@tangle-network/agent-eval 0.215.0 → 0.215.1
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 -0
- package/dist/adapters/http.d.ts +3 -3
- package/dist/{agent-profile-BGyt2z8U.d.ts → agent-profile-DmzL8KaJ.d.ts} +5 -5
- package/dist/{agent-profile-BGyt2z8U.d.ts.map → agent-profile-DmzL8KaJ.d.ts.map} +1 -1
- package/dist/analyst/index.d.ts +13 -13
- package/dist/analyst/index.js +7 -7
- package/dist/{analyze-runs-BlslTkOJ.d.ts → analyze-runs-BhFLOztS.d.ts} +4 -4
- package/dist/{analyze-runs-BlslTkOJ.d.ts.map → analyze-runs-BhFLOztS.d.ts.map} +1 -1
- package/dist/{benchmark-command-CMmeV-_Z.js → benchmark-command-CgpGi7vY.js} +9 -9
- package/dist/{benchmark-command-CMmeV-_Z.js.map → benchmark-command-CgpGi7vY.js.map} +1 -1
- package/dist/benchmarks/index.d.ts +5 -5
- package/dist/benchmarks/index.js +3 -3
- package/dist/campaign/index.d.ts +7 -7
- package/dist/campaign/index.js +4 -4
- package/dist/{campaign-Cuw3DQKF.js → campaign-CpnsCgfz.js} +7 -7
- package/dist/{campaign-Cuw3DQKF.js.map → campaign-CpnsCgfz.js.map} +1 -1
- package/dist/{chat-client-Bnx8NZe4.js → chat-client-DthPOYGd.js} +2 -2
- package/dist/{chat-client-Bnx8NZe4.js.map → chat-client-DthPOYGd.js.map} +1 -1
- package/dist/{chat-json-call-CZI1NoGZ.js → chat-json-call-BrKsyspg.js} +2 -2
- package/dist/{chat-json-call-CZI1NoGZ.js.map → chat-json-call-BrKsyspg.js.map} +1 -1
- package/dist/cli.js +3 -3
- package/dist/contract/index.d.ts +11 -11
- package/dist/contract/index.js +7 -7
- package/dist/{cost-ledger-DbQdN3nO.d.ts → cost-ledger-CO7UyEcQ.d.ts} +3 -8
- package/dist/{cost-ledger-DbQdN3nO.d.ts.map → cost-ledger-CO7UyEcQ.d.ts.map} +1 -1
- package/dist/{cost-ledger-DFOeLeJi.js → cost-ledger-D2FnJPBF.js} +27 -23
- package/dist/cost-ledger-D2FnJPBF.js.map +1 -0
- package/dist/{default-registry-BAiTePcB.d.ts → default-registry-B4B7GMvI.d.ts} +6 -6
- package/dist/{default-registry-BAiTePcB.d.ts.map → default-registry-B4B7GMvI.d.ts.map} +1 -1
- package/dist/{default-registry-BhYGMVbV.js → default-registry-KP4-Tolz.js} +3 -3
- package/dist/{default-registry-BhYGMVbV.js.map → default-registry-KP4-Tolz.js.map} +1 -1
- package/dist/{define-agent-eval-FGtpqnh4.js → define-agent-eval-BdfnqXeP.js} +3 -3
- package/dist/{define-agent-eval-FGtpqnh4.js.map → define-agent-eval-BdfnqXeP.js.map} +1 -1
- package/dist/{define-agent-eval-CZdBPUkk.d.ts → define-agent-eval-DfXvNho5.d.ts} +8 -8
- package/dist/{define-agent-eval-CZdBPUkk.d.ts.map → define-agent-eval-DfXvNho5.d.ts.map} +1 -1
- package/dist/diagnosis/index.d.ts +2 -2
- package/dist/{dspy-rlm-engine-OMBXVN0u.js → dspy-rlm-engine-DZfHM7kT.js} +3 -3
- package/dist/{dspy-rlm-engine-OMBXVN0u.js.map → dspy-rlm-engine-DZfHM7kT.js.map} +1 -1
- package/dist/{engine-BzOR5Ygn.d.ts → engine-C37HS2kr.d.ts} +5 -5
- package/dist/{engine-BzOR5Ygn.d.ts.map → engine-C37HS2kr.d.ts.map} +1 -1
- package/dist/evaluation.d.ts +3 -3
- package/dist/evaluation.js +1 -1
- package/dist/{exact-types-DjhfZglz.d.ts → exact-types-Dlb73wN4.d.ts} +2 -2
- package/dist/{exact-types-DjhfZglz.d.ts.map → exact-types-Dlb73wN4.d.ts.map} +1 -1
- package/dist/experiment/index.d.ts +1 -1
- package/dist/{external-optimizer-process-DCZx5PdG.js → external-optimizer-process-GngofQML.js} +2 -2
- package/dist/{external-optimizer-process-DCZx5PdG.js.map → external-optimizer-process-GngofQML.js.map} +1 -1
- package/dist/{external-optimizer-subprocess-CA3aXT8I.js → external-optimizer-subprocess-BUxifb8g.js} +3 -3
