@tangle-network/agent-eval 0.202.0 → 0.202.2
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 +11 -0
- package/dist/adapters/http.d.ts +2 -2
- package/dist/analyst/index.d.ts +5 -5
- package/dist/analyst/index.js +1 -1
- package/dist/benchmarks/index.d.ts +4 -4
- package/dist/benchmarks/index.js +1 -1
- package/dist/campaign/index.d.ts +5 -5
- package/dist/campaign/index.js +3 -3
- package/dist/{campaign-XXmOU0R0.js → campaign-Btk1yFfw.js} +130 -20
- package/dist/{campaign-XXmOU0R0.js.map → campaign-Btk1yFfw.js.map} +1 -1
- package/dist/{claude-jsonl-CxZZrDJ3.js → claude-jsonl-DoQJBCVR.js} +49 -15
- package/dist/claude-jsonl-DoQJBCVR.js.map +1 -0
- package/dist/cli.js +1 -1
- package/dist/contract/index.d.ts +7 -7
- package/dist/contract/index.js +1 -1
- package/dist/{default-registry-D4qXx2aG.js → default-registry-6QOiTzqi.js} +2 -2
- package/dist/{default-registry-D4qXx2aG.js.map → default-registry-6QOiTzqi.js.map} +1 -1
- package/dist/{default-registry-D1rNS2yB.d.ts → default-registry-DesnDBvl.d.ts} +3 -3
- package/dist/{default-registry-D1rNS2yB.d.ts.map → default-registry-DesnDBvl.d.ts.map} +1 -1
- package/dist/{define-agent-eval-DQXLPqWc.d.ts → define-agent-eval-BDPV3ieA.d.ts} +5 -5
- package/dist/{define-agent-eval-DQXLPqWc.d.ts.map → define-agent-eval-BDPV3ieA.d.ts.map} +1 -1
- package/dist/{engine-DlKIzDKo.d.ts → engine-neXmlmTu.d.ts} +2 -2
- package/dist/{engine-DlKIzDKo.d.ts.map → engine-neXmlmTu.d.ts.map} +1 -1
- package/dist/evaluation.d.ts +1 -1
- package/dist/experiment/index.d.ts +1 -1
- package/dist/{exporters-XMCYJlqT.d.ts → exporters-DjCu6-lX.d.ts} +2 -2
- package/dist/{exporters-XMCYJlqT.d.ts.map → exporters-DjCu6-lX.d.ts.map} +1 -1
- package/dist/{exporters-Coiv-qO2.js → exporters-ISzpUY3e.js} +10 -4
- package/dist/{exporters-Coiv-qO2.js.map → exporters-ISzpUY3e.js.map} +1 -1
- package/dist/{hf-dataset-kjIfIK_e.js → hf-dataset-DdT7nNkf.js} +3 -3
- package/dist/{hf-dataset-kjIfIK_e.js.map → hf-dataset-DdT7nNkf.js.map} +1 -1
- package/dist/{index-D4pC7Tvd.d.ts → index-BCBpEreJ.d.ts} +18 -14
- package/dist/{index-D4pC7Tvd.d.ts.map → index-BCBpEreJ.d.ts.map} +1 -1
- package/dist/{index-giYkbitB.d.ts → index-BPCdNSMW.d.ts} +20 -5
- package/dist/index-BPCdNSMW.d.ts.map +1 -0
- package/dist/{index-Bi2fLITd.d.ts → index-BtcxmabC.d.ts} +4 -4
- package/dist/{index-Bi2fLITd.d.ts.map → index-BtcxmabC.d.ts.map} +1 -1
- package/dist/{index-D1ZX28rV.d.ts → index-CnKgaHKA.d.ts} +6 -6
- package/dist/{index-D1ZX28rV.d.ts.map → index-CnKgaHKA.d.ts.map} +1 -1
- package/dist/index.d.ts +12 -12
- package/dist/index.js +3 -3
- package/dist/{integrity-uwK7XC-8.js → integrity-atFlABhR.js} +2 -2
- package/dist/{integrity-uwK7XC-8.js.map → integrity-atFlABhR.js.map} +1 -1
- package/dist/{jev-protocol-CpKI7XSu.d.ts → jev-protocol-BQLxwbqD.d.ts} +24 -7
- package/dist/jev-protocol-BQLxwbqD.d.ts.map +1 -0
- package/dist/jev-protocol.d.ts +2 -2
- package/dist/jev-protocol.js +19 -4
- package/dist/jev-protocol.js.map +1 -1
- package/dist/{jev-review-D4ynCORp.d.ts → jev-review-DX6MI_jq.d.ts} +2 -2
- package/dist/{jev-review-D4ynCORp.d.ts.map → jev-review-DX6MI_jq.d.ts.map} +1 -1
- package/dist/jev.d.ts +4 -4
- package/dist/jev.js +2 -2
- package/dist/{matrix--IZKROxD.d.ts → matrix-DRXCV70M.d.ts} +2 -2
- package/dist/{matrix--IZKROxD.d.ts.map → matrix-DRXCV70M.d.ts.map} +1 -1
- package/dist/{mint-C9Lojt1V.d.ts → mint-BFTrlq4t.d.ts} +10 -2
- package/dist/mint-BFTrlq4t.d.ts.map +1 -0
- package/dist/{mint-vlRckJEX.js → mint-DE6nQrX_.js} +26 -7
- package/dist/mint-DE6nQrX_.js.map +1 -0
- package/dist/multishot/golden/index.d.ts +1 -1
- package/dist/multishot/index.d.ts +2 -2
- package/dist/openapi.json +1 -1
- package/dist/rl.d.ts +6 -6
- package/dist/rl.js +3 -3
- package/dist/rollout/index.d.ts +5 -5
- package/dist/rollout/index.js +6 -6
- package/dist/{rollout-BUQixCoc.js → rollout-DTunVdJE.js} +41 -21
- package/dist/{rollout-BUQixCoc.js.map → rollout-DTunVdJE.js.map} +1 -1
- package/dist/{schema-B5LtYvn9.d.ts → schema-Bp8S20Ii.d.ts} +25 -5
- package/dist/{schema-B5LtYvn9.d.ts.map → schema-Bp8S20Ii.d.ts.map} +1 -1
- package/dist/{schema-DAMp77zB.js → schema-D08IIzui.js} +27 -1
- package/dist/{schema-DAMp77zB.js.map → schema-D08IIzui.js.map} +1 -1
- package/dist/search/index.d.ts +2 -2
- package/dist/{search-kernel-C9CXvFR7.d.ts → search-kernel-zI-rsHoG.d.ts} +2 -2
- package/dist/{search-kernel-C9CXvFR7.d.ts.map → search-kernel-zI-rsHoG.d.ts.map} +1 -1
- package/dist/{statistical-heldout-HswMvzbP.d.ts → statistical-heldout-BAbd_oXC.d.ts} +2 -2
- package/dist/{statistical-heldout-HswMvzbP.d.ts.map → statistical-heldout-BAbd_oXC.d.ts.map} +1 -1
- package/dist/{store-tool-spans-Dv9x4XYG.d.ts → store-tool-spans-CKRw7JHa.d.ts} +3 -3
- package/dist/{store-tool-spans-Dv9x4XYG.d.ts.map → store-tool-spans-CKRw7JHa.d.ts.map} +1 -1
- package/dist/supervisor-run/index.d.ts +2 -2
- package/dist/supervisor-run/index.d.ts.map +1 -1
- package/dist/supervisor-run/index.js +22 -14
- package/dist/supervisor-run/index.js.map +1 -1
- package/dist/{tool-groups-BY1MdSIE.d.ts → tool-groups-V23aGP_g.d.ts} +2 -2
- package/dist/tool-groups-V23aGP_g.d.ts.map +1 -0
