@tangle-network/agent-eval 0.172.0 → 0.173.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +49 -0
- package/README.md +17 -2
- package/dist/adapters/http.d.ts +2 -2
- package/dist/analyst/index.d.ts +13 -13
- package/dist/analyst/index.js +7 -6
- package/dist/analyst/index.js.map +1 -1
- package/dist/{attestation-CJBGmMVh.d.ts → attestation-c1QvaBdX.d.ts} +2 -2
- package/dist/{attestation-CJBGmMVh.d.ts.map → attestation-c1QvaBdX.d.ts.map} +1 -1
- package/dist/{backend-integrity-e79K3UPD.d.ts → backend-integrity-CeuTgqsd.d.ts} +3 -4
- package/dist/backend-integrity-CeuTgqsd.d.ts.map +1 -0
- package/dist/{benchmark-h-h4bfqj.d.ts → benchmark-BjLGkfnN.d.ts} +3 -3
- package/dist/{benchmark-h-h4bfqj.d.ts.map → benchmark-BjLGkfnN.d.ts.map} +1 -1
- package/dist/{benchmark-command-DoFcisuM.js → benchmark-command-9S20PRel.js} +9 -9
- package/dist/{benchmark-command-DoFcisuM.js.map → benchmark-command-9S20PRel.js.map} +1 -1
- package/dist/benchmarks/index.d.ts +5 -5
- package/dist/benchmarks/index.js +3 -3
- package/dist/{bounded-process-CVOC_D3H.js → bounded-process-BBZob7vl.js} +64 -10
- package/dist/bounded-process-BBZob7vl.js.map +1 -0
- package/dist/builder-eval/index.d.ts +3 -3
- package/dist/builder-eval/index.js +2 -2
- package/dist/campaign/index.d.ts +9 -9
- package/dist/campaign/index.js +6 -6
- package/dist/{campaign-Dp35pBbS.js → campaign-pxS0wmo4.js} +8 -8
- package/dist/{campaign-Dp35pBbS.js.map → campaign-pxS0wmo4.js.map} +1 -1
- package/dist/chat-client-Db4bqYfA.js +115 -0
- package/dist/chat-client-Db4bqYfA.js.map +1 -0
- package/dist/{chat-json-call-5Jxna-aV.js → chat-json-call-C26igCih.js} +16 -6
- package/dist/chat-json-call-C26igCih.js.map +1 -0
- package/dist/cli.js +31 -17
- package/dist/cli.js.map +1 -1
- package/dist/{client-BvwNkIRN.js → client-BlLY6o2w.js} +2 -2
- package/dist/{client-BvwNkIRN.js.map → client-BlLY6o2w.js.map} +1 -1
- package/dist/{client-Df7wdslk.d.ts → client-DlqdbM7n.d.ts} +4 -4
- package/dist/{client-Df7wdslk.d.ts.map → client-DlqdbM7n.d.ts.map} +1 -1
- package/dist/contract/index.d.ts +13 -13
- package/dist/contract/index.js +202 -10
- package/dist/contract/index.js.map +1 -1
- package/dist/{counterfactual-Bee5_BIn.d.ts → counterfactual-CLgrwhkY.d.ts} +4 -4
- package/dist/{counterfactual-Bee5_BIn.d.ts.map → counterfactual-CLgrwhkY.d.ts.map} +1 -1
- package/dist/{chat-client-DI79OPye.js → default-registry-B0bKikCb.js} +3 -39
- package/dist/default-registry-B0bKikCb.js.map +1 -0
- package/dist/{default-registry-XxedTLwu.d.ts → default-registry-BKwc8bN5.d.ts} +6 -6
- package/dist/{default-registry-XxedTLwu.d.ts.map → default-registry-BKwc8bN5.d.ts.map} +1 -1
- package/dist/{define-agent-eval-0wW7gFhr.d.ts → define-agent-eval-CY6qdlGV.d.ts} +6 -6
- package/dist/{define-agent-eval-0wW7gFhr.d.ts.map → define-agent-eval-CY6qdlGV.d.ts.map} +1 -1
- package/dist/{define-agent-eval-jS8xj_Q_.js → define-agent-eval-D_i_s69h.js} +7 -7
- package/dist/{define-agent-eval-jS8xj_Q_.js.map → define-agent-eval-D_i_s69h.js.map} +1 -1
