedDecisionStatistic } from "./paired-promotion-decision-CGzg0cI_.js";
|
|
3
3
|
import { n as FailureClusterReport } from "./failure-cluster-BLURuWG4.js";
|
|
4
4
|
//#region src/held-out-gate.d.ts
|
|
@@ -525,4 +525,4 @@ interface ResearchReport {
|
|
|
525
525
|
declare function researchReport(runs: RunRecord[], opts?: ResearchReportOptions): Promise<ResearchReport>;
|
|
526
526
|
//#endregion
|
|
527
527
|
export { HeldOutGateConfig as C, HeldOutGate as S, SplitCoverage as T, paretoChart as _, ParetoPoint as a, GateDecision as b, ResearchReportCandidate as c, ResearchReportOptions as d, ResearchReportRecommendation as f, gainHistogram as g, SummaryTableRow as h, ParetoFigureSpec as i, ResearchReportDecision as l, SummaryTableOptions as m, GainDistributionFigureSpec as n, RESEARCH_REPORT_HARD_PAIR_FLOOR as o, SummaryTable as p, GainDistributionOptions as r, ResearchReport as s, GainDistributionBin as t, ResearchReportMethodology as u, researchReport as v, HeldOutGateRejectionCode as w, GateEvidence as x, summaryTable as y };
|
|
528
|
-
//# sourceMappingURL=summary-report-
|
|
528
|
+
//# sourceMappingURL=summary-report-B__Y5ub3.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"summary-report-
|
|
1
|
+
{"version":3,"file":"summary-report-B__Y5ub3.d.ts","names":[],"sources":["../src/held-out-gate.ts","../src/summary-report.ts"],"mappings":";;;;KAwIY;;;;;;;;UAiBK;;EAEf;;;EAGA;;;;EAIA;;EAEA;;EAEA;;;EAGA;;UAGe;;;;EAIf;;;;;EAKA;;;;;;;;;;;EAWA;;;;EAIA;;;EAGA;;EAEA;;EAEA;;;EAGA;;;;;;;;;;;;;;;;;EAiBA;;;;;;;;;;;;;EAaA;;;;KAKU,iBAAiB;UAEZ;;EAEf;;EAEA;;EAEA;;;;;;;;EAQA,iBAAiB;;;;;EAKjB,gBAAgB;;;;;EAKhB;;;;;;;;;EASA,gBAAgB;;;EAGhB;;;EAGA;IAAY;IAAa;;;;EAGzB;;;;;;;EAOA;IAAW;IAAW;IAAW;IAAqB;;;;;;EAKtD;;;;;EAKA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;EAKA;;;EAGA;;;;;;;EAOA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA,UAAU;;EAEV;;EAEA,eAAe;;;;;;;cAQJ;mBACM;mBACA;mBACA;mBACA;mBACA;mBACA;mBACA;mBACA;mBACA;mBACA;EAEjB,YAAY,QAAQ;;;;EAoCpB,SAAS,WAAW,aAAa,UAAU,cAAc;;;;;;;UA2SjD;;;;UC7oBO;;;EAGf;;EAEA;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;EACf,MAAM;EACN;EACA;;EAEA;;;;;;;iBAQc,aAAa,MAAM,aAAa,OAAM,sBAA2B;UA6IhE;EACf;;EAEA;;EAEA;;EAEA;;;EAGA;;;EAGA;;UAGe;EACf;EACA;EACA,QAAQ;EACR;IAAQ;IAAc;;;;;;;;iBAQR,YACd,MAAM,aACN;EACE;EACA,gBAAgB,eAAe;IAEhC;UAsDc;;EAEf;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;EACA,MAAM;EACN;EACA;IAAM;IAAa;;;UAGJ;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;iBAQc,cACd,MAAM,aACN,qBACA,oBACA,OAAM,0BACL;KA8ES;;;;;;cAYC;UAEI;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;EASA;;;;;;;EAOA;IAAS;IAAa;;;;;;EAKtB;;;;;EAKA;;EAEA,gBAAgB,eAAe;;EAE/B,kBAAkB;;EAElB;;EAEA;;EAEA;;;;;;EAMA;;UAGe;EACf,UAAU;EACV;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;IAAU;IAAa;;;;;;;EAMvB;EACA;;;;;;;EAOA;EACA;EACA,OAAO;EACP,UAAU;EACV;;UAGe;;;;;EAKf;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;EACA;;;;;;EAMA;EACA;EACA;IAAQ;IAAa;;EACrB;EACA,gBAAgB;EAChB,YAAY;EACZ,SAAS;EACT;IACE,QAAQ;IACR,OAAO;;EAET,aAAa;EACb,kBAAkB;EAClB;EACA;;;;;;;;;;;;;;;;;;;;;iBA6IoB,eACpB,MAAM,aACN,OAAM,wBACL,QAAQ"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { w as TraceAnalysisStore } from "./types-
|
|
2
|
-
import { _ as TraceAnalysisToolDescriptor } from "./engine-
|
|
1
|
+
import { w as TraceAnalysisStore } from "./types-vXyshMwx.js";
|
|
2
|
+
import { _ as TraceAnalysisToolDescriptor } from "./engine-DJqRKbhs.js";
|
|
3
3
|
//#region src/analyst/tool-groups.d.ts
|
|
4
4
|
/** Named tool sets. Kinds pass `tools: TRACE_TOOL_GROUPS.failureForensics` etc. */
|
|
5
5
|
type TraceToolGroupName =
|
|
@@ -25,4 +25,4 @@ type TraceToolGroupName =
|
|
|
25
25
|