- package/dist/{trace-questions-DnqdmwyV.d.ts → trace-questions-B-Wf0W3Q.d.ts} +2 -2
- package/dist/{trace-questions-DnqdmwyV.d.ts.map → trace-questions-B-Wf0W3Q.d.ts.map} +1 -1
- package/dist/trace-repair/index.d.ts +1 -1
- package/dist/traces.d.ts +2 -2
- package/dist/{types-DfvQOe91.d.ts → types-BR1LLIxm.d.ts} +2 -2
- package/dist/{types-DfvQOe91.d.ts.map → types-BR1LLIxm.d.ts.map} +1 -1
- package/dist/{types-C4hH69a9.d.ts → types-BbiT5NPH.d.ts} +2 -2
- package/dist/{types-C4hH69a9.d.ts.map → types-BbiT5NPH.d.ts.map} +1 -1
- package/dist/{types-DRNlLbBR.d.ts → types-DUlbOCwL.d.ts} +2 -2
- package/dist/{types-DRNlLbBR.d.ts.map → types-DUlbOCwL.d.ts.map} +1 -1
- package/docs/campaign-coverage.md +20 -0
- package/docs/changes/jev-recorded-evidence.md +24 -0
- package/docs/jev-replay.md +50 -0
- package/docs/retained-dispatch.md +56 -0
- package/docs/rollout.md +3 -1
- package/package.json +1 -1
- package/dist/claude-jsonl-CxZZrDJ3.js.map +0 -1
- package/dist/index-giYkbitB.d.ts.map +0 -1
- package/dist/jev-protocol-CpKI7XSu.d.ts.map +0 -1
- package/dist/mint-C9Lojt1V.d.ts.map +0 -1
- package/dist/mint-vlRckJEX.js.map +0 -1
- package/dist/tool-groups-BY1MdSIE.d.ts.map +0 -1
|
@@ -4,9 +4,9 @@ import { R as LlmThinkingMode, x as ChatClient } from "./types-C0bWoSKC.js";
|
|
|
4
4
|
import { F as ExternalOptimizerCallbackLimits, V as ExternalOptimizerModelCall, W as ExternalOptimizerModelExecutionObservation, q as ExternalOptimizerRunnerCommand } from "./agent-profile-DDfIjdyS.js";
|
|
5
5
|
import { o as Severity } from "./multi-layer-verifier-BUaQ4C17.js";
|
|
6
6
|
import { l as ExactCapableAnalyst } from "./exact-types-KGxBWH4b.js";
|
|
7
|
-
import { i as TraceAnalystDefinition } from "./default-registry-
|
|
8
|
-
import { n as TraceAnalysisEngine } from "./engine-
|
|
9
|
-
import { C as SupervisorRunSources, w as SupervisorRunTree } from "./types-
|
|
7
|
+
import { i as TraceAnalystDefinition } from "./default-registry-DesnDBvl.js";
|
|
8
|
+
import { n as TraceAnalysisEngine } from "./engine-neXmlmTu.js";
|
|
9
|
+
import { C as SupervisorRunSources, w as SupervisorRunTree } from "./types-DUlbOCwL.js";
|
|
10
10
|
//#region src/semantic-concept-judge.d.ts
|
|
11
11
|
/**
|
|
12
12
|
* Implementation complexity class for weighted scoring.
|
|
@@ -476,4 +476,4 @@ declare const KNOWLEDGE_POISONING_KIND_SPEC: TraceAnalystDefinition;
|
|
|
476
476
|
declare const DEFAULT_TRACE_ANALYST_KINDS: readonly TraceAnalystDefinition[];
|
|
477
477
|
//#endregion
|
|
478
478
|
export { SemanticConceptJudgeInput as A, DspyRlmTraceEngineOptions as C, ConceptFinding as D, createChatTraceEngine as E, SemanticConceptJudgeResult as M, runSemanticConceptJudge as N, ConceptSpec as O, renderFindingSubject as S, ChatTraceEngineOptions as T, FindingSubject as _, FAILURE_MODE_KIND_SPEC as a, findingSubjectGrammarPromptFor as b, emitControlIntegrityFindings as c, FindingsStore as d, PersistedFinding as f, FINDING_SUBJECT_SYNTAX as g, FINDING_SUBJECT_KINDS as h, IMPROVEMENT_KIND_SPEC as i, SemanticConceptJudgeOptions as j, SEMANTIC_CONCEPT_JUDGE_VERSION as k, DiffPolicy as l, diffFindings as m, KNOWLEDGE_POISONING_KIND_SPEC as n, CONTROL_INTEGRITY_ANALYST as o, defaultIsMaterial as p, KNOWLEDGE_GAP_KIND_SPEC as r, ControlIntegrityAnalyst as s, DEFAULT_TRACE_ANALYST_KINDS as t, FindingsDiff as u, FindingSubjectKind as v, createDspyRlmTraceEngine as w, parseFindingSubject as x, KIND_EXPECTED_SUBJECTS as y };
|
|
479
|
-
//# sourceMappingURL=index-
|
|
479
|
+
//# sourceMappingURL=index-BtcxmabC.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index-
|
|
1
|
+
{"version":3,"file":"index-BtcxmabC.d.ts","names":[],"sources":["../src/semantic-concept-judge.ts","../src/analyst/chat-trace-engine.ts","../src/analyst/trace-tool-callback.ts","../src/analyst/dspy-rlm-engine.ts","../src/analyst/finding-subject.ts","../src/analyst/findings-store.ts","../src/analyst/kinds/control-integrity.ts","../src/analyst/kinds/failure-mode.ts","../src/analyst/kinds/improvement.ts","../src/analyst/kinds/knowledge-gap.ts","../src/analyst/kinds/knowledge-poisoning.ts","../src/analyst/kinds/index.ts"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;KAuDY;UAEK;EACf;;EAEA;;EAEA;;EAEA,aAAa;;UAGE;EACf;EACA;;EAEA;EACA;EACA,UAAU;;UAGK;;EAEf;;EAEA;;EAEA,aAAa;IAAQ;IAAc;;;EAEnC,kBAAkB;;EAElB;EACA;;UAGe;EACf;EACA;;EAEA;EACA;EACA;EACA,UAAU;EACV;EACA;EACA;;EAEA;EACA;;;;;;;;KASU;UAQK;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,MAAM;;EAEN,UAAU;EACV,aAAa;EACb;EACA,WAAW;EACX,SAAS;;;;;;EAMT,iBAAiB;;EAEjB,oBAAoB,QAAQ,OAAO;;cAKxB;;;;;;iBAkGS,wBACpB,OAAO,2BACP,SAAS,8BACR,QAAQ;;;UCjMM;;;;;EAKf,MAAM;;EAEN;;EAEA,UAAU;;EAEV;;EAEA;;EAEA,WAAW;;EAEX;;;;;;;;;;EAUA;;;;;;iBAkCc,sBAAsB,SAAS,yBAAyB;;;KCrG5D,0BAA0B;;;UCYrB;;EAEf,MAAM;;EAEN;;EAEA,kBAAkB,aAAa;EAC/B;;EAEA,UAAU;;EAEV;;EAEA;;;;;;EAMA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,kBAAkB,QAAQ;;EAE1B;;;;;;;;;;EAUA;;EAEA,SAAS;;EAET;;;iBAIc,yBAAyB,SAAS,4BAA4B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;KCtClE;EAEN;EAAwB;EAAc;;EACtC;EAAyB;;EACzB;EAAuB;;EACvB;EAAyB;;EAGzB;EAAuB;;EACvB;EAAe;;EACf