- package/dist/{dspy-rlm-engine-CS3qcCEk.js → dspy-rlm-engine-D5byiHn9.js} +5 -6
- package/dist/dspy-rlm-engine-D5byiHn9.js.map +1 -0
- package/dist/{emitter-Bvnu0VzL.d.ts → emitter-Cs0egaFd.d.ts} +3 -3
- package/dist/{emitter-Bvnu0VzL.d.ts.map → emitter-Cs0egaFd.d.ts.map} +1 -1
- package/dist/{engine-BfRay1qD.d.ts → engine-DhFir3Ys.d.ts} +23 -8
- package/dist/{engine-BfRay1qD.d.ts.map → engine-DhFir3Ys.d.ts.map} +1 -1
- package/dist/{eval-campaign-JDTeE6Pl.js → eval-campaign-BeAjdhzC.js} +2 -2
- package/dist/{eval-campaign-JDTeE6Pl.js.map → eval-campaign-BeAjdhzC.js.map} +1 -1
- package/dist/{exact-types-BEecmnWm.d.ts → exact-types-BKOEILRP.d.ts} +2 -2
- package/dist/{exact-types-BEecmnWm.d.ts.map → exact-types-BKOEILRP.d.ts.map} +1 -1
- package/dist/experiment/index.d.ts +4 -4
- package/dist/{external-optimizer-contracts-szBJ_1vh.d.ts → external-optimizer-contracts-CQCpyrIL.d.ts} +2 -2
- package/dist/{external-optimizer-contracts-szBJ_1vh.d.ts.map → external-optimizer-contracts-CQCpyrIL.d.ts.map} +1 -1
- package/dist/{external-optimizer-process-CQxylYeG.js → external-optimizer-process-Cq_Pg15r.js} +2 -2
- package/dist/{external-optimizer-process-CQxylYeG.js.map → external-optimizer-process-Cq_Pg15r.js.map} +1 -1
- package/dist/{external-optimizer-subprocess-Cex8Da2i.js → external-optimizer-subprocess-DgNebftP.js} +2 -2
- package/dist/{external-optimizer-subprocess-Cex8Da2i.js.map → external-optimizer-subprocess-DgNebftP.js.map} +1 -1
- package/dist/failure-cluster-OldNRoAt.d.ts +154 -0
- package/dist/failure-cluster-OldNRoAt.d.ts.map +1 -0
- package/dist/{feedback-trajectory-DIqpCyF0.d.ts → feedback-trajectory-CMnv_uYs.d.ts} +6 -6
- package/dist/{feedback-trajectory-DIqpCyF0.d.ts.map → feedback-trajectory-CMnv_uYs.d.ts.map} +1 -1
- package/dist/{heldout-gate-JgNRDZwZ.d.ts → heldout-gate-Df5hsqmm.d.ts} +7 -7
- package/dist/{heldout-gate-JgNRDZwZ.d.ts.map → heldout-gate-Df5hsqmm.d.ts.map} +1 -1
- package/dist/hosted/index.d.ts +2 -2
- package/dist/hosted/index.js +1 -1
- package/dist/{index-DnglhM0A.d.ts → index-BQqOjerE.d.ts} +94 -13
- package/dist/index-BQqOjerE.d.ts.map +1 -0
- package/dist/{index-_vPrVMRX.d.ts → index-CFDffsKz.d.ts} +11 -11
- package/dist/{index-_vPrVMRX.d.ts.map → index-CFDffsKz.d.ts.map} +1 -1
- package/dist/{index-DDAPhUJJ.d.ts → index-D0Db5X-4.d.ts} +45 -9
- package/dist/index-D0Db5X-4.d.ts.map +1 -0
- package/dist/{index-DMoxLG8P.d.ts → index-e7LXeRVa.d.ts} +3 -3
- package/dist/{index-DMoxLG8P.d.ts.map → index-e7LXeRVa.d.ts.map} +1 -1
- package/dist/index.d.ts +172 -46
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +64 -41
- package/dist/index.js.map +1 -1
- package/dist/{insight-report-08F022xN.d.ts → insight-report-DETqPc_A.d.ts} +4 -4
- package/dist/{insight-report-08F022xN.d.ts.map → insight-report-DETqPc_A.d.ts.map} +1 -1
- package/dist/{integrity-B_EDELom.d.ts → integrity-BKTcA-HP.d.ts} +3 -3
- package/dist/{integrity-B_EDELom.d.ts.map → integrity-BKTcA-HP.d.ts.map} +1 -1
