@tangle-network/agent-eval 0.217.1 → 0.218.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +36 -1
- package/dist/adapters/http.d.ts +2 -2
- package/dist/analyst/index.d.ts +1 -1
- package/dist/analyst/index.d.ts.map +1 -1
- package/dist/analyst/index.js +9 -9
- package/dist/analyst/index.js.map +1 -1
- package/dist/{analyze-runs-DyeNcenW.d.ts → analyze-runs-DjlZOVXx.d.ts} +2 -2
- package/dist/{analyze-runs-DyeNcenW.d.ts.map → analyze-runs-DjlZOVXx.d.ts.map} +1 -1
- package/dist/{benchmark-command-xQK8nOI6.js → benchmark-command-BupdHRyf.js} +661 -562
- package/dist/benchmark-command-BupdHRyf.js.map +1 -0
- package/dist/benchmarks/index.d.ts +4 -4
- package/dist/benchmarks/index.js +3 -3
- package/dist/campaign/index.d.ts +4 -4
- package/dist/campaign/index.js +10 -10
- package/dist/{campaign-CJGkxAQ6.js → campaign-CGvUTnmc.js} +312 -281
- package/dist/{campaign-CJGkxAQ6.js.map → campaign-CGvUTnmc.js.map} +1 -1
- package/dist/{campaign-evidence-CQYw8_0H.js → campaign-evidence-DunzcJGv.js} +2 -2
- package/dist/{campaign-evidence-CQYw8_0H.js.map → campaign-evidence-DunzcJGv.js.map} +1 -1
- package/dist/{chat-client-ldqxYYTN.js → chat-client-B2pXlPdx.js} +2 -2
- package/dist/{chat-client-ldqxYYTN.js.map → chat-client-B2pXlPdx.js.map} +1 -1
- package/dist/{chat-json-call-DavzexSp.js → chat-json-call-Dm9c-b41.js} +2 -2
- package/dist/{chat-json-call-DavzexSp.js.map → chat-json-call-Dm9c-b41.js.map} +1 -1
- package/dist/cli.js +10 -10
- package/dist/{completion-verifier-B_jw6C9p.js → completion-verifier-DbLVf3i8.js} +4 -4
- package/dist/{completion-verifier-B_jw6C9p.js.map → completion-verifier-DbLVf3i8.js.map} +1 -1
- package/dist/contract/index.d.ts +7 -7
- package/dist/contract/index.js +90 -10
- package/dist/contract/index.js.map +1 -1
- package/dist/{default-registry-ClYjq3QE.js → default-registry-BJ-Pq6yL.js} +309 -277
- package/dist/{default-registry-ClYjq3QE.js.map → default-registry-BJ-Pq6yL.js.map} +1 -1
- package/dist/{define-agent-eval-BDxNlP3D.js → define-agent-eval-BGdFKlrc.js} +165 -105
- package/dist/define-agent-eval-BGdFKlrc.js.map +1 -0
- package/dist/{define-agent-eval-BIkH_gAR.d.ts → define-agent-eval-BlH10zEw.d.ts} +6 -6
- package/dist/{define-agent-eval-BIkH_gAR.d.ts.map → define-agent-eval-BlH10zEw.d.ts.map} +1 -1
- package/dist/diagnosis/index.d.ts +1 -1
- package/dist/{dspy-rlm-engine-DhdCr1Dc.js → dspy-rlm-engine-CSMwudLO.js} +2 -2
- package/dist/{dspy-rlm-engine-DhdCr1Dc.js.map → dspy-rlm-engine-CSMwudLO.js.map} +1 -1
- package/dist/eval-campaign-BUMIxKTE.js +393 -0
- package/dist/eval-campaign-BUMIxKTE.js.map +1 -0
- package/dist/{evaluation-jTjC_QQD.d.ts → evaluation-BOZ38vOH.d.ts} +2 -2
- package/dist/{evaluation-jTjC_QQD.d.ts.map → evaluation-BOZ38vOH.d.ts.map} +1 -1
- package/dist/evaluation.d.ts +1 -1
- package/dist/experiment/index.d.ts +1 -1
- package/dist/experiment/index.d.ts.map +1 -1
- package/dist/experiment/index.js +110 -71
- package/dist/experiment/index.js.map +1 -1
- package/dist/{exporters-Cl6IeNAu.js → exporters-DWvM16LE.js} +2 -2
- package/dist/{exporters-Cl6IeNAu.js.map → exporters-DWvM16LE.js.map} +1 -1
- package/dist/{external-optimizer-process-KVFGvYX1.js → external-optimizer-process-C3Wm7Vhn.js} +2 -2
- package/dist/{external-optimizer-process-KVFGvYX1.js.map → external-optimizer-process-C3Wm7Vhn.js.map} +1 -1
- package/dist/{external-optimizer-subprocess-XIHBOSKt.js → external-optimizer-subprocess-B5gdLRZv.js} +3 -3
- package/dist/{external-optimizer-subprocess-XIHBOSKt.js.map → external-optimizer-subprocess-B5gdLRZv.js.map} +1 -1
- package/dist/{findings-store-BMVQtEsm.js → findings-store-BtTzPL0d.js} +5 -5
- package/dist/{findings-store-BMVQtEsm.js.map → findings-store-BtTzPL0d.js.map} +1 -1
- package/dist/{hf-dataset-D9f7-oX7.js → hf-dataset-CKxIJQba.js} +3 -3
- package/dist/{hf-dataset-D9f7-oX7.js.map → hf-dataset-CKxIJQba.js.map} +1 -1
- package/dist/hosted/index.js +1 -1
- package/dist/{index-ZuWx52WG.d.ts → index-B16FNdW5.d.ts} +34 -17
- package/dist/index-B16FNdW5.d.ts.map +1 -0
- package/dist/{index-9jJ2fKSX.d.ts → index-CalAsXMR.d.ts} +7 -5
- package/dist/index-CalAsXMR.d.ts.map +1 -0
- package/dist/{index-BiT6RtNH.d.ts → index-DbMOXYnP.d.ts} +6 -6
- package/dist/{index-BiT6RtNH.d.ts.map → index-DbMOXYnP.d.ts.map} +1 -1
- package/dist/index.d.ts +9 -9
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +611 -688
- package/dist/index.js.map +1 -1
- package/dist/{integrity-DOd9Q86O.js → integrity-C8oUeB0g.js} +322 -197
- package/dist/integrity-C8oUeB0g.js.map +1 -0
- package/dist/{journal-CR4QikYp.js → journal-C7om8Gum.js} +48 -38
- package/dist/journal-C7om8Gum.js.map +1 -0
- package/dist/ledger-core/index.d.ts +1 -1
- package/dist/ledger-core/index.js +2 -2
- package/dist/ledger-core/lock-holder.d.ts +2 -0
- package/dist/ledger-core/lock-holder.js +181 -0
- package/dist/ledger-core/lock-holder.js.map +1 -0
- package/dist/{ledger-core-DrJhbms9.js → ledger-core-CAxUdDk0.js} +2 -2
- package/dist/{ledger-core-DrJhbms9.js.map → ledger-core-CAxUdDk0.js.map} +1 -1
- package/dist/{llm-client-BshCREn6.js → llm-client-_Y6ic-tf.js} +238 -248
- package/dist/llm-client-_Y6ic-tf.js.map +1 -0
- package/dist/{llm-judge-C8jMTNNG.js → llm-judge-Djxh-Uhs.js} +646 -458
- package/dist/llm-judge-Djxh-Uhs.js.map +1 -0
- package/dist/lock-holder-DaJbfpci.d.ts +86 -0
- package/dist/lock-holder-DaJbfpci.d.ts.map +1 -0
- package/dist/{matrix-DHe-dwVJ.d.ts → matrix-CCSBna4q.d.ts} +2 -2
- package/dist/{matrix-DHe-dwVJ.d.ts.map → matrix-CCSBna4q.d.ts.map} +1 -1
- package/dist/{mint-zszyYlpA.js → mint-CbdwjTPW.js} +2 -2
- package/dist/{mint-zszyYlpA.js.map → mint-CbdwjTPW.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/{pareto-BzsxidFD.js → pareto-CTv8vozZ.js} +4 -4