- package/dist/{external-optimizer-subprocess-CA3aXT8I.js.map → external-optimizer-subprocess-BUxifb8g.js.map} +1 -1
- package/dist/{feedback-trajectory-CX01700m.d.ts → feedback-trajectory-BLN_6fEs.d.ts} +3 -3
- package/dist/{feedback-trajectory-CX01700m.d.ts.map → feedback-trajectory-BLN_6fEs.d.ts.map} +1 -1
- package/dist/fuzz.d.ts +1 -1
- package/dist/fuzz.js +1 -1
- package/dist/{index-D_rfXep9.d.ts → index-B3WBm953.d.ts} +10 -10
- package/dist/{index-D_rfXep9.d.ts.map → index-B3WBm953.d.ts.map} +1 -1
- package/dist/{index-BTwxP7L0.d.ts → index-B7m2uzJL.d.ts} +10 -10
- package/dist/{index-BTwxP7L0.d.ts.map → index-B7m2uzJL.d.ts.map} +1 -1
- package/dist/{index-D0_l25We.d.ts → index-CPJJr98o.d.ts} +2 -2
- package/dist/{index-D0_l25We.d.ts.map → index-CPJJr98o.d.ts.map} +1 -1
- package/dist/{index-DNgf5gyG.d.ts → index-Ch9VxCpK.d.ts} +2 -2
- package/dist/{index-DNgf5gyG.d.ts.map → index-Ch9VxCpK.d.ts.map} +1 -1
- package/dist/{index-BWiAAWrk.d.ts → index-tYLUvk_w.d.ts} +8 -8
- package/dist/{index-BWiAAWrk.d.ts.map → index-tYLUvk_w.d.ts.map} +1 -1
- package/dist/index.d.ts +25 -25
- package/dist/index.js +13 -13
- package/dist/{integrity-zloavsM_.d.ts → integrity-CCAxGVK6.d.ts} +2 -2
- package/dist/{integrity-zloavsM_.d.ts.map → integrity-CCAxGVK6.d.ts.map} +1 -1
- package/dist/{judge-gate-B1_O0QIs.d.ts → judge-gate-DNcUrRFA.d.ts} +2 -2
- package/dist/{judge-gate-B1_O0QIs.d.ts.map → judge-gate-DNcUrRFA.d.ts.map} +1 -1
- package/dist/{kind-factory-CMlOx08-.js → kind-factory-AaEhUADH.js} +2 -2
- package/dist/{kind-factory-CMlOx08-.js.map → kind-factory-AaEhUADH.js.map} +1 -1
- package/dist/{llm-client-C_o62yG1.js → llm-client-DQveoxSM.js} +2 -2
- package/dist/{llm-client-C_o62yG1.js.map → llm-client-DQveoxSM.js.map} +1 -1
- package/dist/{llm-judge-Dw0rDpve.js → llm-judge-Cd5s5Ei6.js} +4 -4
- package/dist/{llm-judge-Dw0rDpve.js.map → llm-judge-Cd5s5Ei6.js.map} +1 -1
- package/dist/matrix/index.d.ts +2 -2
- package/dist/{matrix-CQwid3kj.d.ts → matrix-DXVAJMLD.d.ts} +4 -4
- package/dist/{matrix-CQwid3kj.d.ts.map → matrix-DXVAJMLD.d.ts.map} +1 -1
- package/dist/meta-eval/index.d.ts +2 -2
- package/dist/{mint-CDy-brUR.d.ts → mint-DCGoBYhB.d.ts} +2 -2
- package/dist/{mint-CDy-brUR.d.ts.map → mint-DCGoBYhB.d.ts.map} +1 -1
- 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/{prime-protocol-B7et3tI7.d.ts → prime-protocol-BHia9vDz.d.ts} +3 -3
- package/dist/{prime-protocol-B7et3tI7.d.ts.map → prime-protocol-BHia9vDz.d.ts.map} +1 -1
- package/dist/{produced-state-C1cTYK6w.js → produced-state-BGp4BX8t.js} +4 -4
- package/dist/{produced-state-C1cTYK6w.js.map → produced-state-BGp4BX8t.js.map} +1 -1
- package/dist/{registry-BL7H33tj.d.ts → registry-CMYIaDgS.d.ts} +5 -5
- package/dist/{registry-BL7H33tj.d.ts.map → registry-CMYIaDgS.d.ts.map} +1 -1
- package/dist/{release-confidence-C4kI7lQi.d.ts → release-confidence-BK2QCB5M.d.ts} +3 -3
- package/dist/{release-confidence-C4kI7lQi.d.ts.map → release-confidence-BK2QCB5M.d.ts.map} +1 -1
- package/dist/reporting.d.ts +3 -3
- package/dist/{researcher-DuIJ9IaA.d.ts → researcher-f6mZhh58.d.ts} +5 -5
- package/dist/{researcher-DuIJ9IaA.d.ts.map → researcher-f6mZhh58.d.ts.map} +1 -1
- package/dist/{reward-hacking-C6Kp54Av.d.ts → reward-hacking-DHA7c1l8.d.ts} +2 -2
- package/dist/{reward-hacking-C6Kp54Av.d.ts.map → reward-hacking-DHA7c1l8.d.ts.map} +1 -1
- package/dist/rl.d.ts +8 -8
- package/dist/rollout/index.d.ts +2 -2