- package/dist/{kind-factory-DMeEoMQZ.js → kind-factory-gP6lDySe.js} +164 -149
- package/dist/kind-factory-gP6lDySe.js.map +1 -0
- package/dist/{llm-client-BFMRpmqb.js → llm-client-CxQtdtd6.js} +12 -5
- package/dist/llm-client-CxQtdtd6.js.map +1 -0
- package/dist/{llm-judge-aQHIk5_-.js → llm-judge-B2YxbAJb.js} +81 -7
- package/dist/{llm-judge-aQHIk5_-.js.map → llm-judge-B2YxbAJb.js.map} +1 -1
- package/dist/{matrix-Ch8JO1pG.d.ts → matrix-DGu8KhSs.d.ts} +2 -2
- package/dist/{matrix-Ch8JO1pG.d.ts.map → matrix-DGu8KhSs.d.ts.map} +1 -1
- package/dist/meta-eval/index.d.ts +100 -5
- package/dist/meta-eval/index.d.ts.map +1 -1
- package/dist/meta-eval/index.js +200 -2
- package/dist/meta-eval/index.js.map +1 -1
- package/dist/{mint-DjfDUMHr.js → mint-vWOdD8Ae.js} +2 -2
- package/dist/{mint-DjfDUMHr.js.map → mint-vWOdD8Ae.js.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/pipelines/index.d.ts +5 -5
- package/dist/pipelines/index.js +3 -3
- package/dist/{pre-registration-DHz6P_6f.d.ts → pre-registration-BoI4ucR3.d.ts} +2 -2
- package/dist/{pre-registration-DHz6P_6f.d.ts.map → pre-registration-BoI4ucR3.d.ts.map} +1 -1
- package/dist/{produced-state-CxmbFxFd.js → produced-state-7VYDwtkk.js} +3 -3
- package/dist/{produced-state-CxmbFxFd.js.map → produced-state-7VYDwtkk.js.map} +1 -1
- package/dist/{promotion-policy-BBBcz5_3.d.ts → promotion-policy-CvMda3kU.d.ts} +2 -2
- package/dist/{promotion-policy-BBBcz5_3.d.ts.map → promotion-policy-CvMda3kU.d.ts.map} +1 -1
- package/dist/{provenance-Dp-vvyrU.d.ts → provenance-CRY67X50.d.ts} +39 -165
- package/dist/provenance-CRY67X50.d.ts.map +1 -0
- package/dist/{query-BPGMVlbM.js → query-D1nLIKt7.js} +2 -2
- package/dist/{query-BPGMVlbM.js.map → query-D1nLIKt7.js.map} +1 -1
- package/dist/{query-Na5gEIGd.d.ts → query-D6W6MaGx.d.ts} +3 -3
- package/dist/{query-Na5gEIGd.d.ts.map → query-D6W6MaGx.d.ts.map} +1 -1
- package/dist/{registry-xEb_xfns.d.ts → registry-7pOUBrtX.d.ts} +4 -4
- package/dist/{registry-xEb_xfns.d.ts.map → registry-7pOUBrtX.d.ts.map} +1 -1
- package/dist/{release-confidence-D6lQw_o7.d.ts → release-confidence-BAcNYOf1.d.ts} +4 -4
- package/dist/{release-confidence-D6lQw_o7.d.ts.map → release-confidence-BAcNYOf1.d.ts.map} +1 -1
- package/dist/{release-confidence-CzUHc4z4.js → release-confidence-BsGEg_xg.js} +3 -3
- package/dist/{release-confidence-CzUHc4z4.js.map → release-confidence-BsGEg_xg.js.map} +1 -1
- package/dist/reporting.d.ts +3 -3
- package/dist/reporting.js +1 -1
- package/dist/{researcher-CMUTQXD7.d.ts → researcher-jsW1X94L.d.ts} +7 -8
- package/dist/researcher-jsW1X94L.d.ts.map +1 -0
- package/dist/{reward-hacking-SkxYgT0x.js → reward-hacking-CKW4teig.js} +2 -2
- package/dist/{reward-hacking-SkxYgT0x.js.map → reward-hacking-CKW4teig.js.map} +1 -1
- package/dist/{reward-hacking-CgPRUesA.d.ts → reward-hacking-ZXEi9VCq.d.ts} +2 -2
- package/dist/{reward-hacking-CgPRUesA.d.ts.map → reward-hacking-ZXEi9VCq.d.ts.map} +1 -1
- package/dist/rl.d.ts +7 -7
- package/dist/rl.js +4 -4
- package/dist/rollout/index.d.ts +1 -1
