@tangle-network/agent-eval 0.131.1 → 0.133.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +14 -0
- package/dist/analyst/index.d.ts +5 -5
- package/dist/analyst/index.js +3 -3
- package/dist/{analyze-runs-CBnCgfse.d.ts → analyze-runs-DZr7JW-m.d.ts} +5 -5
- package/dist/{analyze-runs-CBnCgfse.d.ts.map → analyze-runs-DZr7JW-m.d.ts.map} +1 -1
- package/dist/benchmarks/index.d.ts +1 -1
- package/dist/benchmarks/index.js +1 -1
- package/dist/{benchmarks-BDtRiOvE.js → benchmarks-DncrVrSr.js} +3 -3
- package/dist/{benchmarks-BDtRiOvE.js.map → benchmarks-DncrVrSr.js.map} +1 -1
- package/dist/campaign/index.d.ts +4 -4
- package/dist/campaign/index.js +2 -2
- package/dist/{campaign-CLG6y9jF.js → campaign-CyQMCCEU.js} +4 -4
- package/dist/{campaign-CLG6y9jF.js.map → campaign-CyQMCCEU.js.map} +1 -1
- package/dist/cli.js +1 -1
- package/dist/{client-w90OcvvR.d.ts → client-D4F9hdzR.d.ts} +4 -4
- package/dist/{client-w90OcvvR.d.ts.map → client-D4F9hdzR.d.ts.map} +1 -1
- package/dist/contract/index.d.ts +188 -22
- package/dist/contract/index.d.ts.map +1 -1
- package/dist/contract/index.js +1663 -65
- package/dist/contract/index.js.map +1 -1
- package/dist/control.d.ts +1 -1
- package/dist/{cost-ledger-DIgQUFZZ.js → cost-ledger-BrJxbrMy.js} +278 -9
- package/dist/cost-ledger-BrJxbrMy.js.map +1 -0
- package/dist/{cost-ledger-Dye6jCgg.d.ts → cost-ledger-fGS_u_O1.d.ts} +18 -2
- package/dist/cost-ledger-fGS_u_O1.d.ts.map +1 -0
- package/dist/{default-registry-Cj1oUpLN.d.ts → default-registry-Cl3pHo4n.d.ts} +4 -4
- package/dist/{default-registry-Cj1oUpLN.d.ts.map → default-registry-Cl3pHo4n.d.ts.map} +1 -1
- package/dist/{default-registry-C-vFCSEc.js → default-registry-D3T9XbuY.js} +4 -13
- package/dist/default-registry-D3T9XbuY.js.map +1 -0
- package/dist/{eval-campaign-C2k-m4aY.js → eval-campaign-CYkUME2T.js} +2 -2
- package/dist/{eval-campaign-C2k-m4aY.js.map → eval-campaign-CYkUME2T.js.map} +1 -1
- package/dist/{task-failure-attributes-CQZlB3et.js → extract-usage-2j25whHw.js} +154 -2
- package/dist/extract-usage-2j25whHw.js.map +1 -0
- package/dist/fuzz.d.ts +1 -1
- package/dist/fuzz.js +1 -1
- package/dist/hosted/index.d.ts +2 -2
- package/dist/{index-B11XCkdf2.d.ts → index-3cdlURSk.d.ts} +2 -2
- package/dist/index-3cdlURSk.d.ts.map +1 -0
- package/dist/{index-VTypFU3t.d.ts → index-B_eUhU9K.d.ts} +7 -7
- package/dist/{index-VTypFU3t.d.ts.map → index-B_eUhU9K.d.ts.map} +1 -1
- package/dist/{index-DNXoNaFF.d.ts → index-C2fkZhv_.d.ts} +2 -2
- package/dist/{index-DNXoNaFF.d.ts.map → index-C2fkZhv_.d.ts.map} +1 -1
- package/dist/{index-NPeSWD98.d.ts → index-CiP2DRCj.d.ts} +5 -5
- package/dist/{index-NPeSWD98.d.ts.map → index-CiP2DRCj.d.ts.map} +1 -1
- package/dist/index.d.ts +69 -23
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +100 -12
- package/dist/index.js.map +1 -1
- package/dist/{llm-client-B_nIBlYo.d.ts → llm-client-BiK4HW0u.d.ts} +2 -2
- package/dist/{llm-client-B_nIBlYo.d.ts.map → llm-client-BiK4HW0u.d.ts.map} +1 -1
- package/dist/{llm-client--GR4JbZE.js → llm-client-ClPW-dWB.js} +2 -2
- package/dist/{llm-client--GR4JbZE.js.map → llm-client-ClPW-dWB.js.map} +1 -1
- package/dist/meta-eval/index.d.ts +216 -3
- package/dist/{index-D_F6VAKe.d.ts.map → meta-eval/index.d.ts.map} +1 -1
- package/dist/meta-eval/index.js +91 -1
- package/dist/meta-eval/index.js.map +1 -1
- package/dist/multishot/index.d.ts +1 -1
- package/dist/openapi.json +1 -1
- package/dist/{release-report-Crg9oFJ0.d.ts → release-report-DfmKSIEE.d.ts} +3 -3
- package/dist/{release-report-Crg9oFJ0.d.ts.map → release-report-DfmKSIEE.d.ts.map} +1 -1
- package/dist/{replay-RE97Ckjl.d.ts → replay-BI6CVKkp.d.ts} +2 -2
- package/dist/{replay-RE97Ckjl.d.ts.map → replay-BI6CVKkp.d.ts.map} +1 -1
- package/dist/{replay-D18-pBAA.js → replay-BMR4TEYY.js} +2 -3
- package/dist/{replay-D18-pBAA.js.map → replay-BMR4TEYY.js.map} +1 -1
- package/dist/reporting.d.ts +4 -4
- package/dist/{researcher-Q5rpPqZY.d.ts → researcher-DMimgHtN.d.ts} +4 -4
- package/dist/{researcher-Q5rpPqZY.d.ts.map → researcher-DMimgHtN.d.ts.map} +1 -1
- package/dist/{reward-hacking-CW-3HN0n.d.ts → reward-hacking-D-QqXvg-.d.ts} +2 -2
- package/dist/{reward-hacking-CW-3HN0n.d.ts.map → reward-hacking-D-QqXvg-.d.ts.map} +1 -1
- package/dist/rl.d.ts +156 -6
- package/dist/rl.d.ts.map +1 -1
- package/dist/rl.js +217 -2
- package/dist/rl.js.map +1 -1
- package/dist/rollout/index.d.ts +1 -1
- package/dist/{rubric-predictive-validity-lXLmashy.d.ts → rubric-predictive-validity-C1dCLcvb.d.ts} +2 -2
