@tangle-network/agent-eval 0.183.0 → 0.184.0
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 +16 -0
- package/dist/adapters/http.d.ts +2 -2
- package/dist/{agent-profile-cell-s__adRnK.d.ts → agent-profile-cell-BHcB-mZX.d.ts} +8 -9
- package/dist/{agent-profile-cell-s__adRnK.d.ts.map → agent-profile-cell-BHcB-mZX.d.ts.map} +1 -1
- package/dist/{agent-profile-cell-Cv6UA-W_.js → agent-profile-cell-DcFW4vOc.js} +16 -15
- package/dist/agent-profile-cell-DcFW4vOc.js.map +1 -0
- package/dist/{agent-profile-CivaSsSy.d.ts → agent-profile-uAIQK3Nr.d.ts} +3 -3
- package/dist/{agent-profile-CivaSsSy.d.ts.map → agent-profile-uAIQK3Nr.d.ts.map} +1 -1
- package/dist/analyst/index.d.ts +10 -10
- package/dist/analyst/index.js +8 -7
- package/dist/analyst/index.js.map +1 -1
- package/dist/{benchmark-DQKzykkO.js → benchmark-BeQJyrRC.js} +3 -3
- package/dist/{benchmark-DQKzykkO.js.map → benchmark-BeQJyrRC.js.map} +1 -1
- package/dist/{benchmark-command-NBrqNwgy.js → benchmark-command-DlSePljH.js} +13 -12
- package/dist/{benchmark-command-NBrqNwgy.js.map → benchmark-command-DlSePljH.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/campaign/index.d.ts +6 -5
- package/dist/campaign/index.js +9 -9
- package/dist/{campaign-BwcEBC91.js → campaign-D6puiGFO.js} +15 -15
- package/dist/{campaign-BwcEBC91.js.map → campaign-D6puiGFO.js.map} +1 -1
- package/dist/{campaign-evidence-PYPUftRy.js → campaign-evidence-aWEmfROB.js} +5 -5
- package/dist/{campaign-evidence-PYPUftRy.js.map → campaign-evidence-aWEmfROB.js.map} +1 -1
- package/dist/{canonical-DPyQ_rpt.js → canonical-BsZVgBhW.js} +2 -4
- package/dist/canonical-BsZVgBhW.js.map +1 -0
- package/dist/cli.js +2 -2
- package/dist/{client-kh2jOjTK.d.ts → client-Bc8LANCG.d.ts} +4 -4
- package/dist/{client-kh2jOjTK.d.ts.map → client-Bc8LANCG.d.ts.map} +1 -1
- package/dist/contract/index.d.ts +10 -10
- package/dist/contract/index.js +14 -14
- package/dist/{default-registry-C0-N4PkG.js → default-registry-CpLko4jB.js} +5 -4
- package/dist/{default-registry-C0-N4PkG.js.map → default-registry-CpLko4jB.js.map} +1 -1
- package/dist/{default-registry-BwDSWVzg.d.ts → default-registry-DYB4aHVn.d.ts} +6 -6
- package/dist/{default-registry-BwDSWVzg.d.ts.map → default-registry-DYB4aHVn.d.ts.map} +1 -1
- package/dist/{define-agent-eval-CwOWWQt_.d.ts → define-agent-eval-DE67ECH1.d.ts} +7 -7
- package/dist/{define-agent-eval-CwOWWQt_.d.ts.map → define-agent-eval-DE67ECH1.d.ts.map} +1 -1
- package/dist/{define-agent-eval-CneiWyuA.js → define-agent-eval-RTekxhUE.js} +8 -8
- package/dist/{define-agent-eval-CneiWyuA.js.map → define-agent-eval-RTekxhUE.js.map} +1 -1
- package/dist/{dspy-rlm-engine-8pgBSiy7.js → dspy-rlm-engine-Da-l3CpD.js} +3 -3
- package/dist/{dspy-rlm-engine-8pgBSiy7.js.map → dspy-rlm-engine-Da-l3CpD.js.map} +1 -1
- package/dist/{engine-DS1cysJy.d.ts → engine-D1ykrAKR.d.ts} +7 -7
- package/dist/{engine-DS1cysJy.d.ts.map → engine-D1ykrAKR.d.ts.map} +1 -1
