@tangle-network/agent-eval 0.144.6 → 0.144.8
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 +30 -0
- package/README.md +2 -0
- package/dist/{benchmark-J9Qe6j2_.d.ts → agent-profile-CgDTo40f.d.ts} +91 -4
- package/dist/agent-profile-CgDTo40f.d.ts.map +1 -0
- package/dist/analyst/index.d.ts +470 -88
- package/dist/analyst/index.d.ts.map +1 -1
- package/dist/analyst/index.js +24 -5
- package/dist/analyst/index.js.map +1 -1
- package/dist/{analyze-runs-DWIvOAGk.js → analyze-runs-BNNK7irB.js} +3 -3
- package/dist/{analyze-runs-DWIvOAGk.js.map → analyze-runs-BNNK7irB.js.map} +1 -1
- package/dist/{baseline-D_fT6277.d.ts → baseline-CavEbRyH.d.ts} +2 -23
- package/dist/baseline-CavEbRyH.d.ts.map +1 -0
- package/dist/{benchmark-command-CQd78YHt.js → benchmark-command-BKENp2s5.js} +662 -605
- package/dist/benchmark-command-BKENp2s5.js.map +1 -0
- package/dist/benchmarks/index.d.ts +1 -1
- package/dist/benchmarks/index.js +1 -1
- package/dist/{benchmarks-BEOkuvIg.js → benchmarks-BlPmjd88.js} +4 -4
- package/dist/{benchmarks-BEOkuvIg.js.map → benchmarks-BlPmjd88.js.map} +1 -1
- package/dist/campaign/index.d.ts +7 -5
- package/dist/campaign/index.js +5 -3
- package/dist/{campaign-CXsdyym7.js → campaign--HVSuvV0.js} +17 -301
- package/dist/campaign--HVSuvV0.js.map +1 -0
- package/dist/cli.js +2 -2
- package/dist/{client-0JI64ovJ.d.ts → client-DjXROWpx.d.ts} +3 -3
- package/dist/{client-0JI64ovJ.d.ts.map → client-DjXROWpx.d.ts.map} +1 -1
- package/dist/{completion-verifier-CBiee74w.d.ts → completion-verifier-foUCLif_.d.ts} +5 -5
- package/dist/{completion-verifier-CBiee74w.d.ts.map → completion-verifier-foUCLif_.d.ts.map} +1 -1
- package/dist/contract/index.d.ts +9 -8
- package/dist/contract/index.d.ts.map +1 -1
- package/dist/contract/index.js +7 -6
- package/dist/contract/index.js.map +1 -1
- package/dist/control.d.ts +2 -2
- package/dist/control.js +1 -1
- package/dist/counterfactual-CWPTrMH7.js +126 -0
- package/dist/counterfactual-CWPTrMH7.js.map +1 -0
- package/dist/counterfactual-CxmxAONP.d.ts +72 -0
- package/dist/counterfactual-CxmxAONP.d.ts.map +1 -0
- package/dist/{default-registry-J9m-_tya.d.ts → default-registry-BZhStdjl.d.ts} +6 -5
- package/dist/default-registry-BZhStdjl.d.ts.map +1 -0
- package/dist/{default-registry-Dta70shL.js → default-registry-BaQXW1Ow.js} +2 -2
- package/dist/{default-registry-Dta70shL.js.map → default-registry-BaQXW1Ow.js.map} +1 -1
- package/dist/{dspy-rlm-engine-19FQEMBK.js → dspy-rlm-engine-BiN49gK6.js} +4 -3
- package/dist/{dspy-rlm-engine-19FQEMBK.js.map → dspy-rlm-engine-BiN49gK6.js.map} +1 -1
- package/dist/{tool-groups-CK0JCkqO.d.ts → engine-nB64f48I.d.ts} +18 -31
- package/dist/engine-nB64f48I.d.ts.map +1 -0
- package/dist/{eval-campaign-CfLQQs9B.js → eval-campaign-DNjCvAm-.js} +7 -6
- package/dist/{eval-campaign-CfLQQs9B.js.map → eval-campaign-DNjCvAm-.js.map} +1 -1
- package/dist/{exact-types-CBYF5MGd.d.ts → exact-types-Djvzosly.d.ts} +2 -2
- package/dist/{exact-types-CBYF5MGd.d.ts.map → exact-types-Djvzosly.d.ts.map} +1 -1
- package/dist/exec-BLtYZdWo.js +49 -0
- package/dist/exec-BLtYZdWo.js.map +1 -0
- package/dist/experiment/index.d.ts +802 -0
- package/dist/experiment/index.d.ts.map +1 -0
- package/dist/experiment/index.js +1108 -0
- package/dist/experiment/index.js.map +1 -0
- package/dist/experiment-tracker-CnRICnMl.js +500 -0
- package/dist/experiment-tracker-CnRICnMl.js.map +1 -0
- package/dist/experiment-tracker-IMntXr6J.d.ts +299 -0
- package/dist/experiment-tracker-IMntXr6J.d.ts.map +1 -0
- package/dist/{external-optimizer-contracts-Q9c0L3eY.d.ts → external-optimizer-contracts-CZuJNcT5.d.ts} +2 -2
- package/dist/{external-optimizer-contracts-Q9c0L3eY.d.ts.map → external-optimizer-contracts-CZuJNcT5.d.ts.map} +1 -1
- package/dist/{feedback-trajectory-GgoS0-MK.d.ts → feedback-trajectory-Rh280oXo.d.ts} +3 -3
- package/dist/{feedback-trajectory-GgoS0-MK.d.ts.map → feedback-trajectory-Rh280oXo.d.ts.map} +1 -1
- package/dist/fuzz.d.ts +1 -1
- package/dist/hosted/index.d.ts +3 -3
- package/dist/{index-BZUe-ODI.d.ts → index-BZ3-y4YL.d.ts} +3 -3
- package/dist/{index-BZUe-ODI.d.ts.map → index-BZ3-y4YL.d.ts.map} +1 -1
- package/dist/{index-DSC51roc.d.ts → index-C3ssXVLv.d.ts} +2 -2
- package/dist/{index-DSC51roc.d.ts.map → index-C3ssXVLv.d.ts.map} +1 -1
- package/dist/{index-4XwggC10.d.ts → index-C5HOo4ZF2.d.ts} +4 -4
- package/dist/index-C5HOo4ZF2.d.ts.map +1 -0
- package/dist/{index-B6-B0zTB.d.ts → index-CvXXlyz7.d.ts} +2 -2
- package/dist/{index-B6-B0zTB.d.ts.map → index-CvXXlyz7.d.ts.map} +1 -1
- package/dist/{index-Dx1kF3Ez.d.ts → index-CwDrUMe0.d.ts} +2 -2
- package/dist/{index-Dx1kF3Ez.d.ts.map → index-CwDrUMe0.d.ts.map} +1 -1