- package/dist/{rubric-predictive-validity-lXLmashy.d.ts.map → rubric-predictive-validity-C1dCLcvb.d.ts.map} +1 -1
- package/dist/{run-evidence-oByzm-dE.d.ts → run-evidence-DokQtX0-.d.ts} +2 -2
- package/dist/{run-evidence-oByzm-dE.d.ts.map → run-evidence-DokQtX0-.d.ts.map} +1 -1
- package/dist/run-record-CN8Zd21B.js.map +1 -1
- package/dist/{run-record-BJnYdTxO.d.ts → run-record-DcObtIGh.d.ts} +4 -14
- package/dist/run-record-DcObtIGh.d.ts.map +1 -0
- package/dist/{semantic-concept-judge-b5m3irbR.js → semantic-concept-judge-BypLt6Fw.js} +4 -4
- package/dist/{semantic-concept-judge-b5m3irbR.js.map → semantic-concept-judge-BypLt6Fw.js.map} +1 -1
- package/dist/{server-m5D9cvnG.js → server-BPqlDBWK.js} +3 -3
- package/dist/{server-m5D9cvnG.js.map → server-BPqlDBWK.js.map} +1 -1
- package/dist/{skill-usage-D5mlWdAJ.d.ts → skill-usage-BaaxFSJR.d.ts} +5 -5
- package/dist/{skill-usage-D5mlWdAJ.d.ts.map → skill-usage-BaaxFSJR.d.ts.map} +1 -1
- package/dist/{skillopt-optimization-method-D0MVjJdP.d.ts → skillopt-optimization-method-CAASpcS3.d.ts} +9 -9
- package/dist/{skillopt-optimization-method-D0MVjJdP.d.ts.map → skillopt-optimization-method-CAASpcS3.d.ts.map} +1 -1
- package/dist/{skillopt-optimization-method-eOJL2570.js → skillopt-optimization-method-DgN7U9iR.js} +3 -3
- package/dist/{skillopt-optimization-method-eOJL2570.js.map → skillopt-optimization-method-DgN7U9iR.js.map} +1 -1
- package/dist/{statistics-Cmj6nynr.d.ts → statistics-DbvkkDPa.d.ts} +2 -2
- package/dist/{statistics-Cmj6nynr.d.ts.map → statistics-DbvkkDPa.d.ts.map} +1 -1
- package/dist/{summary-report-CWwB_LiV.d.ts → summary-report-DnUcjVpV.d.ts} +2 -2
- package/dist/{summary-report-CWwB_LiV.d.ts.map → summary-report-DnUcjVpV.d.ts.map} +1 -1
- package/dist/traces.d.ts +2 -2
- package/dist/traces.js +2 -2
- package/dist/{types-CsD5nTfV.d.ts → types-BokuXvOG.d.ts} +4 -4
- package/dist/{types-CsD5nTfV.d.ts.map → types-BokuXvOG.d.ts.map} +1 -1
- package/dist/{types-DGsxbAEd.d.ts → types-Cc3qbqzj.d.ts} +3 -3
- package/dist/{types-DGsxbAEd.d.ts.map → types-Cc3qbqzj.d.ts.map} +1 -1
- package/dist/wire/index.d.ts +2 -2
- package/dist/wire/index.js +1 -1
- package/package.json +3 -8
- package/dist/belief-state/index.d.ts +0 -622
- package/dist/belief-state/index.d.ts.map +0 -1
- package/dist/belief-state/index.js +0 -1819
- package/dist/belief-state/index.js.map +0 -1
- package/dist/calibration-CNWWA6K8.js +0 -94
- package/dist/calibration-CNWWA6K8.js.map +0 -1
- package/dist/code-agent-session-BjkMTQ7H.js +0 -1390
- package/dist/code-agent-session-BjkMTQ7H.js.map +0 -1
- package/dist/code-agent-session-aYa3SGKz.d.ts +0 -143
- package/dist/code-agent-session-aYa3SGKz.d.ts.map +0 -1
- package/dist/cost-ledger-DIgQUFZZ.js.map +0 -1
- package/dist/cost-ledger-Dye6jCgg.d.ts.map +0 -1
- package/dist/default-registry-C-vFCSEc.js.map +0 -1
- package/dist/extract-usage-BrQ8mCLX.js +0 -155
- package/dist/extract-usage-BrQ8mCLX.js.map +0 -1
- package/dist/index-B11XCkdf2.d.ts.map +0 -1
- package/dist/index-D_F6VAKe.d.ts +0 -217
- package/dist/metrics-C9YY1OcL.js +0 -239
- package/dist/metrics-C9YY1OcL.js.map +0 -1
- package/dist/off-policy-DvgzvtIx.js +0 -220
- package/dist/off-policy-DvgzvtIx.js.map +0 -1
- package/dist/off-policy-mskQw8Mb.d.ts +0 -153
- package/dist/off-policy-mskQw8Mb.d.ts.map +0 -1
- package/dist/run-record-BJnYdTxO.d.ts.map +0 -1
- package/dist/runtime-trajectory-1gyaTOoC.js +0 -93
- package/dist/runtime-trajectory-1gyaTOoC.js.map +0 -1
- package/dist/runtime-trajectory-BXxG4lyi.d.ts +0 -50
- package/dist/runtime-trajectory-BXxG4lyi.d.ts.map +0 -1
- package/dist/task-failure-attributes-CQZlB3et.js.map +0 -1
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { g as JudgeScore } from "./types-
|
|
1
|
+
import { g as JudgeScore } from "./types-Cc3qbqzj.js";
|
|
2
2
|
import { i as ContinuousAgreementOptions, r as ContinuousAgreement } from "./judge-calibration-DFtEMlde.js";
|
|
3
3
|
//#region src/statistics.d.ts
|
|
4
4
|
/** Identity: dimensions already follow "higher = better" by prompt convention
|
|
@@ -511,4 +511,4 @@ declare function eProcess(opts?: EProcessOptions): EProcess;
|
|
|
511
511
|
declare function mulberry32(seed: number): () => number;
|
|
512
512
|
//#endregion
|
|
513
513
|