;EAAkB;EAAc;;EAChC;EAAkB;;EAClB;EAAa;EAAgB;;EAC7B;EAAc;;EACd;EAAkB;;EAClB;EAAkB;;EAClB;EAAwB;;EACxB;EAAuB;;EACvB;EAAc;;EACd;EAAa;EAAgB;;EAC7B;EAAgB;;EAChB;EAAqB;;EACrB;EAAuB;;EAEvB;EAA4B;;EAC5B;EAA2B;;EAE3B;EAAiB;;KAEX,qBAAqB;cAEpB,uBAAuB,cAAc;;;;;;;;;;;;;;;;;iBAgGlC,oBAAoB,iCAAiC;;;;;;;iBAoHrD,qBAAqB,GAAG;;;;;;;;;;cA8D3B,wBAAwB,SAAS,OAAO;;;;;;;;;cAoExC,wBAAwB,eAAe,cAAc;;iBAqElD,+BAA+B;;;;;;;;UCxc9B,yBAAyB;EACxC;;cAGW;WAGiB;mBAFX;EAEjB,YAA4B;EAItB,OAAO,eAAe,UAAU,mBAAmB;;EAQzD,WAAW;;EAkBX,QAAQ,gBAAgB;;UAOT;;EAEf,UAAU;;EAEV,aAAa;;EAEb,WAAW;;;;;EAKX,SAAS;IAAQ,UAAU;IAAkB,SAAS;;;UAGvC;;;;;;;;EAQf,cAAc,UAAU,gBAAgB,SAAS;;;;;;iBAOnC,kBAAkB,GAAG,gBAAgB,GAAG;;;;;iBAWxC,aACd,UAAU,oBACV,SAAS,oBACT,SAAQ,aACP;;;;iBC7Fa,6BACd,OAAO,uBAAuB,mBAC9B,qBACC;;cA4BU,mCACA,oBAAoB,uBAAuB;WAE7C;WACA;WAEA;WACA;IAAS;IAAgC;;WACzC;WACA;aACP;aACA;;EAGI,QACJ,OAAO,uBAAuB,mBAC9B,KAAK,iBACJ,QAAQ;;cAUA,2BAAyB;;;cCHzB,wBAAwB;;;cC3BxB,uBAAuB;;;cCGvB,yBAAyB;;;cCPzB,+BAA+B;;;;;;;;;;;;cCP/B,sCAAsC"}
|
|
@@ -1,13 +1,13 @@
|
|
|
1
1
|
import { c as CostLedgerHandle } from "./cost-ledger-DbQdN3nO.js";
|
|
2
2
|
import { a as RunRecord, c as RunSplitTag, n as RunCostProvenance } from "./run-record-DF-fuu31.js";
|
|
3
3
|
import "./types-CeDfCRXp.js";
|
|
4
|
-
import "./statistical-heldout-
|
|
5
|
-
import "./types-
|
|
6
|
-
import "./index-
|
|
4
|
+
import "./statistical-heldout-BAbd_oXC.js";
|
|
5
|
+
import "./types-BR1LLIxm.js";
|
|
6
|
+
import "./index-BPCdNSMW.js";
|
|
7
7
|
import { p as TraceSpanEvent } from "./client-BzL9niqi.js";
|
|
8
8
|
import { d as InsightReport, t as AnalyzeRunsOptions } from "./analyze-runs-CPiKLJFs.js";
|
|
9
|
-
import "./define-agent-eval-
|
|
10
|
-
import "./default-registry-
|
|
9
|
+
import "./define-agent-eval-BDPV3ieA.js";
|
|
10
|
+
import "./default-registry-DesnDBvl.js";
|
|
11
11
|
import { AgentCandidateBenchmarkCellRef, AgentCandidateBenchmarkSuiteInputs, AgentCandidateBenchmarkTask, AgentCandidateBenchmarkTaskMaterial, AgentCandidateBundle, AgentCandidateEvaluationPolicy, AgentCandidateExperiment, AgentCandidateExperimentMaterial, AgentCandidateExperimentMeasurement, AgentImprovementCost, AgentImprovementMeasuredComparison, AgentProfileImprovementExperiment, AgentProfileImprovementExperimentMaterial, AgentProfileImprovementMeasuredComparison, AgentProfileImprovementMeasurement, AgentProfileImprovementRunCell, AgentProfileImprovementRunReceipt, AgentProfileImprovementSuiteInputs, AgentProfileImprovementTask, AgentProfileImprovementTaskMaterial, CandidateExecutionEvidence, Sha256Digest } from "@tangle-network/agent-interface";
|
|
12
12
|
//#region src/contract/measured-comparison.d.ts
|
|
13
13
|
interface SealCandidateBenchmarkSuiteOptions {
|
|
@@ -687,4 +687,4 @@ interface FromOtelSpansOptions {
|
|
|
687
687
|
declare function fromOtelSpans(opts: FromOtelSpansOptions): RunRecord[];
|
|
688
688
|
//#endregion
|
|
689
689
|
export { FromRunRecordDirResult as $, CodeAgentSessionActionStatus as A, AgentTraceFile as B, fromOpenCodeSession as C, measuredComparisonFromCandidateExperiment as Ct, streamCodeAgentJsonlFile as D, sealCandidateExperiment as Dt, parseCodeAgentJsonlFile as E, sealCandidateBenchmarkTask as Et, CodeAgentSessionTerminalStatus as F, PartitionByAuthoringModelResult as G, AgentTraceRange as H, observeCodeAgentSession as I, EvalReportingSuiteInput as J, parseAgentTrace as K, AgentTraceContributor as L, CodeAgentSessionExecutionReceipt as M, CodeAgentSessionObservation as N, CodeAgentSessionAction as O, verifyCandidateExperiment as Ot, CodeAgentSessionSource as P, FromRunRecordDirOptions as Q, AgentTraceContributorType as R, fromKimiCodeSession as S, evaluatePairedMeasurements as St, parseCodeAgentJsonl as T, sealCandidateBenchmarkSuite as Tt, AgentTraceRecord as U, AgentTraceIndex as V, AuthoringProvenance as W, EvalReportingSuiteResult as X, EvalReportingSuiteOptions as Y, evalReportingSuite as Z, CodeAgentSessionIntakeResult as _, PairedMeasurement as _t, FromFeedbackTableOptions as a, RunAgentProfileImprovementExperimentOptions as at, fromClaudeCodeSession as b, RunCandidateExperimentOptions as bt, CodeAgentStoreObservation as c, runAgentProfileImprovementExperiment as ct, CodeAgentStoreScope as d, sealAgentProfileImprovementTask as dt, RunRecordRejection as et, CodeAgentStoreSession as f, verifyAgentProfileImprovementExperimentComparison as ft, CodeAgentSessionIntakeOptions as g, EvaluatePairedMeasurementsOptions as gt, CodeAgentSessionDiagnostic as h, CompareCandidateExperimentOptions as ht, FeedbackTableRow as i, CompareAgentProfileImprovementExperimentOptions as it, CodeAgentSessionActionSurface as j, CodeAgentSessionActionKind as k, verifyCandidateExperimentComparison as kt, CodeAgentStoreRejection as l, sealAgentProfileImprovementExperiment as lt, CodeAgentJsonlLine as m, CandidateExperimentRun as mt, fromOtelSpans as n, AgentProfileImprovementExperimentExecutionInput as nt, FromFeedbackTableResult as o, SealAgentProfileImprovementSuiteOptions as ot, observeCodeAgentStore as p, CandidateExperimentExecutionInput as pt, partitionRunsByAuthoringModel as q, FeedbackTableMeta as r, AgentProfileImprovementExperimentRun as rt, fromFeedbackTable as s, measuredComparisonFromAgentProfileImprovementExperiment as st, FromOtelSpansOptions as t, fromRunRecordDir as tt, CodeAgentStoreRejections as u, sealAgentProfileImprovementSuite as ut, CodeAgentSessionMetrics as v, PairedMeasurementAdapter as vt, fromPiSession as w, runCandidateExperiment as wt, fromCodexSession as x, SealCandidateBenchmarkSuiteOptions as xt, ParsedCodeAgentJsonl as y, PairedMeasurementEvaluation as yt, AgentTraceConversation as z };
