@tangle-network/agent-eval 0.222.0 → 0.223.0
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 +177 -0
- package/dist/analyst/index.d.ts.map +1 -1
- package/dist/analyst/index.js +1 -1
- package/dist/{analyze-runs-CTheASYV.js → analyze-runs-Ck68mbgj.js} +7 -9
- package/dist/analyze-runs-Ck68mbgj.js.map +1 -0
- package/dist/{benchmark-command-jYWR2gya.js → benchmark-command-CH86twZj.js} +32 -37
- package/dist/benchmark-command-CH86twZj.js.map +1 -0
- package/dist/benchmarks/index.js +3 -3
- package/dist/campaign/index.js +9 -9
- package/dist/{campaign-ddr1N3yD.js → campaign-DdOgQfp8.js} +14 -19
- package/dist/campaign-DdOgQfp8.js.map +1 -0
- package/dist/{campaign-evidence-DN8rWuBj.js → campaign-evidence-MjCjkb5a.js} +3 -3
- package/dist/{campaign-evidence-DN8rWuBj.js.map → campaign-evidence-MjCjkb5a.js.map} +1 -1
- package/dist/cli.js +4 -4
- package/dist/contract/index.js +11 -10
- package/dist/contract/index.js.map +1 -1
- package/dist/{define-agent-eval-BLoP4Lf9.js → define-agent-eval-Cc15qGik.js} +5 -5
- package/dist/{define-agent-eval-BLoP4Lf9.js.map → define-agent-eval-Cc15qGik.js.map} +1 -1
- package/dist/diagnosis/index.js +1 -1
- package/dist/{eval-campaign-Cgixrvf7.js → eval-campaign-C_mNDshv.js} +2 -2
- package/dist/{eval-campaign-Cgixrvf7.js.map → eval-campaign-C_mNDshv.js.map} +1 -1
- package/dist/experiment/index.js +7 -7
- package/dist/index-B7v2QHlD.d.ts.map +1 -1
- package/dist/index-DAk2quYu.d.ts.map +1 -1
- package/dist/index-aDPVYBIM.d.ts.map +1 -1
- package/dist/index.js +13 -12
- package/dist/index.js.map +1 -1
- package/dist/{judge-gate-C2BCOt-M.js → judge-gate-Di_6bifh.js} +4 -4
- package/dist/judge-gate-Di_6bifh.js.map +1 -0
- package/dist/{llm-judge-cgbhTKQp.js → llm-judge-D6JtIqiU.js} +9 -9
- package/dist/{llm-judge-cgbhTKQp.js.map → llm-judge-D6JtIqiU.js.map} +1 -1
- package/dist/mean-BPuUNznQ.js +16 -0
- package/dist/mean-BPuUNznQ.js.map +1 -0
- package/dist/meta-eval/index.js +1 -1
- package/dist/openapi.json +1 -1
- package/dist/{paired-arms-D-54hO-k.js → paired-arms-D5H_58e8.js} +2 -2
- package/dist/{paired-arms-D-54hO-k.js.map → paired-arms-D5H_58e8.js.map} +1 -1
- package/dist/{paired-promotion-decision-FGvHA4p_.js → paired-promotion-decision-BiVumgvM.js} +2 -2
- package/dist/{paired-promotion-decision-FGvHA4p_.js.map → paired-promotion-decision-BiVumgvM.js.map} +1 -1
- package/dist/{mean-CdDYG-us.js → paired-tests-C2zUqgUW.js} +2 -7
- package/dist/paired-tests-C2zUqgUW.js.map +1 -0
- package/dist/{pareto-B9udAt2i.js → pareto-7_LEmYGG.js} +4 -3
- package/dist/{pareto-B9udAt2i.js.map → pareto-7_LEmYGG.js.map} +1 -1
- package/dist/{power-and-mde-ClI6RWw4.js → power-and-mde-CHvyDO_H.js} +2 -2
- package/dist/{power-and-mde-ClI6RWw4.js.map → power-and-mde-CHvyDO_H.js.map} +1 -1
- package/dist/{promotion-policy-CBYeD4zi.js → promotion-policy-D7B4_jEZ.js} +4 -4
- package/dist/{promotion-policy-CBYeD4zi.js.map → promotion-policy-D7B4_jEZ.js.map} +1 -1
- package/dist/reporting.js +3 -3
- package/dist/{reward-hacking-DZzHZlS1.js → reward-hacking-Bjk3MAjh.js} +7 -9
- package/dist/{reward-hacking-DZzHZlS1.js.map → reward-hacking-Bjk3MAjh.js.map} +1 -1
- package/dist/reward-hacking-C9hPVT0Q.d.ts.map +1 -1
- package/dist/rl.js +6 -6
- package/dist/{run-record-CZy4-TnZ.js → run-record-BUm-Jggo.js} +2 -2
- package/dist/{run-record-CZy4-TnZ.js.map → run-record-BUm-Jggo.js.map} +1 -1
- package/dist/search/index.js +2 -2
- package/dist/{search-summary-Ct7SrmVc.js → search-summary-Bt3uuK53.js} +4 -4
- package/dist/{search-summary-Ct7SrmVc.js.map → search-summary-Bt3uuK53.js.map} +1 -1