export { mannWhitneyU as A, pairedSignTest as B, confidenceInterval as C, holm as D, eProcess as E, normalizeScores as F, ranks as G, partialCredit as H, pairedBootstrap as I, spearmanR as J, requiredPairedSampleSize as K, pairedCohensDz as L, mcnemarPower as M, mcnemarRequiredN as N, interRaterReliability as O, mulberry32 as P, wilson as Q, pairedMde as R, cohensD as S, corpusInterRaterAgreementFromJudgeScores as T, passAtK as U, pairedTTest as V, pearsonR as W, weightedMean as X, weightedComposite as Y, wilcoxonSignedRank as Z, WeightedCompositeInput as _, CorpusScoreRecord as a, bonferroni as b, EProcessState as c, PairedBootstrapOptions as d, PairedBootstrapResult as f, SignTestAlternative as g, RiskDifferenceResult as h, CorpusAgreementReport as i, mcnemar as j, interpretCliffs as k, EProcessStep as l, ProportionInterval as m, CorpusAgreementOptions as n, EProcess as o, PairedSignTestResult as p, requiredSampleSize as q, CorpusAgreementPerDimension as r, EProcessOptions as s, CliffsMagnitude as t, McNemarResult as u, WeightedCompositeResult as v, corpusInterRaterAgreement as w, cliffsDelta as x, benjaminiHochberg as y, pairedRiskDifference as z };
|
|
514
|
-
//# sourceMappingURL=statistics-
|
|
514
|
+
//# sourceMappingURL=statistics-DbvkkDPa.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"statistics-
|
|
1
|
+
{"version":3,"file":"statistics-DbvkkDPa.d.ts","names":[],"sources":["../src/statistics.ts"],"mappings":";;;;;cAUa,kBAAe,QAAY,iBAAe;;iBAGvC,aAAa;EAAU;EAAe;;;iBAatC,mBACd,kBACA,qBACA;EAAQ;EAAe;;EACpB;EAAc;EAAe;;;;;;;;iBAsClB,sBAAsB,aAAa;;;;;iBAwDnC,aAAa,aAAa;EAAgB;EAAW;;;iBA+CrD,cAAc,iBAAiB;;;;;;;iBAW/B,YACd,kBACA;EACG;EAAW;EAAY;;;;;;iBAyBZ,mBAAmB,kBAAkB;EAAoB;EAAW;;;;;;;iBAqCpE,QAAQ,aAAa;;;;;;;;;iBAqBrB,eAAe,kBAAkB;KAoBrC;;;;;;;;;;;iBAYI,YAAY,kBAAkB;;;;;;iBAkB9B,gBAAgB,gBAAgB;;;;;iBAuBhC,MAAM;;;;;;iBAmBN,SAAS,aAAa;;;;;iBAuBtB,UAAU,aAAa;UAKtB;;EAEf,MAAM;;;;;EAKN,SAAS;;EAET;;UAGe;EACf;EACA;;;;;;;;;;;iBAYc,kBAAkB,OAAO,yBAAyB;UA4CjD;;EAEf;;EAEA;;EAEA;;EAEA;;UAGe,oCAAoC;EACnD;;EAEA;;EAEA;;UAGe;;EAEf,cAAc;;EAEd;;EAEA;;EAEA;;EAEA;;UAGe,+BAA+B;;;;;;EAM9C;;;;;;EAMA;;;;;;;;;;;;;;;;;;iBAmBc,0BACd,SAAS,qBACT,OAAM,yBACL;;;;;;;;;iBA0Ha,yCACd,aAAa;EAAQ;EAAgB,QAAQ;IAC7C,OAAM,yBACL;;;;;;;iBA+Ga,mBAAmB;EACjC;EACA;EACA;EACA;;;;;;iBAiBc,yBAAyB;EACvC;EACA;EACA;EACA;;;;;;;;iBAkBc,UAAU;EACxB;EACA;EACA;EACA;;;;;;;;;;;;;;;;iBAyBc,iBAAiB;EAC/B;EACA;EACA;EACA;EACA;;;;;;;iBAyBc,aAAa;EAC3B;EACA;EACA;EACA;EACA;;;iBAkBc,WACd,mBACA;EACG;EAAoB;;;;;;;;;;iBAeT,KACd,4BACA;EACG;EAAoB;;;;;;iBA+BT,kBACd,mBACA;EACG;EAAmB;;UAoBP;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;;;;;;;iBASc,gBACd,kBACA,iBACA,OAAM,yBACL;;KAwDS;;UAGK;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,aAAa;;EAEb;;;;;;;;;;;;;;;iBAgBc,eACd,gCACA,aAAa,sBACZ;;UAgDc;;EAEf;;EAEA;;EAEA;;;;;;;;;iBAUc,OAAO,mBAAmB,WAAW,sBAAoB;;UAmBxD;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;;;;iBAec,QACd,SAAS,6BACT,WAAW,8BACV;;UAmBc;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;iBAWc,qBACd,SAAS,6BACT,WAAW,6BACX,sBACC;;;;;;;;;;;iBAyCa,QAAQ,WAAW,WAAW;UAgE7B;;;EAGf;;;;EAIA;;;;EAIA;;UAGe;;EAEf;;EAEA;;EAEA;;UAGe,sBAAsB;EACrC;EACA;EACA;;EAEA;;EAEA;;UAGe;;;EAGf,OAAO,YAAY;EACnB,SAAS;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA8BK,SAAS,OAAM,kBAAuB;;;;;iBAsHtC,WAAW"}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { a as RunRecord } from "./run-record-
|
|
1
|
+
import { a as RunRecord } from "./run-record-DcObtIGh.js";
|
|
2
2
|
import { n as FailureClusterReport } from "./failure-cluster-CqcvCcdR.js";
|
|
3
3
|
//#region src/held-out-gate.d.ts
|
|
4
4
|
type HeldOutGateRejectionCode = 'few_runs' | 'missing_split_scores' | 'missing_cost' | 'negative_delta' | 'overfit_gap' | 'cost_ceiling';
|
|
@@ -404,4 +404,4 @@ interface ResearchReport {
|
|
|
404
404
|
declare function researchReport(runs: RunRecord[], opts?: ResearchReportOptions): Promise<ResearchReport>;
|
|
405
405
|
//#endregion
|
|
406
406
|
export { HeldOutGateConfig as C, HeldOutGate as S, 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 };
|
|
407
|
-
//# sourceMappingURL=summary-report-
|
|
407
|
+
//# sourceMappingURL=summary-report-DnUcjVpV.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"summary-report-
|
|
1
|
+
{"version":3,"file":"summary-report-DnUcjVpV.d.ts","names":[],"sources":["../src/held-out-gate.ts","../src/summary-report.ts"],"mappings":";;;KAuCY;UAQK;;;EAGf;;;EAGA;;;;EAIA;;;EAGA;;EAEA;;EAEA;;;EAGA;;;;;;;;;;;;;EAaA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;IAAY;IAAa;;;EAEzB;;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;EAEjB,YAAY,QAAQ;;;;EAuBpB,SAAS,WAAW,aAAa,UAAU,cAAc;;;;UCpI1C;;;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"}
