@tangle-network/agent-eval 1.0.20 → 1.0.21
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 +4 -6
- package/dist/adapters/http.d.ts +2 -2
- package/dist/analyst/index.d.ts +5 -5
- package/dist/analyst/index.js +5 -5
- package/dist/{analyze-runs-DicleUI_.d.ts → analyze-runs-DYYe18Gw.d.ts} +3 -3
- package/dist/{analyze-runs-DicleUI_.d.ts.map → analyze-runs-DYYe18Gw.d.ts.map} +1 -1
- package/dist/{analyze-runs-e9vfXY--.js → analyze-runs-jLTRKuoK.js} +5 -5
- package/dist/{analyze-runs-e9vfXY--.js.map → analyze-runs-jLTRKuoK.js.map} +1 -1
- package/dist/app-grade/index.d.ts +2 -2
- package/dist/app-grade/index.js +1 -1
- package/dist/{benchmark-command-BGZD77Un.js → benchmark-command-CoNndRCk.js} +3 -3
- package/dist/{benchmark-command-BGZD77Un.js.map → benchmark-command-CoNndRCk.js.map} +1 -1
- package/dist/benchmarks/index.d.ts +4 -4
- package/dist/benchmarks/index.js +3 -3
- package/dist/builder-eval/index.js +1 -1
- package/dist/{calibration-Ze7SjhT9.js → calibration-Cf-wpqqO.js} +2 -2
- package/dist/{calibration-Ze7SjhT9.js.map → calibration-Cf-wpqqO.js.map} +1 -1
- package/dist/campaign/index.d.ts +5 -5
- package/dist/campaign/index.js +9 -9
- package/dist/{campaign-BfDMncQN.js → campaign-Bl07uIaf.js} +11 -11
- package/dist/{campaign-BfDMncQN.js.map → campaign-Bl07uIaf.js.map} +1 -1
- package/dist/{campaign-evidence-BuKlRzSA.js → campaign-evidence-NtH29ctt.js} +3 -3
- package/dist/{campaign-evidence-BuKlRzSA.js.map → campaign-evidence-NtH29ctt.js.map} +1 -1
- package/dist/cli.js +5 -5
- package/dist/{cluster-bootstrap-BYIW7QJ4.js → cluster-bootstrap-B0MTOBRU.js} +2 -2
- package/dist/{cluster-bootstrap-BYIW7QJ4.js.map → cluster-bootstrap-B0MTOBRU.js.map} +1 -1
- package/dist/contract/index.d.ts +8 -8
- package/dist/contract/index.js +10 -10
- package/dist/{decision-point-DJWievVw.js → decision-point-Di2LATNH.js} +3 -3
- package/dist/{decision-point-DJWievVw.js.map → decision-point-Di2LATNH.js.map} +1 -1
- package/dist/{default-registry-CdMtoY-s.js → default-registry-CdFbxPnr.js} +2 -2
- package/dist/{default-registry-CdMtoY-s.js.map → default-registry-CdFbxPnr.js.map} +1 -1
- package/dist/{default-registry-BpGFsSpZ.d.ts → default-registry-aPvwtzEM.d.ts} +2 -2
- package/dist/{default-registry-BpGFsSpZ.d.ts.map → default-registry-aPvwtzEM.d.ts.map} +1 -1
- package/dist/{define-agent-eval-DuQNZajG.js → define-agent-eval-BROmIwn1.js} +5 -5
- package/dist/{define-agent-eval-DuQNZajG.js.map → define-agent-eval-BROmIwn1.js.map} +1 -1
- package/dist/{define-agent-eval-DcK8mg63.d.ts → define-agent-eval-Bx3b9B01.d.ts} +6 -6
- package/dist/{define-agent-eval-DcK8mg63.d.ts.map → define-agent-eval-Bx3b9B01.d.ts.map} +1 -1
- package/dist/diagnosis/index.d.ts +1 -1
- package/dist/diagnosis/index.js +1 -1
- package/dist/{eval-campaign-LmpC_5fm.js → eval-campaign-Dq8DEKmy.js} +2 -2
- package/dist/{eval-campaign-LmpC_5fm.js.map → eval-campaign-Dq8DEKmy.js.map} +1 -1
- package/dist/{evaluation-BCejQgkV.d.ts → evaluation-CFKj_IlJ.d.ts} +2 -2
- package/dist/{evaluation-BCejQgkV.d.ts.map → evaluation-CFKj_IlJ.d.ts.map} +1 -1
- package/dist/evaluation.d.ts +1 -1
- package/dist/experiment/index.d.ts +2 -2
- package/dist/experiment/index.js +6 -6
- package/dist/fuzz.js +1 -1
- package/dist/{index-CeYLLnfi.d.ts → index-7tWvC0aW.d.ts} +16 -16
- package/dist/{index-CeYLLnfi.d.ts.map → index-7tWvC0aW.d.ts.map} +1 -1
- package/dist/{index-CzfdjM6o.d.ts → index-B62BaRzH.d.ts} +2 -2
- package/dist/{index-CzfdjM6o.d.ts.map → index-B62BaRzH.d.ts.map} +1 -1
- package/dist/{index-DYNHOTTB.d.ts → index-BlTbxHlY.d.ts} +2 -2
- package/dist/{index-DYNHOTTB.d.ts.map → index-BlTbxHlY.d.ts.map} +1 -1
- package/dist/{index-D_eAFB3T.d.ts → index-CM12lr9E.d.ts} +3 -3
- package/dist/{index-D_eAFB3T.d.ts.map → index-CM12lr9E.d.ts.map} +1 -1
- package/dist/{index-gSu9DUiw.d.ts → index-DFso1b2W.d.ts} +7 -7
- package/dist/{index-gSu9DUiw.d.ts.map → index-DFso1b2W.d.ts.map} +1 -1
- package/dist/{index-BdLmBInV.d.ts → index-DNXFzSpc.d.ts} +23 -21
- package/dist/{index-BdLmBInV.d.ts.map → index-DNXFzSpc.d.ts.map} +1 -1
- package/dist/{index-Bc8BZfzb.d.ts → index-DfqAy8ct.d.ts} +6 -6
- package/dist/{index-Bc8BZfzb.d.ts.map → index-DfqAy8ct.d.ts.map} +1 -1