declare function buildTraceToolsForGroup(group: TraceToolGroupName, store: TraceAnalysisStore): TraceAnalysisToolDescriptor[];
|
|
26
26
|
//#endregion
|
|
27
27
|
export { buildTraceToolsForGroup as n, TraceToolGroupName as t };
|
|
28
|
-
//# sourceMappingURL=tool-groups-
|
|
28
|
+
//# sourceMappingURL=tool-groups-CZotz-e_.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"tool-groups-CZotz-e_.d.ts","names":[],"sources":["../src/analyst/tool-groups.ts"],"mappings":";;;;KAmBY;;;;;;;;;;;;;;;;;;;;iBAgDI,wBACd,OAAO,oBACP,OAAO,qBACN"}
|
|
@@ -1,8 +1,8 @@
|
|
|
1
1
|
import { r as CaptureIntegrityError } from "../errors-DEE6u6ot.js";
|
|
2
2
|
import { t as DefaultVerdict } from "../verdict-E4eRNf7-.js";
|
|
3
3
|
import { b as CustomTokenPricing, c as CostLedgerHandle, p as CostProvenance } from "../cost-ledger-DbQdN3nO.js";
|
|
4
|
-
import { u as RunTokenUsage } from "../run-record-
|
|
5
|
-
import { a as TraceAnalystLimits, n as TraceAnalysisEngine } from "../engine-
|
|
4
|
+
import { u as RunTokenUsage } from "../run-record-DVV82Gwh.js";
|
|
5
|
+
import { a as TraceAnalystLimits, n as TraceAnalysisEngine } from "../engine-DJqRKbhs.js";
|
|
6
6
|
import { t as PrimeBridgeTransport } from "../prime-bridge-transport-6feEglLf.js";
|
|
7
7
|
import { t as RecordedTrajectoryStep } from "../steps-BArUxhna.js";
|
|
8
8
|
//#region src/trace-repair/mini-swe-scaffold.d.ts
|
package/dist/traces.d.ts
CHANGED
|
@@ -1,17 +1,17 @@
|
|
|
1
1
|
import { c as ValidationError, o as LimitExceededError, r as CaptureIntegrityError, s as NotFoundError, t as AgentEvalError } from "./errors-DEE6u6ot.js";
|
|
2
2
|
import { C as TraceEvent, D as isSandboxSpan, E as isRetrievalSpan, O as isToolSpan, S as ToolSpan, T as isLlmSpan, _ as Span, a as FAILURE_CLASSES, b as SpanStatus, c as JudgeSpan, d as RetrievalSpan, f as Run, g as SandboxSpan, h as RunStatus, i as EventKind, l as LlmSpan, m as RunOutcome, n as BudgetLedgerEntry, o as FailureClass, p as RunLayer, r as BudgetSpec, s as GenericSpan, t as Artifact, u as Message, v as SpanBase, w as isJudgeSpan, x as TRACE_SCHEMA_VERSION, y as SpanKind } from "./schema-BtVldJ3T.js";
|
|
3
|
-
import { a as RunRecord, l as RunTerminalOutcome, s as RunSplitTag, u as RunTokenUsage } from "./run-record-
|
|
4
|
-
import { a as ExtractedUsage, c as extractUsageFromResponse, i as ExtractUsageFromSseOptions, l as extractUsageFromSse, n as CaptureFetchOptions, o as SseUsageMode, r as captureFetchToRawSink, s as extractUsage, t as CaptureFetchContext } from "./capture-fetch-
|
|
3
|
+
import { a as RunRecord, l as RunTerminalOutcome, s as RunSplitTag, u as RunTokenUsage } from "./run-record-DVV82Gwh.js";
|
|
4
|
+
import { a as ExtractedUsage, c as extractUsageFromResponse, i as ExtractUsageFromSseOptions, l as extractUsageFromSse, n as CaptureFetchOptions, o as SseUsageMode, r as captureFetchToRawSink, s as extractUsage, t as CaptureFetchContext } from "./capture-fetch-hykwHfDI.js";
|
|
5
5
|
import { $ as InMemoryRawProviderSinkOptions, Q as InMemoryRawProviderSink, X as FileSystemRawProviderSink, Z as FileSystemRawProviderSinkOptions, at as RawProviderSinkFilter, et as NoopRawProviderSink, it as RawProviderSink, nt as RawProviderDirection, ot as defaultProviderRedactor, rt as RawProviderEvent, st as providerFromBaseUrl, tt as ProviderRedactor } from "./types-Cx3YUh2r.js";
|
|
6
6
|
import { a as RunFilter, i as InMemoryTraceStore, n as FileSystemTraceStore, o as SpanFilter, r as FileSystemTraceStoreOptions, s as TraceStore, t as EventFilter } from "./store-CT9YIIve.js";
|
|
7
7
|
import { a as TraceEmitterOptions, i as TraceEmitter, n as RunCompleteHookContext, o as llmSpanFromProvider, r as SpanHandle, t as RunCompleteHook } from "./emitter-DGQGoLyj.js";