- package/dist/rollout/index.js +2 -2
- package/dist/{rollout-Crypdx8s.js → rollout-C-znbbYg.js} +2 -2
- package/dist/{rollout-Crypdx8s.js.map → rollout-C-znbbYg.js.map} +1 -1
- package/dist/{rubric-predictive-validity-DluJLCKQ.d.ts → rubric-predictive-validity-Dl1dvKCv.d.ts} +2 -2
- package/dist/{rubric-predictive-validity-DluJLCKQ.d.ts.map → rubric-predictive-validity-Dl1dvKCv.d.ts.map} +1 -1
- package/dist/{run-record-DQjRcYwA.d.ts → run-record-DTv1MdjK.d.ts} +2 -2
- package/dist/{run-record-DQjRcYwA.d.ts.map → run-record-DTv1MdjK.d.ts.map} +1 -1
- package/dist/{run-record-DLORoL7t.js → run-record-ZIsR9Fif.js} +2 -2
- package/dist/{run-record-DLORoL7t.js.map → run-record-ZIsR9Fif.js.map} +1 -1
- package/dist/{schema-DID1Cqct.d.ts → schema-CR5cpjQ3.d.ts} +57 -2
- package/dist/{schema-DID1Cqct.d.ts.map → schema-CR5cpjQ3.d.ts.map} +1 -1
- package/dist/{schema-CdIX2aHu.js → schema-CSf6qWgZ.js} +35 -1
- package/dist/{schema-CdIX2aHu.js.map → schema-CSf6qWgZ.js.map} +1 -1
- package/dist/semantic-concept-judge-Ct3QU7t5.js +780 -0
- package/dist/semantic-concept-judge-Ct3QU7t5.js.map +1 -0
- package/dist/{series-convergence-D9WgpXGi.d.ts → series-convergence-DeG33RpC.d.ts} +2 -2
- package/dist/{series-convergence-D9WgpXGi.d.ts.map → series-convergence-DeG33RpC.d.ts.map} +1 -1
- package/dist/{server-CCEnywOR.js → server-BR6onwZB.js} +2 -2
- package/dist/{server-CCEnywOR.js.map → server-BR6onwZB.js.map} +1 -1
- package/dist/{skillopt-optimization-method-LHi02MzH.js → skillopt-optimization-method-BzdphODy.js} +6 -6
- package/dist/{skillopt-optimization-method-LHi02MzH.js.map → skillopt-optimization-method-BzdphODy.js.map} +1 -1
- package/dist/{statistical-heldout-Yldkntvy.d.ts → statistical-heldout-DTyB_6-1.d.ts} +3 -3
- package/dist/{statistical-heldout-Yldkntvy.d.ts.map → statistical-heldout-DTyB_6-1.d.ts.map} +1 -1
- package/dist/{store-Cq9oOrI1.d.ts → store-BErPvYBr.d.ts} +2 -2
- package/dist/{store-Cq9oOrI1.d.ts.map → store-BErPvYBr.d.ts.map} +1 -1
- package/dist/{store-otlp-CHjBvWQY.js → store-otlp-Dow0pk_5.js} +2 -2
- package/dist/{store-otlp-CHjBvWQY.js.map → store-otlp-Dow0pk_5.js.map} +1 -1
- package/dist/{store-tool-spans-BvdUbeOB.d.ts → store-tool-spans-CCZNsihA.d.ts} +8 -8
- package/dist/{store-tool-spans-BvdUbeOB.d.ts.map → store-tool-spans-CCZNsihA.d.ts.map} +1 -1
- package/dist/{store-tool-spans-B9o6tU8f.js → store-tool-spans-CeNj_m2L.js} +3 -3
- package/dist/{store-tool-spans-B9o6tU8f.js.map → store-tool-spans-CeNj_m2L.js.map} +1 -1
- package/dist/storyboard/index.d.ts +1 -1
- package/dist/{summary-report-DRstQNBX.d.ts → summary-report-gMrbYawB.d.ts} +3 -3
- package/dist/{summary-report-DRstQNBX.d.ts.map → summary-report-gMrbYawB.d.ts.map} +1 -1
- package/dist/{task-failure-attributes-CBGtLS_H.js → task-failure-attributes-CZjZeBsY.js} +3 -3
- package/dist/{task-failure-attributes-CBGtLS_H.js.map → task-failure-attributes-CZjZeBsY.js.map} +1 -1
- package/dist/{tool-groups-DjwlMBvW.d.ts → tool-groups-Cp4Xdzrp.d.ts} +3 -3