|
package/dist/traces.d.ts
CHANGED
|
@@ -1,10 +1,10 @@
|
|
|
1
1
|
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";
|
|
2
|
-
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 "./index-
|
|
2
|
+
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 "./index-C2fkZhv_.js";
|
|
3
3
|
import { a as NoopRawProviderSink, c as RawProviderEvent, d as defaultProviderRedactor, f as providerFromBaseUrl, i as InMemoryRawProviderSinkOptions, l as RawProviderSink, n as FileSystemRawProviderSinkOptions, o as ProviderRedactor, r as InMemoryRawProviderSink, s as RawProviderDirection, t as FileSystemRawProviderSink, u as RawProviderSinkFilter } from "./raw-provider-sink-BU29Sh8h.js";
|
|
4
4
|
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";
|
|
5
5
|
import { a as TraceEmitterOptions, i as TraceEmitter, n as RunCompleteHookContext, o as llmSpanFromProvider, r as SpanHandle, t as RunCompleteHook } from "./emitter-DGQGoLyj.js";
|
|
6
6
|
import { a as RunIntegrityReport, i as RunIntegrityIssueCode, n as RunIntegrityExpectations, o as assertRunCaptured, r as RunIntegrityIssue, s as throwIfRunIncomplete, t as RunIntegrityError } from "./integrity-rmVhXWA7.js";
|
|
7
|
-
import { $ as TraceAnalystHookOptions, A as readOtlpStatus, B as TraceInsightQuestion, C as otlpToTraceRunRecords, Ct as traceSpanKindToOpenInferenceKind, D as firstStringAttr, Dt as OtlpSpan, E as extractOtlpAttributes, Et as OtlpResourceSpans, F as TraceInsightContext, G as buildTraceInsightPrompt, H as TraceInsightSuite, I as TraceInsightFinding, J as domainEvidencePattern, K as defaultTraceInsightPanel, L as TraceInsightPanelRole, M as FlattenOtlpOptions, N as OtlpFlatLine, O as inferOtlpKind, Ot as exportRunAsOtlp, P as flattenOtlpExportToNdjson, Q as tokenizeDomainWords, R as TraceInsightPromptInput, S as otlpToRunRecords, St as isOtlpModelCall, T as asString, Tt as OtlpExport, U as TraceInsightTask, V as TraceInsightReadiness, W as buildTraceInsightContext, X as planTraceInsightQuestions, Y as inferDomainKeywords, Z as scoreTraceInsightReadiness, _ as OtlpToRunRecordsOptions, _t as OtlpSpanRole, a as ReplayFetchOptions, at as DEFAULT_REDACTION_RULES, b as otlpRowsToRunRecords, bt as applyToolSpanOtlpAttributes, c as buildTraceAnalystTools, ct as RedactionRule, d as OtlpFileTraceStoreOptions, dt as createOtelTracingStore, et as traceAnalystOnRunComplete, f as SpanNotFoundError, ft as otelRunCompleteHook, g as TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, gt as createOtelExporter, h as TRACE_ANALYST_ACTOR_DESCRIPTION, ht as OtelExporter, i as ReplayCacheStats, it as convertTraceStoresToOtlp, j as stringField, k as projectOtlpFlatLine, l as traceAnalystFunctionGroup, lt as redactString, m as TraceNotFoundError, mt as OtelExportConfig, n as ReplayCacheEntry, nt as TraceStoreToOtlpOptions, o as createReplayFetch, ot as REDACTION_VERSION, p as TraceFileMissingError, pt as ExportableSpan, q as describeTraceInsightScope, r as ReplayCacheMissError, rt as TracesToOtlpResult, s as iterateRawCalls, st as RedactionReport, t as ReplayCache, tt as TraceStoreSource, u as OtlpFileTraceStore, ut as redactValue, v as OtlpTraceRunRecord, vt as OtlpSpanRoleInput, w as ProjectedOtlpSpan, wt as OTEL_AGENT_EVAL_SCOPE, x as otlpRowsToTraceRunRecords, xt as classifyOtlpSpanRole, y as TraceAggregate, yt as ToolSpanOtlpInput, z as TraceInsightQualityGate } from "./replay-
|
|
7
|
+
import { $ as TraceAnalystHookOptions, A as readOtlpStatus, B as TraceInsightQuestion, C as otlpToTraceRunRecords, Ct as traceSpanKindToOpenInferenceKind, D as firstStringAttr, Dt as OtlpSpan, E as extractOtlpAttributes, Et as OtlpResourceSpans, F as TraceInsightContext, G as buildTraceInsightPrompt, H as TraceInsightSuite, I as TraceInsightFinding, J as domainEvidencePattern, K as defaultTraceInsightPanel, L as TraceInsightPanelRole, M as FlattenOtlpOptions, N as OtlpFlatLine, O as inferOtlpKind, Ot as exportRunAsOtlp, P as flattenOtlpExportToNdjson, Q as tokenizeDomainWords, R as TraceInsightPromptInput, S as otlpToRunRecords, St as isOtlpModelCall, T as asString, Tt as OtlpExport, U as TraceInsightTask, V as TraceInsightReadiness, W as buildTraceInsightContext, X as planTraceInsightQuestions, Y as inferDomainKeywords, Z as scoreTraceInsightReadiness, _ as OtlpToRunRecordsOptions, _t as OtlpSpanRole, a as ReplayFetchOptions, at as DEFAULT_REDACTION_RULES, b as otlpRowsToRunRecords, bt as applyToolSpanOtlpAttributes, c as buildTraceAnalystTools, ct