- package/dist/{eval-campaign-CCEkwzoH.js → eval-campaign-XFHEpN86.js} +4 -4
- package/dist/{eval-campaign-CCEkwzoH.js.map → eval-campaign-XFHEpN86.js.map} +1 -1
- package/dist/evaluation.d.ts +75 -0
- package/dist/evaluation.d.ts.map +1 -0
- package/dist/evaluation.js +137 -0
- package/dist/evaluation.js.map +1 -0
- package/dist/{exact-types-BZDe0W2D.d.ts → exact-types-BEHN7tnq.d.ts} +2 -2
- package/dist/{exact-types-BZDe0W2D.d.ts.map → exact-types-BEHN7tnq.d.ts.map} +1 -1
- package/dist/experiment/index.d.ts +2 -2
- package/dist/experiment/index.js +5 -5
- package/dist/{experiment-tracker-C7PfnF4b.d.ts → experiment-tracker-URRGP9ek.d.ts} +2 -2
- package/dist/{experiment-tracker-C7PfnF4b.d.ts.map → experiment-tracker-URRGP9ek.d.ts.map} +1 -1
- package/dist/{external-optimizer-process-DIB4YWEe.js → external-optimizer-process-BbhvXTFx.js} +4 -4
- package/dist/{external-optimizer-process-DIB4YWEe.js.map → external-optimizer-process-BbhvXTFx.js.map} +1 -1
- package/dist/{external-optimizer-subprocess-CjguXN2e.js → external-optimizer-subprocess-DMgmqTB6.js} +3 -3
- package/dist/{external-optimizer-subprocess-CjguXN2e.js.map → external-optimizer-subprocess-DMgmqTB6.js.map} +1 -1
- package/dist/{feedback-trajectory-CXmtITBo.d.ts → feedback-trajectory-CytTR7u4.d.ts} +3 -3
- package/dist/{feedback-trajectory-CXmtITBo.d.ts.map → feedback-trajectory-CytTR7u4.d.ts.map} +1 -1
- package/dist/hosted/index.d.ts +2 -2
- package/dist/{index-DoykkxW0.d.ts → index-58wex7KV.d.ts} +11 -11
- package/dist/{index-DoykkxW0.d.ts.map → index-58wex7KV.d.ts.map} +1 -1
- package/dist/{index--_8_Nlpn.d.ts → index-CKZa_rnt.d.ts} +4 -121
- package/dist/index-CKZa_rnt.d.ts.map +1 -0
- package/dist/{index-DbG8QbH9.d.ts → index-DVrtLO5j.d.ts} +6 -6
- package/dist/{index-DbG8QbH9.d.ts.map → index-DVrtLO5j.d.ts.map} +1 -1
- package/dist/{index-Bp_6sj3x.d.ts → index-ww8736Am.d.ts} +20 -315
- package/dist/index-ww8736Am.d.ts.map +1 -0
- package/dist/index.d.ts +23 -22
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +24 -24
- package/dist/{insight-report-D1qa0HWs.d.ts → insight-report-C3Bfl762.d.ts} +3 -3
- package/dist/{insight-report-D1qa0HWs.d.ts.map → insight-report-C3Bfl762.d.ts.map} +1 -1
- package/dist/jev-protocol.d.ts +74 -0
- package/dist/jev-protocol.d.ts.map +1 -0
- package/dist/jev-protocol.js +91 -0
- package/dist/jev-protocol.js.map +1 -0
- package/dist/jev.d.ts +131 -0
- package/dist/jev.d.ts.map +1 -0
- package/dist/jev.js +312 -0
- package/dist/jev.js.map +1 -0
- package/dist/{journal-Cs9f7385.js → journal-BuRLm6ho.js} +2 -2
- package/dist/{journal-Cs9f7385.js.map → journal-BuRLm6ho.js.map} +1 -1
- package/dist/{kind-factory-DxTzg32u.js → kind-factory-DE088Zhv.js} +4 -3
- package/dist/{kind-factory-DxTzg32u.js.map → kind-factory-DE088Zhv.js.map} +1 -1
- package/dist/ledger-core/index.js +2 -2
- package/dist/{llm-judge-FfaRgz9X.js → llm-judge-C5wbADA5.js} +9 -9