- package/dist/tool-groups-Cp4Xdzrp.d.ts.map +1 -0
- package/dist/{tool-waste-CwGHzBzX.js → tool-waste-B9tdWV6g.js} +253 -6
- package/dist/tool-waste-B9tdWV6g.js.map +1 -0
- package/dist/{tool-waste-BrmLKxMw.d.ts → tool-waste-D23I0zWm.d.ts} +4 -4
- package/dist/{tool-waste-BrmLKxMw.d.ts.map → tool-waste-D23I0zWm.d.ts.map} +1 -1
- package/dist/trace-repair/index.d.ts +2 -2
- package/dist/traces.d.ts +10 -10
- package/dist/traces.js +7 -7
- package/dist/{trajectory-r1bQqvBQ.d.ts → trajectory-D7qrNvaN.d.ts} +3 -3
- package/dist/{trajectory-r1bQqvBQ.d.ts.map → trajectory-D7qrNvaN.d.ts.map} +1 -1
- package/dist/trajectory-replay/index.d.ts +3 -3
- package/dist/{types-CCZ34qmV.d.ts → types-BDV4PiMR.d.ts} +3 -3
- package/dist/{types-CCZ34qmV.d.ts.map → types-BDV4PiMR.d.ts.map} +1 -1
- package/dist/{types-nokrtr7M.d.ts → types-Ba5UQyVD.d.ts} +4 -4
- package/dist/{types-nokrtr7M.d.ts.map → types-Ba5UQyVD.d.ts.map} +1 -1
- package/dist/{types-DMoNFDWi.d.ts → types-DN2WdT5S.d.ts} +3 -3
- package/dist/{types-DMoNFDWi.d.ts.map → types-DN2WdT5S.d.ts.map} +1 -1
- package/dist/types-gvRsyJLh.d.ts +831 -0
- package/dist/types-gvRsyJLh.d.ts.map +1 -0
- package/dist/wire/index.d.ts +3 -3
- package/dist/wire/index.js +1 -1
- package/docs/plants.md +69 -1
- package/docs/public-api.md +6 -4
- package/docs/trace-analysis.md +55 -8
- package/package.json +1 -1
- package/dist/backend-integrity-e79K3UPD.d.ts.map +0 -1
- package/dist/bounded-process-CVOC_D3H.js.map +0 -1
- package/dist/chat-client-DI79OPye.js.map +0 -1
- package/dist/chat-json-call-5Jxna-aV.js.map +0 -1
- package/dist/dspy-rlm-engine-CS3qcCEk.js.map +0 -1
- package/dist/failure-cluster-6YSvsKlp.d.ts +0 -58
- package/dist/failure-cluster-6YSvsKlp.d.ts.map +0 -1
- package/dist/index-DDAPhUJJ.d.ts.map +0 -1
- package/dist/index-DnglhM0A.d.ts.map +0 -1
- package/dist/kind-factory-DMeEoMQZ.js.map +0 -1
- package/dist/llm-client-BFMRpmqb.js.map +0 -1
- package/dist/provenance-Dp-vvyrU.d.ts.map +0 -1
- package/dist/raw-provider-sink-BU29Sh8h.d.ts +0 -134
- package/dist/raw-provider-sink-BU29Sh8h.d.ts.map +0 -1
- package/dist/researcher-CMUTQXD7.d.ts.map +0 -1
- package/dist/semantic-concept-judge-I36eejJx.js +0 -382
- package/dist/semantic-concept-judge-I36eejJx.js.map +0 -1
- package/dist/tool-groups-DjwlMBvW.d.ts.map +0 -1
- package/dist/tool-waste-CwGHzBzX.js.map +0 -1
- package/dist/types-Bfk0uxRj.d.ts +0 -443
- package/dist/types-Bfk0uxRj.d.ts.map +0 -1
|
@@ -1,8 +1,8 @@
|
|
|
1
|
-
import { i as AnalystFinding, n as AnalystContext, w as TraceAnalysisStore } from "./types-
|
|
2
|
-
import { l as ExactCapableAnalyst } from "./exact-types-
|
|
3
|
-
import { n as AnalystRegistry, r as AnalystRegistryOptions } from "./registry-
|
|
4
|
-
import {
|
|
5
|
-
import { t as TraceToolGroupName } from "./tool-groups-
|
|
1
|
+
import { i as AnalystFinding, n as AnalystContext, w as TraceAnalysisStore } from "./types-DN2WdT5S.js";
|
|
2
|
+
import { l as ExactCapableAnalyst } from "./exact-types-BKOEILRP.js";
|
|
3
|
+