- package/dist/{index-BIL5vxxt.d.ts → index-Sh2I0DRc.d.ts} +11 -645
- package/dist/index-Sh2I0DRc.d.ts.map +1 -0
- package/dist/index.d.ts +214 -404
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +233 -649
- package/dist/index.js.map +1 -1
- package/dist/{insight-report-DqEsugpr.d.ts → insight-report-C6h6F_4L.d.ts} +3 -3
- package/dist/{insight-report-DqEsugpr.d.ts.map → insight-report-C6h6F_4L.d.ts.map} +1 -1
- package/dist/{integrity-CNGUaGBY.d.ts → integrity-BuqEKu-x.d.ts} +2 -2
- package/dist/{integrity-CNGUaGBY.d.ts.map → integrity-BuqEKu-x.d.ts.map} +1 -1
- package/dist/integrity-MLzHOfV9.js +141 -0
- package/dist/integrity-MLzHOfV9.js.map +1 -0
- package/dist/kind-factory-BHIgPmzS.js.map +1 -1
- package/dist/{llm-client-Dv5BiKLE.js → llm-client-DzvMUsS_.js} +24 -6
- package/dist/llm-client-DzvMUsS_.js.map +1 -0
- package/dist/matrix/index.d.ts +2 -2
- package/dist/meta-eval/index.d.ts +1 -1
- package/dist/{mint-DD-0oQTA.js → mint-CGEkzPLf.js} +2 -2
- package/dist/{mint-DD-0oQTA.js.map → mint-CGEkzPLf.js.map} +1 -1
- package/dist/{multi-layer-verifier-BHY1gWAc.d.ts → multi-layer-verifier-DnAqwl0h.d.ts} +2 -2
- package/dist/{multi-layer-verifier-BHY1gWAc.d.ts.map → multi-layer-verifier-DnAqwl0h.d.ts.map} +1 -1
- package/dist/multishot/index.d.ts +2 -2
- package/dist/openapi.json +1 -1
- package/dist/paired-promotion-decision-B6zJ3gYM.d.ts +114 -0
- package/dist/paired-promotion-decision-B6zJ3gYM.d.ts.map +1 -0
- package/dist/pipelines/index.d.ts +2 -1
- package/dist/pipelines/index.d.ts.map +1 -1
- package/dist/pre-registration-DakwTRXk.js +96 -0
- package/dist/pre-registration-DakwTRXk.js.map +1 -0
- package/dist/prime-bridge-transport-6feEglLf.d.ts +49 -0
- package/dist/prime-bridge-transport-6feEglLf.d.ts.map +1 -0
- package/dist/prime-protocol-BfSalTfR.js +453 -0
- package/dist/prime-protocol-BfSalTfR.js.map +1 -0
- package/dist/profile-cell.js +242 -1
- package/dist/profile-cell.js.map +1 -0
- package/dist/promotion-policy-ChWhTDBH.d.ts +289 -0
- package/dist/promotion-policy-ChWhTDBH.d.ts.map +1 -0
- package/dist/promotion-policy-CrLrmys8.js +682 -0
- package/dist/promotion-policy-CrLrmys8.js.map +1 -0
- package/dist/{propose-review-control-BciUCZoh.js → propose-review-control-qgWLA6E9.js} +2 -2
- package/dist/{propose-review-control-BciUCZoh.js.map → propose-review-control-qgWLA6E9.js.map} +1 -1
- package/dist/{release-report-ChOgpIoQ.d.ts → release-report-CI8uisI1.d.ts} +2 -2
- package/dist/{release-report-ChOgpIoQ.d.ts.map → release-report-CI8uisI1.d.ts.map} +1 -1
- package/dist/{release-report-Sl0xfkFv.js → release-report-Dy39cbFF.js} +3 -3
- package/dist/{release-report-Sl0xfkFv.js.map → release-report-Dy39cbFF.js.map} +1 -1
- package/dist/{replay-Krvb114g.d.ts → replay-DFf-teiC.d.ts} +5 -4
- package/dist/replay-DFf-teiC.d.ts.map +1 -0
- package/dist/{replay-GW61ezMW.js → replay-DyBLaKFc.js} +3 -3
- package/dist/{replay-GW61ezMW.js.map → replay-DyBLaKFc.js.map} +1 -1
- package/dist/reporting.d.ts +3 -3
- package/dist/reporting.js +2 -2
- package/dist/{researcher-xLeNcpKX.d.ts → researcher-BoaxeCzP.d.ts} +4 -4
- package/dist/{researcher-xLeNcpKX.d.ts.map → researcher-BoaxeCzP.d.ts.map} +1 -1
- package/dist/{reward-hacking-CyuzxKly.js → reward-hacking-B2tPL9a3.js} +12 -3
- package/dist/reward-hacking-B2tPL9a3.js.map +1 -0
- package/dist/{reward-hacking-RZgnGWlx.d.ts → reward-hacking-Cf1PtEOz.d.ts} +33 -3
- package/dist/reward-hacking-Cf1PtEOz.d.ts.map +1 -0
- package/dist/rl.d.ts +17 -7
- package/dist/rl.d.ts.map +1 -1
- package/dist/rl.js +16 -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-C2fD1cf4.js → rollout-2ECTXb2N.js} +2 -2
- package/dist/{rollout-C2fD1cf4.js.map → rollout-2ECTXb2N.js.map} +1 -1
- package/dist/{run-evidence-C6G41MSI.d.ts → run-evidence-BDFFai9R.d.ts} +2 -2
- package/dist/{run-evidence-C6G41MSI.d.ts.map → run-evidence-BDFFai9R.d.ts.map} +1 -1
- package/dist/{run-record-CWN8-VsV.js → run-record-DqOw5X6_.js} +2 -2
- package/dist/{run-record-CWN8-VsV.js.map → run-record-DqOw5X6_.js.map} +1 -1
- package/dist/{schema-Cef2cFmb.d.ts → schema-Cef2cFmb2.d.ts} +1 -1
- package/dist/schema-Cef2cFmb2.d.ts.map +1 -0
- package/dist/{semantic-concept-judge-DwF6n05O.js → semantic-concept-judge-Bmrq6yqU.js} +2 -2
- package/dist/{semantic-concept-judge-DwF6n05O.js.map → semantic-concept-judge-Bmrq6yqU.js.map} +1 -1
- package/dist/sequential-D-BLJBKU.js +299 -0
- package/dist/sequential-D-BLJBKU.js.map +1 -0
- package/dist/{server-D6XJQHw7.js → server-iu0ede49.js} +2 -2
- package/dist/{server-D6XJQHw7.js.map → server-iu0ede49.js.map} +1 -1
- package/dist/{single-run-lock-D5iN0Xzb.js → single-run-lock-BMQEv1wG.js} +7 -161