as RedactionRule, d as OtlpFileTraceStoreOptions, dt as createOtelTracingStore, et as traceAnalystOnRunComplete, f as SpanNotFoundError, ft as otelRunCompleteHook, g as TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, gt as createOtelExporter, h as TRACE_ANALYST_ACTOR_DESCRIPTION, ht as OtelExporter, i as ReplayCacheStats, it as convertTraceStoresToOtlp, j as stringField, k as projectOtlpFlatLine, l as traceAnalystFunctionGroup, lt as redactString, m as TraceNotFoundError, mt as OtelExportConfig, n as ReplayCacheEntry, nt as TraceStoreToOtlpOptions, o as createReplayFetch, ot as REDACTION_VERSION, p as TraceFileMissingError, pt as ExportableSpan, q as describeTraceInsightScope, r as ReplayCacheMissError, rt as TracesToOtlpResult, s as iterateRawCalls, st as RedactionReport, t as ReplayCache, tt as TraceStoreSource, u as OtlpFileTraceStore, ut as redactValue, v as OtlpTraceRunRecord, vt as OtlpSpanRoleInput, w as ProjectedOtlpSpan, wt as OTEL_AGENT_EVAL_SCOPE, x as otlpRowsToTraceRunRecords, xt as classifyOtlpSpanRole, y as TraceAggregate, yt as ToolSpanOtlpInput, z as TraceInsightQualityGate } from "./replay-BI6CVKkp.js";
|
|
8
8
|
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";
|
|
9
9
|
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";
|
|
10
10
|
import { _ as ViewTraceOversized, a as QueryTracesPage, c as SpanMatchRecord, d as TraceAnalystFilters, f as TraceAnalystSpan, g as ViewSpansResult, h as TraceAnalystTraceSummary, i as ErrorCluster, l as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, m as TraceAnalystSpanStatus, n as DEFAULT_TRACE_ANALYST_BUDGETS, o as SearchSpanResult, p as TraceAnalystSpanKind, r as DatasetOverview, s as SearchTraceResult, t as TraceAnalysisStore, u as TraceAnalystByteBudgets, v as ViewTraceResult } from "./store-CxJry_cs.js";
|
package/dist/traces.js
CHANGED
|
@@ -4,9 +4,9 @@ import { S as traceSpanKindToOpenInferenceKind, a as SpanNotFoundError, b as cla
|
|
|
4
4
|
import { n as llmSpanFromProvider, t as TraceEmitter } from "./emitter-CPBAhxum.js";
|
|
5
5
|
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";
|
|
6
6
|
import { n as TRACE_ANALYST_ACTOR_DESCRIPTION, r as TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, t as analyzeTraces } from "./analyst-LsnNpSkm.js";
|
|
7
|
-
import { C as planTraceInsightQuestions, E as traceAnalystOnRunComplete, S as inferDomainKeywords, T as tokenizeDomainWords, _ as buildTraceInsightContext, a as convertTraceStoresToOtlp, b as describeTraceInsightScope, c as otelRunCompleteHook, d as captureFetchToRawSink, f as otlpRowsToRunRecords, g as flattenOtlpExportToNdjson, h as otlpToTraceRunRecords, i as iterateRawCalls, l as OTEL_AGENT_EVAL_SCOPE, m as otlpToRunRecords, n as ReplayCacheMissError, o as createOtelExporter, p as otlpRowsToTraceRunRecords, r as createReplayFetch, s as createOtelTracingStore, t as ReplayCache, u as exportRunAsOtlp, v as buildTraceInsightPrompt, w as scoreTraceInsightReadiness, x as domainEvidencePattern, y as defaultTraceInsightPanel } from "./replay-
|
|
7
|
+
import { C as planTraceInsightQuestions, E as traceAnalystOnRunComplete, S as inferDomainKeywords, T as tokenizeDomainWords, _ as buildTraceInsightContext, a as convertTraceStoresToOtlp, b as describeTraceInsightScope, c as otelRunCompleteHook, d as captureFetchToRawSink, f as otlpRowsToRunRecords, g as flattenOtlpExportToNdjson, h as otlpToTraceRunRecords, i as iterateRawCalls, l as OTEL_AGENT_EVAL_SCOPE, m as otlpToRunRecords, n as ReplayCacheMissError, o as createOtelExporter, p as otlpRowsToTraceRunRecords, r as createReplayFetch, s as createOtelTracingStore, t as ReplayCache, u as exportRunAsOtlp, v as buildTraceInsightPrompt, w as scoreTraceInsightReadiness, x as domainEvidencePattern, y as defaultTraceInsightPanel } from "./replay-BMR4TEYY.js";
|
|
8
|
+
import { n as extractUsageFromResponse, r as extractUsageFromSse, t as extractUsage } from "./extract-usage-2j25whHw.js";
|
|
8
9
|
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";
|
|
9
10
|
import { a as InMemoryTraceStore, i as FileSystemTraceStore, n as assertRunCaptured, r as throwIfRunIncomplete, t as RunIntegrityError } from "./integrity-BzRbCHzi.js";
|
|
10
|
-
import { n as extractUsageFromResponse, r as extractUsageFromSse, t as extractUsage } from "./extract-usage-BrQ8mCLX.js";
|
|
11
11
|
import { i as redactValue, n as REDACTION_VERSION, r as redactString, t as DEFAULT_REDACTION_RULES } from "./redact-7Aq1ukl-.js";