import { n as AnalystRegistry, r as AnalystRegistryOptions } from "./registry-7pOUBrtX.js";
|
|
4
|
+
import { a as TraceAnalystLimits, i as TraceAnalysisEngineResult, l as RawAnalystFinding, n as TraceAnalysisEngine } from "./engine-DhFir3Ys.js";
|
|
5
|
+
import { t as TraceToolGroupName } from "./tool-groups-Cp4Xdzrp.js";
|
|
6
6
|
//#region src/trace-analyst/behavioral-metrics.d.ts
|
|
7
7
|
type SuboptimalCode = 'monotonic-input-growth' | 'output-length-decay' | 'single-tool-dependency' | 'no-self-verification';
|
|
8
8
|
interface SuboptimalSignal {
|
|
@@ -110,4 +110,4 @@ interface DefaultAnalystRegistryOptions {
|
|
|
110
110
|
declare function buildDefaultAnalystRegistry(options?: DefaultAnalystRegistryOptions): AnalystRegistry;
|
|
111
111
|
//#endregion
|
|
112
112
|
export { createTraceAnalyst as a, BehavioralAnalystOptions as c, TraceAnalystDefinition as i, behavioralAnalyst as l, buildDefaultAnalystRegistry as n, renderPriorFindings as o, CreateTraceAnalystOptions as r, runTraceAnalyst as s, DefaultAnalystRegistryOptions as t, deriveEfficiencyFindings as u };
|
|
113
|
-
//# sourceMappingURL=default-registry-
|
|
113
|
+
//# sourceMappingURL=default-registry-BKwc8bN5.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"default-registry-
|
|
1
|
+
{"version":3,"file":"default-registry-BKwc8bN5.d.ts","names":[],"sources":["../src/trace-analyst/behavioral-metrics.ts","../src/analyst/behavioral-analyst.ts","../src/analyst/kind-factory.ts","../src/analyst/default-registry.ts"],"mappings":";;;;;;KAoBY;UAMK;EACf,MAAM;EACN;;EAEA;;EAEA,UAAU;;UAGK;;EAEf;EACA;;EAEA,gBAAgB;;EAEhB;EACA;EACA,eAAe;EACf;EACA;;EAEA;EACA;EACA,SAAS;;UAGM;EACf;EACA;EACA,sBAAsB;EACtB,uBAAuB;;;;UC1BR;;EAEf;;EAEA;;;;;;iBA+Dc,yBACd,SAAS,mBACT;EAAQ;EAAoB;IAC3B;;iBAkCa,kBACd,UAAS,2BACR,oBAAoB;;;;UCrHN;EACf;EACA;EACA;EACA;;EAEA,sBAAsB,SAAS;;EAE/B;;EAEA,WAAW;;EAEX,kBACE,OAAO,oBACP,SAAS,wCACe;EAC1B,SAAS,QAAQ;EACjB,eAAe,KAAK,mBAAmB,SAAS,mBAAmB;EACnE;EACA;;UAGe;EACf,QAAQ;;EAER;EACA;;;iBAIoB,gBAAgB;EACpC,YAAY;EACZ,QAAQ;EACR,OAAO;EACP,SAAS;EACT;IACE,QAAQ;;iBAgFI,mBACd,YAAY,wBACZ,SAAS,4BACR,oBAAoB;iBA+QP,oBAAoB,OAAO;;;UCpZ1B;;;;;;EAMf,SAAS;EACT,uBAAuB;;EAEvB;EACA,aAAa;EACb,WAAW;;iBAGG,4BACd,UAAS,gCACR"}
|
|
@@ -1,10 +1,10 @@
|
|
|
1
1
|
import { c as CostLedgerHandle, f as CostLedgerSummary, m as CostReceipt } from "./cost-ledger-DbQdN3nO.js";
|
|
2
|
-
import { _ as ProposalFinding } from "./types-
|
|
3
|
-
import {
|
|
4
|
-
import {
|
|
2
|
+
import { _ as ProposalFinding } from "./types-DN2WdT5S.js";
|
|
3
|
+
import { H as SurfaceProposer, R as Scenario, S as JudgeConfig, a as CampaignResult, d as DispatchContext, j as MutableSurface, k as LabeledScenarioStore, p as Gate } from "./types-Ba5UQyVD.js";
|
|
4
|
+
import { An as RunEvalOptions, B as OptimizationMethodResult, F as OptimizationMethod, Kn as CampaignStorage, Mt as SearchHistoryReceipt, Nn as CampaignCellRetryPolicy, ct as RunImprovementLoopResult, dt as RunOptimizationOptions, l as LoopProvenanceRecord, ut as PremeasuredOptimizationBaseline, z as OptimizationMethodProvenance } from "./provenance-CRY67X50.js";