|
|
690
|
-
//# sourceMappingURL=index-
|
|
690
|
+
//# sourceMappingURL=index-CnKgaHKA.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index-
|
|
1
|
+
{"version":3,"file":"index-CnKgaHKA.d.ts","names":[],"sources":["../src/contract/measured-comparison.ts","../src/contract/profile-measured-comparison.ts","../src/contract/intake/run-record-dir.ts","../src/contract/eval-reporting-suite.ts","../src/contract/intake/agent-trace.ts","../src/contract/intake/code-agent-observation.ts","../src/contract/intake/code-agent-session.ts","../src/contract/intake/code-agent-store.ts","../src/contract/intake/feedback-table.ts","../src/contract/intake/otel-spans.ts"],"mappings":";;;;;;;;;;;;UAiCiB;EACf,QAAQ,gCAAgC;EACxC;EACA;;UAGe;EACf,YAAY;EACZ;EACA,QAAQ;EACR,MAAM;EACN,eAAe;EACf;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,QAAQ,OAAO,oCAAoC,QAAQ;;EAE3D;;EAEA,aAAa;EACb,SAAS;;UAGM;EACf,cAAc;EACd;IACE;IACA,MAAM;;;UAIO;EACf,YAAY;EACZ,cAAc;EACd;IACE;IACA,MAAM;;EAER,aAAa;EACb;EACA,YAAY;EACZ;EACA,WAAW;;;UAII,kBAAkB;EACjC;EACA,UAAU;EACV,WAAW;;;UAII,yBAAyB;EACxC,MAAM,KAAK;EACX,WAAW,KAAK;IAAkB;IAAc;;EAChD,QAAQ,KAAK;EACb,eAAe,KAAK,OAAO;EAC3B,UAAU,KAAK;EACf,UAAU,KAAK;EACf,OAAO,KAAK;;UAGG,kCAAkC;EACjD,uBAAuB,kBAAkB;EACzC,QAAQ;EACR,SAAS,yBAAyB;;EAElC;;EAEA,kBAAkB;;EAElB,kBAAkB;;;KAIR,8BAA8B,KACxC;EAGA,iBAAiB;EACjB,WAAW;EACX;;;iBAIc,2BACd,UAAU,sCACT;;iBAQa,4BACd,SAAS,qCACR;;iBAiBa,wBACd,UAAU,mCACT;iBAQa,0BAA0B,iBAAiB;;iBAarC,uBACpB,SAAS,gCACR,QAAQ;;;;;;;;iBAkEK,2BAA2B,MACzC,SAAS,kCAAkC,QAC1C;;iBA8Ya,0CACd,SAAS,oCACR;;iBAuEa,oCACd,iBACC;;;UCjrBc;EACf,aAAa;EACb,QAAQ,gCAAgC;EACxC;EACA;;UAGe;EACf,YAAY;EACZ;EACA,aAAa;EACb,MAAM;EACN,SAAS;EACT;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,QACE,OAAO,kDACN,QAAQ;;EAEX;;EAEA,aAAa;EACb,SAAS;;UAGM;EACf,cAAc;EACd;IACE;IACA,MAAM;;;UAIO;EACf,YAAY;EACZ,cAAc;EACd;IACE;IACA,MAAM;;EAER,aAAa;EACb;EACA,YAAY;EACZ;EACA,WAAW;;;iBAIG,gCACd,UAAU,sCACT;;iBAQa,iCACd,SAAS,0CACR;;iBAqBa,sCACd,UAAU,4CACT;;;;;;iBAsBmB,qCACpB,SAAS,8CACR,QAAQ;;iBA0CK,wDACd,SAAS,kDACR;;iBAuEa,kDACd,iBACC;;;;UC7Oc;;EAEf;;EAEA;;EAEA;;UAGe;;;;;;EAMf;;;;;;;EAOA,WAAW;;;;;;EAMX;;UAGe;;EAEf,MAAM;;EAEN,UAAU;;EAEV;;;;;;;;;;iBAkBoB,iBACpB,cACA,UAAS,0BACR,QAAQ;;;;;KC7CC,0BAA0B;UAErB;;;;;EAKf,UAAU,KAAK;;EAEf,OAAO;;;;;;;;;;EAUP;;;;UAKe;;;EAGf,QAAQ;;EAER;;IAEE;;IAEA;;;IAGA;;IAEA;;;IAGA,UAAU;;;EAGZ;;;;;;;iBAUoB,mBACpB,OAAO,yBACP,UAAS,4BACR,QAAQ;;;KChEC;UAEK;EACf,MAAM;;EAEN;;UAGe;EACf;EACA;EACA;;;EAGA,cAAc;;UAGC;EACf;EACA,cAAc;EACd,QAAQ;;UAGO;EACf;EACA,eAAe;;UAGA;EACf;EACA;EACA;EACA;IAAQ;IAAc;;EACtB;IAAS;IAAe;;EACxB,OAAO;;;;UAOQ;EACf;;EAEA;;EAEA;EACA;EACA;;EAEA;;EAEA;;KAGU,kBAAkB,YAAY;;;;;;iBAW1B,gBAAgB,SAAS,qBAAqB;UAmE7C;;;;EAIf,SAAS,YAAY;;;EAGrB,cAAc;;;;;;;;iBASA,8BACd,MAAM,aACN,OAAO,kBACN;;;KCjLS;KAEA;KAEA;KAEA;KASA;UAEK;EACf;EACA;EACA;;UAGe;EACf;EACA;EACA,MAAM;EACN,SAAS;EACT;EACA,QAAQ;EACR;EACA;EACA,UAAU;;UAGK;EACf,QAAQ;EACR;EACA;EACA;IACE,QAAQ;IACR;;EAEF,SAAS;;UAGM;EACf,QAAQ;EACR;EACA;EACA,YAAY;;;;;;;iBAeE,wBACd,SAAS,iCACR;;;UCjCc;EACf;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf,QAAQ;EACR;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;;UAGe;EACf,MAAM;EACN,aAAa;EACb,SAAS;EACT,cAAc;;UAGC;EACf;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;EACA;EACA;EACA;EACA;EACA;;;EAGA,iBAAiB;;;EAGjB,YAAY;;;;KAKF;EACN;EAAe;EAAoB;;EACnC;EAAmB;;iBAcT,oBAAoB,gBAAgB;;;;;;;;;;iBAuB7B,yBAAyB,eAAe,eAAe;;;;;iBA6BxD,wBAAwB,eAAe,QAAQ;iBAUrD,iBACd,SAAS,gCACR;iBAIa,sBACd,SAAS,gCACR;iBAIa,oBACd,SAAS,gCACR;iBAIa,oBACd,SAAS,gCACR;iBAIa,cACd,SAAS,gCACR;;;;UC/Lc;;;;;WAKN;;WAEA,QAAQ;;;;;WAKR;;;;;WAKA;aACE;;aAEA;;;;KAKD;;UAUK;WACN;WACA;WACA;WACA;WACA,aAAa;;;UAIP;WACN;WACA,UAAU,SAAS,QAAQ,OAAO;;WAElC,QAAQ;aACN;aACA,QAAQ;;;;;;;;;KAUT;WAEG;WACA,OAAO;WACP;WACA,mBAAmB;;WAEnB;;WAEA;;WAEA;WACA,WAAW,SAAS,QAAQ,OAAO;WACnC,YAAY;;WAGZ;WACA,OAAO;WACP;;WAEA;WACA,YAAY;;;;;;;;;iBAaL,sBACpB,OAAO,sBACN,QAAQ;;;UCtHM;;;EAGf;;EAEA;;;EAGA;;;EAGA,WAAW;;UAGI;EACf;;;EAGA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA,WAAW;;;EAGX,SAAS;;UAGM;;EAEf,SAAS;;;EAGT,OAAO;;;;EAIP;IAAU;IAAa;;;;;EAIvB;;UAGe;EACf,MAAM;;;EAGN,aAAa;IAAQ;IAAe;IAAe;;;iBAGrC,kBAAkB,MAAM,2BAA2B;;;UC3BlD;EACf,OAAO;;EAEP,eAAe;;EAEf;;;;;;EAMA,eAAe,eAAe,gBAAgB;;iBAGhC,cAAc,MAAM,uBAAuB"}
|
package/dist/index.d.ts
CHANGED
|
@@ -9,15 +9,15 @@ import { a as ContinuousCalibrationResult, d as calibrateJudgeContinuous, f as c
|
|
|
9
9
|
import { a as CorpusAgreementPerDimension, c as corpusInterRaterAgreement, i as CorpusAgreementOptions, l as corpusInterRaterAgreementFromJudgeScores, n as SeriesConvergenceResult, o as CorpusAgreementReport, r as analyzeSeries, s as CorpusScoreRecord, t as SeriesConvergenceOptions, u as interRaterReliability } from "./series-convergence-DY7ha0kx.js";