- package/dist/{pareto-BzsxidFD.js.map → pareto-CTv8vozZ.js.map} +1 -1
- package/dist/{promotion-gate-DaY95yU6.d.ts → promotion-gate-CxEHRsZW.d.ts} +2 -2
- package/dist/{promotion-gate-DaY95yU6.d.ts.map → promotion-gate-CxEHRsZW.d.ts.map} +1 -1
- package/dist/{proposal-findings-BbWidvUk.js → proposal-findings-n-5WBbEx.js} +2 -2
- package/dist/{proposal-findings-BbWidvUk.js.map → proposal-findings-n-5WBbEx.js.map} +1 -1
- package/dist/{render-C-w2EZ6p.js → render-Cs34y0iX.js} +404 -208
- package/dist/render-Cs34y0iX.js.map +1 -0
- package/dist/reporting.d.ts +2 -2
- package/dist/{researcher-0ABVIdGa.d.ts → researcher-GU1tV4dn.d.ts} +2 -2
- package/dist/{researcher-0ABVIdGa.d.ts.map → researcher-GU1tV4dn.d.ts.map} +1 -1
- package/dist/rl.d.ts +2 -2
- package/dist/rl.js +4 -4
- package/dist/rollout/index.js +5 -5
- package/dist/{rollout-D6C50mfA.js → rollout-Dd1ucT_M.js} +5 -5
- package/dist/{rollout-D6C50mfA.js.map → rollout-Dd1ucT_M.js.map} +1 -1
- package/dist/schema-C-ziyJD7.d.ts.map +1 -1
- package/dist/{schema-DxcKTzbi.js → schema-CvhfLjtZ.js} +77 -47
- package/dist/schema-CvhfLjtZ.js.map +1 -0
- package/dist/search/index.d.ts +2 -2
- package/dist/search/index.d.ts.map +1 -1
- package/dist/search/index.js +2 -2
- package/dist/{search-kernel-DxqjBxA-.d.ts → search-kernel-p9JB3SmL.d.ts} +3 -3
- package/dist/{search-kernel-DxqjBxA-.d.ts.map → search-kernel-p9JB3SmL.d.ts.map} +1 -1
- package/dist/{search-ledger-TUvB5gWH.js → search-ledger-ClpN4I3e.js} +4 -4
- package/dist/{search-ledger-TUvB5gWH.js.map → search-ledger-ClpN4I3e.js.map} +1 -1
- package/dist/{search-ledger-file-UKCO2r8a.js → search-ledger-file-Bd5Tioq6.js} +3 -3
- package/dist/{search-ledger-file-UKCO2r8a.js.map → search-ledger-file-Bd5Tioq6.js.map} +1 -1
- package/dist/{search-shipper-Cw2z3-td.js → search-shipper-Bg2HW3Uv.js} +3 -3
- package/dist/{search-shipper-Cw2z3-td.js.map → search-shipper-Bg2HW3Uv.js.map} +1 -1
- package/dist/{search-summary-DQIvJsSB.js → search-summary-CrBuvVnf.js} +3 -3
- package/dist/{search-summary-DQIvJsSB.js.map → search-summary-CrBuvVnf.js.map} +1 -1
- package/dist/{server-DoVbqqtJ.js → server-C6M0qR8h.js} +2 -2
- package/dist/{server-DoVbqqtJ.js.map → server-C6M0qR8h.js.map} +1 -1
- package/dist/{single-run-lock-U58yf8N6.js → single-run-lock-Dekd8_Wv.js} +7 -7
- package/dist/single-run-lock-Dekd8_Wv.js.map +1 -0
- package/dist/{skillopt-optimization-method-kyL94-ri.js → skillopt-optimization-method-5dlex7MJ.js} +1339 -1059
- package/dist/skillopt-optimization-method-5dlex7MJ.js.map +1 -0
- package/dist/{statistical-heldout-C9AbgREN.d.ts → statistical-heldout-BHfLZHrm.d.ts} +2 -2
- package/dist/{statistical-heldout-C9AbgREN.d.ts.map → statistical-heldout-BHfLZHrm.d.ts.map} +1 -1
- package/dist/{summary-report-DM7tlml1.d.ts → summary-report-CkbRcSuu.d.ts} +16 -1
- package/dist/summary-report-CkbRcSuu.d.ts.map +1 -0
- package/dist/supervisor-run/index.d.ts.map +1 -1
- package/dist/supervisor-run/index.js +3 -3
- package/dist/{systemone-Bp_Ac0p1.d.ts → systemone-BMzDJb1i.d.ts} +3 -3
- package/dist/{systemone-Bp_Ac0p1.d.ts.map → systemone-BMzDJb1i.d.ts.map} +1 -1
- package/dist/systemone.d.ts +2 -2
- package/dist/{terminal-record-78g6E4vh.js → terminal-record-D4Rv-j1W.js} +241 -193
- package/dist/{terminal-record-78g6E4vh.js.map → terminal-record-D4Rv-j1W.js.map} +1 -1
- package/dist/trajectory-replay/index.d.ts +4 -1
- package/dist/trajectory-replay/index.d.ts.map +1 -1
- package/dist/trajectory-replay/index.js +834 -824
- package/dist/trajectory-replay/index.js.map +1 -1
- package/dist/{tree-BGRglkKd.js → tree-B5e1uAeL.js} +122 -14
- package/dist/tree-B5e1uAeL.js.map +1 -0
- package/dist/{types-DtMbFw6U.d.ts → types-BKCzzmit.d.ts} +2 -2
- package/dist/{types-DtMbFw6U.d.ts.map → types-BKCzzmit.d.ts.map} +1 -1
- package/dist/{types-D8GhJoHk.d.ts → types-DtkO1G9L.d.ts} +2 -2
- package/dist/{types-D8GhJoHk.d.ts.map → types-DtkO1G9L.d.ts.map} +1 -1
- package/dist/wire/index.js +1 -1
- package/docs/design/statistics-decisions.md +1 -1
- package/docs/eval-surface-map.md +1 -0
- package/docs/profile-json-identity.md +1 -2
- package/docs/public-api.md +27 -16
- package/package.json +7 -2
- package/dist/benchmark-command-xQK8nOI6.js.map +0 -1
- package/dist/define-agent-eval-BDxNlP3D.js.map +0 -1
- package/dist/eval-campaign-BLEfBGbA.js +0 -343
- package/dist/eval-campaign-BLEfBGbA.js.map +0 -1
- package/dist/index-9jJ2fKSX.d.ts.map +0 -1
- package/dist/index-ZuWx52WG.d.ts.map +0 -1
- package/dist/integrity-DOd9Q86O.js.map +0 -1
- package/dist/journal-CR4QikYp.js.map +0 -1
- package/dist/llm-client-BshCREn6.js.map +0 -1
- package/dist/llm-judge-C8jMTNNG.js.map +0 -1
- package/dist/render-C-w2EZ6p.js.map +0 -1
- package/dist/schema-DxcKTzbi.js.map +0 -1
- package/dist/single-run-lock-U58yf8N6.js.map +0 -1
- package/dist/skillopt-optimization-method-kyL94-ri.js.map +0 -1
- package/dist/summary-report-DM7tlml1.d.ts.map +0 -1
- package/dist/tree-BGRglkKd.js.map +0 -1
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"define-agent-eval-
|
|
1
|
+
{"version":3,"file":"define-agent-eval-BlH10zEw.d.ts","names":["_$__0.FinalEvidenceUse","_$__0.GateResult"],"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":";;;;;;;;;;UAaU,uBAAuB,kBAAkB,UAAU,mBACnD,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;;;;;;;;;;;;kBAA5CA;;;;;;;;;;;UCAS;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;;;;;UCpBhB;;;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;;;;;;;;;;;;;EAcA,eAAe;;;;EAKf,eAAe;;EAGf;;;;;;;;EASA;;;;;;;;;EAUA,oBAAoB,uBAAuB,WAAW;;;EAItD,WAAW;;;;;EAMX,SAAS,uBAAuB,WAAW;;;EAI3C,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;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA0NtB,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;;;KC9nB5B,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"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import { a as FailureClassification, i as FailureBlame } from "../failure-cluster-DN0t63Z0.js";