- package/dist/index.d.ts +16 -16
- package/dist/index.js +14 -14
- package/dist/{judge-gate-BliY81A0.js → judge-gate-CrN0PaAL.js} +2 -2
- package/dist/{judge-gate-BliY81A0.js.map → judge-gate-CrN0PaAL.js.map} +1 -1
- package/dist/{llm-judge-Bnxuwjxl.js → llm-judge-C813vmpc.js} +9 -9
- package/dist/{llm-judge-Bnxuwjxl.js.map → llm-judge-C813vmpc.js.map} +1 -1
- package/dist/{matrix-D6zEwLvE.d.ts → matrix-C-GxIgeJ.d.ts} +2 -2
- package/dist/{matrix-D6zEwLvE.d.ts.map → matrix-C-GxIgeJ.d.ts.map} +1 -1
- package/dist/meta-eval/index.d.ts +2 -2
- package/dist/meta-eval/index.js +6 -6
- 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/{outcome-observations-BpicIVQi.js → outcome-observations-DPkH-LHQ.js} +2 -2
- package/dist/{outcome-observations-BpicIVQi.js.map → outcome-observations-DPkH-LHQ.js.map} +1 -1
- package/dist/{paired-arms-Dep5tbsw.js → paired-arms-DjMEtYI9.js} +2 -2
- package/dist/{paired-arms-Dep5tbsw.js.map → paired-arms-DjMEtYI9.js.map} +1 -1
- package/dist/{paired-promotion-decision-Nt4CkkOT.js → paired-promotion-decision-Bi9__RQS.js} +2 -2
- package/dist/{paired-promotion-decision-Nt4CkkOT.js.map → paired-promotion-decision-Bi9__RQS.js.map} +1 -1
- package/dist/{paired-promotion-decision-BNzPfXDw.d.ts → paired-promotion-decision-DJU3fZxH.d.ts} +2 -2
- package/dist/{paired-promotion-decision-BNzPfXDw.d.ts.map → paired-promotion-decision-DJU3fZxH.d.ts.map} +1 -1
- package/dist/{pareto-C445lnEH.js → pareto-fJZg8pXm.js} +3 -3
- package/dist/{pareto-C445lnEH.js.map → pareto-fJZg8pXm.js.map} +1 -1
- package/dist/pipelines/index.js +1 -1
- package/dist/{promotion-gate-DjAsmBSZ.d.ts → promotion-gate-Da0xV4sa.d.ts} +2 -2
- package/dist/{promotion-gate-DjAsmBSZ.d.ts.map → promotion-gate-Da0xV4sa.d.ts.map} +1 -1
- package/dist/{promotion-policy-CUWOv01m.js → promotion-policy-DCgftbnd.js} +4 -4
- package/dist/{promotion-policy-CUWOv01m.js.map → promotion-policy-DCgftbnd.js.map} +1 -1
- package/dist/{release-confidence-DDrkmWxT.js → release-confidence-DGtuPDxJ.js} +2 -2
- package/dist/{release-confidence-DDrkmWxT.js.map → release-confidence-DGtuPDxJ.js.map} +1 -1
- package/dist/reporting.d.ts +3 -3
- package/dist/reporting.js +4 -4
- package/dist/{researcher-D35yo0B8.d.ts → researcher-CSlc84Bp.d.ts} +2 -2
- package/dist/{researcher-D35yo0B8.d.ts.map → researcher-CSlc84Bp.d.ts.map} +1 -1
- package/dist/{reward-hacking-DRb8K6N1.js → reward-hacking-BYiDFbIl.js} +2 -2
- package/dist/{reward-hacking-DRb8K6N1.js.map → reward-hacking-BYiDFbIl.js.map} +1 -1
- package/dist/rl.d.ts +4 -4
- package/dist/rl.js +7 -7
- package/dist/rollout/index.d.ts +2 -2
- package/dist/{rubric-predictive-validity-BMS7SUyh.js → rubric-predictive-validity-DL7gJ51F.js} +2 -2
- package/dist/{rubric-predictive-validity-BMS7SUyh.js.map → rubric-predictive-validity-DL7gJ51F.js.map} +1 -1
- package/dist/{run-record-CIflWk6i.js → run-record-CFbnkE1K.js} +2 -2
- package/dist/{run-record-CIflWk6i.js.map → run-record-CFbnkE1K.js.map} +1 -1
- package/dist/{runtime-rollouts-CfRyjk60.js → runtime-rollouts-BNrm5nC_.js} +2 -2
- package/dist/{runtime-rollouts-CfRyjk60.js.map → runtime-rollouts-BNrm5nC_.js.map} +1 -1
- package/dist/search/index.d.ts +2 -2
- package/dist/search/index.js +2 -2
- package/dist/{search-kernel-Ds2zILpY.d.ts → search-kernel-CQGLbmgn.d.ts} +3 -3
- package/dist/{search-kernel-Ds2zILpY.d.ts.map → search-kernel-CQGLbmgn.d.ts.map} +1 -1
- package/dist/{search-summary-BSftVrGY.js → search-summary-sPWiYtSl.js} +4 -4
- package/dist/{search-summary-BSftVrGY.js.map → search-summary-sPWiYtSl.js.map} +1 -1
- package/dist/{sequential-CqG6aLWS.js → sequential-9TVb5Nu5.js} +3 -3
- package/dist/{sequential-CqG6aLWS.js.map → sequential-9TVb5Nu5.js.map} +1 -1
- package/dist/{skillopt-optimization-method-DYwEndON.js → skillopt-optimization-method-QO1nU8Kz.js} +6 -6
- package/dist/{skillopt-optimization-method-DYwEndON.js.map → skillopt-optimization-method-QO1nU8Kz.js.map} +1 -1
- package/dist/{statistical-heldout--0b1lgx4.d.ts → statistical-heldout-CvOwNqt9.d.ts} +4 -4
- package/dist/{statistical-heldout--0b1lgx4.d.ts.map → statistical-heldout-CvOwNqt9.d.ts.map} +1 -1
- package/dist/{statistical-heldout-D6i58CJ5.js → statistical-heldout-DLO4Ebos.js} +3 -3