- package/dist/single-run-lock-BMQEv1wG.js.map +1 -0
- package/dist/{skill-usage-GlOphAhX.d.ts → skill-usage-CJlWEUFt.d.ts} +10 -10
- package/dist/{skill-usage-GlOphAhX.d.ts.map → skill-usage-CJlWEUFt.d.ts.map} +1 -1
- package/dist/{skillopt-optimization-method-7S43rbDB.d.ts → skillopt-optimization-method-BO7NAl3b.d.ts} +8 -294
- package/dist/skillopt-optimization-method-BO7NAl3b.d.ts.map +1 -0
- package/dist/{skillopt-optimization-method-Bfb-vBKe.js → skillopt-optimization-method-CkaI2ly4.js} +19 -654
- package/dist/skillopt-optimization-method-CkaI2ly4.js.map +1 -0
- package/dist/statistical-heldout-Dn9ruizm.d.ts +555 -0
- package/dist/statistical-heldout-Dn9ruizm.d.ts.map +1 -0
- package/dist/{statistics-C-dm-J6H.d.ts → statistics-D6Uebe_4.d.ts} +2 -2
- package/dist/{statistics-C-dm-J6H.d.ts.map → statistics-D6Uebe_4.d.ts.map} +1 -1
- package/dist/steps-BArUxhna.d.ts +51 -0
- package/dist/steps-BArUxhna.d.ts.map +1 -0
- package/dist/{integrity-fdt8XPAv.js → store-DNe_Uv1Q.js} +2 -140
- package/dist/store-DNe_Uv1Q.js.map +1 -0
- package/dist/{summary-report-B0cAyA7N.d.ts → summary-report-DuUS_i7W.d.ts} +3 -114
- package/dist/summary-report-DuUS_i7W.d.ts.map +1 -0
- package/dist/{summary-report-9A5y7EsK.js → summary-report-Lf-5I7xh.js} +2 -2
- package/dist/{summary-report-9A5y7EsK.js.map → summary-report-Lf-5I7xh.js.map} +1 -1
- package/dist/supervisor-run/index.d.ts +2 -2
- package/dist/supervisor-run/index.js +1 -1
- package/dist/{supervisor-run-DiyQVczd.js → supervisor-run-D_sokXcO.js} +32 -8
- package/dist/{supervisor-run-DiyQVczd.js.map → supervisor-run-D_sokXcO.js.map} +1 -1
- package/dist/tool-groups-DIVBnJyl.d.ts +28 -0
- package/dist/tool-groups-DIVBnJyl.d.ts.map +1 -0
- package/dist/trace-repair/index.d.ts +2102 -0
- package/dist/trace-repair/index.d.ts.map +1 -0
- package/dist/trace-repair/index.js +3878 -0
- package/dist/trace-repair/index.js.map +1 -0
- package/dist/traces.d.ts +6 -6
- package/dist/traces.js +3 -2
- package/dist/trajectory-YC15QDYQ.d.ts +24 -0
- package/dist/trajectory-YC15QDYQ.d.ts.map +1 -0
- package/dist/trajectory-replay/index.d.ts +781 -0
- package/dist/trajectory-replay/index.d.ts.map +1 -0
- package/dist/trajectory-replay/index.js +2103 -0
- package/dist/trajectory-replay/index.js.map +1 -0
- package/dist/{types-XMVEdrE_.d.ts → types-D216SgwM.d.ts} +24 -6
- package/dist/types-D216SgwM.d.ts.map +1 -0
- package/dist/{types-yLK8gXE9.d.ts → types-D4mog56g.d.ts} +2 -2
- package/dist/{types-yLK8gXE9.d.ts.map → types-D4mog56g.d.ts.map} +1 -1
- package/dist/{types-BhP9q0Fq.d.ts → types-DF_Udrp-.d.ts} +52 -3
- package/dist/{types-BhP9q0Fq.d.ts.map → types-DF_Udrp-.d.ts.map} +1 -1
- package/dist/{types-DOZyvsFU.d.ts → types-DYuNHo9R.d.ts} +3 -3
- package/dist/{types-DOZyvsFU.d.ts.map → types-DYuNHo9R.d.ts.map} +1 -1
- package/dist/usage-receipt-EVI8B8Xu.js.map +1 -1
- package/dist/verdict-DExhxfgR.d.ts +201 -0
- package/dist/verdict-DExhxfgR.d.ts.map +1 -0
- package/dist/verdict-cache-BCcOh0kF.js +159 -0
- package/dist/verdict-cache-BCcOh0kF.js.map +1 -0
- package/dist/wire/index.d.ts +2 -2
- package/dist/wire/index.js +1 -1
- package/docs/campaign-proposers.md +5 -0
- package/docs/charter.md +112 -0
- package/docs/experiment.md +104 -0
- package/docs/prime-analyst.md +1 -0
- package/docs/trace-analysis.md +26 -0
- package/docs/trace-repair-admission.md +194 -0
- package/docs/trace-repair-analyst-arms.md +121 -0
- package/docs/trace-repair-continuation.md +107 -0
- package/docs/trace-repair-grader.md +163 -0
- package/docs/trajectory-replay.md +110 -0
- package/docs/verification-strategies.md +103 -0
- package/package.json +19 -2
- package/dist/agent-profile-cell-CbfBm2g6.js +0 -335
- package/dist/agent-profile-cell-CbfBm2g6.js.map +0 -1
- package/dist/baseline-D_fT6277.d.ts.map +0 -1
- package/dist/benchmark-J9Qe6j2_.d.ts.map +0 -1
- package/dist/benchmark-command-CQd78YHt.js.map +0 -1
- package/dist/campaign-CXsdyym7.js.map +0 -1
- package/dist/default-registry-J9m-_tya.d.ts.map +0 -1
- package/dist/index-4XwggC10.d.ts.map +0 -1
- package/dist/index-BIL5vxxt.d.ts.map +0 -1
- package/dist/integrity-fdt8XPAv.js.map +0 -1
- package/dist/llm-client-Dv5BiKLE.js.map +0 -1
- package/dist/replay-Krvb114g.d.ts.map +0 -1
- package/dist/reward-hacking-CyuzxKly.js.map +0 -1
- package/dist/reward-hacking-RZgnGWlx.d.ts.map +0 -1
- package/dist/schema-Cef2cFmb.d.ts.map +0 -1
- package/dist/single-run-lock-D5iN0Xzb.js.map +0 -1
- package/dist/skillopt-optimization-method-7S43rbDB.d.ts.map +0 -1
- package/dist/skillopt-optimization-method-Bfb-vBKe.js.map +0 -1
- package/dist/summary-report-B0cAyA7N.d.ts.map +0 -1