- package/dist/{llm-judge-FfaRgz9X.js.map → llm-judge-C5wbADA5.js.map} +1 -1
- package/dist/{matrix-CyhW-vgJ.d.ts → matrix-CPKsRY8Y.d.ts} +2 -2
- package/dist/{matrix-CyhW-vgJ.d.ts.map → matrix-CPKsRY8Y.d.ts.map} +1 -1
- package/dist/meta-eval/index.d.ts +54 -2
- package/dist/meta-eval/index.d.ts.map +1 -1
- package/dist/meta-eval/index.js +52 -3
- package/dist/meta-eval/index.js.map +1 -1
- package/dist/{mint-ySIIkKlV.js → mint-B4zgSAJW.js} +63 -14
- package/dist/mint-B4zgSAJW.js.map +1 -0
- package/dist/mint-BI14xehJ.d.ts +285 -0
- package/dist/mint-BI14xehJ.d.ts.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/pipelines/index.js +2 -2
- package/dist/{produced-state-CGoVInrl.js → produced-state-D3uWsY8t.js} +23 -19
- package/dist/{produced-state-CGoVInrl.js.map → produced-state-D3uWsY8t.js.map} +1 -1
- package/dist/profile-cell.d.ts +1 -1
- package/dist/profile-cell.js +1 -1
- package/dist/{promotion-policy-BKFrl7CC.js → promotion-policy-CD-bUeyR.js} +3 -3
- package/dist/{promotion-policy-BKFrl7CC.js.map → promotion-policy-CD-bUeyR.js.map} +1 -1
- package/dist/{query-D1nLIKt7.js → query-DlV0pdx4.js} +2 -2
- package/dist/{query-D1nLIKt7.js.map → query-DlV0pdx4.js.map} +1 -1
- package/dist/{registry-BRbB6Y0v.d.ts → registry-D5Ly3GRV.d.ts} +3 -3
- package/dist/{registry-BRbB6Y0v.d.ts.map → registry-D5Ly3GRV.d.ts.map} +1 -1
- package/dist/{release-confidence-BcqeQTHW.d.ts → release-confidence-DsETiO4f.d.ts} +3 -3
- package/dist/{release-confidence-BcqeQTHW.d.ts.map → release-confidence-DsETiO4f.d.ts.map} +1 -1
- package/dist/{release-confidence-DMg8n18l.js → release-confidence-RanRUXRW.js} +2 -2
- package/dist/{release-confidence-DMg8n18l.js.map → release-confidence-RanRUXRW.js.map} +1 -1
- package/dist/reporting.d.ts +3 -3
- package/dist/reporting.js +2 -2
- package/dist/{researcher-64T49THL.d.ts → researcher-CJkH-6Ty.d.ts} +4 -4
- package/dist/{researcher-64T49THL.d.ts.map → researcher-CJkH-6Ty.d.ts.map} +1 -1
- package/dist/{reward-hacking-uzO_ihep.d.ts → reward-hacking-BIQ2BWLG.d.ts} +2 -2
- package/dist/{reward-hacking-uzO_ihep.d.ts.map → reward-hacking-BIQ2BWLG.d.ts.map} +1 -1
- package/dist/rl.d.ts +10 -6
- package/dist/rl.d.ts.map +1 -1
- package/dist/rl.js +21 -7
- package/dist/rl.js.map +1 -1
- package/dist/rollout/index.d.ts +2 -2
- package/dist/rollout/index.js +2 -2
- package/dist/{rollout-D_54FZVu.js → rollout-C0Ifk3vK.js} +2 -2
- package/dist/{rollout-D_54FZVu.js.map → rollout-C0Ifk3vK.js.map} +1 -1
- package/dist/{rubric-predictive-validity-Bmj2_cll.d.ts → rubric-predictive-validity-CTU8y8Cv.d.ts} +2 -2
- package/dist/{rubric-predictive-validity-Bmj2_cll.d.ts.map → rubric-predictive-validity-CTU8y8Cv.d.ts.map} +1 -1
- package/dist/{run-record-DPa2PeHx.js → run-record-BhMVsHR1.js} +2 -2
- package/dist/{run-record-DPa2PeHx.js.map → run-record-BhMVsHR1.js.map} +1 -1
- package/dist/{run-record-BiTWauyO.d.ts → run-record-D9wS35fK.d.ts} +2 -2
- package/dist/{run-record-BiTWauyO.d.ts.map → run-record-D9wS35fK.d.ts.map} +1 -1