|
|
10
10
|
import { a as partialCredit, c as spearmanR, d as weightedMean, i as confidenceInterval, l as summarizeNumberSeries, n as WeightedCompositeInput, o as pearsonR, r as WeightedCompositeResult, s as ranks, t as SeriesDistribution, u as weightedComposite } from "./descriptive-B2iPaT9J.js";
|
|
11
11
|
import { n as bonferroni, r as holm, t as benjaminiHochberg } from "./multiplicity-DIWHvysC.js";
|
|
12
|
-
import { A as PairArmsOptions, Bt as RiskDifferenceResult, C as hashJson, Ct as verifyAttestation, D as ComparePairedArmsOptions, F as PairedCorrectness, Ft as EProcessStep, Gt as pairedRiskDifference, Ht as isBinaryOutcomeVector, I as PairedMetricDelta, It as eProcess, Jt as passAtK, Kt as pairedRiskDifferenceExact, L as comparePairedArms, Lt as ExactRiskDifferenceResult, M as PairRunRecordsResult, Mt as EProcess, N as PairedArmRow, Nt as EProcessOptions, O as MatchedPair, P as PairedArmsComparison, Pt as EProcessState, R as pairArms, Rt as McNemarResult, St as attest, Ut as mcnemar, Vt as ScoreRiskDifferenceResult, Wt as pairedBinaryScale, Yt as wilson, bt as AttestationVerification, et as Objective, j as PairArmsResult, k as MatchedRunRecordPair, nt as dominates, qt as pairedRiskDifferenceScore, rt as paretoFrontier, tt as ParetoResult, vt as ATTESTATION_ALGORITHM, w as manifestContentDigest, xt as AttestedReport, yt as AttestationProvenance, z as pairRunRecords, zt as ProportionInterval } from "./statistical-heldout-
|
|
12
|
+
import { A as PairArmsOptions, Bt as RiskDifferenceResult, C as hashJson, Ct as verifyAttestation, D as ComparePairedArmsOptions, F as PairedCorrectness, Ft as EProcessStep, Gt as pairedRiskDifference, Ht as isBinaryOutcomeVector, I as PairedMetricDelta, It as eProcess, Jt as passAtK, Kt as pairedRiskDifferenceExact, L as comparePairedArms, Lt as ExactRiskDifferenceResult, M as PairRunRecordsResult, Mt as EProcess, N as PairedArmRow, Nt as EProcessOptions, O as MatchedPair, P as PairedArmsComparison, Pt as EProcessState, R as pairArms, Rt as McNemarResult, St as attest, Ut as mcnemar, Vt as ScoreRiskDifferenceResult, Wt as pairedBinaryScale, Yt as wilson, bt as AttestationVerification, et as Objective, j as PairArmsResult, k as MatchedRunRecordPair, nt as dominates, qt as pairedRiskDifferenceScore, rt as paretoFrontier, tt as ParetoResult, vt as ATTESTATION_ALGORITHM, w as manifestContentDigest, xt as AttestedReport, yt as AttestationProvenance, z as pairRunRecords, zt as ProportionInterval } from "./statistical-heldout-BAbd_oXC.js";
|
|
13
13
|
import { _ as pairedSignTest, a as PairedPromotionDecision, c as BOOTSTRAP_GATE_MIN_N, d as PairedBootstrapResult, f as PairedSignTestResult, g as pairedDeltaTieFraction, h as pairedBootstrap, i as PairedMcNemarEvidence, l as DECISION_PAIRED_DELTA_STATISTIC, m as SignTestAlternative, n as PairedDecisionShape, o as PairedPromotionDecisionOptions, p as PairedTTestResult, r as PairedDecisionStatistic, s as decidePairedPromotion, t as PairedDecisionMethod, u as PairedBootstrapOptions, v as pairedTTest } from "./paired-promotion-decision-DPKz8ZnF.js";
|
|
14
14
|
import { a as requiredPairedSampleSize, i as pairedMde, n as mcnemarPower, o as requiredSampleSize, r as mcnemarRequiredN, t as mulberry32 } from "./random-DGEyr3zG.js";
|
|
15
15
|
import { C as RankTestMethod, D as WilcoxonSignedRankResult, E as WILCOXON_EXACT_MAX_N, O as mannWhitneyU, S as MannWhitneyResult, T as RankTestOptions, _ as bootstrapCi, a as ReleaseConfidenceIssue, b as MANN_WHITNEY_EXACT_MAX_STATES, f as evaluateReleaseConfidence, g as Verdict, i as ReleaseConfidenceInput, k as wilcoxonSignedRank, m as BootstrapResult, o as ReleaseConfidenceMetrics, p as BootstrapOptions, s as ReleaseConfidenceScorecard, t as ActionableSideInfo, u as ReleaseTraceEvidence, w as RankTestMethodRequest, x as MANN_WHITNEY_EXACT_MAX_WORK, y as DEFAULT_PERMUTATIONS } from "./release-confidence-CkoXFx9I.js";
|
|
16
16
|
import { a as RunFilter, i as InMemoryTraceStore, n as FileSystemTraceStore, o as SpanFilter, s as TraceStore, t as EventFilter } from "./store-C4avFjIi.js";
|
|
17
|
-
import { a as MintedRolloutLine } from "./schema-
|
|
18
|
-
import { a as mintRolloutRows, c as ScorePreference, i as RolloutSearchLineage, n as MintRolloutResult, t as MintRolloutOptions } from "./mint-
|
|
19
|
-
import { R as Scenario$1, S as JudgeConfig, _ as GateDecision, a as CampaignResult, w as JudgeScore$1 } from "./types-
|
|
20
|
-
import { Ar as DEFAULT_RED_TEAM_CORPUS, At as RequirementCheck, Br as CanaryKind, Ct as extractProducedState, Dt as LlmCorrectnessCheckerOpts, Et as CorrectnessChecker, Fr as redTeamDataset, Ft as verifyCompletion, Hr as CanaryReport, Ir as redTeamReport, Ji as llmJudge, Lr as scoreRedTeamOutput, Mr as RedTeamCategory, Mt as TaskGold, Nr as RedTeamFinding, Nt as completionVerdict, Ot as ProducedProposal, Pr as RedTeamReport, Pt as createLlmCorrectnessChecker, Qi as transientDispatchFailure, Rr as CanaryAlert, St as ToolCallEventLike, Tt as CompletionVerdict, Ur as runCanaries, Vr as CanaryOptions, Zi as quotaExhaustedUntil, _a as runCampaign, _t as assertRealBackend, bt as ProposalEventLike, ga as RunCampaignOptions, gt as BackendIntegrityReport, ha as CampaignCellRetryPolicy, ht as BackendIntegrityError, jr as RedTeamCase, jt as SatisfiedBy, kt as ProducedState, ma as CampaignCellFailureReceipt, qi as LlmJudgeOptions, vt as summarizeBackendIntegrity, wt as CompletionRequirement, xt as RuntimeEventLike, yt as ArtifactEventLike, zr as CanaryEvaluation } from "./index-