- package/dist/{statistical-heldout-D6i58CJ5.js.map → statistical-heldout-DLO4Ebos.js.map} +1 -1
- package/dist/{statistics-DCZIC-qT.js → statistics-CvvC-MPv.js} +28 -21
- package/dist/statistics-CvvC-MPv.js.map +1 -0
- package/dist/statistics.d.ts +2 -2
- package/dist/statistics.js +2 -2
- package/dist/{summary-report-D_uussI3.js → summary-report-BEhqoxrA.js} +3 -3
- package/dist/{summary-report-D_uussI3.js.map → summary-report-BEhqoxrA.js.map} +1 -1
- package/dist/{summary-report-Cv1AC-iv.d.ts → summary-report-DyZquTaH.d.ts} +2 -2
- package/dist/{summary-report-Cv1AC-iv.d.ts.map → summary-report-DyZquTaH.d.ts.map} +1 -1
- package/dist/supervisor-run/index.d.ts +1 -1
- package/dist/supervisor-run/index.js +1 -1
- package/dist/{systemone-C3xyEKvG.d.ts → systemone-DVk-Crtu.d.ts} +4 -4
- package/dist/{systemone-C3xyEKvG.d.ts.map → systemone-DVk-Crtu.d.ts.map} +1 -1
- package/dist/systemone.d.ts +2 -2
- package/dist/systemone.js +1 -1
- package/dist/{tool-waste-DEhKfiIE.js → tool-waste-BqSmmPUz.js} +2 -2
- package/dist/{tool-waste-DEhKfiIE.js.map → tool-waste-BqSmmPUz.js.map} +1 -1
- package/dist/trace-repair/index.js +1 -1
- package/dist/{tree-B2y1UR8o.js → tree-C-N2kP-8.js} +4 -4
- package/dist/{tree-B2y1UR8o.js.map → tree-C-N2kP-8.js.map} +1 -1
- package/dist/{types-BwF--ZRz.d.ts → types-BsdcSWnf.d.ts} +2 -2
- package/dist/{types-BwF--ZRz.d.ts.map → types-BsdcSWnf.d.ts.map} +1 -1
- package/dist/{types-BAJ2Hb-d.d.ts → types-DUVhW5V1.d.ts} +2 -2
- package/dist/{types-BAJ2Hb-d.d.ts.map → types-DUVhW5V1.d.ts.map} +1 -1
- package/dist/{types-CeMZYsNF.d.ts → types-ip2D3-Ib.d.ts} +2 -2
- package/dist/{types-CeMZYsNF.d.ts.map → types-ip2D3-Ib.d.ts.map} +1 -1
- package/dist/wire/index.d.ts +2 -2
- package/docs/design/statistics-decisions.md +7 -0
- package/package.json +1 -1
- package/dist/statistics-DCZIC-qT.js.map +0 -1
package/CHANGELOG.md
CHANGED
|
@@ -4,6 +4,10 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
|
|
|
4
4
|
|
|
5
5
|
---
|
|
6
6
|
|
|
7
|
+
## [1.0.21] — 2026-11-06
|
|
8
|
+
|
|
9
|
+
- Add `independentRiskDifference`: Newcombe/Wilson intervals for independent before/after counts, with explicit uncertainty for empty arms.
|
|
10
|
+
|
|
7
11
|
## [1.0.20] — 2026-11-06
|
|
8
12
|
|
|
9
13
|
- docs(statistics): trim whitespace in interval guidance
|
|
@@ -89,19 +93,13 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
|
|
|
89
93
|
## [1.0.11] — 2026-10-30
|
|
90
94
|
|
|
91
95
|
- fix(statistics): export summarizeDistribution from the package root
|
|
92
|
-
- chore(release): 1.0.10
|
|
93
|
-
- chore(release): 1.0.9
|
|
94
96
|
- feat(statistics): summarizeDistribution + percentileNearestRank (consolidation home)
|
|
95
|
-
- chore(release): 1.0.8
|
|
96
97
|
- chore(analyst): re-pin the implementation manifest on merged main
|
|
97
98
|
- feat(statistics,harness): two-sample Welch evidence + lifted sandbox assertion library
|
|
98
|
-
- chore(release): 1.0.7 (#1172)
|
|
99
99
|
- docs(statistics): keep inference policy and exports readable (#1173)
|
|
100
100
|
- feat(statistics): own independent effect summaries and ROC area (#1171)
|
|
101
|
-
- chore(release): 1.0.6 (#1170)
|
|
102
101
|
- refactor(reporting): share text table layout and simplify summaries (#1169)
|
|
103
102
|
- refactor(harness): retain unavailable verdicts through cleanup (#1168)
|
|
104
|
-
- chore(release): 1.0.5 (#1167)
|
|
105
103
|
- refactor(harness): normalize compact named expectations (#1166)
|
|
106
104
|
- chore(release): 1.0.4 (#1165)
|
|
107
105
|
- refactor(harness): own resource acquisition and readiness cleanup (#1164)
|
package/dist/adapters/http.d.ts
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
import { d as CostReceipt } from "../cost-ledger-Czro9wmo.js";
|
|
2
|
-
import { R as Scenario, d as DispatchContext, f as DispatchFn } from "../types-
|
|
3
|
-
import "../index-
|
|
2
|
+
import { R as Scenario, d as DispatchContext, f as DispatchFn } from "../types-ip2D3-Ib.js";
|
|
3
|
+
import "../index-DFso1b2W.js";
|
|
4
4
|
export interface HttpDispatchOptions<TScenario extends Scenario, _TArtifact> {
|
|
5
5
|