- package/dist/{run-record-DualPTn2.js → run-record-tg_5H2IK.js} +2 -2
- package/dist/{run-record-DualPTn2.js.map → run-record-tg_5H2IK.js.map} +1 -1
- package/dist/search-ledger-types-D7KDVg-o.d.ts +302 -0
- package/dist/search-ledger-types-D7KDVg-o.d.ts.map +1 -0
- package/dist/{semantic-concept-judge-_3Fp6Q_3.js → semantic-concept-judge-5r9hDEBA.js} +4 -4
- package/dist/{semantic-concept-judge-_3Fp6Q_3.js.map → semantic-concept-judge-5r9hDEBA.js.map} +1 -1
- package/dist/{sequential-D_xHrBZe.js → sequential-BfDkCBGo.js} +3 -3
- package/dist/{sequential-D_xHrBZe.js.map → sequential-BfDkCBGo.js.map} +1 -1
- package/dist/{server-BZQR1f0C.js → server-DmpcFiVK.js} +2 -2
- package/dist/{server-BZQR1f0C.js.map → server-DmpcFiVK.js.map} +1 -1
- package/dist/{skillopt-optimization-method-DY1aQ-aL.js → skillopt-optimization-method-B2zxoDk0.js} +7 -7
- package/dist/{skillopt-optimization-method-DY1aQ-aL.js.map → skillopt-optimization-method-B2zxoDk0.js.map} +1 -1
- package/dist/{statistical-heldout-CpVd6FmY.d.ts → statistical-heldout-Bu4oNOJh.d.ts} +3 -3
- package/dist/{statistical-heldout-CpVd6FmY.d.ts.map → statistical-heldout-Bu4oNOJh.d.ts.map} +1 -1
- package/dist/{store-otlp-DAaEbMLf.js → store-otlp-rU5PFlOB.js} +2 -2
- package/dist/{store-otlp-DAaEbMLf.js.map → store-otlp-rU5PFlOB.js.map} +1 -1
- package/dist/{store-tool-spans-Dt-YdAuE.d.ts → store-tool-spans-DwdokbHg.d.ts} +54 -6
- package/dist/store-tool-spans-DwdokbHg.d.ts.map +1 -0
- package/dist/{store-tool-spans-BZ8MPKFe.js → store-tool-spans-zrtftcRe.js} +68 -6
- package/dist/store-tool-spans-zrtftcRe.js.map +1 -0
- package/dist/{summary-report-IGAOtgPd.js → summary-report-BMvou2oZ.js} +2 -2
- package/dist/{summary-report-IGAOtgPd.js.map → summary-report-BMvou2oZ.js.map} +1 -1
- package/dist/{summary-report-D1h4dlrK.d.ts → summary-report-BzFS5UhZ.d.ts} +2 -2
- package/dist/{summary-report-D1h4dlrK.d.ts.map → summary-report-BzFS5UhZ.d.ts.map} +1 -1
- package/dist/{task-failure-attributes-B_JjrOVo.js → task-failure-attributes-BOxjfbhr.js} +2 -2
- package/dist/{task-failure-attributes-B_JjrOVo.js.map → task-failure-attributes-BOxjfbhr.js.map} +1 -1
- package/dist/{tool-groups-B2bSNaJB.d.ts → tool-groups-qHMydbAH.d.ts} +3 -3
- package/dist/tool-groups-qHMydbAH.d.ts.map +1 -0
- package/dist/{tool-waste-C7MU9u1e.js → tool-waste-nbHPAxei.js} +2 -2
- package/dist/{tool-waste-C7MU9u1e.js.map → tool-waste-nbHPAxei.js.map} +1 -1
- package/dist/trace-repair/index.d.ts +2 -2
- package/dist/trace-repair/index.js +3 -3
- package/dist/traces.d.ts +5 -5
- package/dist/traces.js +7 -7
- package/dist/trajectory-replay/index.js +1 -1
- package/dist/{types-CS0qk_Yp.d.ts → types-BTRcIqFr.d.ts} +3 -3
- package/dist/{types-CS0qk_Yp.d.ts.map → types-BTRcIqFr.d.ts.map} +1 -1
- package/dist/{types-D7gEdPoQ.d.ts → types-BjB2wZM9.d.ts} +2 -2
- package/dist/{types-D7gEdPoQ.d.ts.map → types-BjB2wZM9.d.ts.map} +1 -1