- package/dist/tool-groups-CK0JCkqO.d.ts.map +0 -1
- package/dist/types-XMVEdrE_.d.ts.map +0 -1
- package/dist/verdict-Dps8_okt.d.ts +0 -37
- package/dist/verdict-Dps8_okt.d.ts.map +0 -1
package/dist/cli.js
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
#!/usr/bin/env node
|
|
2
2
|
import { o as runRolloutReleaseCli } from "./hf-dataset-XggBupCr.js";
|
|
3
|
-
import { n as runAnalystBenchmarkCommand } from "./benchmark-command-
|
|
4
|
-
import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-
|
|
3
|
+
import { n as runAnalystBenchmarkCommand } from "./benchmark-command-BKENp2s5.js";
|
|
4
|
+
import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-iu0ede49.js";
|
|
5
5
|
import { writeFileSync } from "node:fs";
|
|
6
6
|
//#region src/cli-config.ts
|
|
7
7
|
function resolveCliLlmConfig(env = process.env) {
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
import { l as RunTerminalOutcome } from "./run-record-DdSa93_W.js";
|
|
2
|
-
import { _ as GateDecision, j as MutableSurface } from "./types-
|
|
3
|
-
import { o as InsightReport } from "./insight-report-
|
|
2
|
+
import { _ as GateDecision, j as MutableSurface } from "./types-DYuNHo9R.js";
|
|
3
|
+
import { o as InsightReport } from "./insight-report-C6h6F_4L.js";
|
|
4
4
|
//#region src/hosted/types.d.ts
|
|
5
5
|
declare const HOSTED_WIRE_VERSION: '2026-07-24.v1';
|
|
6
6
|
type HostedWireVersion = typeof HOSTED_WIRE_VERSION;
|
|
@@ -199,4 +199,4 @@ declare function hostedClientFromEnv(overrides?: Partial<HostedTenant> & {
|
|
|
199
199
|
}): HostedClient | undefined;
|
|
200
200
|
//#endregion
|
|
201
201
|
export { UnixNanoTimestamp as _, hostedTenantFromEnv as a, EvalRunGenerationSnapshot as c, HostedIngestHeaders as d, HostedWireVersion as f, TraceSpanEvent as g, IngestTracesRequest as h, hostedClientFromEnv as i, EvalRunStatus as l, IngestResponse as m, HostedTenant as n, EvalRunCellScore as o, IngestEvalRunsRequest as p, createHostedClient as r, EvalRunEvent as s, HostedClient as t, HOSTED_WIRE_VERSION as u };
|
|
202
|
-
//# sourceMappingURL=client-
|
|
202
|
+
//# sourceMappingURL=client-DjXROWpx.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"client-
|
|
1
|
+
{"version":3,"file":"client-DjXROWpx.d.ts","names":[],"sources":["../src/hosted/types.ts","../src/hosted/client.ts"],"mappings":";;;;cAgCa;KACD,2BAA2B;;UAKtB;;EAEf;;EAEA;;EAEA,yBAAyB;;EAEzB;;;KAMU;UAQK;;EAEf;;EAEA;;EAEA;;EAEA,YAAY,eAAe;;EAE3B,iBAAiB;;EAEjB;;EAEA;;UAGe;;EAEf;;;EAGA;;;EAGA,UAAU;;EAEV,OAAO;;EAEP;;EAEA;;EAEA;;;;;;;UAQe;;EAEf;;EAEA;;EAEA;;EAEA,QAAQ;;EAER,QAAQ;;EAER,WAAW;;EAEX,aAAa;;EAEb,eAAe;;EAEf;;EAEA;;EAEA;;EAEA;;;;;;EAMA,gBAAgB;;;;;;KASN;;;;;;UAOK;EACf;EACA;EACA;EACA;EACA,mBAAmB;EACnB,iBAAiB;EACjB,YAAY;EACZ,SAAS;IACP,cAAc;IACd;IACA,aAAa;;EAEf;IAAW;IAAgC;;;EAE3C;;EAEA;;EAEA;;EAEA;;UAKe;EACf,aAAa;EACb,QAAQ;;UAGO;EACf,aAAa;EACb,OAAO;;UAGQ;;EAEf;;EAEA,UAAU;IAAQ;IAAe;;;;;UC1JlB;;EAEf;;EAEA;;EAEA;;EAEA,mBAAmB;;EAEnB;;EAEA;;UAGe;EACf,cAAc,OAAO,cAAc,0BAA0B,QAAQ;EACrE,eAAe,QAAQ,gBAAgB,0BAA0B,QAAQ;EACzE,aAAa,OAAO,kBAAkB,0BAA0B,QAAQ;WAC/D,QAAQ;WACR,aAAa;;iBAyGR,mBAAmB,QAAQ,eAAe;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAsE1C,oBACd,YAAW,QAAQ;EAAkB,MAAM;IAC1C;iBAiBa,oBACd,YAAW,QAAQ;EAAkB,MAAM;IAC1C"}
|
|
@@ -1,8 +1,8 @@
|
|
|
1
|
-
import { t as DefaultVerdict } from "./verdict-
|
|
1
|
+
import { t as DefaultVerdict } from "./verdict-DExhxfgR.js";
|
|
2
2
|
import { c as CostLedgerHandle } from "./cost-ledger-Bv_e8XHY.js";
|
|
3
|
-
import { A as ChatClient,
|
|
4
|
-
import { c as AnalystRunInputs, i as AnalystFinding, l as AnalystRunResult, n as AnalystContext, s as AnalystRunEvent, t as Analyst, u as AnalystRunSummary } from "./types-
|
|
5
|
-
import { i as ExactAnalystRunEvent, o as ExactAnalystRunResult, u as ExactExecutionComponentIdentity } from "./exact-types-
|
|
3
|
+
import { A as ChatClient, Ft as RawProviderSink } from "./types-D216SgwM.js";
|
|
4
|
+
import { c as AnalystRunInputs, i as AnalystFinding, l as AnalystRunResult, n as AnalystContext, s as AnalystRunEvent, t as Analyst, u as AnalystRunSummary } from "./types-DF_Udrp-.js";
|
|
5
|
+
import { i as ExactAnalystRunEvent, o as ExactAnalystRunResult, u as ExactExecutionComponentIdentity } from "./exact-types-Djvzosly.js";
|
|
6
6
|
//#region src/analyst/registry.d.ts
|
|
7
7
|
interface AnalystHooks {
|
|
8
8
|
/** Legacy runs may mutate ctx; exact runs provide a frozen observational context. */
|
|
@@ -411,4 +411,4 @@ declare function createTokenRecallChecker(opts?: {
|
|
|
411
411
|
}): CorrectnessChecker;
|
|
412
412
|
//#endregion
|
|
413
413
|