/** Static endpoint URL. Mutually exclusive with `resolveUrl`. */
|
|
6
6
|
url?: string;
|
package/dist/analyst/index.d.ts
CHANGED
|
@@ -1,19 +1,19 @@
|
|
|
1
1
|
import { m as CustomTokenPricing, o as CostLedgerHandle } from "../cost-ledger-Czro9wmo.js";
|
|
2
2
|
import { A as FindingParseOutcome, B as FindingTreatment, C as CalibrationInterval, D as FindingLayer, E as FindingEvidenceUnit, F as independentUnitCount, H as TreatmentLayer, I as lineageRoots, L as parseAnalystFinding, M as FindingScope, N as FindingScopeReading, O as FindingMeasure, P as ProposalFindingSchema, R as readFindingScope, S as makeProposalFinding, T as FindingCalibration, V as TREATMENT_LAYERS, Y as TraceAnalysisStore, _ as ExecutionProbeRequest, a as AnalystFindingSchemaVersion, b as computeFindingId, c as AnalystRunEvent, d as AnalystRunSummary, f as AnalystSeverity, g as ExecutionProbeOutcome, h as ExecutionProbe, i as AnalystFinding, j as FindingReaderLabel, k as FindingMethod, l as AnalystRunInputs, m as EvidenceRef, n as AnalystContext, o as AnalystInputKind, ot as TraceAnalystSpan, p as AnalystUsageReceipt, r as AnalystCost, s as AnalystRequirements, t as Analyst, u as AnalystRunResult, v as ProposalFinding, w as FINDING_LAYERS, x as makeFinding, y as ProposalFindingOrigin, z as FindingConditions } from "../types-kBMNdo1X.js";
|
|
3
3
|
import { A as ChatResponse, D as ChatCallOpts, F as OpenAiCompatibleTransportOpts, I as SandboxSdkTransportOpts, L as createChatClient, M as CreateChatClientOpts, N as CustomTransportOpts, O as ChatClient, P as MockTransportOpts, j as ChatTransport, k as ChatRequest, t as BootstrapInterval } from "../cluster-Cd0MRbr9.js";
|
|
4
|
-
import {
|
|
4
|
+
import { Ft as NominalAgreementOptions } from "../index-DNXFzSpc.js";
|
|
5
5
|
import { A as traceStoreEvidenceResolver, C as AnalystEvidenceResolutionError, D as AnalystLatencyDistribution, E as AnalystIssueExpectation, G as ExternalOptimizerRunnerCommand, H as ExternalOptimizerModelExecutionObservation, O as RunAnalystBenchmarkOptions, S as AnalystEvidenceResolution, T as AnalystFindingScore, _ as AnalystBenchmarkProvenance, b as AnalystBenchmarkSummary, d as AnalystBenchmarkDatasetRef, f as AnalystBenchmarkDescriptor, g as AnalystBenchmarkOutput, h as AnalystBenchmarkObservation, k as runAnalystBenchmark, m as AnalystBenchmarkLabelState, p as AnalystBenchmarkError, t as AgentProfile, u as AnalystBenchmarkCase, v as AnalystBenchmarkResult, w as AnalystEvidenceResolver, x as AnalystEvidenceExpectation, y as AnalystBenchmarkRunner, z as ExternalOptimizerModelCall } from "../agent-profile-BO8ruI6R.js";
|
|
6
|
-
import { C as BehavioralWindowPolicy, S as BehavioralTraceWindow, _ as BehavioralTraceEventMap, a as createTraceAnalyst, b as BehavioralTraceQuestionsOptions, c as BehavioralAnalystOptions, d as pinBehavioralFailureClass, f as BEHAVIORAL_TRACE_QUESTIONS, g as BehavioralTraceEvent, h as BehavioralTraceAnswer, i as TraceAnalystDefinition, l as assertFailureClasses, m as BehavioralQuestionCalibration, n as buildDefaultAnalystRegistry, o as runTraceAnalyst, p as BEHAVIORAL_TRACE_QUESTIONS_VERSION, r as CreateTraceAnalystOptions, s as BEHAVIORAL_FAILURE_CLASS, t as DefaultAnalystRegistryOptions, u as behavioralAnalyst, v as BehavioralTraceLayer, w as behavioralTraceQuestionDigest, x as BehavioralTraceSupport, y as BehavioralTraceQuestionId } from "../default-registry-
|
|
6
|
+
import { C as BehavioralWindowPolicy, S as BehavioralTraceWindow, _ as BehavioralTraceEventMap, a as createTraceAnalyst, b as BehavioralTraceQuestionsOptions, c as BehavioralAnalystOptions, d as pinBehavioralFailureClass, f as BEHAVIORAL_TRACE_QUESTIONS, g as BehavioralTraceEvent, h as BehavioralTraceAnswer, i as TraceAnalystDefinition, l as assertFailureClasses, m as BehavioralQuestionCalibration, n as buildDefaultAnalystRegistry, o as runTraceAnalyst, p as BEHAVIORAL_TRACE_QUESTIONS_VERSION, r as CreateTraceAnalystOptions, s as BEHAVIORAL_FAILURE_CLASS, t as DefaultAnalystRegistryOptions, u as behavioralAnalyst, v as BehavioralTraceLayer, w as behavioralTraceQuestionDigest, x as BehavioralTraceSupport, y as BehavioralTraceQuestionId } from "../default-registry-aPvwtzEM.js";
|
|
7