- package/dist/types-C8VBbiw6.js +91 -0
- package/dist/types-C8VBbiw6.js.map +1 -0
- package/dist/{types-BvZoPTGa.d.ts → types-c3ZRTbw9.d.ts} +3 -3
- package/dist/{types-BvZoPTGa.d.ts.map → types-c3ZRTbw9.d.ts.map} +1 -1
- package/dist/{types-DQ0e2E7y.js → usage-receipt-Cx3xNeyB.js} +2 -89
- package/dist/usage-receipt-Cx3xNeyB.js.map +1 -0
- package/dist/{verdict-BQ3pCFf8.js → verdict-DhPoXApk.js} +2 -2
- package/dist/{verdict-BQ3pCFf8.js.map → verdict-DhPoXApk.js.map} +1 -1
- package/dist/{verdict-cache-B3eCVQtY.js → verdict-cache-BQEsVy9n.js} +2 -2
- package/dist/{verdict-cache-B3eCVQtY.js.map → verdict-cache-BQEsVy9n.js.map} +1 -1
- package/dist/wire/index.d.ts +1 -1
- package/dist/wire/index.js +1 -1
- package/docs/changes/evaluation-definition-identity.md +27 -0
- package/docs/changes/jev-evidence-review.md +18 -0
- package/docs/changes/jev-product-adoption.md +18 -0
- package/docs/changes/probability-policy-audit.md +81 -0
- package/docs/jev-behavior-review.md +174 -0
- package/docs/jev-integrations.md +50 -0
- package/docs/jev.md +140 -0
- package/docs/produced-state.md +13 -0
- package/docs/profile-json-identity.md +33 -0
- package/package.json +16 -1
- package/dist/agent-profile-cell-Cv6UA-W_.js.map +0 -1
- package/dist/canonical-DPyQ_rpt.js.map +0 -1
- package/dist/exporters-D98GPi1n.d.ts +0 -161
- package/dist/exporters-D98GPi1n.d.ts.map +0 -1
- package/dist/index--_8_Nlpn.d.ts.map +0 -1
- package/dist/index-Bp_6sj3x.d.ts.map +0 -1
- package/dist/mint-ySIIkKlV.js.map +0 -1
- package/dist/store-tool-spans-BZ8MPKFe.js.map +0 -1
- package/dist/store-tool-spans-Dt-YdAuE.d.ts.map +0 -1
- package/dist/tool-groups-B2bSNaJB.d.ts.map +0 -1
- package/dist/types-DQ0e2E7y.js.map +0 -1
|
@@ -1,8 +1,8 @@
|
|
|
1
|
-
import { D as TraceAnalysisStore, i as AnalystFinding, n as AnalystContext } from "./types-
|
|
2
|
-
import { l as ExactCapableAnalyst } from "./exact-types-
|
|
3
|
-
import { n as AnalystRegistry, r as AnalystRegistryOptions } from "./registry-
|
|
4
|
-
import { a as TraceAnalystLimits, i as TraceAnalysisEngineResult, l as RawAnalystFinding, n as TraceAnalysisEngine } from "./engine-
|
|
5
|
-
import { t as TraceToolGroupName } from "./tool-groups-
|
|
1
|
+
import { D as TraceAnalysisStore, i as AnalystFinding, n as AnalystContext } from "./types-BjB2wZM9.js";
|
|
2
|
+
import { l as ExactCapableAnalyst } from "./exact-types-BEHN7tnq.js";
|
|
3
|
+
import { n as AnalystRegistry, r as AnalystRegistryOptions } from "./registry-D5Ly3GRV.js";
|
|
4
|
+
import { a as TraceAnalystLimits, i as TraceAnalysisEngineResult, l as RawAnalystFinding, n as TraceAnalysisEngine } from "./engine-D1ykrAKR.js";
|
|
5
|
+
import { t as TraceToolGroupName } from "./tool-groups-qHMydbAH.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-DYB4aHVn.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"default-registry-
|
|
1
|
+