|
|
2
|
-
import { r as ExecutionReport } from "../analyze-runs-
|
|
2
|
+
import { r as ExecutionReport } from "../analyze-runs-DjlZOVXx.js";
|
|
3
3
|
import { l as PrimeRawUsage } from "../prime-protocol-BWG2DQIk.js";
|
|
4
4
|
import { i as nodeHttpPrimeBridgeTransport, n as PrimeBridgeTransportRequest, r as PrimeBridgeTransportResult, t as PrimeBridgeTransport } from "../prime-bridge-transport-6feEglLf.js";
|
|
5
5
|
import { a as isContentAttribute, i as ingestSpans, n as IngestReport, r as epochMillis, t as DiagnosisSpan } from "../spans-CVngGij3.js";
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import { a as resolveModelPricing } from "./metrics-BKaPBtIG.js";
|
|
2
|
-
import { _ as resolveExternalOptimizerCallbackLimits, a as closeServer, c as waitForActiveHandlers, d as assertExternalOptimizerModelBudget, g as removeCredentialEnvironment, i as startExternalOptimizerModelProxy, o as listenLocal, r as runWithCleanup, s as sendJsonIfOpen, t as runExternalOptimizerProcess, v as resolveExternalOptimizerProcessLimits } from "./external-optimizer-subprocess-
|
|
2
|
+
import { _ as resolveExternalOptimizerCallbackLimits, a as closeServer, c as waitForActiveHandlers, d as assertExternalOptimizerModelBudget, g as removeCredentialEnvironment, i as startExternalOptimizerModelProxy, o as listenLocal, r as runWithCleanup, s as sendJsonIfOpen, t as runExternalOptimizerProcess, v as resolveExternalOptimizerProcessLimits } from "./external-optimizer-subprocess-B5gdLRZv.js";
|
|
3
3
|
import { n as isRecord } from "./json-BhdsJaOQ.js";
|
|
4
4
|
import { A as RawAnalystFindingSchema, B as parseFindingSubject, P as coerceJson } from "./kind-factory-Gy4fXixd.js";
|
|
5
5
|
import { randomBytes } from "node:crypto";
|
|
@@ -457,4 +457,4 @@ function sanitizedRunner(runner) {
|
|
|
457
457
|
//#endregion
|
|
458
458
|
export { decodeRawFindingArray as n, createDspyRlmTraceEngine as t };
|
|
459
459
|
|
|
460
|
-
//# sourceMappingURL=dspy-rlm-engine-
|
|
460
|
+
//# sourceMappingURL=dspy-rlm-engine-CSMwudLO.js.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"dspy-rlm-engine-DhdCr1Dc.js","names":[],"sources":["../src/analyst/finding-codec.ts","../src/analyst/trace-tool-callback.ts","../src/analyst/dspy-rlm-engine.ts"],"sourcesContent":["/**\n * The raw-finding wire contract — one decoder for both sides of the process\n * boundary.\n *\n * A finding array crosses a process boundary twice: an optimizer bridge writes\n * it in Python, TypeScript reads it back. Two decoders written independently\n * accept different shapes, so Python could report a completed investigation\n * whose rows TypeScript then dropped — the paid work disappearing between two\n * \"successes\". This module is the TypeScript half; `finding_contract.json`\n * carries the same rules to the Python half, generated from the schema here.\n *\n * Decoding never invents an empty result. A value that is not an array of rows\n * fails with the exact type it was, and a malformed row is reported with its\n * index and path while its valid siblings survive.\n */\n\nimport { type RawAnalystFinding, RawAnalystFindingSchema } from './finding-signature'\nimport { parseFindingSubject } from './finding-subject'\nimport { coerceJson } from './parse-tolerant'\n\n/**\n * Version of the wire contract both languages implement. Bump when the\n * accepted row shape changes; the Python package pins the same value, and the\n * emitted contract carries it so a mismatched pair is visible.\n */\nexport const FINDING_WIRE_CONTRACT_VERSION = 1\n\n/** Why one row was refused. Stable across languages: Python reports the same codes. */\ntype FindingRejectionCode = 'not-an-object' | 'schema' | 'invalid-subject' | 'row-limit'\n\n/** A refused row, named precisely enough to repair or report. */\ninterface RejectedFindingRow {\n /** Position in the submitted array. */\n readonly index: number\n /** Dotted path to the offending field, `''` for the row itself. */\n readonly path: string\n readonly code: FindingRejectionCode\n readonly message: string\n}\n\ninterface DecodedFindingArray {\n readonly accepted: RawAnalystFinding[]\n readonly rejected: RejectedFindingRow[]\n /**\n * Set when the value was not a finding array at all — the array itself is\n * the failure, so `accepted` and `rejected` are both empty and a caller must\n * not read the result as \"no findings\".\n */\n readonly topLevelError?: string\n}\n\n/**\n * Rows past this count are refused rather than validated. A model that emits\n * thousands of rows has lost the plot, and the diagnostics for them would\n * dwarf the answer they came with.\n */\nexport const MAX_FINDING_ROWS = 500\n\n/**\n * Decode a submitted findings value into accepted rows plus per-row\n * diagnostics.\n *\n * Accepts an array of rows, or a string carrying one (a JSON array, possibly\n * fenced). Anything else — an object, a number, null, undefined — sets\n * `topLevelError` naming the type received: a caller that asked for findings\n * and got a number has a defect to report, not an empty result to record.