|
|
8
8
|
import { a as RunIntegrityReport, i as RunIntegrityIssueCode, n as RunIntegrityExpectations, o as assertRunCaptured, r as RunIntegrityIssue, s as throwIfRunIncomplete, t as RunIntegrityError } from "./integrity-OrcI9Nau.js";
|
|
9
|
-
import { A as createBoundedTraceAnalysisStore, B as OtlpSpan, C as scoreTraceInsightReadiness, D as AnalyzeTracesResult, E as AnalyzeTracesOptions, F as redactString, I as redactValue, L as OTEL_AGENT_EVAL_SCOPE, M as REDACTION_VERSION, N as RedactionReport, O as analyzeTraces, P as RedactionRule, R as OtlpExport, S as planTraceInsightQuestions, T as AnalyzeTracesInput, V as exportRunAsOtlp, _ as buildTraceInsightPrompt, a as ToolSpansToTraceAnalysisStoreOptions, b as domainEvidencePattern, c as TraceInsightFinding, d as TraceInsightQualityGate, f as TraceInsightQuestion, g as buildTraceInsightContext, h as TraceInsightTask, i as OtlpFileTraceStoreOptions, j as DEFAULT_REDACTION_RULES, k as OtlpFlatLine, l as TraceInsightPanelRole, m as TraceInsightSuite, n as toolSpansToTraceAnalysisStore, o as otlpTextToTraceAnalysisStore, p as TraceInsightReadiness, r as OtlpFileTraceStore, s as TraceInsightContext, t as ToolTraceMissingError, u as TraceInsightPromptInput, v as defaultTraceInsightPanel, w as tokenizeDomainWords, x as inferDomainKeywords, y as describeTraceInsightScope, z as OtlpResourceSpans } from "./store-tool-spans-
|
|
9
|
+
import { A as createBoundedTraceAnalysisStore, B as OtlpSpan, C as scoreTraceInsightReadiness, D as AnalyzeTracesResult, E as AnalyzeTracesOptions, F as redactString, I as redactValue, L as OTEL_AGENT_EVAL_SCOPE, M as REDACTION_VERSION, N as RedactionReport, O as analyzeTraces, P as RedactionRule, R as OtlpExport, S as planTraceInsightQuestions, T as AnalyzeTracesInput, V as exportRunAsOtlp, _ as buildTraceInsightPrompt, a as ToolSpansToTraceAnalysisStoreOptions, b as domainEvidencePattern, c as TraceInsightFinding, d as TraceInsightQualityGate, f as TraceInsightQuestion, g as buildTraceInsightContext, h as TraceInsightTask, i as OtlpFileTraceStoreOptions, j as DEFAULT_REDACTION_RULES, k as OtlpFlatLine, l as TraceInsightPanelRole, m as TraceInsightSuite, n as toolSpansToTraceAnalysisStore, o as otlpTextToTraceAnalysisStore, p as TraceInsightReadiness, r as OtlpFileTraceStore, s as TraceInsightContext, t as ToolTraceMissingError, u as TraceInsightPromptInput, v as defaultTraceInsightPanel, w as tokenizeDomainWords, x as inferDomainKeywords, y as describeTraceInsightScope, z as OtlpResourceSpans } from "./store-tool-spans-C9c4R7ca.js";
|
|
10
10
|
import { C as TOOL_LATENCY_MS, D as asNumber, E as applyLlmSpanOtlpAttributes, O as contextInputTokens, S as TOOL_ARGS_CAPTURED, T as TOOL_NAME_ATTR_KEYS, _ as LlmSpanOtlpInput, a as LLM_CACHE_WRITE_TOKEN_ATTR_KEYS, b as RUN_COST_ATTR_KEYS, c as LLM_COST_USD, d as LLM_MODEL_ATTR_KEYS, f as LLM_MODEL_NAME, g as LLM_REASONING_TOKEN_ATTR_KEYS, h as LLM_REASONING_TOKENS, i as LLM_CACHE_WRITE_TOKENS, k as firstNumberAttr, l as LLM_INPUT_TOKENS, m as LLM_OUTPUT_TOKEN_ATTR_KEYS, n as LLM_CACHED_TOKENS, o as LLM_CONTEXT_TOKENS, p as LLM_OUTPUT_TOKENS, r as LLM_CACHED_TOKEN_ATTR_KEYS, s as LLM_COST_ATTR_KEYS, t as INPUT_VALUE, u as LLM_INPUT_TOKEN_ATTR_KEYS, v as OPENINFERENCE_SPAN_KIND, w as TOOL_NAME, x as SPAN_KIND_ATTR_KEYS, y as OUTPUT_VALUE } from "./attribute-vocabulary-DLJ6303h.js";
|
|
11
11
|
import { a as judgeSpans, c as runsForScenario, i as hasCapturedToolArgs, l as toolSpans, n as argHash, o as llmSpans, r as groupBy, s as runFailureClass, t as aggregateLlm } from "./query-CJ_DX8vl.js";
|
|
12
|
-
import "./index-
|
|
13
|
-