export { ExactAnalystRunExecutionError as A, jsonHasKeys as C, AnalystRegistryOptions as D, AnalystRegistry as E, RegistryRunOpts as M, assertExactRegistryRunOpts as N, BudgetPolicy as O, containsAll as S, AnalystHooks as T, ValidationContext as _, ProducedProposal as a, byteLengthRange as b, SatisfiedBy as c, createLlmCorrectnessChecker as d, createTokenRecallChecker as f, ArtifactValidator as g, Artifact as h, LlmCorrectnessCheckerOpts as i, ExactRegistryRunOpts as j, ExactAnalystBudgetPolicy as k, TaskGold as l, verifyCompletion as m, CompletionVerdict as n, ProducedState as o, parseCorrectnessResponse as p, CorrectnessChecker as r, RequirementCheck as s, CompletionRequirement as t, completionVerdict as u, ValidationIssue as v, regexMatch as w, composeValidators as x, ValidationResult as y };
|
|
414
|
-
//# sourceMappingURL=completion-verifier-
|
|
414
|
+
//# sourceMappingURL=completion-verifier-foUCLif_.d.ts.map
|
package/dist/{completion-verifier-CBiee74w.d.ts.map → completion-verifier-foUCLif_.d.ts.map}
RENAMED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"completion-verifier-
|
|
1
|
+
{"version":3,"file":"completion-verifier-foUCLif_.d.ts","names":[],"sources":["../src/analyst/registry.ts","../src/artifact-validator.ts","../src/completion-verifier.ts"],"mappings":";;;;;;UAqDiB;;EAEf,iBAAiB;IACf,SAAS;IACT,KAAK;IACL;aACS;;EAEX,gBAAgB;IACd,SAAS;IACT,SAAS;IACT,UAAU;IACV;aACS;;;;;;EAMX,SAAS;IACP,SAAS;IACT,OAAO;IACP;MACE,+BAA+B,QAAQ;;EAE3C,YAAY;IAAQ,QAAQ;aAA4B;;UAGzC;;EAEf;;EAEA,UAAU;;;;;;;EAOV,YAAY;IACV,SAAS;IACT;IACA;IACA;;;UAIa;;EAEf,OAAO;;EAEP,OAAO,aAAa,SAAS;;EAE7B,QAAQ;;EAER,gBAAgB;;EAEhB,gBAAgB;;EAEhB,eAAe;;UAGA;;EAEf;;EAEA;;EAEA,SAAS;;EAET;;EAEA,SAAS;;EAET,aAAa;;EAEb;;EAEA,OAAO;;;;;;;;;EASP,gBAAgB,cAAc,kBAAkB,eAAe,cAAc;;;;;;EAM7E;;;KAIU;WAEG;WACA;;WAGA;WACA;;WAEA,SAAS,SAAS;;;;;;;;;;UAWhB;WACN;WACA,QAAQ;WACR;WACA,QAAQ;WACR,YAAY;WACZ,oBAAoB;WACpB;WACA,MAAM,SAAS;WACf,eACL,cAAc,kBACd,SAAS,eAAe,cAAc;WAEjC;WACA;WACA;WACA;;;cAqCE,sCAAsC;WACxC;WACA,QAAQ;EAEjB,YAAY,iBAAiB,QAAQ,uBAAuB,UAAU;;cAU3D;mBACM;mBACA;EAEjB,YAAY,UAAS;EAIrB,SAAS,SAAS;EAsBlB,QAAQ;IACN;IACA;IACA;IACA,MAAM;;EAUF,IACJ,eACA,QAAQ,kBACR,UAAS,kBACR,QAAQ;;EAUL,SACJ,eACA,QAAQ,kBACR,SAAS,uBACR,QAAQ;;EAQJ,eACL,eACA,QAAQ,kBACR,SAAS,uBACR,eAAe;;;;;;;;;;;;EAmBX,UACL,eACA,QAAQ,kBACR,UAAS,kBACR,eAAe;UAIV;UA8BA;UAwFO;UA2aP;UAaA;UAUA;;;iBAiGM,2BAA2B,yBAAyB,SAAS;;;;;;;;;;;;;;;;UC79B5D;;EAEf;;EAEA;;EAEA;;EAEA,QAAQ;;EAER,WAAW;;UAGI;EACf;EACA;;EAEA,iBAAiB;;EAEjB,QAAQ;;UAGO;EACf;EACA;;EAEA;;UAGe;EACf;;EAEA;EACA,QAAQ;;EAER,WAAW;;UAGI;;EAEf;;EAEA;;EAEA,SAAS,UAAU,UAAU,SAAS,oBAAoB,QAAQ;;;;;;iBAWpD,kBACd,YAAY,qBACZ;EAAY;EAAe;IAC1B;;iBAgCa,WAAW,cAAc,SAAS,SAAS;;iBAkB3C,YAAY,cAAc,0BAA0B;;iBAuCpD,gBAAgB,cAAc,aAAa,cAAc;;iBAoBzD,YACd,cACA,oBACA;EAAY;IACX;;;;KClJS;UAEK;;EAEf;;EAEA;;EAEA;;EAEA,cAAc;;UAGC;EACf;EACA,cAAc;;UAGC;EACf;EACA;EACA;;EAEA;;;UAIe;;EAEf,WAAW;;EAEX,WAAW;;EAEX;;UAGe;EACf;EACA;;EAEA;;;;;;EAMA;;EAEA;;;;;;;;;EASA;;EAEA;;EAEA;;;;;UAMe,0BAA0B;EACzC;EACA,cAAc;;EAEd;;EAEA;;EAEA;;;;;;;;;iBAUc,kBAAkB;EAChC;EACA,cAAc;IACZ;;;;;;KAoCQ,sBACV,aAAa,uBACb,oBACG;EAAU;EAAkB;;;;;;;;;iBAqLX,iBACpB,MAAM,UACN,OAAO,eACP,kBAAkB,qBACjB,QAAQ;UAoFM;EACf;;EAEA,aAAa;EACb;EACA,WAAW;EACX,SAAS;;EAET;;;;;;EAMA;;;;;;EAMA,UAAU;;;;;;;;;iBAUI,yBAAyB;EAAgB;EAAkB;;;;;;;;iBAiC3D,4BACd,MAAM,YACN,OAAM,4BACL;;;;;;;;;;;;;;;iBAyJa,yBACd;EAAQ;EAAoB;IAC3B"}
|
package/dist/contract/index.d.ts
CHANGED
|
@@ -1,13 +1,14 @@
|
|
|
1
1
|
import { c as CostLedgerHandle, f as CostLedgerSummary, m as CostReceipt } from "../cost-ledger-Bv_e8XHY.js";
|
|
2
2
|
import { a as RunRecord, n as RunCostProvenance, s as RunSplitTag } from "../run-record-DdSa93_W.js";
|
|
3
|
-
import { A as ChatClient, F as CreateChatClientOpts, V as createChatClient } from "../types-
|
|
4
|
-
import {
|
|
5
|
-
import { n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-
|
|
6
|
-
import { C as JudgeDimension, H as SurfaceProposer, M as OptimizerConfig, R as Scenario, S as JudgeConfig, V as SessionScript, _ as GateDecision, a as CampaignResult, b as GenerationRecord, c as CampaignTraceWriter, d as DispatchContext, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, m as GateCheckStatus, n as CampaignArtifactWriter, p as Gate, r as CampaignCellResult, t as CampaignAggregates, v as GateResult, w as JudgeScore, y as GenerationCandidate } from "../types-
|
|
7
|
-
import { a as FailureClusterInsight, c as JudgeInsight, d as Recommendation, f as ReleaseSummary, i as FailureClassTally, l as LiftInsight, m as TokenUsageInsight, n as ExecutionErrorOutcomeCell, o as InsightReport, p as ScalarDistribution, r as ExecutionInsight, s as InterRaterInsight, t as CostProvenanceSummary, u as OutcomeCorrelationInsight } from "../insight-report-