\n */\nexport function decodeRawFindingArray(value: unknown): DecodedFindingArray {\n const rows = findingRows(value)\n if (typeof rows === 'string') return { accepted: [], rejected: [], topLevelError: rows }\n\n const accepted: RawAnalystFinding[] = []\n const rejected: RejectedFindingRow[] = []\n for (const [index, row] of rows.entries()) {\n if (index >= MAX_FINDING_ROWS) {\n rejected.push({\n index,\n path: '',\n code: 'row-limit',\n message: `findings array exceeds ${MAX_FINDING_ROWS} rows`,\n })\n continue\n }\n const decoded = decodeRow(row, index)\n if ('finding' in decoded) accepted.push(decoded.finding)\n else rejected.push(decoded.rejection)\n }\n return { accepted, rejected }\n}\n\nfunction decodeRow(\n row: unknown,\n index: number,\n): { finding: RawAnalystFinding } | { rejection: RejectedFindingRow } {\n if (row === null || typeof row !== 'object' || Array.isArray(row)) {\n return {\n rejection: {\n index,\n path: '',\n code: 'not-an-object',\n message: `finding row must be an object, received ${describe(row)}`,\n },\n }\n }\n const parsed = RawAnalystFindingSchema.safeParse(row)\n if (!parsed.success) {\n const issue = parsed.error.issues[0]\n const path = (issue?.path ?? []).join('.')\n // The schema refines `subject` against the grammar; name that refusal\n // separately so a repair turn can be told which rule the row broke.\n const code: FindingRejectionCode =\n path === 'subject' && parseFindingSubject((row as { subject?: string }).subject) === null\n ? 'invalid-subject'\n : 'schema'\n return {\n rejection: { index, path, code, message: issue?.message ?? 'row does not match the schema' },\n }\n }\n return { finding: parsed.data }\n}\n\n/**\n * Narrow a submitted value to candidate rows, or return the message naming why\n * it is not a findings array at all.\n */\nfunction findingRows(value: unknown): unknown[] | string {\n if (Array.isArray(value)) return value\n if (typeof value === 'string') {\n const parsed = coerceJson(value)\n if (parsed === undefined) {\n return 'findings must be a JSON array; the string received is not JSON'\n }\n if (Array.isArray(parsed)) return parsed\n const unwrapped = findingsProperty(parsed)\n if (unwrapped !== undefined) return unwrapped\n return `findings must be a JSON array, received ${describe(parsed)}`\n }\n const unwrapped = findingsProperty(value)\n if (unwrapped !== undefined) return unwrapped\n return `findings must be an array, received ${describe(value)}`\n}\n\n/** `{ findings: [...] }` is the one wrapper models emit often enough to unwrap. */\nfunction findingsProperty(value: unknown): unknown[] | undefined {\n if (value === null || typeof value !== 'object' || Array.isArray(value)) return undefined\n const inner = (value as Record<string, unknown>).findings\n return Array.isArray(inner) ? inner : undefined\n}\n\nfunction describe(value: unknown): string {\n if (value === null) return 'null'\n if (Array.isArray(value)) return 'an array'\n return `a ${typeof value}`\n}\n\n\n","import { randomBytes } from 'node:crypto'\nimport { createServer, type IncomingMessage, type ServerResponse } from 'node:http'\nimport {\n type ExternalOptimizerCallbackLimits,\n resolveExternalOptimizerCallbackLimits,\n} from '../campaign/external-optimizer-contracts'\nimport {\n closeServer,\n listenLocal,\n sendJsonIfOpen,\n waitForActiveHandlers,\n} from '../campaign/external-optimizer-http'\nimport type { TraceAnalysisToolDescriptor } from '../trace-analyst/tools'\n\nimport { isRecord } from '../llm-client/json'\ninterface TraceToolCallback {\n url: string\n token: string\n calls: () => number\n close: () => Promise<void>\n}\n\nexport type TraceToolCallbackLimits = ExternalOptimizerCallbackLimits\n\n/** Expose one bounded trace-tool set only on an authenticated loopback socket. */\nexport async function startTraceToolCallback(args: {\n tools: readonly TraceAnalysisToolDescriptor[]\n maxCalls: number\n /** Trace-tool request/response byte limits. Omitted fields use finite defaults. */\n limits?: Partial<TraceToolCallbackLimits>\n signal?: AbortSignal\n}): Promise<TraceToolCallback> {\n if (!Number.isSafeInteger(args.maxCalls) || args.maxCalls <= 0) {\n throw new TypeError('trace tool callback maxCalls must be a positive safe integer')\n }\n const limits = resolveExternalOptimizerCallbackLimits(args.limits, 'trace tool callback limits')\n args.signal?.throwIfAborted()\n const byName = new Map(args.tools.map((tool) => [tool.name, tool]))\n if (byName.size !== args.tools.length) {\n throw new Error('trace tool callback received duplicate tool names')\n }\n\n const token = randomBytes(32).toString('hex')\n let calls = 0\n let accepting = true\n let closePromise: Promise<void> | undefined\n const activeControllers = new Set<AbortController>()\n const activeHandlers = new Set<Promise<void>>()\n const server = createServer((request, response) => {\n if (!accepting) {\n sendJsonIfOpen(response, 503, { error: 'trace tool callback is closing' })\n return\n }\n const controller = new AbortController()\n const abortRequest = (): void => {\n request.destroy()\n response.destroy()\n }\n activeControllers.add(controller)\n controller.signal.addEventListener('abort', abortRequest, { once: true })\n\n let handler!: Promise<void>\n handler = handleRequest(request, response, controller.signal).finally(() => {\n controller.signal.removeEventListener('abort', abortRequest)\n activeControllers.delete(controller)\n activeHandlers.delete(handler)\n })\n activeHandlers.add(handler)\n void handler.catch(() => undefined)\n })\n const port = await listenLocal(server)\n const close = (): Promise<void> => {\n closePromise ??= closeCallback()\n return closePromise\n }\n const onAbort = (): void => {\n void close().catch(() => undefined)\n }\n args.signal?.addEventListener('abort', onAbort, { once: true })\n if (args.signal?.aborted) onAbort()\n\n return {\n url: `http://127.0.0.1:${port}/call`,\n token,\n calls: () => calls,\n close,\n }\n\n async function handleRequest(\n request: IncomingMessage,\n response: ServerResponse,\n signal: AbortSignal,\n ): Promise<void> {\n try {\n if (request.method !