|
|
17
|
+
import { a as MintedRolloutLine } from "./schema-Bp8S20Ii.js";
|
|
18
|
+
import { a as mintRolloutRows, c as ScorePreference, i as RolloutSearchLineage, n as MintRolloutResult, t as MintRolloutOptions } from "./mint-BFTrlq4t.js";
|
|
19
|
+
import { R as Scenario$1, S as JudgeConfig, _ as GateDecision, a as CampaignResult, w as JudgeScore$1 } from "./types-BR1LLIxm.js";
|
|
20
|
+
import { Ar as DEFAULT_RED_TEAM_CORPUS, At as RequirementCheck, Br as CanaryKind, Ct as extractProducedState, Dt as LlmCorrectnessCheckerOpts, Et as CorrectnessChecker, Fr as redTeamDataset, Ft as verifyCompletion, Hr as CanaryReport, Ir as redTeamReport, Ji as llmJudge, Lr as scoreRedTeamOutput, Mr as RedTeamCategory, Mt as TaskGold, Nr as RedTeamFinding, Nt as completionVerdict, Ot as ProducedProposal, Pr as RedTeamReport, Pt as createLlmCorrectnessChecker, Qi as transientDispatchFailure, Rr as CanaryAlert, St as ToolCallEventLike, Tt as CompletionVerdict, Ur as runCanaries, Vr as CanaryOptions, Zi as quotaExhaustedUntil, _a as runCampaign, _t as assertRealBackend, bt as ProposalEventLike, ga as RunCampaignOptions, gt as BackendIntegrityReport, ha as CampaignCellRetryPolicy, ht as BackendIntegrityError, jr as RedTeamCase, jt as SatisfiedBy, kt as ProducedState, ma as CampaignCellFailureReceipt, qi as LlmJudgeOptions, vt as summarizeBackendIntegrity, wt as CompletionRequirement, xt as RuntimeEventLike, yt as ArtifactEventLike, zr as CanaryEvaluation } from "./index-BPCdNSMW.js";
|
|
21
21
|
import { a as ProfileAxisSpec, c as expandProfileAxes, i as HarnessType, l as harnessAxisOf, n as CODING_HARNESSES, o as agentProfileHash, r as HARNESS_NATIVE_MODEL, s as agentProfileId, t as AgentProfile } from "./agent-profile-DDfIjdyS.js";
|
|
22
22
|
import { i as DatasetSplit, n as DatasetManifest, r as DatasetScenario, t as Dataset } from "./dataset-DQqhOCPt.js";
|
|
23
23
|
import { C as defineEquivalenceCheck, S as buildEquivalenceRecord, _ as StrategyChecker, a as CheckerIdentity, b as VerificationStrategyProfile, c as EquivalenceCheckDefinition, d as EquivalenceCheckerInput, f as EquivalenceCheckerResult, g as EquivalenceRecord, h as EquivalenceProtocolError, i as equivalenceVerdict, l as EquivalenceCheckSpec, m as EquivalenceObligationStatus, n as VerdictCertification, o as CheckerOutcome, p as EquivalenceObligation, r as certificationEvidenceDigest, s as EquivalenceArm, t as DefaultVerdict, u as EquivalenceChecker, v as VERIFICATION_STRATEGIES, w as runEquivalenceCheck, x as VerificationStrategySource, y as VERIFICATION_STRATEGY_SOURCES } from "./verdict-E4eRNf7-.js";
|
|
@@ -27,11 +27,11 @@ import { a as FAILURE_BLAME, c as FailureContext, d as INFRA_FAILURE_BLAMES, f a
|
|
|
27
27
|
import { d as InsightReport, i as analyzeRuns, t as AnalyzeRunsOptions } from "./analyze-runs-CPiKLJFs.js";
|
|
28
28
|
import { i as ExactAnalystRunEvent, l as ExactCapableAnalyst, o as ExactAnalystRunResult } from "./exact-types-KGxBWH4b.js";
|
|
29
29
|
import { c as RegistryRunOpts, i as BudgetPolicy, n as AnalystRegistry, r as AnalystRegistryOptions, s as ExactRegistryRunOpts } from "./registry-Bu3lNole.js";
|
|
30
|
-
import { _ as SelfImproveMethodResult, a as DefinedAgentEval, c as SelfImproveMethodOptions, d as SelfImproveProposerOptions, f as SelfImproveProposerResult, g as SelfImproveMethodProvenance, h as selfImprove, i as DefineAgentEvalOptions, l as SelfImproveOptions, n as AgentEvalEvaluateOptions, o as defineAgentEval, p as SelfImproveResult, r as AgentEvalImproveOptions, s as SelfImproveBudget, t as AgentEvalAgent, u as SelfImproveProgressEvent } from "./define-agent-eval-
|
|
31
|
-
import { a as createTraceAnalyst, i as TraceAnalystDefinition, n as buildDefaultAnalystRegistry, r as CreateTraceAnalystOptions, t as DefaultAnalystRegistryOptions } from "./default-registry-
|
|
32
|
-
import { i as TraceAnalysisEngineResult, l as RawAnalystFinding, n as TraceAnalysisEngine, t as DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS } from "./engine-
|
|
33
|
-
import { A as SemanticConceptJudgeInput, C as DspyRlmTraceEngineOptions, D as ConceptFinding, E as createChatTraceEngine, M as SemanticConceptJudgeResult, N as runSemanticConceptJudge, O as ConceptSpec, T as ChatTraceEngineOptions, _ as FindingSubject, a as FAILURE_MODE_KIND_SPEC, d as FindingsStore, f as PersistedFinding, j as SemanticConceptJudgeOptions, k as SEMANTIC_CONCEPT_JUDGE_VERSION, l as DiffPolicy, m as diffFindings, t as DEFAULT_TRACE_ANALYST_KINDS, u as FindingsDiff, v as FindingSubjectKind, w as createDspyRlmTraceEngine } from "./index-
|
|
34
|
-