|
|
8
|
-
import {
|
|
9
|
-
import { A as RunEvalOptions,
|
|
10
|
-
import { c as
|
|
3
|
+
import { A as ChatClient, F as CreateChatClientOpts, V as createChatClient } from "../types-D216SgwM.js";
|
|
4
|
+
import { _ as ProposalFinding, i as AnalystFinding, v as ProposalFindingOrigin, x as makeProposalFinding } from "../types-DF_Udrp-.js";
|
|
5
|
+
import { n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-BZhStdjl.js";
|
|
6
|
+
import { C as JudgeDimension, H as SurfaceProposer, M as OptimizerConfig, R as Scenario, S as JudgeConfig, V as SessionScript, _ as GateDecision, a as CampaignResult, b as GenerationRecord, c as CampaignTraceWriter, d as DispatchContext, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, m as GateCheckStatus, n as CampaignArtifactWriter, p as Gate, r as CampaignCellResult, t as CampaignAggregates, v as GateResult, w as JudgeScore, y as GenerationCandidate } from "../types-DYuNHo9R.js";
|
|
7
|
+
import { a as FailureClusterInsight, c as JudgeInsight, d as Recommendation, f as ReleaseSummary, i as FailureClassTally, l as LiftInsight, m as TokenUsageInsight, n as ExecutionErrorOutcomeCell, o as InsightReport, p as ScalarDistribution, r as ExecutionInsight, s as InterRaterInsight, t as CostProvenanceSummary, u as OutcomeCorrelationInsight } from "../insight-report-C6h6F_4L.js";
|
|
8
|
+
import { D as analyzeRuns, E as SummarizeExecutionOptions, O as summarizeExecution, T as ExecutionReport, w as AnalyzeRunsOptions } from "../engine-nB64f48I.js";
|
|
9
|
+
import { A as RunEvalOptions, At as compareOptimizationMethods, B as OptimizerModelBudget, C as RunImprovementLoopOptions, Ct as OptimizationMethodProvenance, D as RunOptimizationOptions, Dt as OptimizationPackageSource, E as PremeasuredOptimizationBaseline, F as GepaOptimizationMethodConfig, G as DefaultProductionRewardHackingOptions, Gt as inMemoryCampaignStorage, H as heldOutGate, Ht as CampaignStorage, I as GepaOptimizationRecipe, K as defaultProductionGate, L as GepaRunnerCommand, Lt as RunCampaignOptions, M as GepaAdaptiveEngineRun, N as GepaEngineOptions, Nt as CampaignCellFailureReceipt, Ot as OptimizationTokenUsage, P as GepaEngineRun, R as gepaOptimizationMethod, T as runImprovementLoop, U as DefaultProductionGateCheck, V as HeldOutGateOptions, W as DefaultProductionGateOptions, Wt as fsCampaignStorage, Yt as campaignSplitDigest, _t as CompareOptimizationMethodsOptions, an as ReferenceEquivalenceScenario, bt as OptimizationMethodComparison, cn as LlmJudgeDimension, dt as externalTextOptimizationMethod, en as REFERENCE_EQUIVALENCE_INPUT_LIMITS, ft as ExternalTextOptimizationMethodConfig, gt as ExternalTextEvaluationResponse, ht as ExternalOptimizationExample, i as skillOptOptimizationMethod, in as ReferenceEquivalenceJudgeResult, j as runEval, ln as LlmJudgeOptions, mt as ExternalTextOptimizerResult, n as SkillOptRunnerCommand, nn as ReferenceEquivalenceJudgeInput, on as createReferenceEquivalenceJudge, p as LoopProvenanceRecord, pt as ExternalTextOptimizerContext, r as SkillOptTrainerConfig, rn as ReferenceEquivalenceJudgeOptions, sn as runReferenceEquivalenceJudge, t as SkillOptOptimizationMethodConfig, tn as REFERENCE_EQUIVALENCE_JUDGE_VERSION, un as llmJudge, ut as composeGate, vt as ComparisonCost, w as RunImprovementLoopResult, wt as OptimizationMethodResult, xt as OptimizationMethodInput, yt as OptimizationMethod, z as OpenAICompatibleOptimizerModel, zt as runCampaign } from "../skillopt-optimization-method-BO7NAl3b.js";
|
|
10
|
+
import { a as ObjectiveSource, b as PowerPreflight, c as PromotionPolicy, d as paretoSignificanceGate, i as EvidenceVector, l as buildEvidenceVector, n as AxisVerdict, o as ParetoSignificanceGateOptions, r as BuildEvidenceVectorOptions, s as PromotionObjective, t as AxisEvidence, u as paretoPolicy } from "../promotion-policy-ChWhTDBH.js";
|
|
11
|
+
import { c as EvalRunGenerationSnapshot, g as TraceSpanEvent, n as HostedTenant, o as EvalRunCellScore, s as EvalRunEvent } from "../client-DjXROWpx.js";
|
|
11
12
|
import { i as InMemoryOutcomeStore, n as FileSystemOutcomeStore, o as OutcomeStore, r as FileSystemOutcomeStoreOptions, t as DeploymentOutcome } from "../outcome-store-BYHIuO0e.js";
|
|
12
13
|
import { AgentCandidateBenchmarkCellRef, AgentCandidateBenchmarkSuiteInputs, AgentCandidateBenchmarkTask, AgentCandidateBenchmarkTaskMaterial, AgentCandidateBundle, AgentCandidateEvaluationPolicy, AgentCandidateExperiment, AgentCandidateExperimentMaterial, AgentCandidateExperimentMeasurement, AgentImprovementCost, AgentImprovementMeasuredComparison, AgentProfileImprovementExperiment, AgentProfileImprovementExperimentMaterial, AgentProfileImprovementMeasuredComparison, AgentProfileImprovementMeasurement, AgentProfileImprovementRunCell, AgentProfileImprovementRunReceipt, AgentProfileImprovementSuiteInputs, AgentProfileImprovementTask, AgentProfileImprovementTaskMaterial, CandidateExecutionEvidence, Sha256Digest } from "@tangle-network/agent-interface";