import { A as SearchSpanResult, B as ViewSpansResult, C as TRACE_ANALYSIS_LIMITS, D as DatasetOverview, E as DEFAULT_TRACE_ANALYST_BUDGETS, F as TraceAnalystFilters, H as ViewTraceResult, I as TraceAnalystSpan, L as TraceAnalystSpanKind, M as SpanMatchRecord, N as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, O as ErrorCluster, P as TraceAnalystByteBudgets, R as TraceAnalystSpanStatus, S as BoundedTraceAnalysisStoreOptions, T as TraceAnalysisStoreContext, V as ViewTraceOversized, j as SearchTraceResult, k as QueryTracesPage, w as TraceAnalysisStore, z as TraceAnalystTraceSummary } from "./types-
|
|
14
|
-
import { _ as TraceAnalysisToolDescriptor, g as TRACE_ANALYST_TOOL_NAMESPACE, h as BuildTraceAnalysisToolsOptions, v as buildTraceAnalysisToolDescriptors, y as traceAnalystFunctionGroup } from "./engine-
|
|
12
|
+
import "./index-BKShPTwZ.js";
|
|
13
|
+
import { A as SearchSpanResult, B as ViewSpansResult, C as TRACE_ANALYSIS_LIMITS, D as DatasetOverview, E as DEFAULT_TRACE_ANALYST_BUDGETS, F as TraceAnalystFilters, H as ViewTraceResult, I as TraceAnalystSpan, L as TraceAnalystSpanKind, M as SpanMatchRecord, N as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, O as ErrorCluster, P as TraceAnalystByteBudgets, R as TraceAnalystSpanStatus, S as BoundedTraceAnalysisStoreOptions, T as TraceAnalysisStoreContext, V as ViewTraceOversized, j as SearchTraceResult, k as QueryTracesPage, w as TraceAnalysisStore, z as TraceAnalystTraceSummary } from "./types-vXyshMwx.js";
|
|
14
|
+
import { _ as TraceAnalysisToolDescriptor, g as TRACE_ANALYST_TOOL_NAMESPACE, h as BuildTraceAnalysisToolsOptions, v as buildTraceAnalysisToolDescriptors, y as traceAnalystFunctionGroup } from "./engine-DJqRKbhs.js";
|
|
15
15
|
//#region src/trace/otlp-attributes.d.ts
|
|
16
16
|
type ToolSpanOtlpInput = Pick<ToolSpan, 'toolName' | 'args' | 'argsCaptured' | 'result' | 'latencyMs'>;
|
|
17
17
|
type OtlpSpanRole = 'AGENT' | 'CHAIN' | 'EVALUATOR' | 'GUARDRAIL' | 'LLM' | 'SPAN' | 'TOOL' | 'UNKNOWN';
|
package/dist/traces.js
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
import { a as isRetrievalSpan, i as isLlmSpan, n as TRACE_SCHEMA_VERSION, o as isSandboxSpan, r as isJudgeSpan, s as isToolSpan, t as FAILURE_CLASSES } from "./schema-CRhEY1SO.js";
|
|
2
2
|
import { a as judgeSpans, c as runsForScenario, i as hasCapturedToolArgs, l as toolSpans, n as argHash, o as llmSpans, r as groupBy, s as runFailureClass, t as aggregateLlm } from "./query-Di7eEQ79.js";
|
|
3
|
-
import {
|
|
3
|
+
import { c as validateRunRecord, i as modelHasSnapshot } from "./run-record-D2lDdSAz.js";
|
|
4
4
|
import { i as redactValue, n as REDACTION_VERSION, r as redactString, t as DEFAULT_REDACTION_RULES } from "./redact-7Aq1ukl-.js";
|
|
5
5
|
import { n as llmSpanFromProvider, t as TraceEmitter } from "./emitter-CPBAhxum.js";
|
|
6
6
|
import { $ as classifyOtlpSpanRole, C as TraceNotFoundError, G as extractOtlpAttributes, J as projectOtlpFlatLine, K as firstStringAttr, Q as applyToolSpanOtlpAttributes, S as TraceFileTooLargeError, U as asString, W as compareSpanTime, X as spanEpochMillis, Y as readOtlpStatus, Z as stringField, _ as TraceAnalysisLimitError, b as TraceFileMalformedError, c as traceAnalystFunctionGroup, et as isOtlpModelCall, g as SpanNotFoundError, h as TRACE_ANALYSIS_LIMITS, l as createBoundedTraceAnalysisStore, m as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, o as TRACE_ANALYST_TOOL_NAMESPACE, p as DEFAULT_TRACE_ANALYST_BUDGETS, q as inferOtlpKind, s as buildTraceAnalysisToolDescriptors, tt as traceSpanKindToOpenInferenceKind, v as TraceAnalysisStoreContractError, x as TraceFileMissingError, y as TraceAnalysisValidationError } from "./kind-factory-CPmSd58s.js";
|