== 'POST' || request.url !== '/call') {\n sendJsonIfOpen(response, 404, { error: 'not found' })\n return\n }\n if (request.headers.authorization !== `Bearer ${token}`) {\n sendJsonIfOpen(response, 401, { error: 'unauthorized' })\n return\n }\n if (calls >= args.maxCalls) {\n sendJsonIfOpen(response, 429, { error: 'trace tool call limit reached' })\n return\n }\n const body = await readJson(request, limits.maxRequestBytes)\n if (!isRecord(body) || typeof body.name !== 'string' || !('args' in body)) {\n sendJsonIfOpen(response, 400, { error: 'name and args are required' })\n return\n }\n const tool = byName.get(body.name)\n if (!tool) {\n sendJsonIfOpen(response, 404, { error: `unknown trace tool '${body.name}'` })\n return\n }\n calls += 1\n const result = await tool.handler(body.args, { signal })\n const encoded = JSON.stringify({ result })\n if (Buffer.byteLength(encoded) > limits.maxResponseBytes) {\n sendJsonIfOpen(response, 413, { error: 'trace tool response too large' })\n return\n }\n response.writeHead(200, {\n 'content-type': 'application/json; charset=utf-8',\n 'content-length': String(Buffer.byteLength(encoded)),\n })\n response.end(encoded)\n } catch (error) {\n sendJsonIfOpen(response, signal.aborted ? 499 : 400, {\n error: error instanceof Error ? error.message : String(error),\n })\n }\n }\n\n async function closeCallback(): Promise<void> {\n args.signal?.removeEventListener('abort', onAbort)\n accepting = false\n const closingServer = closeServer(server)\n server.closeIdleConnections?.()\n for (const controller of activeControllers) controller.abort()\n const [serverResult] = await Promise.allSettled([\n closingServer,\n waitForActiveHandlers(activeHandlers),\n ])\n if (activeControllers.size !== 0 || activeHandlers.size !== 0) {\n throw new Error('trace tool callback closed with active requests')\n }\n if (serverResult?.status === 'rejected') throw serverResult.reason\n }\n}\n\nfunction readJson(request: IncomingMessage, maxRequestBytes: number): Promise<unknown> {\n return new Promise((resolve, reject) => {\n let size = 0\n const chunks: Buffer[] = []\n request.on('data', (chunk: Buffer) => {\n size += chunk.length\n if (size > maxRequestBytes) {\n reject(new Error('trace tool request too large'))\n request.destroy()\n return\n }\n chunks.push(chunk)\n })\n request.on('error', reject)\n request.on('end', () => {\n try {\n resolve(JSON.parse(Buffer.concat(chunks).toString('utf8')))\n } catch (error) {\n reject(error)\n }\n })\n })\n}\n","import {\n assertExternalOptimizerModelBudget,\n type ExternalOptimizerModelCall,\n type ExternalOptimizerModelExecutionObservation,\n type ExternalOptimizerModelProxy,\n type ExternalOptimizerRunnerCommand,\n removeCredentialEnvironment,\n resolveExternalOptimizerCallbackLimits,\n resolveExternalOptimizerProcessLimits,\n} from '../campaign/external-optimizer-contracts'\nimport { startExternalOptimizerModelProxy } from '../campaign/external-optimizer-model-proxy'\nimport { runWithCleanup } from '../campaign/external-optimizer-resources'\nimport { runExternalOptimizerProcess } from '../campaign/external-optimizer-subprocess'\nimport type { CustomTokenPricing } from '../cost-ledger'\nimport { resolveModelPricing } from '../metrics'\nimport {\n DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS,\n type TraceAnalysisEngine,\n type TraceAnalysisEngineResult,\n} from './engine'\nimport { decodeRawFindingArray } from './finding-codec'\nimport { startTraceToolCallback, type TraceToolCallbackLimits } from './trace-tool-callback'\n\nimport { isRecord } from '../llm-client/json'\nconst DEFAULT_TIMEOUT_MS = 10 * 60_000\nconst DEFAULT_MAX_COST_USD = 1\nconst DEFAULT_MAX_MODEL_REQUEST_BYTES = 16 * 1024 * 1024\nconst DEFAULT_MAX_MODEL_RESPONSE_BYTES = 4 * 1024 * 1024\nconst DEFAULT_TRACE_TOOL_TIMEOUT_MS = 60_000\nconst MAX_TIMER_DELAY_MS = 2_147_483_647\nconst BRIDGE_MODULE = 'agent_eval_rpc.dspy_rlm_bridge'\n/** Bumped whenever this engine's execution behavior changes. */\nconst DSPY_RLM_ENGINE_VERSION = '1.0.0'\n\nexport interface DspyRlmTraceEngineOptions {\n /** Caller-owned execution path. Agent Eval never receives provider credentials. */\n call: ExternalOptimizerModelCall\n /** Stable public identity for the caller-owned path, such as an AgentProfile digest. */\n callRef: string\n /** Persist the finite execution record returned for every admitted call. */\n recordExecution: (observation: ExternalOptimizerModelExecutionObservation) => void\n model: string\n /** Exact provider rates. Required when the model is absent from the pricing table. */\n pricing?: CustomTokenPricing\n /** Maximum provider spend for one investigation. Default: 1 USD. */\n maxCostUsd?: number\n /** Controller response cap. Default: {@link DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS}. */\n maxOutputTokens?: number\n /**\n * Thinking tokens one controller turn may bill on top of its completion.\n * A reasoning model bills these beyond `maxOutputTokens`, so the cost\n * reservation must cover them. Default: four times the completion cap.\n */\n maxReasoningTokens?: number\n /** Maximum caller-owned model invocations. Default derives from the analysis limits. */\n maxModelRequests?: number\n /** Maximum model request bytes. Default: 16 MiB. */\n maxModelRequestBytes?: number\n /** Maximum model response bytes. Default: 4 MiB. */\n maxModelResponseBytes?: number\n /** Deadline for one caller-owned model invocation. Default: the whole analysis deadline. */\n modelRequestTimeoutMs?: number\n /** Trace-tool loopback request/response byte limits. */\n traceToolLimits?: Partial<TraceToolCallbackLimits>\n /** Deadline for one Python-to-Node trace-tool call. Default: 60 seconds. */\n traceToolTimeoutMs?: number\n /**\n * How the controller's reasoning and code fields are obtained.