- package/dist/{rubric-predictive-validity-BiNNEjpG.d.ts → rubric-predictive-validity-2SgHxoa5.d.ts} +2 -2
- package/dist/{rubric-predictive-validity-BiNNEjpG.d.ts.map → rubric-predictive-validity-2SgHxoa5.d.ts.map} +1 -1
- package/dist/{run-record-p2AYTrCn.d.ts → run-record-D1Hw-Jml.d.ts} +2 -2
- package/dist/{run-record-p2AYTrCn.d.ts.map → run-record-D1Hw-Jml.d.ts.map} +1 -1
- package/dist/search/index.d.ts +2 -2
- package/dist/{search-kernel-BWp7hfJn.d.ts → search-kernel-BqyBZ9Wm.d.ts} +4 -4
- package/dist/{search-kernel-BWp7hfJn.d.ts.map → search-kernel-BqyBZ9Wm.d.ts.map} +1 -1
- package/dist/{semantic-concept-judge-sDBNg_mI.js → semantic-concept-judge-BPqbwIVs.js} +6 -6
- package/dist/{semantic-concept-judge-sDBNg_mI.js.map → semantic-concept-judge-BPqbwIVs.js.map} +1 -1
- package/dist/{server-BRpadnVt.js → server-DbcC7xMl.js} +3 -3
- package/dist/{server-BRpadnVt.js.map → server-DbcC7xMl.js.map} +1 -1
- package/dist/{skillopt-optimization-method-CvaFGm95.js → skillopt-optimization-method-Dcepr6QL.js} +5 -5
- package/dist/{skillopt-optimization-method-CvaFGm95.js.map → skillopt-optimization-method-Dcepr6QL.js.map} +1 -1
- package/dist/{statistical-heldout-DZaTo3gZ.d.ts → statistical-heldout-CJTj2EoS.d.ts} +3 -3
- package/dist/{statistical-heldout-DZaTo3gZ.d.ts.map → statistical-heldout-CJTj2EoS.d.ts.map} +1 -1
- package/dist/{store-otlp-0AiS29kg.js → store-otlp-DWVhUgEj.js} +2 -2
- package/dist/{store-otlp-0AiS29kg.js.map → store-otlp-DWVhUgEj.js.map} +1 -1
- package/dist/{store-tool-spans-BAcBBSM8.js → store-tool-spans-BNICrhsb.js} +3 -3
- package/dist/{store-tool-spans-BAcBBSM8.js.map → store-tool-spans-BNICrhsb.js.map} +1 -1
- package/dist/{store-tool-spans-DuvZ4YWb.d.ts → store-tool-spans-BNzls3OD.d.ts} +7 -7
- package/dist/{store-tool-spans-DuvZ4YWb.d.ts.map → store-tool-spans-BNzls3OD.d.ts.map} +1 -1
- package/dist/{summary-report-CK0ueoJd.d.ts → summary-report-C-CtItCN.d.ts} +2 -2
- package/dist/{summary-report-CK0ueoJd.d.ts.map → summary-report-C-CtItCN.d.ts.map} +1 -1
- package/dist/{systemone-review-BReh3mvm.d.ts → systemone-review-C4dZgrgN.d.ts} +2 -2
- package/dist/{systemone-review-BReh3mvm.d.ts.map → systemone-review-C4dZgrgN.d.ts.map} +1 -1
- package/dist/systemone.d.ts +4 -4
- package/dist/{tool-groups-DOriCbkR.d.ts → tool-groups-BKRrtqy4.d.ts} +3 -3
- package/dist/tool-groups-BKRrtqy4.d.ts.map +1 -0
- package/dist/{trace-questions-RCaYLkwH.d.ts → trace-questions-CLx8vt6t.d.ts} +4 -4
- package/dist/{trace-questions-RCaYLkwH.d.ts.map → trace-questions-CLx8vt6t.d.ts.map} +1 -1
- package/dist/trace-repair/index.d.ts +3 -3
- package/dist/traces.d.ts +6 -6
- package/dist/traces.js +3 -3
- package/dist/{types-CMnPcbW5.d.ts → types-CGEmVYOk.d.ts} +6 -6
- package/dist/{types-CMnPcbW5.d.ts.map → types-CGEmVYOk.d.ts.map} +1 -1
- package/dist/{types-Lub7o6bN.d.ts → types-DBZ7keha.d.ts} +2 -2
- package/dist/{types-Lub7o6bN.d.ts.map → types-DBZ7keha.d.ts.map} +1 -1
- package/dist/{types-KEkKkETv.d.ts → types-DHb5TI36.d.ts} +4 -4
- package/dist/{types-KEkKkETv.d.ts.map → types-DHb5TI36.d.ts.map} +1 -1
- package/dist/{types-Dj0Qv2qi.d.ts → types-DinHtRJp.d.ts} +3 -3
- package/dist/{types-Dj0Qv2qi.d.ts.map → types-DinHtRJp.d.ts.map} +1 -1
- package/dist/wire/index.d.ts +3 -3
- package/dist/wire/index.js +1 -1
- package/docs/campaign-proposers.md +3 -0
- package/package.json +1 -1
- package/dist/cost-ledger-DFOeLeJi.js.map +0 -1
- package/dist/tool-groups-DOriCbkR.d.ts.map +0 -1
package/dist/index.js