7
|
import { a as ExactAnalystRunPolicySnapshot, c as ExactAnalystSnapshot, d as ExactExecutionComponentSnapshot, i as ExactAnalystRunEvent, l as ExactCapableAnalyst, n as ExactAnalystExecutionPlanSnapshot, o as ExactAnalystRunResult, r as ExactAnalystRunCompletion, s as ExactAnalystRunSummary, t as ExactAnalystBudgetSnapshot, u as ExactExecutionComponentIdentity } from "../exact-types-VAFjtOQN.js";
|
|
8
8
|
import { g as SystemOneState } from "../systemone-protocol-C4t7JH4Y.js";
|
|
9
|
-
import { B as DiffPolicy, F as FindingsTimeStep, H as FindingsStore, I as LedgerFinding, L as diffFindingsOverTime, M as FindingsLedgerOutcome, N as FindingsLedgerRead, P as FindingsQuery, R as openFindingsLedger, U as PersistedFinding, V as FindingsDiff, W as diffFindings, j as FindingsLedger, r as SystemOneEvaluatorOptions, z as queryFindings } from "../systemone-
|
|
9
|
+
import { B as DiffPolicy, F as FindingsTimeStep, H as FindingsStore, I as LedgerFinding, L as diffFindingsOverTime, M as FindingsLedgerOutcome, N as FindingsLedgerRead, P as FindingsQuery, R as openFindingsLedger, U as PersistedFinding, V as FindingsDiff, W as diffFindings, j as FindingsLedger, r as SystemOneEvaluatorOptions, z as queryFindings } from "../systemone-DVk-Crtu.js";
|
|
10
10
|
import { a as RAW_FINDING_SCHEMA_PROMPT, c as RawAnalystFindingSchema, i as TraceAnalystLimits, l as evidenceRefsFromRawFinding, n as TraceAnalysisEngineRequest, o as RawAnalystEvidence, r as TraceAnalysisEngineResult, s as RawAnalystFinding, t as TraceAnalysisEngine, u as parseRawFinding } from "../engine-BF8naBn7.js";
|
|
11
11
|
import { n as buildTraceToolsForGroup, t as TraceToolGroupName } from "../tool-groups-R5RO9OHn.js";
|
|
12
12
|
import { a as BudgetPolicy, c as RegistryRunOpts, i as AnalystRegistryOptions, n as ExactAnalystRunExecutionError, o as ExactAnalystBudgetPolicy, r as AnalystHooks, s as ExactRegistryRunOpts, t as AnalystRegistry } from "../registry-BAcEbzKm.js";
|
|
13
|
-
import { T as SemanticConceptJudgeOptions, _ as DspyRlmTraceEngineOptions, a as FAILURE_MODE_KIND_SPEC, b as createChatTraceEngine, c as emitControlIntegrityFindings, d as FindingSubject, f as FindingSubjectKind, g as renderFindingSubject, h as parseFindingSubject, i as IMPROVEMENT_KIND_SPEC, l as FINDING_SUBJECT_KINDS, m as findingSubjectGrammarPromptFor, n as KNOWLEDGE_POISONING_KIND_SPEC, o as CONTROL_INTEGRITY_ANALYST, p as KIND_EXPECTED_SUBJECTS, r as KNOWLEDGE_GAP_KIND_SPEC, s as ControlIntegrityAnalyst, t as DEFAULT_TRACE_ANALYST_KINDS, u as FINDING_SUBJECT_SYNTAX, v as createDspyRlmTraceEngine, y as ChatTraceEngineOptions } from "../index-
|
|
13
|
+
import { T as SemanticConceptJudgeOptions, _ as DspyRlmTraceEngineOptions, a as FAILURE_MODE_KIND_SPEC, b as createChatTraceEngine, c as emitControlIntegrityFindings, d as FindingSubject, f as FindingSubjectKind, g as renderFindingSubject, h as parseFindingSubject, i as IMPROVEMENT_KIND_SPEC, l as FINDING_SUBJECT_KINDS, m as findingSubjectGrammarPromptFor, n as KNOWLEDGE_POISONING_KIND_SPEC, o as CONTROL_INTEGRITY_ANALYST, p as KIND_EXPECTED_SUBJECTS, r as KNOWLEDGE_GAP_KIND_SPEC, s as ControlIntegrityAnalyst, t as DEFAULT_TRACE_ANALYST_KINDS, u as FINDING_SUBJECT_SYNTAX, v as createDspyRlmTraceEngine, y as ChatTraceEngineOptions } from "../index-CM12lr9E.js";
|
|
14
14
|
import { C as runPrimeExchange, D as ReplyRowDecoded, E as ReplyEnvelope, O as ReplyRowRejection, S as projectPrimeTrajectory, T as ReplyContract, _ as buildPrimePrompt, a as PrimeProjectionOutcome, b as normalizePrimeUsage, c as PrimeProtocolIdentity, d as PrimeRepairPromptSpec, f as PrimeRepairState, g as analystUsageReceiptFromPrimeUsage, h as PrimeTurnRecord, i as PrimeProjectionDelivery, l as PrimeRawUsage, m as PrimeRowDecoded, n as PrimeExchangeOutcome, o as PrimeProjectionSource, p as PrimeReplyContract, r as PrimeFailure, s as PrimePromptSpec, t as PrimeExchangeOptions, u as PrimeRejectedRow, v as buildPrimeRepairPrompt, w as DecodedReply, x as primeProtocolSha256, y as mergePrimeRawUsage } from "../prime-protocol-D1NkJRuD.js";
|
|
15
15
|