\n *\n * `tolerant` parses marker output strictly first, then recovers the fields\n * deterministically from prose plus a fenced code block — the shape coding\n * models naturally emit — at no extra model cost. `two-step` extracts with a\n * second call per turn. `chat` accepts marker output only. Default:\n * `tolerant`.\n */\n controlAdapter?: 'chat' | 'two-step' | 'tolerant'\n /** Python command used to load agent-eval-rpc[dspy]. Default: python. */\n runner?: ExternalOptimizerRunnerCommand\n /** Whole investigation deadline. Default: 10 minutes. */\n timeoutMs?: number\n}\n\n/** Use the official DSPy RLM as a bounded recursive trace-analysis engine. */\nexport function createDspyRlmTraceEngine(options: DspyRlmTraceEngineOptions): TraceAnalysisEngine {\n assertOptions(options)\n const maxOutputTokens = options.maxOutputTokens ?? DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS\n if (!Number.isSafeInteger(maxOutputTokens) || maxOutputTokens <= 0) {\n throw new TypeError('DSPy RLM maxOutputTokens must be a positive safe integer')\n }\n const controlAdapter = options.controlAdapter ?? 'tolerant'\n const maxReasoningTokens = options.maxReasoningTokens ?? maxOutputTokens * 4\n if (!Number.isSafeInteger(maxReasoningTokens) || maxReasoningTokens < 0) {\n throw new TypeError('DSPy RLM maxReasoningTokens must be a non-negative safe integer')\n }\n const timeoutMs = options.timeoutMs ?? DEFAULT_TIMEOUT_MS\n assertTimerDelay(timeoutMs, 'timeoutMs')\n const maxCostUsd = options.maxCostUsd ?? DEFAULT_MAX_COST_USD\n if (!Number.isFinite(maxCostUsd) || maxCostUsd <= 0) {\n throw new TypeError('DSPy RLM maxCostUsd must be positive and finite')\n }\n const pricing = options.pricing ?? pricingForModel(options.model)\n const maxModelRequestBytes = options.maxModelRequestBytes ?? DEFAULT_MAX_MODEL_REQUEST_BYTES\n const maxModelResponseBytes = options.maxModelResponseBytes ?? DEFAULT_MAX_MODEL_RESPONSE_BYTES\n const modelRequestTimeoutMs = options.modelRequestTimeoutMs ?? timeoutMs\n const traceToolTimeoutMs = options.traceToolTimeoutMs ?? DEFAULT_TRACE_TOOL_TIMEOUT_MS\n assertExternalOptimizerModelBudget(\n {\n maxCostUsd,\n maxRequests: options.maxModelRequests ?? 1,\n maxRequestBytes: maxModelRequestBytes,\n maxResponseBytes: maxModelResponseBytes,\n maxOutputTokensPerRequest: maxOutputTokens,\n maxReasoningTokensPerRequest: maxReasoningTokens,\n pricing,\n requestTimeoutMs: modelRequestTimeoutMs,\n },\n 'DSPy RLM model limits',\n )\n resolveExternalOptimizerCallbackLimits(options.traceToolLimits, 'DSPy RLM trace tool limits')\n assertTimerDelay(traceToolTimeoutMs, 'traceToolTimeoutMs')\n\n const runner = sanitizedRunner(options.runner)\n const processLimits = resolveExternalOptimizerProcessLimits(runner?.limits)\n return {\n id: 'dspy-rlm',\n description: 'Official DSPy RLM with bounded trace tools and metered model calls.',\n model: options.model,\n version: DSPY_RLM_ENGINE_VERSION,\n executionConfig: {\n bridge_module: BRIDGE_MODULE,\n call_ref: options.callRef,\n model: options.model,\n pricing: { ...pricing },\n max_cost_usd: maxCostUsd,\n max_output_tokens: maxOutputTokens,\n max_reasoning_tokens: maxReasoningTokens,\n control_adapter: controlAdapter,\n timeout_ms: timeoutMs,\n max_model_requests: options.maxModelRequests ?? null,\n max_request_bytes: maxModelRequestBytes,\n max_response_bytes: maxModelResponseBytes,\n model_request_timeout_ms: modelRequestTimeoutMs,\n trace_tool_limits: options.traceToolLimits ?? null,\n trace_tool_timeout_ms: traceToolTimeoutMs,\n process_limits: processLimits,\n runner: runner ? 'caller-supplied' : 'default',\n runner_command: runner?.command ?? null,\n },\n async analyze(request) {\n const callback = await startTraceToolCallback({\n tools: request.tools,\n maxCalls: request.limits.maxToolCalls,\n ...(options.traceToolLimits ? { limits: options.traceToolLimits } : {}),\n ...(request.signal ? { signal: request.signal } : {}),\n })\n let modelProxy: ExternalOptimizerModelProxy | undefined\n const modelExecutions: ExternalOptimizerModelExecutionObservation[] = []\n const result = await runWithCleanup({\n label: 'DSPy RLM trace-analysis resources',\n run: async () => {\n modelProxy = await startExternalOptimizerModelProxy({\n call: options.call,\n callRef: options.callRef,\n recordExecution: (observation) => {\n modelExecutions.push(structuredClone(observation))\n options.recordExecution(observation)\n },\n model: options.model,\n budget: {\n maxCostUsd,\n maxRequests: resolveMaxModelRequests(options.maxModelRequests, request.limits),\n maxRequestBytes: maxModelRequestBytes,\n maxResponseBytes: maxModelResponseBytes,\n maxOutputTokensPerRequest: maxOutputTokens,\n maxReasoningTokensPerRequest: maxReasoningTokens,\n requestTimeoutMs: modelRequestTimeoutMs,\n pricing,\n },\n costLedger: request.costLedger,\n channel: 'analyst',\n phase: request.costPhase,\n actor: request.analystId,\n ...(request.costTags ? { tags: request.costTags } : {}),\n ...(request.signal ? { signal: request.signal } : {}),\n })\n request.log?.