{"version":3,"file":"default-registry-DYB4aHVn.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;;iBAgHI,mBACd,YAAY,wBACZ,SAAS,4BACR,oBAAoB;iBA2RP,oBAAoB,OAAO;;;UChc1B;;;;;;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 { B as PowerPreflight, Mt as EvaluationClaim, ft as EvidenceReceipt, it as CampaignEvidenceContext } from "./statistical-heldout-
|
|
4
|
-
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, v as GateResult } from "./types-
|
|
5
|
-
import {
|
|
6
|
-
import { o as InsightReport } from "./insight-report-
|
|
7
|
-
import { n as HostedTenant } from "./client-
|
|
2
|
+
import { _ as ProposalFinding } from "./types-BjB2wZM9.js";
|
|
3
|
+
import { B as PowerPreflight, Mt as EvaluationClaim, ft as EvidenceReceipt, it as CampaignEvidenceContext } from "./statistical-heldout-Bu4oNOJh.js";
|
|
4
|
+
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, v as GateResult } from "./types-BTRcIqFr.js";
|
|
5
|
+
import { Aa as openAutoPr, Fa as RunCampaignOptions, Fi as RunImprovementLoopResult, Li as PremeasuredOptimizationBaseline, Pa as CampaignCellRetryPolicy, Qa as FinalEvidenceUse, Qr as OptimizationMethod, Ri as RunOptimizationOptions, Yn as LoopProvenanceRecord, Za as FinalEvidencePolicy, di as ComparisonCost, ia as SearchHistoryCoverageRow, ii as OptimizationMethodResult, ja as RunEvalOptions, oa as SearchHistoryReceipt, or as CampaignComparisonUnits, qa as CampaignStorage, ri as OptimizationMethodProvenance, ta as SearchHistoryAdmissionOptions } from "./index-ww8736Am.js";
|
|
6
|
+
import { o as InsightReport } from "./insight-report-C3Bfl762.js";
|
|
7
|
+
import { n as HostedTenant } from "./client-Bc8LANCG.js";
|
|
8
8
|
//#region src/campaign/presets/run-final-comparison.d.ts
|
|
9
9
|
interface FinalComparisonOptions<TScenario extends Scenario, TArtifact> extends Omit<RunCampaignOptions<TScenario, TArtifact>, 'dispatch'> {
|
|
10
10
|
baselineSurface: MutableSurface;
|
|
@@ -523,4 +523,4 @@ interface DefinedAgentEval<TScenario extends Scenario, TArtifact> {
|
|
|
523
523
|
declare function defineAgentEval<TScenario extends Scenario, TArtifact>(defaults: DefineAgentEvalOptions<TScenario, TArtifact>): DefinedAgentEval<TScenario, TArtifact>;
|
|
524
524
|
//#endregion
|
|
525
525
|
export { SelfImproveMethodResult as _, DefinedAgentEval as a, SelfImproveMethodOptions as c, SelfImproveProposerOptions as d, SelfImproveProposerResult as f, SelfImproveMethodProvenance as g, selfImprove as h, DefineAgentEvalOptions as i, SelfImproveOptions as l, SelfImproveRunError as m, AgentEvalEvaluateOptions as n, defineAgentEval as o, SelfImproveResult as p, AgentEvalImproveOptions as r, SelfImproveBudget as s, AgentEvalAgent as t, SelfImproveProgressEvent as u };
|
|
526
|
-
//# sourceMappingURL=define-agent-eval-
|
|
526
|
+
//# sourceMappingURL=define-agent-eval-DE67ECH1.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"define-agent-eval-
|
|
1
|
+