|
|
13
14
|
//#region src/contract/self-improve.d.ts
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/contract/self-improve.ts","../../src/contract/define-agent-eval.ts","../../src/contract/measured-comparison.ts","../../src/contract/profile-measured-comparison.ts","../../src/contract/intake/run-record-dir.ts","../../src/contract/eval-reporting-suite.ts","../../src/contract/diff.ts","../../src/contract/intake/agent-trace.ts","../../src/contract/intake/code-agent-observation.ts","../../src/contract/intake/code-agent-session.ts","../../src/contract/intake/feedback-table.ts","../../src/contract/intake/otel-spans.ts"],"mappings":"
|
|
1
|
+
{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/contract/self-improve.ts","../../src/contract/define-agent-eval.ts","../../src/contract/measured-comparison.ts","../../src/contract/profile-measured-comparison.ts","../../src/contract/intake/run-record-dir.ts","../../src/contract/eval-reporting-suite.ts","../../src/contract/diff.ts","../../src/contract/intake/agent-trace.ts","../../src/contract/intake/code-agent-observation.ts","../../src/contract/intake/code-agent-session.ts","../../src/contract/intake/feedback-table.ts","../../src/contract/intake/otel-spans.ts"],"mappings":";;;;;;;;;;;;;;UA+DiB;;;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;;;EAIX,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;;;EAIA,cAAc,OAAO;;;;EAKrB;EACA;EACA;;;;;;;;;;;;;;EAeA,eAAe;;;EAIf,eAAe;;;;EAKf,eAAe;;EAGf;;;;;;;;EASA;;;;;;;;;EAUA,oBAAoB,uBAAuB,WAAW;;;EAItD,WAAW;;;;;;;EAQX,mBAAmB,uBAAuB,WAAW;;UAGtC,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;;EAEV;IACE;IACA,MAAM;IACN;IACA,aAAa;;;;;;;;EAQf,SAAS;;;;;EAKT,QAAQ;;;;;;EAMR,KAAK,yBAAyB,WAAW;;;cAI9B,4BAA4B;WAC9B,MAAM;WACN,UAAU;EAEnB,YAAY,gBAAgB,QAAQ;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA+LhB,YAAY,kBAAkB,UAAU,WAC5D,MAAM,mBAAmB,WAAW,aACnC,QAAQ,kBAAkB,WAAW;;;KCviB5B,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;;;UCvDd;EACf,QAAQ,gCAAgC;EACxC;EACA;;UAGe;EACf,YAAY;EACZ;EACA,QAAQ;EACR,MAAM;EACN,eAAe;EACf;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,QAAQ,OAAO,oCAAoC,QAAQ;;EAE3D;;EAEA,aAAa;EACb,SAAS;;UAGM;EACf,cAAc;EACd;IACE;IACA,MAAM;;;UAIO;EACf,YAAY;EACZ,cAAc;EACd;IACE;IACA,MAAM;;EAER,aAAa;EACb;EACA,YAAY;EACZ;EACA,WAAW;;;UAII,kBAAkB;EACjC;EACA,UAAU;EACV,WAAW;;;UAII,yBAAyB;EACxC,MAAM,KAAK;EACX,WAAW,KAAK;IAAkB;IAAc;;EAChD,QAAQ,KAAK;EACb,eAAe,KAAK,OAAO;EAC3B,UAAU,KAAK;EACf,UAAU,KAAK;EACf,OAAO,KAAK;;UAGG,kCAAkC;EACjD,uBAAuB,kBAAkB;EACzC,QAAQ;EACR,SAAS,yBAAyB;;EAElC;;EAEA,kBAAkB;;EAElB,kBAAkB;;;KAIR,8BAA8B,KACxC;EAGA,iBAAiB;EACjB,WAAW;EACX;;;iBAIc,2BACd,UAAU,sCACT;;iBAQa,4BACd,SAAS,qCACR;;iBAiBa,wBACd,UAAU,mCACT;iBAQa,0BAA0B,iBAAiB;;iBAarC,uBACpB,SAAS,gCACR,QAAQ;;;;;;;;iBAkEK,2BAA2B,MACzC,SAAS,kCAAkC,QAC1C;;iBA8Ya,0CACd,SAAS,oCACR;;iBAuEa,oCACd,iBACC;iBAkHa,6BAA6B,iBAAiB;iBAM9C,oCACd,iBACC;iBAkBa,8BAA8B;;;;;;;;;;UC7zB7B;EACf,aAAa;EACb,QAAQ,gCAAgC;EACxC;EACA;;UAGe;EACf,YAAY;EACZ;EACA,aAAa;EACb,MAAM;EACN,SAAS;EACT;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,QACE,OAAO,kDACN,QAAQ;;EAEX;;EAEA,aAAa;EACb,SAAS;;UAGM;EACf,cAAc;EACd;IACE;IACA,MAAM;;;UAIO;EACf,YAAY;EACZ,cAAc;EACd;IACE;IACA,MAAM;;EAER,aAAa;EACb;EACA,YAAY;EACZ;EACA,WAAW;;;iBAIG,gCACd,UAAU,sCACT;;iBAQa,iCACd,SAAS,0CACR;;iBAqBa,sCACd,UAAU,4CACT;iBAOa,kCAAkC,iBAAiB;iBAInD,yCACd,iBACC;iBAIa,wCACd,iBACC;;;;;;iBASmB,qCACpB,SAAS,8CACR,QAAQ;;iBA0CK,wDACd,SAAS,kDACR;;iBAuEa,kDACd,iBACC;;;;UCnPc;;EAEf;;EAEA;;EAEA;;UAGe;;;;;;EAMf;;;;;;;EAOA,WAAW;;;;;;EAMX;;UAGe;;EAEf,MAAM;;EAEN,UAAU;;EAEV;;;;;;;;;;iBAkBoB,iBACpB,cACA,UAAS,0BACR,QAAQ;;;;;KC7CC,0BAA0B;UAErB;;;;;EAKf,UAAU,KAAK;;EAEf,OAAO;;;;;;;;;;EAUP;;;;UAKe;;;EAGf,QAAQ;;EAER;;IAEE;;IAEA;;;IAGA;;IAEA;;;IAGA,UAAU;;;EAGZ;;;;;;;iBAUoB,mBACpB,OAAO,yBACP,UAAS,4BACR,QAAQ;;;;;;UC5DM;EACf;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;;;EAGA,YAAY,eAAe,eAAe;;;;UAK3B;EACf;EACA;EACA;EACA;EACA;;EAEA,SAAS;;EAET,SAAS;;EAET,OAAO;;EAEP;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;;;;UAMe;EACf;EACA;EACA;EACA;EACA,oBAAoB;EACpB,mBAAmB;EACnB;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA,cAAc;;;;EAId,aAAa;;;;;;;;;iBAoDC,gBACd,QAAQ,2BACR,OAAO,4BACN;;;;;;iBAqEa,SAAS,QAAQ,cAAc,OAAO,eAAe;;;;;;;iBAsCrD,wBAAwB,KAAK,eAAe;;;KC1OhD;UAEK;EACf,MAAM;;EAEN;;UAGe;EACf;EACA;EACA;;;EAGA,cAAc;;