|
|
5
5
|
import { t as PowerPreflight } from "./power-preflight-Ptse_Kq7.js";
|
|
6
|
-
import { o as InsightReport } from "./insight-report-
|
|
7
|
-
import { n as HostedTenant } from "./client-
|
|
6
|
+
import { o as InsightReport } from "./insight-report-DETqPc_A.js";
|
|
7
|
+
import { n as HostedTenant } from "./client-DlqdbM7n.js";
|
|
8
8
|
//#region src/contract/self-improve.d.ts
|
|
9
9
|
interface SelfImproveBudget {
|
|
10
10
|
/** Hard spend cap across the full run. Each paid call reserves its enforced
|
|
@@ -404,4 +404,4 @@ interface DefinedAgentEval<TScenario extends Scenario, TArtifact> {
|
|
|
404
404
|
declare function defineAgentEval<TScenario extends Scenario, TArtifact>(defaults: DefineAgentEvalOptions<TScenario, TArtifact>): DefinedAgentEval<TScenario, TArtifact>;
|
|
405
405
|
//#endregion
|
|
406
406
|
export { DefinedAgentEval as a, SelfImproveOptions as c, SelfImproveRunError as d, selfImprove as f, DefineAgentEvalOptions as i, SelfImproveProgressEvent as l, AgentEvalEvaluateOptions as n, defineAgentEval as o, AgentEvalImproveOptions as r, SelfImproveBudget as s, AgentEvalAgent as t, SelfImproveResult as u };
|
|
407
|
-
//# sourceMappingURL=define-agent-eval-
|
|
407
|
+
//# sourceMappingURL=define-agent-eval-CY6qdlGV.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"define-agent-eval-
|
|
1
|
+
{"version":3,"file":"define-agent-eval-CY6qdlGV.d.ts","names":[],"sources":["../src/contract/self-improve.ts","../src/contract/define-agent-eval.ts"],"mappings":";;;;;;;;UAiEiB;;;EAGf;;;;EAIA;;EAEA;;EAEA;;;EAGA;;;EAGA;;;;EAIA;;EAEA,mBAAmB;;;;;;;;;EASnB;;EAEA;;;;;EAKA;;KAGU;EACN;EAA0B;;EAC1B;EAA4B;EAAuB;;EACnD;EAA4B;EAAe;;EAC3C;EAA8B;EAAe;EAAuB;;EAGpE;EAAsB;EAAkB;;EACxC;EAAyB;EAAW;EAAY;EAAa;;UAElD,mBAAmB,kBAAkB,UAAU;;;;;;;;;;;;;;EAc9D,QAAQ,SAAS,gBAAgB,UAAU,WAAW,KAAK,oBAAoB,QAAQ;;;;;;;EAQvF;;;EAIA,WAAW;;EAGX,OAAO,YAAY,WAAW;;;EAI9B,iBAAiB;;EAGjB,SAAS;;;;;;;;;;;;EAaT,sBAAsB,gCAAgC,WAAW;;;;;EAMjE,WAAW,gBAAgB;;;;;;EAO3B,SAAS,mBAAmB,WAAW;;;EAIvC,qBAAqB;;;EAIrB,OAAO,KAAK,WAAW;;;;;;;;EASvB,cAAc,eAAe,gBAAgB,iBAAiB,mBAAmB;;;;EAKjF,UAAU;;;;EAKV;;;EAIA,gBAAgB,QAAQ;;;;;EAMxB,iBAAiB;IACf,UAAU;IACV;IACA;;;;EAKF;;;;;;;EAQA,YAAY;;;EAIZ,cAAc,OAAO;;;;EAKrB;EACA;EACA;;;;;;;;;;;;;;EAeA,eAAe;;;EAIf,eAAe;;;;EAKf,eAAe;;EAGf;;;;;;;;EASA;;;;;;;;;EAUA,oBAAoB,uBAAuB,WAAW;;;EAItD,WAAW;;;;;;;EAQX,mBAAmB,uBAAuB,WAAW;;;;;;EAOrD,eAAe,uBAAuB,WAAW;;;;EAKjD,eAAe,uBAAuB,WAAW;;UAGlC,kBAAkB,kBAAkB,UAAU;;;;EAI7D;IACE;IACA,aAAa;;;;;EAKf;IACE;IACA,aAAa;IACb,SAAS;;;IAGT;;;;IAIA;;;;;;EAMF;;;EAGA;;;;EAIA,YAAY;;EAEZ;;;EAGA;;EAEA;;EAEA;;EAEA,MAAM;;;EAGN,UAAU;;;EAGV,gBAAgB;;EAEhB;IACE;IACA,MAAM;IACN;IACA,aAAa;;;;;;;;EAQf,SAAS;;;;;EAKT,QAAQ;;;;;;EAMR,KAAK,yBAAyB,WAAW;;;cAI9B,4BAA4B;WAC9B,MAAM;WACN,UAAU;EAEnB,YAAY,gBAAgB,QAAQ;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAuMhB,YAAY,kBAAkB,UAAU,WAC5D,MAAM,mBAAmB,WAAW,aACnC,QAAQ,kBAAkB,WAAW;;;KCvkB5B,eAAe,kBAAkB,UAAU,cACrD