{"version":3,"file":"define-agent-eval-DE67ECH1.d.ts","names":[],"sources":["../src/campaign/presets/run-final-comparison.ts","../src/contract/self-improve-method.ts","../src/contract/self-improve.ts","../src/contract/define-agent-eval.ts"],"mappings":";;;;;;;;UAaiB,uBAAuB,kBAAkB,UAAU,mBAC1D,KAAK,mBAAmB,WAAW;EAC3C,iBAAiB;EACjB,eAAe;EACf,sBACE,SAAS,gBACT,UAAU,WACV,KAAK,WAAW,mBAAmB,WAAW,+BAC3C,QAAQ;EACb,MAAM,KAAK,WAAW;EACtB,QAAQ;EACR,gBAAgB;EAChB;EACA;EACA,cAAc,QAAQ,gBAAgB,UAAU,mBAAmB;;;iBAI/C,mBAAmB,kBAAkB,UAAU,WACnE,MAAM,uBAAuB,WAAW,aAAU;;;;;;;;;;;;kBAA5C;;;;;;;;;;;UCKS;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,oBAAoB,YAAY,0BAA0B;EAC1D,QAAQ;EACR;IACE;IACA;IACA;IACA;IACA;IACA;IACA;IACA,sBAAsB;;EAExB,MAAM,QAAQ,kBAAkB;EAChC;EACA;EACA;EACA;EACA,MAAM;EACN;EACA;;UAGe,wBAAwB,kBAAkB,UAAU,mBAC3D,KACN,0BAA0B,WAAW;EAWvC;IAAa,UAAU;IAAiB,QAAQ;;EAChD,uBAAuB;EACvB;;EAEA,UAAU,0BAA0B,WAAW;EAC/C,QAAQ,KAAK,0BAA0B,WAAW;IAChD;;EAEF,YAAY;EACZ,cAAc,YAAY,0BAA0B,WAAW;;EAE/D,MAAM;;EAEN,YAAY;EACZ,UAAU,0BAA0B,WAAW;EAC/C,KAAK,QAAQ,kBAAkB,mBAAmB,WAAW;IAC3D;IACA,iBAAiB,0BAA0B,WAAW;IACtD,eAAe,0BAA0B,WAAW;IACpD;IACA,QAAQ;IACR,MAAM;IACN,WAAW,kBAAkB;;;;;UCnBhB;;;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,mBACtD;;;;;;;;;;;;;;EAcR,QAAQ,SAAS,gBAAgB,UAAU,WAAW,KAAK,oBAAoB,QAAQ;;;;;;;EAQvF;;EAEA;;;EAIA,WAAW;;EAGX,OAAO,YAAY,WAAW;;;EAI9B,iBAAiB;;EAGjB,SAAS;;;;;;;;;;;;;EAcT,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,uBAAuB;;;;;EAM/C,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;;EAEjD,WAAW;;EAEX,QAAQ;;EAER,gBAAgB;;UAGD,0BAA0B,kBAAkB,UAAU;EACrE;EACA,QAAQ;EACR,gBAAgB;;;;EAIhB;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;;;;EAIT,QAAQ;;;;;;EAMR,KAAK,yBAAyB,WAAW;;KAG/B,kBAAkB,kBAAkB,UAAU,aACtD,0BAA0B,WAAW,aACrC,wBAAwB,WAAW;KAE3B,yBAAyB,kBAAkB,UAAU,aAAa,KAC5E,mBAAmB,WAAW;EAG9B,QAAQ,mBAAmB,WAAW;EACtC;EACA,gBAAgB,QAAQ;;KAGd,2BAA2B,kBAAkB,UAAU,aAAa,KAC9E,mBAAmB,WAAW;EAG9B;EACA,gBAAgB,QAAQ;;;cAIb,4BAA4B;WAC9B,MAAM;WACN,UAAU;EAEnB,YAAY,gBAAgB,QAAQ;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAsNtB,YAAY,kBAAkB,UAAU,WACtD,MAAM,yBAAyB,WAAW,aACzC,QAAQ,wBAAwB,WAAW;iBAC9B,YAAY,kBAAkB,UAAU,WACtD,MAAM,2BAA2B,WAAW,aAC3C,QAAQ,0BAA0B,WAAW;iBAChC,YAAY,kBAAkB,UAAU,WACtD,MAAM,mBAAmB,WAAW,aACnC,QAAQ,kBAAkB,WAAW;;;KC/oB5B,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"}
|
|
@@ -6,14 +6,14 @@ import { i as requiredPairedSampleSize, r as pairedMde } from "./power-and-mde-B
|
|
|
6
6
|
import { r as pairRunRecords } from "./paired-arms-JSDcs2zx.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 { a as campaignCellToRunRecord, i as campaignCellTaskScore, n as campaignCellExecutionEvidence, r as campaignCellJudgeDimensions } from "./run-record-
|
|
10
|
-
import { B as surfaceDispatchRef, N as defineEvaluationClaim, T as pairedCampaignComposites, V as surfaceHash, Z as campaignSplitDigest, f as campaignMeasurementDigest, h as loopProvenanceArgsFromResult, k as powerPreflight, m as emitLoopProvenance, p as canonicalDigest, t as createCampaignEvidenceReceipt, tt as aggregatePairedHoldout, z as surfaceContentHash } from "./campaign-evidence-