import { i as BenchmarkEvaluation } from "./types-
|
|
30
|
+
import { _ as SelfImproveMethodResult, a as DefinedAgentEval, c as SelfImproveMethodOptions, d as SelfImproveProposerOptions, f as SelfImproveProposerResult, g as SelfImproveMethodProvenance, h as selfImprove, i as DefineAgentEvalOptions, l as SelfImproveOptions, n as AgentEvalEvaluateOptions, o as defineAgentEval, p as SelfImproveResult, r as AgentEvalImproveOptions, s as SelfImproveBudget, t as AgentEvalAgent, u as SelfImproveProgressEvent } from "./define-agent-eval-BDPV3ieA.js";
|
|
31
|
+
import { a as createTraceAnalyst, i as TraceAnalystDefinition, n as buildDefaultAnalystRegistry, r as CreateTraceAnalystOptions, t as DefaultAnalystRegistryOptions } from "./default-registry-DesnDBvl.js";
|
|
32
|
+
import { i as TraceAnalysisEngineResult, l as RawAnalystFinding, n as TraceAnalysisEngine, t as DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS } from "./engine-neXmlmTu.js";
|
|
33
|
+
import { A as SemanticConceptJudgeInput, C as DspyRlmTraceEngineOptions, D as ConceptFinding, E as createChatTraceEngine, M as SemanticConceptJudgeResult, N as runSemanticConceptJudge, O as ConceptSpec, T as ChatTraceEngineOptions, _ as FindingSubject, a as FAILURE_MODE_KIND_SPEC, d as FindingsStore, f as PersistedFinding, j as SemanticConceptJudgeOptions, k as SEMANTIC_CONCEPT_JUDGE_VERSION, l as DiffPolicy, m as diffFindings, t as DEFAULT_TRACE_ANALYST_KINDS, u as FindingsDiff, v as FindingSubjectKind, w as createDspyRlmTraceEngine } from "./index-BtcxmabC.js";
|
|
34
|
+
import { i as BenchmarkEvaluation } from "./types-BbiT5NPH.js";
|
|
35
35
|
import { a as SpanHandle, n as CaptureWrite, o as TraceEmitter, s as TraceEmitterOptions, t as CaptureErrorHandler } from "./emitter-HcCmFQl_.js";
|
|
36
36
|
import { n as SandboxDriver } from "./sandbox-harness-BlSOu4LX.js";
|
|
37
37
|
import { S as shareAllowed, b as redactForShare, d as ShareSafetyVerdict, f as assessShareSafety, i as RedactOptions, o as RedactionProfile, r as REDACTION_VERSION, s as RedactionReport, u as ShareSafetyStatus, x as redactText, y as redact } from "./redact-Cveoq6fJ.js";
|
|
@@ -40,11 +40,11 @@ import { n as TrajectoryStep, r as buildTrajectory, t as Trajectory } from "./tr
|
|
|
40
40
|
import { a as runCounterfactual, i as CounterfactualRunner, n as CounterfactualMutation, r as CounterfactualResult, t as CounterfactualContext } from "./counterfactual-CPaWN99-.js";
|
|
41
41
|
import { A as AnalystFindingDigest, B as ControlRunResult, C as createFeedbackTrajectory, D as renderPreferenceMemoryMarkdown, E as feedbackTrajectoryToOptimizerRow, F as ControlActionOutcome, G as StopDecision, H as ControlRuntimeError, I as ControlBudget, J as subjectiveEval, K as objectiveEval, L as ControlContext, M as AnalystRunDigest, N as analystFindingDigest, O as summarizePreferenceMemory, P as analystRunDigest, R as ControlDecision, S as controlRunToFeedbackTrajectory, T as feedbackTrajectoriesToOptimizerRows, U as ControlSeverity, V as ControlRuntimeConfig, W as ControlStep, _ as PreferenceMemoryEntry, a as FeedbackLabel, b as analystRunToReviewRequests, c as FeedbackOptimizerRow, d as FeedbackTask, f as FeedbackTrajectory, g as InMemoryFeedbackTrajectoryStore, h as FileSystemFeedbackTrajectoryStore, i as FeedbackAttempt, j as AnalystReviewDecision, k as withAssignedFeedbackSplit, l as FeedbackOutcome, m as FeedbackTrajectoryStore, n as AnalystReviewRequest, o as FeedbackLabelKind, p as FeedbackTrajectoryFilter, q as runAgentControlLoop, r as FeedbackArtifactType, s as FeedbackLabelSource, t as AnalystFeedbackTrajectoryOptions, u as FeedbackSplitPolicy, v as ProposedSideEffect, w as feedbackTrajectoriesToDatasetScenarios, x as assignFeedbackSplit, y as analystRunToFeedbackTrajectory, z as ControlEvalResult } from "./feedback-trajectory-BXCGTRkz.js";
|
|
42
42
|
import { y as OUTPUT_VALUE } from "./attribute-vocabulary-CcXqHODT.js";
|
|
43
|
-
import { C as inferDomainKeywords, H as captureFetchToRawSink, I as OtlpExport, J as extractUsageFromSse, K as extractUsage, M as AnalyzeTracesResult, N as analyzeTraces, O as scoreTraceInsightReadiness, P as OtlpFlatLine, R as OtlpSpan, S as domainEvidencePattern, W as ExtractedUsage, g as TraceInsightSuite, h as TraceInsightReadiness, k as tokenizeDomainWords, n as toolSpansToTraceAnalysisStore, o as otlpTextToTraceAnalysisStore, r as OtlpFileTraceStore, v as buildTraceInsightContext, x as describeTraceInsightScope, y as buildTraceInsightPrompt, z as exportRunAsOtlp } from "./store-tool-spans-
|
|
43
|
+
import { C as inferDomainKeywords, H as captureFetchToRawSink, I as OtlpExport, J as extractUsageFromSse, K as extractUsage, M as AnalyzeTracesResult, N as analyzeTraces, O as scoreTraceInsightReadiness, P as OtlpFlatLine, R as OtlpSpan, S as domainEvidencePattern, W as ExtractedUsage, g as TraceInsightSuite, h as TraceInsightReadiness, k as tokenizeDomainWords, n as toolSpansToTraceAnalysisStore, o as otlpTextToTraceAnalysisStore, r as OtlpFileTraceStore, v as buildTraceInsightContext, x as describeTraceInsightScope, y as buildTraceInsightPrompt, z as exportRunAsOtlp } from "./store-tool-spans-CKRw7JHa.js";