- package/dist/{sequential-B1I0tRfL.js → sequential-De7BTAlV.js} +2 -2
- package/dist/{sequential-B1I0tRfL.js.map → sequential-De7BTAlV.js.map} +1 -1
- package/dist/{skillopt-optimization-method-qB1M554K.js → skillopt-optimization-method-BKFbJEyz.js} +9 -12
- package/dist/skillopt-optimization-method-BKFbJEyz.js.map +1 -0
- package/dist/{statistical-heldout-TP6ndL0B.js → statistical-heldout-C9hz_AF9.js} +3 -3
- package/dist/{statistical-heldout-TP6ndL0B.js.map → statistical-heldout-C9hz_AF9.js.map} +1 -1
- package/dist/storyboard/index.d.ts +17 -0
- package/dist/storyboard/index.d.ts.map +1 -1
- package/dist/storyboard/index.js +154 -107
- package/dist/storyboard/index.js.map +1 -1
- package/dist/{summary-report-CPfsrLac.js → summary-report-BLnTyQKH.js} +5 -4
- package/dist/{summary-report-CPfsrLac.js.map → summary-report-BLnTyQKH.js.map} +1 -1
- package/dist/trace-repair/index.js +2 -1
- package/dist/trace-repair/index.js.map +1 -1
- package/dist/{tree-DrIChF4a.js → tree-DnBkWXun.js} +5 -4
- package/dist/{tree-DrIChF4a.js.map → tree-DnBkWXun.js.map} +1 -1
- package/package.json +2 -2
- package/dist/analyze-runs-CTheASYV.js.map +0 -1
- package/dist/benchmark-command-jYWR2gya.js.map +0 -1
- package/dist/campaign-ddr1N3yD.js.map +0 -1
- package/dist/judge-gate-C2BCOt-M.js.map +0 -1
- package/dist/mean-CdDYG-us.js.map +0 -1
- package/dist/skillopt-optimization-method-qB1M554K.js.map +0 -1
|
@@ -1,9 +1,9 @@
|
|
|
1
1
|
import { t as AgentEvalError } from "./errors-DeEEV9Vx.js";
|
|
2
2
|
import { a as hashCanonical, i as compareCodeUnits, r as canonicalString } from "./canonical-CicJlINA.js";
|
|
3
|
-
import { n as aggregatePairedHoldout, o as pairHoldout } from "./statistical-heldout-
|
|
3
|
+
import { n as aggregatePairedHoldout, o as pairHoldout } from "./statistical-heldout-C9hz_AF9.js";
|
|
4
4
|
import { t as FileLedgerJournal } from "./journal-C1uLJi0w.js";
|
|
5
5
|
import { n as contentHash } from "./verdict-cache-BVH-fl18.js";
|
|
6
|
-
import { r as projectCampaignCellQuality } from "./run-record-
|
|
6
|
+
import { r as projectCampaignCellQuality } from "./run-record-BUm-Jggo.js";
|
|
7
7
|
import { l as surfaceHash, o as renderSurfaceDiff, s as surfaceContentHash, t as autoCloseTruncatedJson, u as surfaceHashMatches } from "./json-recovery-DK4MDm5r.js";
|
|
8
8
|
import { t as defineEvaluationClaim } from "./claim-Dod_mjk3.js";
|
|
9
9
|
import { createHash } from "node:crypto";
|
|
@@ -1428,4 +1428,4 @@ function createCampaignEvidenceReceipt(input) {
|
|
|
1428
1428
|
}
|
|
1429
1429
|
export { campaignSplitDigest as A, assertRealBackend as C, assertCompleteCampaign as D, assertCampaignSplitIdentity as E, formatCoverageFailures as M, campaignCoverage as O, BackendIntegrityError as S, assertCampaignDesign as T, parseReflectionResponse as _, ATTESTATION_ALGORITHM as a, FinalEvidenceError as b, campaignMeasurementDigest as c, loopProvenanceArgsFromResult as d, campaignBreakdown as f, buildReflectionPrompt as g, pairedCampaignComposites as h, verifyEvidenceReceipt as i, campaignSplitDigestFromIdentities as j, campaignScenarioIdentity as k, canonicalDigest as l, campaignMeanCompositeOrNull as m, createEvidenceReceipt as n, attest as o, campaignMeanComposite as p, isIndependentEvidence as r, verifyAttestation as s, createCampaignEvidenceReceipt as t, emitLoopProvenance as u, powerPreflight as v, summarizeBackendIntegrity as w, openFinalEvidenceLedger as x, FinalEvidenceConflictError as y };
|
|
1430
1430
|
|
|
1431
|
-
//# sourceMappingURL=campaign-evidence-
|
|
1431
|
+
//# sourceMappingURL=campaign-evidence-MjCjkb5a.js.map
|