|
|
12
12
|
export { DEFAULT_REDACTION_RULES, DEFAULT_TRACE_ANALYST_BUDGETS, FAILURE_CLASSES, FileSystemRawProviderSink, FileSystemTraceStore, INPUT_VALUE, InMemoryRawProviderSink, InMemoryTraceStore, LLM_CACHED_TOKENS, LLM_CACHED_TOKEN_ATTR_KEYS, LLM_CACHE_WRITE_TOKENS, LLM_CACHE_WRITE_TOKEN_ATTR_KEYS, LLM_CONTEXT_TOKENS, LLM_COST_ATTR_KEYS, LLM_COST_USD, LLM_INPUT_TOKENS, LLM_INPUT_TOKEN_ATTR_KEYS, LLM_MODEL_ATTR_KEYS, LLM_MODEL_NAME, LLM_OUTPUT_TOKENS, LLM_OUTPUT_TOKEN_ATTR_KEYS, LLM_REASONING_TOKENS, LLM_REASONING_TOKEN_ATTR_KEYS, NoopRawProviderSink, OPENINFERENCE_SPAN_KIND, OTEL_AGENT_EVAL_SCOPE, OUTPUT_VALUE, OtlpFileTraceStore, REDACTION_VERSION, RUN_COST_ATTR_KEYS, ReplayCache, ReplayCacheMissError, RunIntegrityError, SPAN_KIND_ATTR_KEYS, SpanNotFoundError, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, TOOL_NAME_ATTR_KEYS, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, TRACE_SCHEMA_VERSION, TraceEmitter, TraceFileMissingError, TraceNotFoundError, aggregateLlm, analyzeTraces, applyLlmSpanOtlpAttributes, applyToolSpanOtlpAttributes, argHash, asNumber, asString, assertRunCaptured, buildTraceAnalystTools, buildTraceInsightContext, buildTraceInsightPrompt, captureFetchToRawSink, classifyOtlpSpanRole, contextInputTokens, convertTraceStoresToOtlp, createOtelExporter, createOtelTracingStore, createReplayFetch, defaultProviderRedactor, defaultTraceInsightPanel, describeTraceInsightScope, domainEvidencePattern, exportRunAsOtlp, extractOtlpAttributes, extractUsage, extractUsageFromResponse, extractUsageFromSse, firstNumberAttr, firstStringAttr, flattenOtlpExportToNdjson, groupBy, hasCapturedToolArgs, inferDomainKeywords, inferOtlpKind, isJudgeSpan, isLlmSpan, isOtlpModelCall, isRetrievalSpan, isSandboxSpan, isToolSpan, iterateRawCalls, judgeSpans, llmSpanFromProvider, llmSpans, otelRunCompleteHook, otlpRowsToRunRecords, otlpRowsToTraceRunRecords, otlpToRunRecords, otlpToTraceRunRecords, planTraceInsightQuestions, projectOtlpFlatLine, providerFromBaseUrl, readOtlpStatus, redactString, redactValue, runFailureClass, runsForScenario, scoreTraceInsightReadiness, stringField, throwIfRunIncomplete, tokenizeDomainWords, toolSpans, traceAnalystFunctionGroup, traceAnalystOnRunComplete, traceSpanKindToOpenInferenceKind };
|
|
@@ -1,6 +1,6 @@
|
|
|
1
|
-
import {
|
|
2
|
-
import {
|
|
3
|
-
import { n as LlmCallMetadata } from "./llm-client-
|
|
1
|
+
import { S as PaidCallResult, T as RunPaidCallInput, a as CostChannel, c as CostLedgerHandle, f as CostLedgerSummary } from "./cost-ledger-fGS_u_O1.js";
|
|
2
|
+
import { u as RunTokenUsage } from "./run-record-DcObtIGh.js";
|
|
3
|
+
import { n as LlmCallMetadata } from "./llm-client-BiK4HW0u.js";
|
|
4
4
|
//#region src/campaign/types.d.ts
|
|
5
5
|
/** Stable identifier + kind tag for any scenario. Consumers
|
|
6
6
|
* extend with their per-domain payload (persona, task, requirement, ...). */
|
|
@@ -637,4 +637,4 @@ interface CampaignResult<TArtifact = unknown, TScenario extends Scenario = Scena
|
|
|
637
637
|
}
|
|
638
638
|
//#endregion
|
|
639
639
|
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 };
|
|
640
|
-
//# sourceMappingURL=types-
|
|
640
|
+
//# sourceMappingURL=types-BokuXvOG.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"types-
|
|
1
|
+
{"version":3,"file":"types-BokuXvOG.d.ts","names":[],"sources":["../src/campaign/types.ts"],"mappings":";;;;;;UA8BiB;EACf;EACA;EACA;;;;;EAKA;;;UAIe,iCAAiC,KAAK;EACrD;;;;;UAMe;EACf;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;;;;;;;;;UAUa,eAAe;EAC9B,gBAAgB;EAChB,SAAS;EACT,UAAU;;EAEV;EACA;EACA,QAAQ;;EAER,QAAQ;;;EAGR,kBAAkB;;;EAGlB,mBAAmB;;;EAGnB;;;EAGA,UAAU;;;;EAIV;;;;;;;EAOA,gBAAgB;;EAEhB,aAAa;EACb;;;;;;;;EAQA;;;;;;;;;;;;;;UAee,gBAAgB;EAC/B;;;;;EAKA,QAAQ,KAAK,eAAe,aAAa,QAAQ,MAAM,iBAAiB;;;EAGxE,QAAQ;IAAQ,SAAS;;IAAyB;IAAe;;;UAGlD;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;EAC5B;;EAEA;;EAEA;;;EAGA,YAAY;;;;EAIZ;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 { c as CostLedgerHandle, f as CostLedgerSummary } from "./cost-ledger-
|
|
2
|
-
import { i as LlmCallResult, r as LlmCallRequest } from "./llm-client-
|
|
1
|
+
import { c as CostLedgerHandle, f as CostLedgerSummary } from "./cost-ledger-fGS_u_O1.js";
|
|
2
|
+
import { i as LlmCallResult, r as LlmCallRequest } from "./llm-client-BiK4HW0u.js";
|
|
3
3
|
//#region src/analyst/chat-client.d.ts
|
|
4
4
|
/**
|
|
5
5
|
* Unified chat interface using the package's canonical LLM request and result.