import { i as nodeHttpPrimeBridgeTransport, n as PrimeBridgeTransportRequest, r as PrimeBridgeTransportResult, t as PrimeBridgeTransport } from "../prime-bridge-transport-Cfqir_Im.js";
|
|
16
|
-
import { b as SupervisorRunSources } from "../types-
|
|
16
|
+
import { b as SupervisorRunSources } from "../types-DUVhW5V1.js";
|
|
17
17
|
import { a as askTraceQuestions, i as VerifiedTraceFinding, n as TraceQuestionOutcome, r as TraceQuestionSpec, t as AskTraceQuestionsOptions } from "../trace-questions-BY6taxnl.js";
|
|
18
18
|
import { z } from "zod";
|
|
19
19
|
interface SemanticConceptJudgeAdapterOpts {
|
package/dist/analyst/index.js
CHANGED
|
@@ -1,20 +1,20 @@
|
|
|
1
1
|
import { n as assertConsistentFindingConditions, o as mergeFindingComposition, r as findingCompositionKey, t as TREATMENT_LAYERS } from "../finding-composition-Belh_qU3.js";
|
|
2
|
-
import {
|
|
2
|
+
import { G as nominalAgreement, V as rocAuc, gt as clusterBootstrap } from "../statistics-CvvC-MPv.js";
|
|
3
3
|
import { t as FileLedgerJournal } from "../journal-CTX5uK-n.js";
|
|
4
4
|
import { r as merkleRoot } from "../ledger-core-24IxvmVY.js";
|
|
5
5
|
import { A as spanningWindow, C as UNKNOWN_LINEAGE, D as mostSevere, E as mergeEvidenceUnits, O as parseAnalystFinding, S as ProposalFindingSchema, T as lineageRoots, b as AnalystFindingSchema, k as readFindingScope, w as independentUnitCount, x as FINDING_LAYERS } from "../external-optimizer-subprocess-_C_2yriN.js";
|
|
6
6
|
import { a as makeProposalFinding, i as makeFinding, r as computeFindingId } from "../types-CL1_N7AZ.js";
|
|
7
7
|
import { t as assertProposalFindings } from "../proposal-findings-DWs94yT8.js";
|
|
8
8
|
import { A as RawAnalystFindingSchema, B as parseFindingSubject, F as stripCodeFences, I as FINDING_SUBJECT_KINDS, L as FINDING_SUBJECT_SYNTAX, M as parseRawFinding, N as parseTraceSpanEvidenceUri, R as KIND_EXPECTED_SUBJECTS, V as renderFindingSubject, j as evidenceRefsFromRawFinding, k as RAW_FINDING_SCHEMA_PROMPT, n as runTraceAnalyst, r as buildTraceToolsForGroup, t as createTraceAnalyst, z as findingSubjectGrammarPromptFor } from "../kind-factory-CV0FNCMM.js";
|
|
9
|
-
import { _ as behavioralTraceSupportSchema, a as KNOWLEDGE_POISONING_KIND_SPEC, b as behavioralTraceQuestionDigest, c as FAILURE_MODE_KIND_SPEC, d as emitControlIntegrityFindings, f as BEHAVIORAL_FAILURE_CLASS, g as validBehavioralCalibration, h as pinBehavioralFailureClass, i as DEFAULT_TRACE_ANALYST_KINDS, l as CONTROL_INTEGRITY_ANALYST, m as behavioralAnalyst, n as AnalystRegistry, o as KNOWLEDGE_GAP_KIND_SPEC, p as assertFailureClasses, r as ExactAnalystRunExecutionError, s as IMPROVEMENT_KIND_SPEC, t as buildDefaultAnalystRegistry, u as ControlIntegrityAnalyst, v as BEHAVIORAL_TRACE_QUESTIONS, y as BEHAVIORAL_TRACE_QUESTIONS_VERSION } from "../default-registry-
|
|
9
|
+
import { _ as behavioralTraceSupportSchema, a as KNOWLEDGE_POISONING_KIND_SPEC, b as behavioralTraceQuestionDigest, c as FAILURE_MODE_KIND_SPEC, d as emitControlIntegrityFindings, f as BEHAVIORAL_FAILURE_CLASS, g as validBehavioralCalibration, h as pinBehavioralFailureClass, i as DEFAULT_TRACE_ANALYST_KINDS, l as CONTROL_INTEGRITY_ANALYST, m as behavioralAnalyst, n as AnalystRegistry, o as KNOWLEDGE_GAP_KIND_SPEC, p as assertFailureClasses, r as ExactAnalystRunExecutionError, s as IMPROVEMENT_KIND_SPEC, t as buildDefaultAnalystRegistry, u as ControlIntegrityAnalyst, v as BEHAVIORAL_TRACE_QUESTIONS, y as BEHAVIORAL_TRACE_QUESTIONS_VERSION } from "../default-registry-CdFbxPnr.js";
|
|
10
10
|
import { t as createDspyRlmTraceEngine } from "../dspy-rlm-engine-cWyn8G1a.js";
|
|
11
11
|
import { n as diffFindings, r as createChatTraceEngine, t as FindingsStore } from "../findings-store-BSuaV3dN.js";
|
|
12
|
-
import { m as validDecisionIdentity, o as systemOneAnalyst, t as DECISION_POINT_AUC_BAR } from "../decision-point-
|
|
12
|
+
import { m as validDecisionIdentity, o as systemOneAnalyst, t as DECISION_POINT_AUC_BAR } from "../decision-point-Di2LATNH.js";
|
|
13
13
|
import { t as createChatClient } from "../chat-client-Zz97O3NB.js";
|
|
14
|
-