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import { S as PaidCallResult, T as RunPaidCallInput, a as CostChannel, c as CostLedgerHandle, f as CostLedgerSummary, p as CostProvenance } from "./cost-ledger-DbQdN3nO.js";
|
|
2
|
-
import { u as RunTokenUsage } from "./run-record-
|
|
2
|
+
import { u as RunTokenUsage } from "./run-record-DVV82Gwh.js";
|
|
3
3
|
import { E as LlmCallMetadata } from "./types-Cx3YUh2r.js";
|
|
4
|
-
import { _ as ProposalFinding } from "./types-
|
|
4
|
+
import { _ as ProposalFinding } from "./types-vXyshMwx.js";
|
|
5
5
|
//#region src/campaign/types.d.ts
|
|
6
6
|
/** Stable identifier + kind tag for any scenario. Consumers
|
|
7
7
|
* extend with their per-domain payload (persona, task, requirement, ...). */
|
|
@@ -626,4 +626,4 @@ interface CampaignResult<TArtifact = unknown, TScenario extends Scenario = Scena
|
|
|
626
626
|
}
|
|
627
627
|
//#endregion
|
|
628
628
|
export { LabeledScenarioWrite as A, ScoredSurfaceOutcome as B, JudgeDimension as C, LabeledScenarioSampleArgs as D, LabeledScenarioRecord as E, ProposeContext as F, labelTrustRank as G, SurfaceProposer as H, ProposedCandidate as I, RedactionStatus as L, OptimizerConfig as M, ParetoParent as N, LabeledScenarioSource as O, ProposalTrackContext as P, Scenario as R, JudgeConfig as S, LabelTrust as T, TraceSpan as U, SessionScript as V, isProposedCandidate as W, GateDecision as _, CampaignResult as a, GenerationRecord as b, CampaignTraceWriter as c, DispatchContext as d, DispatchFn as f, GateContribution as g, GateContext as h, CampaignCostMeter as i, MutableSurface as j, LabeledScenarioStore as k, CodeSurface as l, GateCheckStatus as m, CampaignArtifactWriter as n, CampaignScenarioIdentity as o, Gate as p, CampaignCellResult as r, CampaignTokenUsage as s, CampaignAggregates as t, ComponentSurface as u, GateResult as v, JudgeScore as w, JudgeAggregate as x, GenerationCandidate as y, ScenarioAggregate as z };
|
|
629
|
-
//# sourceMappingURL=types-
|
|
629
|
+
//# sourceMappingURL=types-BjsNDR49.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"types-
|
|
1
|
+
{"version":3,"file":"types-BjsNDR49.d.ts","names":[],"sources":["../src/campaign/types.ts"],"mappings":";;;;;;;UAgCiB;EACf;EACA;EACA;;;;;EAKA;;;UAIe,iCAAiC,KAAK;EACrD;;;;;UAMe;EACf;;EAEA;EACA;EACA;EACA;EACA,QAAQ;EACR,OAAO;EACP,WAAW;EACX,MAAM;;EAEN;;EAEA;;;;;;;;EAQA;;;;KAKU,WAAW,kBAAkB,UAAU,cACjD,UAAU,WACV,KAAK,oBACF,QAAQ;;;;UAOI,cAAc,WAAW;EACxC;EACA;EACA;;EAEA;;;EAGA,sBAAsB,UAAU,WAAW,sBAAsB,UAAU,cAAc;;UAK1E;;EAEf;;EAEA;;;;;;;;;UAUe,YAAY,WAAW,kBAAkB,WAAW;EACnE;EACA,YAAY;;;;EAIZ;;;EAGA,MAAM;IACJ,UAAU;IACV,UAAU;IACV,QAAQ;;IAER,aAAa;IACb;IACA,WAAW;MACT,aAAa,QAAQ;EACzB,aAAa,UAAU;;;;;;;;;;;UAYR;EACf,YAAY;EACZ;EACA;;EAEA,UAAU;;;;EAIV;;;EAGA;;EAEA;;EAEA,WAAW,eAAe;;;;;;;UAUX;WACN;;;WAGA;;WAEA;;WAEA;;WAEA;;WAEA;;WAEA;;;WAGA;aACE;aACA;aACA;;;WAGF;;;UAIM;WACN;WACA,YAAY,SAAS;;;;;;;;;;;KAYpB,0BAA0B,mBAAmB;;;;;;;UAQxC;EACf,SAAS;;EAET;;;;EAIA;;;;iBAKc,oBACd,OAAO,iBAAiB,oBACvB,SAAS;;;;;;;;;UAkBK;EACf,SAAS;EACT;;;EAGA,YAAY;;;EAGZ;;EAEA;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;;;;;UAMe;;;EAGf;;EAEA;EACA;EACA;EACA,YAAY;;;;;EAKZ,WAAW;IAAQ;IAAoB;IAAmB;IAAgB;;EAC1E;IACE;IACA;;;;;UAMa,eAAe,YAAY;WACjC,gBAAgB;WAChB,SAAS,cAAc;WACvB,UAAU,cAAc;;WAExB;WACA;WACA,QAAQ;;WAER,QAAQ;;;WAGR,kBAAkB;;;WAGlB,mBAAmB;;;;WAInB;;;;;;;WAOA,gBAAgB,cAAc;;WAE9B,aAAa;WACb;;;;;;;;;;;;;;UAeM,gBAAgB,YAAY;EAC3C;;;;;EAKA,QAAQ,KAAK,eAAe,aAAa,QAAQ,MAAM,iBAAiB;;;EAGxE,QAAQ;IAAQ,SAAS,cAAc;;IAAwB;IAAe;;;UAG/D;EACf;EACA;EACA,mBAAmB,qBAAqB;;UAGzB,wBAAwB;EACvC,UAAU;;;KAMA;;KAGA;UAEK;EACf;EACA,QAAQ;EACR;;UAGe,YAAY,WAAW,kBAAkB;EACxD,oBAAoB,YAAY;EAChC,oBAAoB,YAAY;;EAEhC,aAAa,YAAY,eAAe;;;;;EAKxC,sBAAsB,YAAY,eAAe;;;;;;;EAOjD,yBAAyB,YAAY,eAAe;;