UAGC;EACf;EACA,cAAc;EACd,QAAQ;;UAGO;EACf;EACA,eAAe;;UAGA;EACf;EACA;EACA;EACA;IAAQ;IAAc;;EACtB;IAAS;IAAe;;EACxB,OAAO;;;;UAOQ;EACf;;EAEA;;EAEA;EACA;EACA;;EAEA;;EAEA;;KAGU,kBAAkB,YAAY;;;;;;iBAW1B,gBAAgB,SAAS,qBAAqB;UAmE7C;;;;EAIf,SAAS,YAAY;;;EAGrB,cAAc;;;;;;;;iBASA,8BACd,MAAM,aACN,OAAO,kBACN;;;KCjLS;KAEA;KAEA;KAEA;KASA;UAEK;EACf;EACA;EACA;;UAGe;EACf;EACA;EACA,MAAM;EACN,SAAS;EACT;EACA,QAAQ;EACR;EACA;EACA,UAAU;;UAGK;EACf,QAAQ;EACR;EACA;EACA;IACE,QAAQ;IACR;;EAEF,SAAS;;UAGM;EACf,QAAQ;EACR;EACA;EACA,YAAY;;;;;;;iBAeE,wBACd,SAAS,iCACR;;;UCrCc;EACf;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf,QAAQ;EACR;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;;UAGe;EACf,MAAM;EACN,aAAa;EACb,SAAS;EACT,cAAc;;UAGC;EACf;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;EACA;EACA;EACA;EACA;EACA;;;EAGA,iBAAiB;;;EAGjB,YAAY;;iBAGE,oBAAoB,gBAAgB;iBAepC,iBACd,SAAS,gCACR;iBAIa,sBACd,SAAS,gCACR;iBAIa,oBACd,SAAS,gCACR;iBAIa,oBACd,SAAS,gCACR;iBAIa,cACd,SAAS,gCACR;cAIU,2BAAkB;;;UCpJd;;;EAGf;;EAEA;;;EAGA;;;EAGA,WAAW;;UAGI;EACf;;;EAGA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA,WAAW;;;EAGX,SAAS;;UAGM;;EAEf,SAAS;;;EAGT,OAAO;;;;EAIP;IAAU;IAAa;;;;;EAIvB;;UAGe;EACf,MAAM;;;EAGN,aAAa;IAAQ;IAAe;IAAe;;;iBAGrC,kBAAkB,MAAM,2BAA2B;;;UCvBlD;EACf,OAAO;;EAEP,eAAe;;EAEf;;;;;;EAMA,eAAe,eAAe,gBAAgB;;iBAGhC,cAAc,MAAM,uBAAuB"}
|
package/dist/contract/index.js
CHANGED
|
@@ -1,18 +1,19 @@
|
|
|
1
1
|
import { c as ValidationError } from "../errors-D-LKuDhb.js";
|
|
2
2
|
import { i as estimateCost, o as isModelPriced } from "../metrics-C9YY1OcL.js";
|
|
3
3
|
import { i as CostLedger } from "../cost-ledger-DMFxsLKr.js";
|
|
4
|
-
import { C as createChatClient, t as buildDefaultAnalystRegistry } from "../default-registry-
|
|
4
|
+
import { C as createChatClient, t as buildDefaultAnalystRegistry } from "../default-registry-BaQXW1Ow.js";
|
|
5
5
|
import { LLM_MODEL_ATTR_KEYS, SPAN_KIND_ATTR_KEYS } from "../trace-attributes.js";
|
|
6
6
|
import { nt as classifyOtlpSpanRole, rt as isOtlpModelCall } from "../kind-factory-BHIgPmzS.js";
|
|
7
7
|
import { s as makeProposalFinding } from "../usage-receipt-EVI8B8Xu.js";
|
|
8
|
-
import { i as inMemoryCampaignStorage, n as createRunCostLedger, r as fsCampaignStorage } from "../single-run-lock-
|
|
8
|
+
import { i as inMemoryCampaignStorage, n as createRunCostLedger, r as fsCampaignStorage } from "../single-run-lock-BMQEv1wG.js";
|
|
9
9
|
import { m as mapConcurrentRange } from "../ledger-core-DXZIqu17.js";
|
|
10
|
-
import {
|
|
10
|
+
import { $ as runReferenceEquivalenceJudge, H as resolveRunDir, M as surfaceContentHash, N as surfaceHash, Q as createReferenceEquivalenceJudge, S as assertOptimizationResult, V as runCampaign, X as REFERENCE_EQUIVALENCE_INPUT_LIMITS, Z as REFERENCE_EQUIVALENCE_JUDGE_VERSION, _ as heldOutGate, a as emitLoopProvenance, b as composeGate, ct as llmJudge, d as runImprovementLoop, g as gepaOptimizationMethod, o as loopProvenanceArgsFromResult, p as runEval, q as campaignSplitDigest, t as skillOptOptimizationMethod, v as defaultProductionGate, w as compareOptimizationMethods, x as externalTextOptimizationMethod } from "../skillopt-optimization-method-CkaI2ly4.js";
|
|
11
11
|
import { E as pairedBootstrap } from "../statistics-ByxzSiOM.js";
|
|
12
|
-
import { i as parseRunRecordSafe, r as modelHasSnapshot } from "../run-record-
|
|
12
|
+
import { i as parseRunRecordSafe, r as modelHasSnapshot } from "../run-record-DqOw5X6_.js";
|
|
13
|
+
import { c as heldoutSignificance, i as powerPreflight, n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector, u as decidePairedPromotion } from "../promotion-policy-CrLrmys8.js";
|
|
13
14
|
import { a as recordAggregateMeasurements, i as readTaskFailureLabels, o as summarizeExecutionMeasurements, s as summarizeTraceErrors, t as extractUsage } from "../extract-usage-BW27f3XW.js";
|
|
14
|
-
import { n as summarizeExecution, t as analyzeRuns } from "../analyze-runs-
|
|
15
|
-
import { c as campaignCellTaskScore, l as campaignCellToRunRecord, o as campaignCellExecutionEvidence, s as campaignCellJudgeDimensions } from "../reward-hacking-
|
|
15
|
+
import { n as summarizeExecution, t as analyzeRuns } from "../analyze-runs-BNNK7irB.js";
|
|
16
|
+
import { c as campaignCellTaskScore, l as campaignCellToRunRecord, o as campaignCellExecutionEvidence, s as campaignCellJudgeDimensions } from "../reward-hacking-B2tPL9a3.js";
|
|
16
17
|
import { n as InMemoryOutcomeStore, t as FileSystemOutcomeStore } from "../outcome-store-ChBKlTd_.js";
|
|
17
18
|
import { t as createHostedClient } from "../client-LIuo-KPv.js";
|
|
18
19
|
import { dirname, join } from "node:path";
|