import { A as agentRxBenchmarkCase, C as ANALYST_BENCHMARK_COST_LEDGER_FILE, D as AGENT_RX_UPSTREAM_REVISION, E as ANALYST_BENCHMARK_OBSERVATIONS_FILE, F as runAnalystBenchmark, I as traceStoreEvidenceResolver, M as normalizeAgentRxCategory, N as externalId, O as codeTraceBenchCase, P as positiveStep, S as ANALYST_BENCHMARK_IMPLEMENTATION_FILES, T as ANALYST_BENCHMARK_MANIFEST_FILE, _ as readAnalystBenchmarkArtifact, a as runInlineAnalystDefinition, b as ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, c as publicRlmAnalystDefinition, d as publicDirectAnalystDefinition, f as runChunkedAnalystDefinition, g as expandCodeTraceFailureBlocks, h as analystDefinitionProtocolSha256, i as primeCodeTraceAnalystDefinition, j as agentRxPredictionsToFindings, k as codeTracerPredictionsToFindings, l as runReplVariableAnalystDefinition, m as analystDefinitionAsymmetries, n as createPrimeBenchmarkRunner, o as loadPublicBenchmarkRows, p as AnalystExpressivenessError, r as primeAnalystProtocolSha256, s as createPublicBenchmarkRlmRunner, t as runAnalystBenchmarkCommand, u as createPublicBenchmarkDirectRunner, v as CODE_TRACE_BENCH_ANALYST_PROMPT, w as ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, x as ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, y as publicBenchmarkProtocolSha256 } from "../benchmark-command-
|
|
14
|
+
import { A as agentRxBenchmarkCase, C as ANALYST_BENCHMARK_COST_LEDGER_FILE, D as AGENT_RX_UPSTREAM_REVISION, E as ANALYST_BENCHMARK_OBSERVATIONS_FILE, F as runAnalystBenchmark, I as traceStoreEvidenceResolver, M as normalizeAgentRxCategory, N as externalId, O as codeTraceBenchCase, P as positiveStep, S as ANALYST_BENCHMARK_IMPLEMENTATION_FILES, T as ANALYST_BENCHMARK_MANIFEST_FILE, _ as readAnalystBenchmarkArtifact, a as runInlineAnalystDefinition, b as ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, c as publicRlmAnalystDefinition, d as publicDirectAnalystDefinition, f as runChunkedAnalystDefinition, g as expandCodeTraceFailureBlocks, h as analystDefinitionProtocolSha256, i as primeCodeTraceAnalystDefinition, j as agentRxPredictionsToFindings, k as codeTracerPredictionsToFindings, l as runReplVariableAnalystDefinition, m as analystDefinitionAsymmetries, n as createPrimeBenchmarkRunner, o as loadPublicBenchmarkRows, p as AnalystExpressivenessError, r as primeAnalystProtocolSha256, s as createPublicBenchmarkRlmRunner, t as runAnalystBenchmarkCommand, u as createPublicBenchmarkDirectRunner, v as CODE_TRACE_BENCH_ANALYST_PROMPT, w as ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, x as ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, y as publicBenchmarkProtocolSha256 } from "../benchmark-command-CoNndRCk.js";
|
|
15
15
|
import { a as normalizePrimeUsage, c as runPrimeExchange, i as mergePrimeRawUsage, n as buildPrimePrompt, o as primeProtocolSha256, r as buildPrimeRepairPrompt, s as projectPrimeTrajectory, t as analystUsageReceiptFromPrimeUsage } from "../prime-protocol-CrhpXRz9.js";
|
|
16
16
|
import { t as nodeHttpPrimeBridgeTransport } from "../prime-bridge-transport-Csd5SQg7.js";
|
|
17
|
-
import { t as clusterBootstrapMean } from "../cluster-bootstrap-
|
|
17
|
+
import { t as clusterBootstrapMean } from "../cluster-bootstrap-B0MTOBRU.js";
|
|
18
18
|
import { z } from "zod";
|
|
19
19
|
import { existsSync, readFileSync, readdirSync, statSync } from "node:fs";
|
|
20
20
|
import { join } from "node:path";
|
|
@@ -1,8 +1,8 @@
|
|
|
1
1
|
import { o as FailureClass } from "./schema-CkIvnmLU.js";
|
|
2
2
|
import { a as RunRecord, u as RunTerminalOutcome } from "./run-record-D2BFMPGl.js";
|
|
3
|
-
import {
|
|
3
|
+
import { Ht as ContinuousAgreement } from "./index-DNXFzSpc.js";
|
|
4
4
|
import { r as DatasetScenario } from "./dataset-DThC1S6G.js";
|
|
5
|
-
import { f as ParetoFigureSpec, l as GainDistributionBin } from "./summary-report-
|
|
5
|
+
import { f as ParetoFigureSpec, l as GainDistributionBin } from "./summary-report-DyZquTaH.js";
|
|
6
6
|
import { t as AnalystRegistry } from "./registry-BAcEbzKm.js";
|
|
7
7
|
interface InsightReport {
|
|
8
8
|
/** Number of runs analyzed. */
|
|
@@ -480,4 +480,4 @@ interface AnalyzeRunsOptions {
|
|
|
480
480
|
}
|
|
481
481
|
declare function analyzeRuns(opts: AnalyzeRunsOptions): Promise<InsightReport>;
|
|
482
482
|