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import { t as __exportAll } from "./rolldown-runtime-8H4AJuhK.js";
|
|
2
2
|
import { i as JudgeError, n as CaptureIntegrityError, o as NotFoundError, r as ConfigError, s as ValidationError, t as AgentEvalError } from "./errors-Dngq5h35.js";
|
|
3
3
|
import { a as hashCanonical, r as canonicalString } from "./canonical-BsZVgBhW.js";
|
|
4
|
-
import { a as CODING_HARNESSES, c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, n as completionVerdict, o as HARNESS_NATIVE_MODEL, r as createLlmCorrectnessChecker, s as agentProfileHash, t as extractProducedState, u as expandProfileAxes } from "./produced-state-
|
|
4
|
+
import { a as CODING_HARNESSES, c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, n as completionVerdict, o as HARNESS_NATIVE_MODEL, r as createLlmCorrectnessChecker, s as agentProfileHash, t as extractProducedState, u as expandProfileAxes } from "./produced-state-BGp4BX8t.js";
|
|
5
5
|
import { d as verifyAgentProfileCell, l as toAgentProfileJson, m as manifestContentDigest, o as buildAgentProfileCell, p as hashJson, r as agentProfileCellKey, s as groupRunsByAgentProfileCell, t as AGENT_PROFILE_KINDS, u as validateAgentProfileCell } from "./agent-profile-cell-BYc0bxsL.js";
|
|
6
6
|
import { t as interpolatedQuantile } from "./quantile--0yb07pD.js";
|
|
7
7
|
import { a as medianInPlace, i as makeRng, t as assertFiniteSample } from "./internal-CRHwS9iz.js";
|
|
@@ -25,24 +25,24 @@ import { r as observedSplitScore, t as isRealnessGated } from "./reward-nw2xZGZG
|
|
|
25
25
|
import { a as parseRunRecordSafe, c as validateRunRecord, i as modelHasSnapshot, n as UNKNOWN_MODEL, o as runTaskScore, r as isRunRecord, s as searchCellRunId, t as RunRecordValidationError } from "./run-record-DlTZ_lqe.js";
|
|
26
26
|
import { a as summaryTable, n as gainHistogram, r as paretoChart } from "./summary-report-DtdRucM0.js";
|
|
27
27
|
import { n as contentHash, r as fileVerdictCache, t as canonicalJson } from "./verdict-cache-BQEsVy9n.js";
|
|
28
|
-
import { J as redTeamDataset, X as scoreRedTeamOutput, Y as redTeamReport, Z as runCanaries, a as transientDispatchFailure, i as quotaExhaustedUntil, o as aggregateRunScore, q as DEFAULT_RED_TEAM_CORPUS, rt as runCampaign, s as clamp01, t as llmJudge } from "./llm-judge-
|
|
28
|
+
import { J as redTeamDataset, X as scoreRedTeamOutput, Y as redTeamReport, Z as runCanaries, a as transientDispatchFailure, i as quotaExhaustedUntil, o as aggregateRunScore, q as DEFAULT_RED_TEAM_CORPUS, rt as runCampaign, s as clamp01, t as llmJudge } from "./llm-judge-Cd5s5Ei6.js";
|
|
29
29
|
import { a as resolveModelPricing, i as isModelPriced, n as estimateCost, r as estimateTokens, t as MODEL_PRICING } from "./metrics-BKaPBtIG.js";
|
|
30
|
-
import { a as CostLedgerPersistenceError, c as costForTokenPricing, i as CostLedger, l as costForUsage, n as CostCallConflictError, o as CostReceiptCaptureError, r as CostCeilingReachedError, s as CostReservationExceededError, t as CostAccountingIncompleteError, u as modelPriceKey } from "./cost-ledger-
|
|
30
|
+
import { a as CostLedgerPersistenceError, c as costForTokenPricing, i as CostLedger, l as costForUsage, n as CostCallConflictError, o as CostReceiptCaptureError, r as CostCeilingReachedError, s as CostReservationExceededError, t as CostAccountingIncompleteError, u as modelPriceKey } from "./cost-ledger-D2FnJPBF.js";
|
|
31
31
|