;EAGpD,uBAAuB,YAAY;EACnC,WAAW;EACX;IAAQ;IAAmB;;;EAE3B,aAAa;EACb;EACA,QAAQ;;UAGO;EACf,UAAU;EACV;EACA,mBAAmB;EACnB;;;UAIe,KAAK,qBAAqB,kBAAkB,WAAW;EACtE;EACA,OAAO,KAAK,YAAY,WAAW,aAAa,QAAQ;;;;UAOzC;EACf,KAAK,cAAc,aAAa,0BAA0B;EAC1D,SAAS;;UAGM;EACf,IAAI,aAAa;EACjB,aAAa,aAAa;;;;UAKX;EACf,MAAM,cAAc,kBAAkB,aAAa;EACnD,UAAU,cAAc,iBAAiB;;;;;;KAO/B,qBAAqB;;;;;UAMhB;;EAEf,YAAY,GACV,OAAO,KAAK,iBAAiB;IAC3B,UAAU;MAEX,QAAQ,eAAe;;;;;KAQhB;KAOA;;;;;;;;;;;;;;;KAgBA;;iBASI,eAAe,OAAO;;;;UAOrB,qBAAqB,kBAAkB,WAAW,UAAU;EAC3E,UAAU;EACV,UAAU;EACV,aAAa,eAAe;EAC5B,QAAQ;EACR;EACA;EACA,iBAAiB;;;;;EAKjB,aAAa;;EAEb;;UAGe,sBAAsB,kBAAkB,WAAW,UAAU,6BACpE,qBAAqB,WAAW;;EAExC;;;EAGA;;UAGe;EACf;;EAEA;;;;EAIA;EACA;IACE;IACA,SAAS,wBAAwB;IACjC;IACA;;;;;IAKA,WAAW;;;UAIE;EACf,QAAQ,OAAO,uBAAuB;EACtC,OAAO,MAAM,4BAA4B,QAAQ;EACjD,QAAQ;IACN;IACA;IACA,UAAU;;;IAGV,SAAS,OAAO;;;UAMH,mBAAmB;;;EAGlC;EACA;EACA;EACA;EACA;EACA,UAAU;EACV,aAAa,eAAe;;;EAG5B;;EAEA,gBAAgB;;EAEhB;;;EAGA,YAAY;;;EAGZ;;;EAGA;EACA;EACA;EACA;;EAEA;;EAEA;EACA;;UAGe;EACf;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;;UAGe;EACf;EACA,YAAY;EACZ;;;;;;UAOe;EACf;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA;;;;EAIA;;;;EAIA;IACE;IACA;IACA,iBAAiB;MAAQ;MAAgB;;;;;EAI3C,YAAY;;;;;;;;;;EAUZ,WAAW;IAAQ;IAAoB;IAAmB;IAAgB;;;;EAG1E;;;;EAIA;;UAGe;EACf,SAAS,eAAe;EACxB,YAAY,eAAe;;EAE3B,MAAM;;EAEN;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe,eAAe,qBAAqB,kBAAkB,WAAW;;EAEhF;;EAEA;EACA;;EAEA;EACA;EACA;EACA;EACA,OAAO,MAAM,mBAAmB;EAChC,YAAY;EACZ;IACE,aAAa;IACb;;EAEF,OAAO;EACP;EACA;EACA,iBAAiB;;;EAGjB,WAAW,MAAM,2BAA2B,KAAK"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { s as RunSplitTag } from "./run-record-
|
|
2
|
-
import { R as Scenario, d as DispatchContext } from "./types-
|
|
1
|
+
import { s as RunSplitTag } from "./run-record-DVV82Gwh.js";
|
|
2
|
+
import { R as Scenario, d as DispatchContext } from "./types-BjsNDR49.js";
|
|
3
3
|
//#region src/benchmarks/types.d.ts
|
|
4
4
|
type BenchmarkTaskKind = 'retrieval' | 'rag-answer' | 'hallucination' | 'kb-improvement' | 'routing' | 'custom';
|
|
5
5
|
type BenchmarkFamily = 'beir' | 'mteb-retrieval' | 'msmarco' | 'trec-dl' | 'miracl' | 'lotte' | 'bright' | 'crag' | 'hotpotqa' | 'kilt' | 'ragtruth' | 'faithbench' | 'first-party' | 'custom';
|
|
@@ -90,4 +90,4 @@ declare const BENCHMARK_SPLIT_SEED = "agent-eval-v1";
|
|
|
90
90
|
declare function deterministicSplit(itemId: string, seed?: string): RunSplitTag;
|
|
91
91
|
//#endregion
|
|
92
92
|
export { BenchmarkFamily as a, BenchmarkSource as c, BenchmarkEvaluation as i, BenchmarkTaskKind as l, BenchmarkAdapter as n, BenchmarkResponder as o, BenchmarkDatasetItem as r, BenchmarkScenario as s, BENCHMARK_SPLIT_SEED as t, deterministicSplit as u };
|
|
93
|
-
//# sourceMappingURL=types-
|
|
93
|
+
//# sourceMappingURL=types-C4bSVIr7.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"types-
|
|
1
|
+
{"version":3,"file":"types-C4bSVIr7.d.ts","names":[],"sources":["../src/benchmarks/types.ts"],"mappings":";;;KASY;KAQA;UAgBK,qBAAqB;;;EAGpC;;EAEA,SAAS;;EAET,QAAQ;;EAER,SAAS;;EAET,WAAW;;EAEX;;EAEA,SAAS;EACT,WAAW;;UAGI;;;;EAIf;;EAEA;;EAEA,aAAa;;;EAGb,KAAK;EACL;;UAGe;EACf;EACA;EACA;EACA;EACA;;;UAQe,iBAAiB,kBAAkB,oBAAoB;;EAEtE;EACA,SAAS;EACT,WAAW;EACX;EACA,SAAS;EACT;;;;;EAKA,YAAY,OAAO,cAAc,QAAQ,qBAAqB;;EAE9D,SAAS,MAAM,qBAAqB,WAAW,UAAU,YAAY,QAAQ;;;;EAI7E,YAAY,iBAAiB;;UAGd,kBAAkB,4BAA4B;EAC7D;EACA;EACA,QAAQ;EACR,UAAU;EACV,UAAU;EACV,MAAM,qBAAqB;;KAGjB,mBAAmB,oBAAoB,uBAAuB;EACxE,UAAU,kBAAkB;EAC5B,MAAM,qBAAqB;EAC3B,SAAS;MACL,QAAQ,aAAa;;;cAoBd;;;;;;;;;iBAUG,mBACd,gBACA,gBACC"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import { c as CostLedgerHandle } from "./cost-ledger-DbQdN3nO.js";
|
|
2
|
-