('trace analyst engine started', {\n engine: 'dspy-rlm',\n model: options.model,\n tools: request.tools.map((tool) => tool.name),\n limits: request.limits,\n })\n const raw = await runExternalOptimizerProcess<unknown>({\n label: 'DSPy RLM trace analysis',\n tempPrefix: 'agent-eval-dspy-rlm-',\n module: BRIDGE_MODULE,\n input: {\n operation: 'analyze',\n question: request.question,\n instructions: request.instructions,\n modelProxy: {\n baseUrl: modelProxy.baseUrl,\n apiKey: modelProxy.apiKey,\n model: options.model,\n maxOutputTokens,\n },\n toolCallback: {\n url: callback.url,\n token: callback.token,\n timeoutMs: traceToolTimeoutMs,\n },\n toolSpecs: request.tools.map(({ name, description, parameters }) => ({\n name,\n description,\n parameters,\n })),\n controlAdapter,\n limits: {\n maxIterations: request.limits.maxIterations,\n maxLlmCalls: request.limits.maxLlmCalls,\n maxOutputChars: request.limits.maxOutputChars,\n },\n // Omitted entirely when the caller supplied none, so a request\n // without structured inputs sends the payload it always sent.\n ...(request.taskInputs ? { taskInputs: request.taskInputs } : {}),\n },\n ...(runner ? { runner } : {}),\n additionalArgs: ['--max-input-bytes', String(processLimits.maxInputBytes)],\n timeoutMs,\n ...(request.signal ? { signal: request.signal } : {}),\n })\n const parsed = parseBridgeOutput(raw, (index, reason) => {\n request.log?.('finding rejected: bridge row failed schema validation', {\n engine: 'dspy-rlm',\n index,\n reason,\n })\n })\n const successfulCompletions = modelProxy.successfulCompletions()\n const requestAttempts = modelProxy.requestAttempts()\n if (parsed.modelCalls !== successfulCompletions) {\n throw new Error(\n `DSPy RLM reported ${parsed.modelCalls} model calls, but the provider proxy recorded ${successfulCompletions}`,\n )\n }\n modelProxy.assertExecutionComplete()\n return {\n ...parsed,\n toolCalls: callback.calls(),\n runtime: {\n ...parsed.runtime,\n modelRequestAttempts: requestAttempts,\n modelSuccessfulCompletions: successfulCompletions,\n modelExecutions,\n },\n } satisfies TraceAnalysisEngineResult\n },\n cleanup: async () => {\n const results = await Promise.allSettled([\n ...(modelProxy ? [modelProxy.close()] : []),\n callback.close(),\n ])\n const errors = results.flatMap((entry) =>\n entry.status === 'rejected' ? [entry.reason] : [],\n )\n if (errors.length > 0) {\n throw new AggregateError(errors, 'DSPy RLM resource cleanup failed')\n }\n },\n })\n request.log?.('trace analyst engine completed', {\n engine: 'dspy-rlm',\n model_calls: result.modelCalls,\n model_request_attempts: result.runtime.modelRequestAttempts,\n tool_calls: result.toolCalls,\n findings: result.findings.length,\n })\n return result\n },\n }\n}\n\nfunction parseBridgeOutput(\n value: unknown,\n onRejectedFinding: (index: number, reason: string) => void,\n): Omit<TraceAnalysisEngineResult, 'toolCalls'> {\n if (!isRecord(value)) throw new Error('DSPy RLM bridge output must be an object')\n if (typeof value.answer !== 'string' || !value.answer.trim()) {\n throw new Error('DSPy RLM bridge returned no answer')\n }\n // Findings are model output: one malformed row is model noise, not a bridge\n // fault, and the rest of the paid investigation must survive it. The codec\n // is the same decoder the Python side runs, so a row this bridge accepts is\n // a row the optimizer could report, and vice versa.\n const decoded = decodeRawFindingArray(value.findings)\n if (decoded.topLevelError !== undefined) {\n throw new Error(`DSPy RLM bridge findings must be an array: ${decoded.topLevelError}`)\n }\n const findings = decoded.accepted\n const rejectedFindings = decoded.rejected.length\n for (const rejection of decoded.rejected) {\n onRejectedFinding(\n rejection.index,\n `${rejection.code}${rejection.path ? ` at ${rejection.path}` : ''}: ${rejection.message}`,\n )\n }\n if (!Array.isArray(value.trajectory)) {\n throw new Error('DSPy RLM bridge trajectory must be an array')\n }\n if (!Number.isSafeInteger(value.modelCalls) || (value.modelCalls as number) <= 0) {\n throw new Error('DSPy RLM bridge modelCalls must be a positive safe integer')\n }\n if (!isRecord(value.runtime)) {\n throw new Error('DSPy RLM bridge runtime must be an object')\n }\n return {\n answer: value.answer,\n findings,\n trajectory: value.trajectory,\n modelCalls: value.modelCalls as number,\n runtime: { ...value.runtime, rejectedFindings },\n }\n}\n\nfunction assertOptions(options: DspyRlmTraceEngineOptions): void {\n for (const [name, value] of [\n ['callRef', options.callRef],\n ['model', options.model],\n ] as const) {\n if (typeof value !== 'string' || !value.trim() || value !== value.trim()) {\n throw new TypeError(`DSPy RLM ${name} must be a trimmed non-empty string`)\n }\n }\n if (typeof options.call !== 'function') {\n throw new TypeError('DSPy RLM call must be a function')\n }\n if (typeof options.recordExecution !== 'function') {\n throw new TypeError('DSPy RLM recordExecution must be a function')\n }\n if (\n options.maxModelRequests !== undefined &&\n (!Number.isSafeInteger(options.maxModelRequests) || options.maxModelRequests <= 0)\n ) {\n throw new TypeError('DSPy RLM maxModelRequests must be a positive safe integer')\n }\n}\n\nfunction resolveMaxModelRequests(\n configured: number | undefined,\n limits: { maxIterations: number; maxLlmCalls: number },\n): number {\n if (configured !== undefined) return configured\n const derived = limits.maxIterations + limits.maxLlmCalls + 1\n if (!Number.isSafeInteger(derived) || derived <= 0) {\n throw new Error('DSPy RLM analysis limits produce an invalid model request limit')\n }\n return derived\n}\n\nfunction assertTimerDelay(value: number, field: string): void {\n if (!Number.isSafeInteger(value) || value <= 0 || value > MAX_TIMER_DELAY_MS) {\n throw new TypeError(`DSPy RLM ${field} must be between 1 and ${MAX_TIMER_DELAY_MS}`)\n }\n}\n\nfunction pricingForModel(model: string): CustomTokenPricing {\n const pricing = resolveModelPricing(model)\n if (!pricing) {\n throw new Error(\n `no pricing is configured for '${model}'; provide DspyRlmTraceEngineOptions.pricing`,\n )\n }\n return {\n inputUsdPerMillion: pricing.input * 1_000,\n outputUsdPerMillion: pricing.output * 1_000,\n }\n}\n\nfunction sanitizedRunner(\n runner: ExternalOptimizerRunnerCommand | undefined,\n): ExternalOptimizerRunnerCommand | undefined {\n if (!runner) return undefined\n return {\n ...