import { C as parseReflectionResponse, L as BackendIntegrityError, N as surfaceContentHash, R as assertRealBackend, S as buildReflectionPrompt, c as verifyAttestation, o as ATTESTATION_ALGORITHM, s as attest, z as summarizeBackendIntegrity } from "./campaign-evidence-OcUciD-N.js";
|
|
32
32
|
import { d as redactForShare, f as redactText, n as REDACTION_VERSION, p as shareAllowed, r as assessShareSafety, u as redact } from "./redact-Dk1ZdJEm.js";
|
|
33
33
|
import { D as dominates, O as paretoFrontier, f as pairedPromotionPower, p as requiredPairsForPairedPromotion } from "./search-kernel-BztlERLv.js";
|
|
34
|
-
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-
|
|
35
|
-
import { r as selfImprove, t as defineAgentEval } from "./define-agent-eval-
|
|
34
|
+
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-DQveoxSM.js";
|
|
35
|
+
import { r as selfImprove, t as defineAgentEval } from "./define-agent-eval-BdfnqXeP.js";
|
|
36
36
|
import { n as equivalenceVerdict, t as certificationEvidenceDigest } from "./verdict-DhPoXApk.js";
|
|
37
37
|
import { t as TraceEmitter } from "./emitter-DbMXUIVP.js";
|
|
38
38
|
import { t as buildTrajectory } from "./trajectory-D_7rLrvE.js";
|
|
39
39
|
import { t as runCounterfactual } from "./counterfactual-DtMWvn7D.js";
|
|
40
|
-
import { G as DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS, U as parseFindingSubject, h as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, l as createBoundedTraceAnalysisStore, m as DEFAULT_TRACE_ANALYST_BUDGETS, t as createTraceAnalyst } from "./kind-factory-
|
|
41
|
-
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-
|
|
40
|
+
import { G as DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS, U as parseFindingSubject, h as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, l as createBoundedTraceAnalysisStore, m as DEFAULT_TRACE_ANALYST_BUDGETS, t as createTraceAnalyst } from "./kind-factory-AaEhUADH.js";
|
|
41
|
+
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-KP4-Tolz.js";
|
|
42
42
|
import { OUTPUT_VALUE } from "./trace-attributes.js";
|
|
43
43
|
import { r as extractUsageFromSse, t as extractUsage } from "./extract-usage-BrnzRJ6H.js";
|
|
44
44
|
import { n as InMemoryRawProviderSink, r as NoopRawProviderSink, t as FileSystemRawProviderSink } from "./raw-provider-sink-rB2DYxr6.js";
|
|
45
|
-
import { S as exportRunAsOtlp, T as captureFetchToRawSink, _ as tokenizeDomainWords, d as inferDomainKeywords, g as scoreTraceInsightReadiness, l as describeTraceInsightScope, n as toolSpansToTraceAnalysisStore, o as buildTraceInsightContext, s as buildTraceInsightPrompt, u as domainEvidencePattern, v as analyzeTraces } from "./store-tool-spans-
|
|
45
|
+
import { S as exportRunAsOtlp, T as captureFetchToRawSink, _ as tokenizeDomainWords, d as inferDomainKeywords, g as scoreTraceInsightReadiness, l as describeTraceInsightScope, n as toolSpansToTraceAnalysisStore, o as buildTraceInsightContext, s as buildTraceInsightPrompt, u as domainEvidencePattern, v as analyzeTraces } from "./store-tool-spans-BNICrhsb.js";
|
|
46
46
|
import { n as assertRunCaptured, t as RunIntegrityError } from "./integrity-BvnVARVl.js";
|
|
47
47
|
import { n as InMemoryTraceStore, t as FileSystemTraceStore } from "./store-BhNuLdJw.js";
|
|
48
48
|
import { t as packageVersion$1 } from "./package-version-D7lQHt_-.js";
|
|
@@ -52,14 +52,14 @@ import { t as runEvalCampaign } from "./eval-campaign-4byTtqrf.js";