import { a as RunRecord, n as RunCostProvenance, u as RunTokenUsage } from "./run-record-
|
|
2
|
+
import { a as RunRecord, n as RunCostProvenance, u as RunTokenUsage } from "./run-record-DVV82Gwh.js";
|
|
3
3
|
import { a as JudgeInput, p as ChatClient } from "./types-Cx3YUh2r.js";
|
|
4
4
|
import { RE2JS } from "re2js";
|
|
5
5
|
//#region src/trace-analyst/types.d.ts
|
|
@@ -594,4 +594,4 @@ type AnalystRunEvent = {
|
|
|
594
594
|
};
|
|
595
595
|
//#endregion
|
|
596
596
|
export { SearchSpanResult as A, ViewSpansResult as B, TRACE_ANALYSIS_LIMITS as C, DatasetOverview as D, DEFAULT_TRACE_ANALYST_BUDGETS as E, TraceAnalystFilters as F, ViewTraceResult as H, TraceAnalystSpan as I, TraceAnalystSpanKind as L, SpanMatchRecord as M, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX as N, ErrorCluster as O, TraceAnalystByteBudgets as P, TraceAnalystSpanStatus as R, BoundedTraceAnalysisStoreOptions as S, TraceAnalysisStoreContext as T, ViewTraceOversized as V, ProposalFinding as _, AnalystInputKind as a, makeFinding as b, AnalystRunInputs as c, AnalystSeverity as d, AnalystUsageReceipt as f, ExecutionProbeRequest as g, ExecutionProbeOutcome as h, AnalystFinding as i, SearchTraceResult as j, QueryTracesPage as k, AnalystRunResult as l, ExecutionProbe as m, AnalystContext as n, AnalystRequirements as o, EvidenceRef as p, AnalystCost as r, AnalystRunEvent as s, Analyst as t, AnalystRunSummary as u, ProposalFindingOrigin as v, TraceAnalysisStore as w, makeProposalFinding as x, computeFindingId as y, TraceAnalystTraceSummary as z };
|
|
597
|
-
//# sourceMappingURL=types-
|
|
597
|
+
//# sourceMappingURL=types-vXyshMwx.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"types-
|
|
1
|
+
{"version":3,"file":"types-vXyshMwx.d.ts","names":[],"sources":["../src/trace-analyst/types.ts","../src/trace-analyst/store-contract.ts","../src/analyst/types.ts"],"mappings":";;;;;;;;;;;;;;;;;;;;KAgBY;KAUA;;;;UAKK;EACf;EACA;EACA;EACA;EACA,MAAM;EACN;EACA;EACA;EACA,QAAQ;EACR;EACA;EACA;EACA;EACA;;;EAGA,YAAY;;UAGG;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA;;;;;;;UAQe;;EAEf;;EAEA;;EAEA;;EAEA;EACA;EACA;;EAEA;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;;EAEA;EACA;IAAU;IAAqB;;;;;;EAK/B,gBAAgB;EAChB;IAAc;IAAkB;;;UAGjB;EACf,QAAQ;EACR;EACA;;;;;UAMe;EACf;EACA,QAAQ;EACR,YAAY;;UAGG;EACf;;EAEA,gBAAgB;;EAEhB;EACA;;UAGe;EACf;EACA,OAAO;;EAEP;;EAEA;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA,WAAW;;;EAGX;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA,MAAM;EACN;;UAGe;EACf;EACA;EACA,MAAM;EACN;;;UAIe;;;EAGf;;;EAGA;;;EAGA;;;EAGA;;cAGW,+BAA+B;;;cAS/B;;;cC7MA;WACX;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;UAGe;EACf,SAAS;;;;;;;;;UAUM;EACf,SAAS,kBAAkB,UAAU,4BAA4B;EAEjE,SACE;IAAS;IAAkB;KAC3B,UAAU,4BACT;EAEH,YACE,UAAU,qBACV,UAAU,4BACT,QAAQ;EAEX,YACE;IAAS,UAAU;IAAqB;IAAe;KACvD,UAAU,4BACT,QAAQ;EAEX,YAAY,UAAU,qBAAqB,UAAU,4BAA4B;EAEjF,UACE;IACE;IACA;KAEF,UAAU,4BACT,QAAQ;EAEX,UACE;IACE;IACA;IACA;KAEF,UAAU,4BACT,QAAQ;EAEX,YACE;IACE;IACA;IACA;KAEF,UAAU,4BACT,QAAQ;EAEX,WACE;IACE;IACA;IACA;IACA;KAEF,UAAU,4BACT,QAAQ;;UAGI;EACf,UAAU,QAAQ;;;;;;;;UC/DH;EACf;;;;;;;EAOA;EACA;EACA;EACA,UAAU;;;;;;;EAOV;EACA;EACA;EACA,eAAe;EACf;EACA;;EAEA;;;;;;EAMA;;;;EAIA;;EAEA,WAAW;;KAGD;;KAGA;;KAGA,kBAAkB;WACnB,iBAAiB;;UAGX;;;;;;;EAOf;EACA;EACA;;;;;;;;KAWU;UAOK;;EAEf;;EAEA;;EAEA;;EAEA;;UAGe;;EAEf;;EAEA;;;;;;UAOe;EACf,aAAa;EACb;EACA,YAAY;EACZ,aAAa;;EAEb,SAAS;;UAGM;EACf;;EAEA;;EAEA;;EAEA;;EAEA,aAAa;;EAEb;;;;;;;EAOA,OAAO;;;;;;;;;;EAUP,gBAAgB,cAAc;;;;;;;EAO9B,mBAAmB,cAAc;;;;;EAKjC,eAAe,SAAS;;EAExB,OAAO;;EAEP,OAAO,aAAa,SAAS;;EAE7B,SAAS;;;;;;;;EAQT,QAAQ;;;UAMO;EACf;;EAEA;;EAEA;;EAEA;EACA,SAAS;;;;;;;KAQC;EAEN;EACA;EACA;EACA;EACA;;EAEA;;EAEA;EAAkB;IAAS;IAAe;;;;;;;;UAO/B;;WAEN;EACT,QAAQ,SAAS,wBAAwB,QAAQ;;;;;;;;UASlC,QAAQ;;WAEd;;WAEA;WACA,WAAW;WACX,MAAM;WACN,WAAW;;WAEX;EACT,QAAQ,OAAO,QAAQ,KAAK,iBAAiB,QAAQ;;;UAItC;;EAEf;;EAEA,QAAQ;;EAER,MAAM;;EAEN;;;;;;;;EAQA;IAAkB;IAAsB;;;;;;;;EAOxC;;;;;;;;;;iBAac,iBAAiB;EAC/B;EACA;EACA;EACA;;EAEA;;;;;;iBAyBc,YACd,MAAM,KAAK;EACT;EACA;IAED;;iBAiBa,oBACd,MAAM,KAAK;EACT;EACA;IAED;UAOc;EACf;EACA;;EAEA;EACA;EACA;;EAEA,OAAO;;EAEP;IAAU;IAAe;;;UAGV;EACf;EACA;EACA;EACA;EACA,UAAU;EACV,aAAa;;EAEb;;;;;EAKA,wBAAwB;;;;;;;;;;;;;;;KAkBd;EAEN;EACA;EACA;EACA;;EAEA,aAAa;;EAGb;EACA,SAAS;;EAGT;EACA;EACA;;EAGA;;EAEA,SAAS;EACT,UAAU,cAAc;;EAGxB;EACA,QAAQ"}
|
package/dist/wire/index.d.ts
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import { c as CostLedgerHandle } from "../cost-ledger-DbQdN3nO.js";
|
|
2
2
|
import { A as LlmClientOptions, N as LlmRouteRequirements } from "../types-Cx3YUh2r.js";
|
|
3
3
|
import { s as TraceStore } from "../store-CT9YIIve.js";
|
|
4
|
-
import { m as FeedbackTrajectoryStore } from "../feedback-trajectory-
|
|
4
|
+
import { m as FeedbackTrajectoryStore } from "../feedback-trajectory-D9vYSob_.js";
|
|
5
5
|
import { z } from "zod";
|
|
6
6
|
import { ServerType } from "@hono/node-server";
|
|
7
7
|
import { Hono } from "hono";
|
package/docs/eval-surface-map.md
CHANGED
|
@@ -19,7 +19,8 @@ composition point already exists (see Produced-state grading below).**
|
|
|
19
19
|
When variants of the same task run inside one `runCampaign`, give those scenarios the same `seedGroup` so each repetition uses common randomness.
|
|
20
20
|
Use `runProfileMatrix` instead when profiles are separate campaign axes.
|
|
21
21
|
Set `maxConcurrency` for cases within one profile and `maxProfileConcurrency` for independent profile columns; results retain caller order regardless of completion order.
|
|
22
|
-
Every paid-call receipt must match the profile, and a moving alias must resolve to one snapshot across the entire profile column.
|
|
22
|
+
Every paid-call receipt must match the profile, and a successful moving alias must resolve to one snapshot across the entire profile column.
|
|
23
|
+
Failed cells with no served model remain durable with explicit unknown model, cost, and usage fields.
|
|
23
24
|
The caller commit and profile config are always part of cache identity; set `dispatchRef` when execution behavior can change without a new commit.
|
|
24
25
|
A failed profile cancels active sibling columns before the matrix rejects.
|
|
25
26
|
|
package/docs/wire-protocol.md
CHANGED
|
@@ -18,6 +18,10 @@ your code (any language)
|
|
|
18
18
|
|
|
19
19
|
Both transports talk to identical handlers. If you need a sustained connection (live agent paths, high-frequency calls), use HTTP. If you need a one-shot (cron, CI, batch), use stdio RPC. The wire shape is the same.
|
|
20
20
|
|
|
21
|
+
This versioned `/v1` protocol is separate from the source-only distributed `Dispatch` adapter.
|
|
22
|
+
The adapter's `/dispatch` endpoint is documented in [`docs/distributed-driver.md`](./distributed-driver.md).
|
|
23
|
+
Its request body includes the required `runAttemptId`, which the worker must preserve unchanged when it builds its `DispatchContext`.
|
|
24
|
+
|
|
21
25
|
## Two transports, one contract
|
|
22
26
|
|
|
23
27
|
| | HTTP | stdio RPC |
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@tangle-network/agent-eval",
|
|
3
|
-
"version": "0.145.
|
|
3
|
+
"version": "0.145.8",
|
|
4
4
|
"description": "Evaluate and improve AI agents from runs, traces, judges, and feedback. Compare candidates, cluster failures, measure lift, and gate releases.",
|
|
5
5
|
"homepage": "https://github.com/tangle-network/agent-eval#readme",
|
|
6
6
|
"repository": {
|