|
|
11
|
-
import "./query-
|
|
9
|
+
import { a as campaignCellToRunRecord, i as campaignCellTaskScore, n as campaignCellExecutionEvidence, r as campaignCellJudgeDimensions } from "./run-record-BhMVsHR1.js";
|
|
10
|
+
import { B as surfaceDispatchRef, N as defineEvaluationClaim, T as pairedCampaignComposites, V as surfaceHash, Z as campaignSplitDigest, f as campaignMeasurementDigest, h as loopProvenanceArgsFromResult, k as powerPreflight, m as emitLoopProvenance, p as canonicalDigest, t as createCampaignEvidenceReceipt, tt as aggregatePairedHoldout, z as surfaceContentHash } from "./campaign-evidence-aWEmfROB.js";
|
|
11
|
+
import "./query-DlV0pdx4.js";
|
|
12
12
|
import { r as observedSplitScore } from "./reward-nw2xZGZG.js";
|
|
13
|
-
import { c as validateRunRecord, i as modelHasSnapshot } from "./run-record-
|
|
14
|
-
import { r as paretoChart } from "./summary-report-
|
|
15
|
-
import { C as assertSearchHistoryAdmissionOptions, G as assertIndependentEvaluationSplit, I as runFinalComparison, J as reserveFinalEvidence, K as captureFinalEvidencePolicy, L as openAutoPr, Q as resolveRunDir, R as captureJudge, Y as runEval, d as combineComparisonCosts, f as costFromLedgerSummary, h as runImprovementLoop, q as evaluationUnitMap, u as executeOptimizationMethod, z as defaultProductionGate } from "./llm-judge-
|
|
16
|
-
import { C as inMemoryCampaignStorage, S as fsCampaignStorage, x as createRunCostLedger } from "./external-optimizer-subprocess-
|
|
13
|
+
import { c as validateRunRecord, i as modelHasSnapshot } from "./run-record-tg_5H2IK.js";
|
|
14
|
+
import { r as paretoChart } from "./summary-report-BMvou2oZ.js";
|
|
15
|
+
import { C as assertSearchHistoryAdmissionOptions, G as assertIndependentEvaluationSplit, I as runFinalComparison, J as reserveFinalEvidence, K as captureFinalEvidencePolicy, L as openAutoPr, Q as resolveRunDir, R as captureJudge, Y as runEval, d as combineComparisonCosts, f as costFromLedgerSummary, h as runImprovementLoop, q as evaluationUnitMap, u as executeOptimizationMethod, z as defaultProductionGate } from "./llm-judge-C5wbADA5.js";
|
|
16
|
+
import { C as inMemoryCampaignStorage, S as fsCampaignStorage, x as createRunCostLedger } from "./external-optimizer-subprocess-DMgmqTB6.js";
|
|
17
17
|
import { t as createHostedClient } from "./client-CXE-U1SA.js";
|
|
18
18
|
//#region src/contamination-guard.ts
|
|
19
19
|
function checkCanaries(output, scenarios) {
|
|
@@ -1781,4 +1781,4 @@ function requirePositiveInteger(value, field) {
|
|
|
1781
1781
|
//#endregion
|
|
1782
1782
|
export { summarizeExecution as a, analyzeRuns as i, SelfImproveRunError as n, checkCanaries as o, selfImprove as r, defineAgentEval as t };
|
|
1783
1783
|
|
|
1784
|
-
//# sourceMappingURL=define-agent-eval-
|
|
1784
|
+
//# sourceMappingURL=define-agent-eval-RTekxhUE.js.map
|