,SAAS,gBACT,UAAU,WACV,KAAK,oBACF,QAAQ;KAED,uBAAuB,kBAAkB,UAAU,aAAa,mBAC1E,WACA;UAGe,yBAAyB,kBAAkB,UAAU,mBAC5D,KACN,eAAe,WAAW;;EAI5B,YAAY;;EAEZ,UAAU;;EAEV,QAAQ,eAAe,WAAW;;EAElC,QAAQ,YAAY,WAAW;;EAE/B,SAAS,YAAY,WAAW;;EAEhC;;KAGU,wBAAwB,kBAAkB,UAAU,aAAa,KAC3E,QAAQ,mBAAmB,WAAW;EAGtC,SAAS,QAAQ;EACjB,eAAe,QAAQ;;UAGR,iBAAiB,kBAAkB,UAAU;;WAEnD,oBAAoB;;WAEpB,iBAAiB;;;;;EAK1B,SACE,OAAO,yBAAyB,WAAW,aAC1C,QAAQ,eAAe,WAAW;;;;;;;EAOrC,QACE,OAAO,wBAAwB,WAAW,aACzC,QAAQ,kBAAkB,WAAW;;;;;;;;;iBAU1B,gBAAgB,kBAAkB,UAAU,WAC1D,UAAU,uBAAuB,WAAW,aAC3C,iBAAiB,WAAW"}
|
|
@@ -6,15 +6,15 @@ import { i as requiredPairedSampleSize, r as pairedMde } from "./power-and-mde-B
|
|
|
6
6
|
import { r as pairRunRecords } from "./paired-arms-D4aeIHUy.js";
|
|
7
7
|
import { o as pairedTTest, r as pairedBootstrap } from "./paired-tests-C8iCsioC.js";
|
|
8
8
|
import { r as welchsTTest } from "./baseline-BC-eBZ7U.js";
|
|
9
|
-
import "./query-
|
|
9
|
+
import "./query-D1nLIKt7.js";
|
|
10
10
|
import { r as observedSplitScore } from "./reward-nw2xZGZG.js";
|
|
11
|
-
import { c as validateRunRecord, i as modelHasSnapshot } from "./run-record-
|
|
11
|
+
import { c as validateRunRecord, i as modelHasSnapshot } from "./run-record-ZIsR9Fif.js";
|
|
12
12
|
import { r as paretoChart } from "./summary-report-Bgh8CpNK.js";
|
|
13
|
-
import {
|
|
14
|
-
import { c as campaignCellTaskScore, l as campaignCellToRunRecord, o as campaignCellExecutionEvidence, s as campaignCellJudgeDimensions } from "./reward-hacking-
|
|
15
|
-
import { C as inMemoryCampaignStorage, S as fsCampaignStorage, x as createRunCostLedger } from "./external-optimizer-subprocess-
|
|
13
|
+
import { D as assertOptimizationResult, I as surfaceContentHash, L as surfaceHash, Q as defaultProductionGate, _ as runImprovementLoop, d as emitLoopProvenance, f as loopProvenanceArgsFromResult, it as runEval, st as resolveRunDir } from "./llm-judge-B2YxbAJb.js";
|
|
14
|
+
import { c as campaignCellTaskScore, l as campaignCellToRunRecord, o as campaignCellExecutionEvidence, s as campaignCellJudgeDimensions } from "./reward-hacking-CKW4teig.js";
|
|
15
|
+
import { C as inMemoryCampaignStorage, S as fsCampaignStorage, x as createRunCostLedger } from "./external-optimizer-subprocess-DgNebftP.js";
|
|
16
16
|
import { t as powerPreflight } from "./power-preflight-CFXm0Vjo.js";
|
|
17
|
-
import { t as createHostedClient } from "./client-
|
|
17
|
+
import { t as createHostedClient } from "./client-BlLY6o2w.js";
|
|
18
18
|
//#region src/contamination-guard.ts
|
|
19
19
|
function checkCanaries(output, scenarios) {
|
|
20
20
|
const leaks = [];
|
|
@@ -1509,4 +1509,4 @@ function requirePositiveInteger(value, field) {
|
|
|
1509
1509
|
//#endregion
|
|
1510
1510
|
export { summarizeExecution as a, analyzeRuns as i, SelfImproveRunError as n, checkCanaries as o, selfImprove as r, defineAgentEval as t };
|
|
1511
1511
|
|
|
1512
|
-
//# sourceMappingURL=define-agent-eval-
|
|
1512
|
+
//# sourceMappingURL=define-agent-eval-D_i_s69h.js.map
|