|
|
@@ -384,4 +384,4 @@ interface EvalResult {
|
|
|
384
384
|
}
|
|
385
385
|
//#endregion
|
|
386
386
|
export { ChatClient as A, SandboxSdkTransportOpts as B, ScenarioFile as C, TurnMetrics as D, Turn as E, CreateChatClientOpts as F, CustomTransportOpts as I, DirectProviderTransportOpts as L, ChatResponse as M, ChatTransport as N, TurnResult as O, CliBridgeTransportOpts as P, MockTransportOpts as R, Scenario as S, TestResult as T, createChatClient as V, PersonaConfig as _, CheckResult as a, RouteMap as b, DriverResult as c, FeedbackPattern as d, JudgeConfig as f, JudgeScore as g, JudgeRubric as h, BenchmarkRunnerConfig as i, ChatRequest as j, ChatCallOpts as k, DriverState as l, JudgeInput as m, ArtifactResult as n, CollectedArtifacts as o, JudgeFn as p, BenchmarkReport as r, CompletionCriterion as s, ArtifactCheck as t, EvalResult as u, PersonaRigor as v, ScenarioResult as w, RubricDimension as x, ProductClientConfig as y, RouterTransportOpts as z };
|
|
387
|
-
//# sourceMappingURL=types-
|
|
387
|
+
//# sourceMappingURL=types-Cc3qbqzj.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"types-
|
|
1
|
+
{"version":3,"file":"types-Cc3qbqzj.d.ts","names":[],"sources":["../src/analyst/chat-client.ts","../src/types.ts"],"mappings":";;;;;;UAkBiB;;WAEN,WAAW;;WAEX;;WAEA;;EAGT,KAAK,KAAK,aAAa,OAAO,eAAe,QAAQ;;KAG3C;UAQK,oBAAoB,KAAK;;EAExC;;KAGU,eAAe;UAEV;;EAEf,SAAS;;EAET;;EAEA;;EAEA;;KAKU,uBACR,sBACA,yBACA,8BACA,0BACA,sBACA;UAEM;EACR;;EAEA;;UAGe,4BAA4B;EAC3C;EACA;EACA;;UAGe,+BAA+B;EAC9C;EACA;EACA;;UAGe,oCAAoC;EACnD;EACA;EACA;;;;;;UAOe,gCAAgC;EAC/C;EACA,OAAO,KAAK,aAAa,OAAO,iBAAiB,QAAQ;;;UAI1C,4BAA4B;EAC3C;EACA,OAAO,KAAK,aAAa,OAAO,iBAAiB,QAAQ;;;;;;UAO1C,0BAA0B;EACzC;EACA,UAAU,KAAK,aAAa,OAAO,iBAAiB,QAAQ;;;;;;iBAO9C,iBAAiB,MAAM,uBAAuB;;;UCjH7C;EACf;EACA;EACA;EACA;EACA;EACA,OAAO;EACP,gBAAgB;EAChB;;UAGe;EACf;EACA;EACA;EACA;;UAKe;EACf;EAQA;EACA;EACA;EACA;;UAKe;EACf;EACA;EACA,QAAQ;;UAGO;EACf;EACA;EACA,YAAY;;UAGG;EACf;EACA;EACA;EACA;EACA;;UAKe;EACf;EACA;EACA,OAAO;EACP,iBAAiB;EACjB,aAAa;EACb;EACA;EACA;EACA,WAAW;;EAEX,OAAO;;UAGQ;EACf;EACA;EACA;EACA;EACA;IAAmB;IAAc;;EACjC;EACA;;UAGe;EACf,OAAO;EACP;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;;UAGe;EACf;IAAc;IAAc;;EAC5B;IAAmB;IAAc,QAAQ;;EACzC;IAAc;IAAkB;;EAChC;;UAKe;EACf;EACA;EACA;EACA;EACA,SAAS;EACT,OAAO;EACP;IACE;IACA,WAAW;MAAiB;MAAa;MAAgB;;IACzD,aAAa;MAAiB;MAAa;;IAC3C;MAAW;MAAkB;MAAe;;IAC5C;MAAa;MAAkB;MAAe;;;;UAMjC;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;GACC;;UAGc;EACf;EACA,QAAQ;;EAER;;UAKe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;IACE;MACE;MACA;MACA;;;EAGJ,OAAO;EACP,gBAAgB;;UAKD;EACf;EACA,QAAQ,OAAO;EACf,YAAY,OAAO;;UAGJ;EACf;EACA;;;;;;;;;;KAWU;UAEK;EACf;EACA;EACA;EACA,oBAAoB;EACpB,mBAAmB;EACnB;EACA;;EAEA,QAAQ;;;;;;;EAOR;;;;;;;EAOA;;;;;;EAMA;;UAGe;EACf;EACA;EACA;IAAa;IAAiB;IAAkB;;EAChD;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;IAAa;IAAiB;IAAkB;;EAChD;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;;EAEA;;EAEA;;;;;;EAMA;EACA;EACA,SAAS;EACT,YAAY;EACZ;EACA;EACA;;UAKe;EACf,WAAW;EACX,QAAQ;EACR;EACA;EACA;EACA;EACA;EACA;;EAEA,aAAa;;UAGE;EACf,UAAU;EACV,OAAO;EACP,WAAW;;EAEX,aAAa;EACb;EACA,WAAW;EACX,SAAS;;KAGC,WAAW,MAAM,YAAY,OAAO,eAAe,QAAQ;UAItD;EACf;EACA;EACA;EACA;EACA,QAAQ;;UAGO;EACf;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA"}
|
package/dist/wire/index.d.ts
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
|
+
import { c as CostLedgerHandle } from "../cost-ledger-fGS_u_O1.js";
|
|
1
2
|
import { s as TraceStore } from "../store-CT9YIIve.js";
|
|
2
|
-
import {
|
|
3
|
-
import { o as LlmClientOptions, u as LlmRouteRequirements } from "../llm-client-B_nIBlYo.js";
|
|
3
|
+
import { o as LlmClientOptions, u as LlmRouteRequirements } from "../llm-client-BiK4HW0u.js";
|
|
4
4
|