export { ReleaseSummary as _, summarizeExecution as a, ExecutionInsight as c, InsightReport as d, InterRaterInsight as f, Recommendation as g, OutcomeCorrelationInsight as h, SummarizeExecutionOptions as i, FailureClassTally as l, LiftInsight as m, analyzeRuns as n, CostProvenanceSummary as o, JudgeInsight as p, ExecutionReport as r, ExecutionErrorOutcomeCell as s, AnalyzeRunsOptions as t, FailureClusterInsight as u, ScalarDistribution as v, TokenUsageInsight as y };
|
|
483
|
-
//# sourceMappingURL=analyze-runs-
|
|
483
|
+
//# sourceMappingURL=analyze-runs-DYYe18Gw.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"analyze-runs-
|
|
1
|
+
{"version":3,"file":"analyze-runs-DYYe18Gw.d.ts","names":[],"sources":["../src/contract/insight-report.ts","../src/contract/execution-insight.ts","../src/contract/analyze-runs.ts"],"mappings":";;;;;;UAqCiB;;EAEf;;;;EAKA,WAAW;;EAGX,WAAW;;;EAIX,cAAc,eAAe;;EAG7B;IACE,MAAM;IACN,QAAQ;;;;;IAKR,aAAa;;;;;IAKb;MAAa;MAAe;;;;;;EAM9B,QAAQ,eAAe;;;;EAKvB,aAAa;;;;EAKb,OAAO;;;EAIP,kBAAkB;;;EAIlB,gBAAgB;;;;;EAMhB,qBAAqB;;;;;EAMrB,SAAS;;;;;;EAOT,wBAAwB;;;;EAKxB,iBAAiB;;;EAIjB,iBAAiB;;UAGF;EACf;IAAY;IAAW;;EACvB;IAAa;IAAW;;;;EAGxB;IAAc;IAAW;;EACzB;IAAc;;EACd;;UAGe;;EAEf,YAAY;;EAEZ,SAAS;;;EAGT,YAAY;;;EAGZ;IACE;IACA,YAAY;IACZ,SAAS;IACT;;;EAGF,QAAQ;IAAQ;IAAe;;;;;EAI/B;IACE;IACA;IACA;;;;;EAKF;IACE;;;IAGA;;IAEA;;IAEA;;IAEA;;IAEA;;;;;IAKA,mBAAmB,OAAO,oBAAoB;;;;EAIhD;IACE;IACA;IACA;IACA;IACA;;;UAIa;;EAEf;;EAEA;;EAEA;;UAGe;EACf,OAAO;EACP,QAAQ;EACR,WAAW;EACX,QAAQ;EACR,YAAY;EACZ;IACE;IACA;IACA;IACA;IACA;;;;UAOa;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,WAAW;;;;;EAKX,WAAW;IAAQ;IAAe;;;UAGnB;;EAEf;;EAEA;;;EAGA,cAAc;;;EAGd;;;EAGA;;;EAGA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,SAAS;;EAET,mBAAmB;IACjB;IACA,SAAS;MAAQ;MAAe;;IAChC;;;UAIa;EACf;EACA;;EAEA;;EAEA;;;EAGA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;EAWA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;;EAEf,UAAU;IACR;IACA;;IAEA;;IAEA;;IAEA;;EAEF;;;;;UAMe;;EAEf,cAAc;;EAEd;;EAEA;;UAGe;;EAEf;;;EAGA;EACA,UAAU;IAAQ;IAAe;IAAgB;;;UAGlC;;;EAGf;;EAEA;;EAEA;;EAEA;;EAEA;IACE;IACA;IACA;;;UAIa;;;EAGf;EACA,MAAM;IACJ;IACA;IACA;;;;EAIF;;UAGQ;;EAER;;EAEA;;;;EAIA;;EAEA;EACA;;KAGU,eACP;EACC;;EAEA;;EAEA;;EAEA;;EAEA;MAED;EACC;;EAEA;EACA;EACA;EACA;;UAGW;;EAEf;EACA;;EAEA;;;;EAIA,SAAS,eAAe;;;EAGxB;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;;EAEA;;UCzbe;EACf,MAAM;EACN;;UAGe;EACf,WAAW;EACX,gBAAgB;;;iBAIF,mBAAmB,MAAM,4BAA4B;UC4BpD;;EAEf,MAAM;;;EAGN;;;;;EAKA;EACA;;;EAGA,8BAA8B;;;EAG9B,kBAAkB;;;EAGlB,UAAU;;;;EAIV;IACE;IACA,cAAc;;;;;EAKhB,cAAc;IAAQ;IAAe;IAAe;;;EAEpD;;;;EAIA;;;;;;;;;EASA,eAAe;;;EAGf;;iBAGoB,YAAY,MAAM,qBAAqB,QAAQ"}
|
|
@@ -1,10 +1,10 @@
|
|
|
1
1
|
import { r as observedSplitScore } from "./reward-DqMfaEax.js";
|
|
2
2
|
import { c as validateRunRecord } from "./run-record-CpLS4Uid.js";
|
|
3
|
-
import { r as paretoChart } from "./summary-report-
|
|
3
|
+
import { r as paretoChart } from "./summary-report-BEhqoxrA.js";
|
|
4
4
|
import { t as interpolatedQuantile } from "./quantile-CncXP-Ss.js";
|
|
5
|
-
import { r as pairRunRecords } from "./paired-arms-
|
|
6
|
-
import {
|
|
7
|
-
import { n as aggregatePairedHoldout } from "./statistical-heldout-
|
|
5
|
+
import { r as pairRunRecords } from "./paired-arms-DjMEtYI9.js";
|
|
6
|
+
import { I as cohensD, J as continuousAgreement, N as studentTTwoSidedPValue, R as pairedCohensDz, c as pairedMde, ct as mean$1, g as pairedTTest, l as requiredPairedSampleSize, p as pairedBootstrap, rt as spearmanR, tt as pearsonR, ut as meanOrZero, yt as studentTQuantile } from "./statistics-CvvC-MPv.js";
|
|
7
|
+
import { n as aggregatePairedHoldout } from "./statistical-heldout-DLO4Ebos.js";
|
|
8
8
|
import "./query-CGXzO9d4.js";
|
|
9
9
|
// src/baseline.ts
|
|
10
10
|
/** Detect baseline regressions with Welch inference, effect size and per-side stability. */
|
|
@@ -1059,4 +1059,4 @@ async function analyzeRuns(opts) {
|
|
|
1059
1059
|
}
|
|
1060
1060
|
export { welchsTTest as a, iqr as i, checkCanaries as n, summarizeExecution as r, analyzeRuns as t };
|
|
1061
1061
|
|
|
1062
|
-
//# sourceMappingURL=analyze-runs-
|
|
1062
|
+
//# sourceMappingURL=analyze-runs-jLTRKuoK.js.map
|