|
|
|
52
52
|
import "./rollout-D9bKKBnZ.js";
|
|
53
53
|
import { t as mintRolloutRows } from "./mint-Csor1BLJ.js";
|
|
54
54
|
import { n as pairedEvalueSequence, t as evaluateInterimReleaseConfidence } from "./sequential-CzK5DarL.js";
|
|
55
|
-
import { n as paidJsonChat } from "./chat-json-call-
|
|
55
|
+
import { n as paidJsonChat } from "./chat-json-call-BrKsyspg.js";
|
|
56
56
|
import { i as makeProposalFinding, n as computeFindingId, r as makeFinding } from "./types-CW5b90JH.js";
|
|
57
|
-
import { t as createDspyRlmTraceEngine } from "./dspy-rlm-engine-
|
|
58
|
-
import { a as diffFindings, n as runSemanticConceptJudge, o as createChatTraceEngine, r as FindingsStore, t as SEMANTIC_CONCEPT_JUDGE_VERSION } from "./semantic-concept-judge-
|
|
57
|
+
import { t as createDspyRlmTraceEngine } from "./dspy-rlm-engine-DZfHM7kT.js";
|
|
58
|
+
import { a as diffFindings, n as runSemanticConceptJudge, o as createChatTraceEngine, r as FindingsStore, t as SEMANTIC_CONCEPT_JUDGE_VERSION } from "./semantic-concept-judge-BPqbwIVs.js";
|
|
59
59
|
import { _ as analyzeSeries, a as judgeGateDecision, c as snapshotFromJudgeAgreement, i as assertJudgeMayGate, n as JudgeGateRefusedError, o as measureJudgeAgreement, r as addCalibrationVerdicts, s as registerCalibrationSet, t as DEFAULT_JUDGE_GATE_POLICY } from "./judge-gate-B1qYnkpi.js";
|
|
60
|
-
import { a as spanRecordsToTraceAnalysisStore, i as otlpTextToTraceAnalysisStore, n as OtlpFileTraceStore } from "./store-otlp-
|
|
60
|
+
import { a as spanRecordsToTraceAnalysisStore, i as otlpTextToTraceAnalysisStore, n as OtlpFileTraceStore } from "./store-otlp-DWVhUgEj.js";
|
|
61
61
|
import { n as evaluateReleaseConfidence, r as bootstrapCi } from "./release-confidence-Dbe9AS_X.js";
|
|
62
|
-
import { t as createChatClient } from "./chat-client-
|
|
62
|
+
import { t as createChatClient } from "./chat-client-DthPOYGd.js";
|
|
63
63
|
import { createHash } from "node:crypto";
|
|
64
64
|
import { accessSync, appendFileSync, constants, cpSync, existsSync, mkdirSync, promises, readFileSync, readdirSync, statSync, writeFileSync } from "node:fs";
|
|
65
65
|
import { basename, delimiter, dirname, extname, isAbsolute, join, relative, resolve } from "node:path";
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import { r as CaptureIntegrityError } from "./errors-DEE6u6ot.js";
|
|
2
|
-
import { rt as RawProviderSink } from "./types-
|
|
2
|
+
import { rt as RawProviderSink } from "./types-DBZ7keha.js";
|
|
3
3
|
import { s as TraceStore } from "./store-C4avFjIi.js";
|
|
4
4
|
//#region src/trace/integrity.d.ts
|
|
5
5
|
interface RunIntegrityExpectations {
|
|
@@ -58,4 +58,4 @@ declare function assertRunCaptured(store: TraceStore, runId: string, expectation
|
|
|
58
58
|
declare function throwIfRunIncomplete(report: RunIntegrityReport): void;
|
|
59
59
|
//#endregion
|
|
60
60
|
export { RunIntegrityReport as a, RunIntegrityIssueCode as i, RunIntegrityExpectations as n, assertRunCaptured as o, RunIntegrityIssue as r, throwIfRunIncomplete as s, RunIntegrityError as t };
|
|
61
|
-
//# sourceMappingURL=integrity-
|
|
61
|
+
//# sourceMappingURL=integrity-CCAxGVK6.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"integrity-
|
|
1
|
+