import { h as FeedbackTrajectoryStore } from "../feedback-trajectory-CVaeREXV.js";
|
|
5
5
|
import { z } from "zod";
|
|
6
6
|
import { ServerType } from "@hono/node-server";
|
package/dist/wire/index.js
CHANGED
|
@@ -1,2 +1,2 @@
|
|
|
1
|
-
import { A as TraceEventSchema, C as HealthResponseSchema, D as RubricDimensionSchema, E as ListRubricsResponseSchema, F as hashRubric, M as TracesIngestResponseSchema, N as VersionResponseSchema, O as RubricInfoSchema, P as WIRE_VERSION, S as FeedbackTrajectorySchema, T as JudgeResultSchema, _ as ErrorResponseSchema, a as runRpcBatch, b as FeedbackIngestResponseSchema, c as WireError, d as handleListRubrics, f as handleTracesIngest, g as listBuiltinRubrics, h as getBuiltinRubric, i as dispatchRpc, j as TracesIngestRequestSchema, k as RubricSchema, l as handleFeedbackIngest, m as BUILTIN_RUBRICS, n as startServer, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi, t as createApp, u as handleJudge, v as FailureModeSchema, w as JudgeRequestSchema, x as FeedbackLabelSchema, y as FeedbackAttemptSchema } from "../server-
|
|
1
|
+
import { A as TraceEventSchema, C as HealthResponseSchema, D as RubricDimensionSchema, E as ListRubricsResponseSchema, F as hashRubric, M as TracesIngestResponseSchema, N as VersionResponseSchema, O as RubricInfoSchema, P as WIRE_VERSION, S as FeedbackTrajectorySchema, T as JudgeResultSchema, _ as ErrorResponseSchema, a as runRpcBatch, b as FeedbackIngestResponseSchema, c as WireError, d as handleListRubrics, f as handleTracesIngest, g as listBuiltinRubrics, h as getBuiltinRubric, i as dispatchRpc, j as TracesIngestRequestSchema, k as RubricSchema, l as handleFeedbackIngest, m as BUILTIN_RUBRICS, n as startServer, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi, t as createApp, u as handleJudge, v as FailureModeSchema, w as JudgeRequestSchema, x as FeedbackLabelSchema, y as FeedbackAttemptSchema } from "../server-BPqlDBWK.js";
|
|
2
2
|
export { BUILTIN_RUBRICS, ErrorResponseSchema, FailureModeSchema, FeedbackAttemptSchema, FeedbackIngestResponseSchema, FeedbackLabelSchema, FeedbackTrajectorySchema, HealthResponseSchema, JudgeRequestSchema, JudgeResultSchema, ListRubricsResponseSchema, RubricDimensionSchema, RubricInfoSchema, RubricSchema, TraceEventSchema, TracesIngestRequestSchema, TracesIngestResponseSchema, VersionResponseSchema, WIRE_VERSION, WireError, buildOpenApi, createApp, dispatchRpc, getBuiltinRubric, handleFeedbackIngest, handleJudge, handleListRubrics, handleTracesIngest, handleVersion, hashRubric, listBuiltinRubrics, runRpcBatch, runRpcOnce, startServer, startServerAsync };
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@tangle-network/agent-eval",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.133.0",
|
|
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": {
|
|
@@ -114,11 +114,6 @@
|
|
|
114
114
|
"import": "./dist/authenticity/index.js",
|
|
115
115
|
"default": "./dist/authenticity/index.js"
|
|
116
116
|
},
|
|
117
|
-
"./belief-state": {
|
|
118
|
-
"types": "./dist/belief-state/index.d.ts",
|
|
119
|
-
"import": "./dist/belief-state/index.js",
|
|
120
|
-
"default": "./dist/belief-state/index.js"
|
|
121
|
-
},
|
|
122
117
|
"./contract": {
|
|
123
118
|
"types": "./dist/contract/index.d.ts",
|
|
124
119
|
"import": "./dist/contract/index.js",
|
|
@@ -173,8 +168,8 @@
|
|
|
173
168
|
"@asteasolutions/zod-to-openapi": "^9.1.0",
|
|
174
169
|
"@ax-llm/ax": "^23.0.5",
|
|
175
170
|
"@hono/node-server": "^2.0.12",
|
|
176
|
-
"@tangle-network/agent-core": "0.4.
|
|
177
|
-
"@tangle-network/agent-interface": "0.
|
|
171
|
+
"@tangle-network/agent-core": "0.4.25",
|
|
172
|
+
"@tangle-network/agent-interface": "0.36.0",
|
|
178
173
|
"hono": "^4.12.32",
|
|
179
174
|
"zod": "^4.4.3"
|
|
180
175
|
},
|