|
|
44
44
|
import { a as RunIntegrityReport, o as assertRunCaptured, t as RunIntegrityError } from "./integrity-CaeNeusS.js";
|
|
45
45
|
import { c as judgeSpans, f as runsForScenario, i as argHash } from "./query-DBUQhehH.js";
|
|
46
46
|
import { a as SteeringChange, c as CampaignRunContext, d as CampaignVariant, f as EvalCampaignOptions, h as runEvalCampaign, i as Researcher, l as CampaignRunOutcome, m as FailedRun, n as ExperimentResult, o as CampaignFactoryParams, p as EvalCampaignResult, r as FailureMode, s as CampaignIntegrityPolicy, t as ExperimentPlan, u as CampaignScenario } from "./researcher-DojEMb7E.js";
|
|
47
|
-
import "./index-
|
|
47
|
+
import "./index-BCBpEreJ.js";
|
|
48
48
|
import { _ as iqr, a as ToolStats, c as computeToolUseMetrics, d as judgeAgreementView, i as toolWasteView, m as budgetBreachView, o as ToolUseMetrics, s as ToolUseOptions } from "./tool-waste-CP5DO383.js";
|
|
49
49
|
import { SpanKind } from "@tangle-network/agent-trace-contract";
|
|
50
50
|
//#region src/statistics/effect-sizes.d.ts
|
package/dist/index.js
CHANGED
|
@@ -35,7 +35,7 @@ import { t as TraceEmitter } from "./emitter-DbMXUIVP.js";
|
|
|
35
35
|
import { t as buildTrajectory } from "./trajectory-D_7rLrvE.js";
|
|
36
36
|
import { t as runCounterfactual } from "./counterfactual-DtMWvn7D.js";
|
|
37
37
|
import { G as DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS, U as parseFindingSubject, h as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, m as DEFAULT_TRACE_ANALYST_BUDGETS, t as createTraceAnalyst, u as createBoundedTraceAnalysisStore } from "./kind-factory-CdPPHtGZ.js";
|
|
38
|
-
import { _ as completedAnalystReviewQuality, b as validateAnalystReviewDecisions, c as FAILURE_MODE_KIND_SPEC, g as assertUniqueFindingIds, h as analystRunDigest, i as DEFAULT_TRACE_ANALYST_KINDS, m as analystFindingDigest, n as AnalystRegistry, t as buildDefaultAnalystRegistry, v as readAnalystReview, y as snapshotAnalystRun } from "./default-registry-
|
|
38
|
+
import { _ as completedAnalystReviewQuality, b as validateAnalystReviewDecisions, c as FAILURE_MODE_KIND_SPEC, g as assertUniqueFindingIds, h as analystRunDigest, i as DEFAULT_TRACE_ANALYST_KINDS, m as analystFindingDigest, n as AnalystRegistry, t as buildDefaultAnalystRegistry, v as readAnalystReview, y as snapshotAnalystRun } from "./default-registry-6QOiTzqi.js";
|
|
39
39
|
import { OUTPUT_VALUE } from "./trace-attributes.js";
|
|
40
40
|
import { r as extractUsageFromSse, t as extractUsage } from "./extract-usage-DFATEuWW.js";
|
|
41
41
|
import { n as InMemoryRawProviderSink, r as NoopRawProviderSink, t as FileSystemRawProviderSink } from "./raw-provider-sink-rB2DYxr6.js";
|
|
@@ -46,8 +46,8 @@ import { t as packageVersion$1 } from "./package-version-D7lQHt_-.js";
|
|
|
46
46
|
import { a as classifyFailure, i as INFRA_FAILURE_BLAMES, o as classifyFailureReason, r as FAILURE_BLAME, s as failureBlame, t as DEFAULT_FAILURE_REASON_RULES } from "./failure-taxonomy-CjabTATj.js";
|
|
47
47
|
import { n as runBoundedProcess, t as DEFAULT_BOUNDED_PROCESS_TIMEOUT_MS } from "./bounded-process-Cj9YjXbE.js";
|
|
48
48
|
import { t as runEvalCampaign } from "./eval-campaign-ha_aWtk7.js";
|
|
49
|
-
import "./rollout-
|
|
50
|
-
import { t as mintRolloutRows } from "./mint-
|
|
49
|
+
import "./rollout-DTunVdJE.js";
|
|
50
|
+
import { t as mintRolloutRows } from "./mint-DE6nQrX_.js";
|
|
51
51
|
import { n as pairedEvalueSequence, t as evaluateInterimReleaseConfidence } from "./sequential-CzK5DarL.js";
|
|
52
52
|
import { S as judgeFamily, _ as assertServedModels, b as CrossFamilyError, d as stripFencedJson, f as ModelSubstitutionError, h as assertServedModel, i as backoffMs, l as isTransientLlmError, m as assertCrossFamilyServed, o as costReceiptFromLlm, p as ServedCrossFamilyError, r as LlmResponseError, s as costReceiptFromLlmError, t as LlmCallError, u as maximumChargeForLlmRequest, v as checkServedModel, x as assertCrossFamily, y as servedModelAcceptable } from "./llm-client-Bhqoglar.js";
|
|
53
53
|
import { n as paidJsonChat } from "./chat-json-call-Dx_sHzqm.js";
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import { c as unscreenedRewardFields } from "./reward-nw2xZGZG.js";
|
|
2
|
-
import { f as validateRolloutLine, r as ROLLOUT_SCHEMA, u as isRolloutLine } from "./schema-
|
|
2
|
+
import { f as validateRolloutLine, r as ROLLOUT_SCHEMA, u as isRolloutLine } from "./schema-D08IIzui.js";
|
|
3
3
|
import { a as parseJson, f as isUnavailable, i as asRecord, r as readTerminalRecord, s as parseSupervisorTree } from "./terminal-record-BHocwFw1.js";
|
|
4
4
|
//#region src/supervisor-run/integrity-issues.ts
|
|
5
5
|
function evidence(path, value) {
|
|
@@ -1161,4 +1161,4 @@ function analyzeSupervisorRunIntegrity(input, options = {}) {
|
|
|
1161
1161
|
//#endregion
|
|
1162
1162
|
export { supervisorRunRolloutLines as n, SUPERVISOR_RUN_INTEGRITY_SCHEMA as r, analyzeSupervisorRunIntegrity as t };
|
|
1163
1163
|
|
|
1164
|
-
//# sourceMappingURL=integrity-
|
|
1164
|
+
//# sourceMappingURL=integrity-atFlABhR.js.map
|