{"version":3,"file":"integrity-CCAxGVK6.d.ts","names":[],"sources":["../src/trace/integrity.ts"],"mappings":";;;;UAyBiB;;EAEf;;EAEA;;EAEA;;;;;EAKA,UAAU;;EAEV;;;;;;EAMA;;EAEA;;KAGU;UAWK;EACf,MAAM;EACN;EACA,SAAS;;UAGM;EACf;EACA;EACA;EACA;EACA;EACA;;;;;;EAMA;IAAmB;IAAiB;;EACpC,QAAQ;;cAGG,0BAA0B;WACT,QAAQ;EAApC,YAA4B,QAAQ;;iBAOhB,kBACpB,OAAO,YACP,eACA,eAAc,2BACb,QAAQ;;iBAuHK,qBAAqB,QAAQ"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import { a as JudgeError } from "./errors-DEE6u6ot.js";
|
|
2
|
-
import { u as JudgeScore } from "./types-
|
|
2
|
+
import { u as JudgeScore } from "./types-DBZ7keha.js";
|
|
3
3
|
import { a as ContinuousCalibrationResult, i as ContinuousAgreementOptions, n as CandidateScore, o as GoldenItem, r as ContinuousAgreement, t as CalibrationResult } from "./judge-calibration-DFtEMlde.js";
|
|
4
4
|
import { a as ComputedInterval } from "./ast-hI-vjW6J.js";
|
|
5
5
|
//#region src/statistics/agreement-irr.d.ts
|
|
@@ -451,4 +451,4 @@ declare function snapshotFromJudgeAgreement(agreement: JudgeAgreement, options?:
|
|
|
451
451
|
}): SentinelSnapshot;
|
|
452
452
|
//#endregion
|
|
453
453
|
export { SentinelThresholds as A, validateSentinelSnapshot as B, JudgeSentinelOptions as C, SentinelSetOptions as D, SentinelReport as E, inMemorySentinelStore as F, CorpusAgreementPerDimension as G, SeriesConvergenceResult as H, judgeSentinelReport as I, corpusInterRaterAgreement as J, CorpusAgreementReport as K, snapshotFromAgreement as L, SnapshotMeta as M, evalHealthStamp as N, SentinelSnapshot as O, fileSentinelStore as P, snapshotFromCalibration as R, EvalHealthStamp as S, SentinelMetrics as T, analyzeSeries as U, SeriesConvergenceOptions as V, CorpusAgreementOptions as W, interRaterReliability as X, corpusInterRaterAgreementFromJudgeScores as Y, assertJudgeMayGate as _, JudgeCalibrationSet as a, registerCalibrationSet as b, JudgeGateDecision as c, JudgeGateRefusedError as d, JudgeGateStatus as f, addCalibrationVerdicts as g, OwnerVerdict as h, JudgeAgreement as i, SentinelTrend as j, SentinelStore as k, JudgeGateInput as l, JudgeRunMeta as m, CalibrationMerge as n, JudgeConfusion as o, JudgeRun as p, CorpusScoreRecord as q, DEFAULT_JUDGE_GATE_POLICY as r, JudgeExampleAgreement as s, CalibrationExample as t, JudgeGatePolicy as u, judgeGateDecision as v, SentinelMetricName as w, snapshotFromJudgeAgreement as x, measureJudgeAgreement as y, snapshotFromSentinelSet as z };
|
|
454
|
-
//# sourceMappingURL=judge-gate-
|
|
454
|
+
//# sourceMappingURL=judge-gate-DNcUrRFA.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"judge-gate-
|
|
1
|
+
{"version":3,"file":"judge-gate-DNcUrRFA.d.ts","names":[],"sources":["../src/statistics/agreement-irr.ts","../src/series-convergence.ts","../src/meta-eval/sentinel.ts","../src/meta-eval/judge-gate.ts"],"mappings":";;;;;;;;;;;;;;;;;;;;;;iBAyBgB,sBAAsB,aAAa;UA0ElC;;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;;;;;;;;;;;;;UCvRc;;EAEf;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;;EAEA;;EAEA;;iBAGc,cACd,kBACA,UAAS,2BACR;;;cCoBG;KAEM,6BAA6B;UAExB;;EAEf;;EAEA;;EAEA;;UAGe;;EAEf;EACA;;EAEA;;;;;;;EAOA,SAAS;;;UAIM;EACf;EACA;EACA;;;iBAcc,yBAAyB,UAAU,kBAAkB;;;;;;iBAiCrD,wBACd,QAAQ,oBAAoB,6BAC5B,MAAM,eACL;;;;;;;;iBAmBa,sBACd,QAAQ,wBAAwB,qBAChC,MAAM,eACL;UAYc;;EAEf;;;;;;;;;iBAUc,wBACd,QAAQ,kBACR,QAAQ,cACR,MAAM,cACN,UAAS,qBACR;UA4Cc;EACf,OAAO,UAAU,mBAAmB;;EAEpC,QAAQ,mBAAmB,QAAQ;;iBAGrB,sBAAsB,UAAS,qBAA0B;;;;;;;iBAqBzD,kBAAkB,eAAe;UA0ChC;;EAEf;;EAEA;;EAEA;;EAEA;;UAGe;;EAEf;EACA,aAAa;;EAEb,cAAc;;UAGC;EACf;EACA,QAAQ;;EAER,OAAO;;EAEP;;EAEA;;EAEA;EACA;EACA;;UAGe;EACf,UAAU;EACV;EACA;;;;;;;EAOA;;iBAKc,oBACd,SAAS,oBACT,MAAM,uBACL;UA6Gc;EACf;EACA;;;;;;;;;;iBAWc,gBAAgB,QAAQ,iBAAiB;;;KChb7C;;UAGK;;EAEf;EACA,SAAS;;EAET;;EAEA;;EAEA;;EAEA;EACA;;UAGe;EACf;;EAEA;EACA,mBAAmB;;;;;;iBAuDL,uBAAuB,KAAK,sBAAsB;UAwBjD;EACf,KAAK;;EAEL;;EAEA;;EAEA;;;;;;;;iBASc,uBACd,KAAK,qBACL,mBAAmB,uBAClB;;UAkCc;EACf;;EAEA,SAAS;;EAET,mBAAmB;;UAGJ;EACf;;EAEA;EACA;EACA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;UAGe;EACf;EACA,SAAS;EACT;;EAEA;;EAEA;;UAGe,uBAAuB;EACtC;;EAEA;;EAEA;;EAEA,iBAAiB,OAAO;EACxB;;EAEA;;;;;;EAMA;IAAmB;IAAgB;IAAgB,UAAU;;;;;;EAK7D;;EAEA;;EAEA;EACA,WAAW;EACX;;IAEE;;IAEA;;IAEA;;;EAGF,eAAe;;EAEf;;;;;;;iBAyBc,sBACd,KAAK,qBACL,eAAe,YACf,MAAM,eACL;UA+Gc;;;;;EAKf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,QAAQ;;cAGG,2BAA2B,SAAS;KAUrC;UAEK;EACf;EACA;EACA;;EAEA;EACA,QAAQ;;EAER;EACA,WAAW;EACX,QAAQ;;UAGO;EACf;;EAEA;;EAEA;;EAEA,WAAW;EACX;;EAEA,mBAAmB;EACnB,SAAS,QAAQ;;;iBAUH,kBAAkB,OAAO,iBAAiB;;cA2G7C,8BAA8B;WAChC,UAAU;EAEnB,YAAY,UAAU;;;iBASR,mBAAmB,UAAU;;;;;;;;iBAW7B,2BACd,WAAW,gBACX;EAAW,UAAU;IACpB"}
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
import { a as LimitExceededError, n as CaptureIntegrityError, o as NotFoundError, s as ValidationError, t as AgentEvalError } from "./errors-Dngq5h35.js";
|
|
2
2
|
import { a as hashCanonical, r as canonicalString } from "./canonical-BsZVgBhW.js";
|
|
3
|
-
import { i as CostLedger } from "./cost-ledger-
|
|
3
|
+
import { i as CostLedger } from "./cost-ledger-D2FnJPBF.js";
|
|
4
4
|
import { t as deepFreezeCanonicalJson } from "./deep-freeze-CurcOQg3.js";
|
|
5
5
|
import { i as validateUsageSettlementTimeout, n as settleUsageReceiptFromCostLedger } from "./usage-receipt-Cx3xNeyB.js";
|
|
6
6
|
import { LLM_MODEL_ATTR_KEYS, TOOL_NAME_ATTR_KEYS as TOOL_NAME_ATTR_KEYS$1 } from "./trace-attributes.js";
|
|
@@ -2081,4 +2081,4 @@ function truncateForContext(value, max) {
|
|
|
2081
2081
|
//#endregion
|
|
2082
2082
|
export { spanEpochMillis as A, FINDING_SUBJECT_SYNTAX as B, TraceFileTooLargeError as C, extractOtlpAttributes as D, compareSpanTime as E, parseRawFinding as F, DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS as G, findingSubjectGrammarPromptFor as H, parseTraceSpanEvidenceUri as I, snapshotExactExecutionPlan as J, resolveTraceAnalystLimits as K, coerceJson as L, RAW_FINDING_SCHEMA_PROMPT as M, RawAnalystFindingSchema as N, firstStringAttr$1 as O, evidenceRefsFromRawFinding as P, stripCodeFences as R, TraceFileMissingError as S, asString as T, parseFindingSubject as U, KIND_EXPECTED_SUBJECTS as V, renderFindingSubject as W, SpanNotFoundError as _, TRACE_ANALYST_TOOL_NAMESPACE as a, TraceAnalysisValidationError as b, bindSpanSourceReader as c, compileSearchRegex as d, truncateForBudget as f, TRACE_ANALYSIS_LIMITS as g, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX as h, buildTraceToolsForGroup as i, stringField as j, projectOtlpFlatLine as k, createBoundedTraceAnalysisStore as l, DEFAULT_TRACE_ANALYST_BUDGETS as m, renderPriorFindings as n, UNTRUSTED_TRACE_TEXT as o, validateInteger as p, snapshotExactExecutionComponentIdentity as q, runTraceAnalyst as r, buildTraceAnalysisToolDescriptors as s, createTraceAnalyst as t, traceSpanSchema as u, TraceAnalysisLimitError as v, TraceNotFoundError as w, TraceFileMalformedError as x, TraceAnalysisStoreContractError as y, FINDING_SUBJECT_KINDS as z };
|
|
2083
2083
|
|
|
2084
|
-
//# sourceMappingURL=kind-factory-
|
|
2084
|
+
//# sourceMappingURL=kind-factory-AaEhUADH.js.map
|