(runner.command ? { command: runner.command } : {}),\n ...(runner.args ? { args: runner.args } : {}),\n ...(runner.env ? { env: removeCredentialEnvironment(runner.env) } : {}),\n ...(runner.limits ? { limits: { ...runner.limits } } : {}),\n }\n}\n\n"],"mappings":";;;;;;;;;;;;;;;AAmEA,SAAgB,sBAAsB,OAAqC;CACzE,MAAM,OAAO,YAAY,KAAK;CAC9B,IAAI,OAAO,SAAS,UAAU,OAAO;EAAE,UAAU,CAAC;EAAG,UAAU,CAAC;EAAG,eAAe;CAAK;CAEvF,MAAM,WAAgC,CAAC;CACvC,MAAM,WAAiC,CAAC;CACxC,KAAK,MAAM,CAAC,OAAO,QAAQ,KAAK,QAAQ,GAAG;EACzC,IAAI,SAAA,KAA2B;GAC7B,SAAS,KAAK;IACZ;IACA,MAAM;IACN,MAAM;IACN,SAAS;GACX,CAAC;GACD;EACF;EACA,MAAM,UAAU,UAAU,KAAK,KAAK;EACpC,IAAI,aAAa,SAAS,SAAS,KAAK,QAAQ,OAAO;OAClD,SAAS,KAAK,QAAQ,SAAS;CACtC;CACA,OAAO;EAAE;EAAU;CAAS;AAC9B;AAEA,SAAS,UACP,KACA,OACoE;CACpE,IAAI,QAAQ,QAAQ,OAAO,QAAQ,YAAY,MAAM,QAAQ,GAAG,GAC9D,OAAO,EACL,WAAW;EACT;EACA,MAAM;EACN,MAAM;EACN,SAAS,2CAA2C,SAAS,GAAG;CAClE,EACF;CAEF,MAAM,SAAS,wBAAwB,UAAU,GAAG;CACpD,IAAI,CAAC,OAAO,SAAS;EACnB,MAAM,QAAQ,OAAO,MAAM,OAAO;EAClC,MAAM,QAAQ,OAAO,QAAQ,CAAC,EAAA,CAAG,KAAK,GAAG;EAOzC,OAAO,EACL,WAAW;GAAE;GAAO;GAAM,MAJ1B,SAAS,aAAa,oBAAqB,IAA6B,OAAO,MAAM,OACjF,oBACA;GAE4B,SAAS,OAAO,WAAW;EAAgC,EAC7F;CACF;CACA,OAAO,EAAE,SAAS,OAAO,KAAK;AAChC;;;;;AAMA,SAAS,YAAY,OAAoC;CACvD,IAAI,MAAM,QAAQ,KAAK,GAAG,OAAO;CACjC,IAAI,OAAO,UAAU,UAAU;EAC7B,MAAM,SAAS,WAAW,KAAK;EAC/B,IAAI,WAAW,KAAA,GACb,OAAO;EAET,IAAI,MAAM,QAAQ,MAAM,GAAG,OAAO;EAClC,MAAM,YAAY,iBAAiB,MAAM;EACzC,IAAI,cAAc,KAAA,GAAW,OAAO;EACpC,OAAO,2CAA2C,SAAS,MAAM;CACnE;CACA,MAAM,YAAY,iBAAiB,KAAK;CACxC,IAAI,cAAc,KAAA,GAAW,OAAO;CACpC,OAAO,uCAAuC,SAAS,KAAK;AAC9D;;AAGA,SAAS,iBAAiB,OAAuC;CAC/D,IAAI,UAAU,QAAQ,OAAO,UAAU,YAAY,MAAM,QAAQ,KAAK,GAAG,OAAO,KAAA;CAChF,MAAM,QAAS,MAAkC;CACjD,OAAO,MAAM,QAAQ,KAAK,IAAI,QAAQ,KAAA;AACxC;AAEA,SAAS,SAAS,OAAwB;CACxC,IAAI,UAAU,MAAM,OAAO;CAC3B,IAAI,MAAM,QAAQ,KAAK,GAAG,OAAO;CACjC,OAAO,KAAK,OAAO;AACrB;;;;AChIA,eAAsB,uBAAuB,MAMd;CAC7B,IAAI,CAAC,OAAO,cAAc,KAAK,QAAQ,KAAK,KAAK,YAAY,GAC3D,MAAM,IAAI,UAAU,8DAA8D;CAEpF,MAAM,SAAS,uCAAuC,KAAK,QAAQ,4BAA4B;CAC/F,KAAK,QAAQ,eAAe;CAC5B,MAAM,SAAS,IAAI,IAAI,KAAK,MAAM,KAAK,SAAS,CAAC,KAAK,MAAM,IAAI,CAAC,CAAC;CAClE,IAAI,OAAO,SAAS,KAAK,MAAM,QAC7B,MAAM,IAAI,MAAM,mDAAmD;CAGrE,MAAM,QAAQ,YAAY,EAAE,CAAC,CAAC,SAAS,KAAK;CAC5C,IAAI,QAAQ;CACZ,IAAI,YAAY;CAChB,IAAI;CACJ,MAAM,oCAAoB,IAAI,IAAqB;CACnD,MAAM,iCAAiB,IAAI,IAAmB;CAC9C,MAAM,SAAS,cAAc,SAAS,aAAa;EACjD,IAAI,CAAC,WAAW;GACd,eAAe,UAAU,KAAK,EAAE,OAAO,iCAAiC,CAAC;GACzE;EACF;EACA,MAAM,aAAa,IAAI,gBAAgB;EACvC,MAAM,qBAA2B;GAC/B,QAAQ,QAAQ;GAChB,SAAS,QAAQ;EACnB;EACA,kBAAkB,IAAI,UAAU;EAChC,WAAW,OAAO,iBAAiB,SAAS,cAAc,EAAE,MAAM,KAAK,CAAC;EAExE,IAAI;EACJ,UAAU,cAAc,SAAS,UAAU,WAAW,MAAM,CAAC,CAAC,cAAc;GAC1E,WAAW,OAAO,oBAAoB,SAAS,YAAY;GAC3D,kBAAkB,OAAO,UAAU;GACnC,eAAe,OAAO,OAAO;EAC/B,CAAC;EACD,eAAe,IAAI,OAAO;EAC1B,QAAa,YAAY,KAAA,CAAS;CACpC,CAAC;CACD,MAAM,OAAO,MAAM,YAAY,MAAM;CACrC,MAAM,cAA6B;EACjC,iBAAiB,cAAc;EAC/B,OAAO;CACT;CACA,MAAM,gBAAsB;EAC1B,MAAW,CAAC,CAAC,YAAY,KAAA,CAAS;CACpC;CACA,KAAK,QAAQ,iBAAiB,SAAS,SAAS,EAAE,MAAM,KAAK,CAAC;CAC9D,IAAI,KAAK,QAAQ,SAAS,QAAQ;CAElC,OAAO;EACL,KAAK,oBAAoB,KAAK;EAC9B;EACA,aAAa;EACb;CACF;CAEA,eAAe,cACb,SACA,UACA,QACe;EACf,IAAI;GACF,IAAI,QAAQ,WAAW,UAAU,QAAQ,QAAQ,SAAS;IACxD,eAAe,UAAU,KAAK,EAAE,OAAO,YAAY,CAAC;IACpD;GACF;GACA,IAAI,QAAQ,QAAQ,kBAAkB,UAAU,SAAS;IACvD,eAAe,UAAU,KAAK,EAAE,OAAO,eAAe,CAAC;IACvD;GACF;GACA,IAAI,SAAS,KAAK,UAAU;IAC1B,eAAe,UAAU,KAAK,EAAE,OAAO,gCAAgC,CAAC;IACxE;GACF;GACA,MAAM,OAAO,MAAM,SAAS,SAAS,OAAO,eAAe;GAC3D,IAAI,CAAC,SAAS,IAAI,KAAK,OAAO,KAAK,SAAS,YAAY,EAAE,UAAU,OAAO;IACzE,eAAe,UAAU,KAAK,EAAE,OAAO,6BAA6B,CAAC;IACrE;GACF;GACA,MAAM,OAAO,OAAO,IAAI,KAAK,IAAI;GACjC,IAAI,CAAC,MAAM;IACT,eAAe,UAAU,KAAK,EAAE,OAAO,uBAAuB,KAAK,KAAK,GAAG,CAAC;IAC5E;GACF;GACA,SAAS;GACT,MAAM,SAAS,MAAM,KAAK,QAAQ,KAAK,MAAM,EAAE,OAAO,CAAC;GACvD,MAAM,UAAU,KAAK,UAAU,EAAE,OAAO,CAAC;GACzC,IAAI,OAAO,WAAW,OAAO,IAAI,OAAO,kBAAkB;IACxD,eAAe,UAAU,KAAK,EAAE,OAAO,gCAAgC,CAAC;IACxE;GACF;GACA,SAAS,UAAU,KAAK;IACtB,gBAAgB;IAChB,kBAAkB,OAAO,OAAO,WAAW,OAAO,CAAC;GACrD,CAAC;GACD,SAAS,IAAI,OAAO;EACtB,SAAS,OAAO;GACd,eAAe,UAAU,OAAO,UAAU,MAAM,KAAK,EACnD,OAAO,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK,EAC9D,CAAC;EACH;CACF;CAEA,eAAe,gBAA+B;EAC5C,KAAK,QAAQ,oBAAoB,SAAS,OAAO;EACjD,YAAY;EACZ,MAAM,gBAAgB,YAAY,MAAM;EACxC,OAAO,uBAAuB;EAC9B,KAAK,MAAM,cAAc,mBAAmB,WAAW,MAAM;EAC7D,MAAM,CAAC,gBAAgB,MAAM,QAAQ,WAAW,CAC9C,eACA,sBAAsB,cAAc,CACtC,CAAC;EACD,IAAI,kBAAkB,SAAS,KAAK,eAAe,SAAS,GAC1D,MAAM,IAAI,MAAM,iDAAiD;EAEnE,IAAI,cAAc,WAAW,YAAY,MAAM,aAAa;CAC9D;AACF;AAEA,SAAS,SAAS,SAA0B,iBAA2C;CACrF,OAAO,IAAI,SAAS,SAAS,WAAW;EACtC,IAAI,OAAO;EACX,MAAM,SAAmB,CAAC;EAC1B,QAAQ,GAAG,SAAS,UAAkB;GACpC,QAAQ,MAAM;GACd,IAAI,OAAO,iBAAiB;IAC1B,uBAAO,IAAI,MAAM,8BAA8B,CAAC;IAChD,QAAQ,QAAQ;IAChB;GACF;GACA,OAAO,KAAK,KAAK;EACnB,CAAC;EACD,QAAQ,GAAG,SAAS,MAAM;EAC1B,QAAQ,GAAG,aAAa;GACtB,IAAI;IACF,QAAQ,KAAK,MAAM,OAAO,OAAO,MAAM,CAAC,CAAC,SAAS,MAAM,CAAC,CAAC;GAC5D,SAAS,OAAO;IACd,OAAO,KAAK;GACd;EACF,CAAC;CACH,CAAC;AACH;;;ACtJA,MAAM,qBAAqB;AAC3B,MAAM,uBAAuB;AAC7B,MAAM,kCAAkC;AACxC,MAAM,mCAAmC;AACzC,MAAM,gCAAgC;AACtC,MAAM,qBAAqB;AAC3B,MAAM,gBAAgB;;AAEtB,MAAM,0BAA0B;;AAmDhC,SAAgB,yBAAyB,SAAyD;CAChG,cAAc,OAAO;CACrB,MAAM,kBAAkB,QAAQ,mBAAA;CAChC,IAAI,CAAC,OAAO,cAAc,eAAe,KAAK,mBAAmB,GAC/D,MAAM,IAAI,UAAU,0DAA0D;CAEhF,MAAM,iBAAiB,QAAQ,kBAAkB;CACjD,MAAM,qBAAqB,QAAQ,sBAAsB,kBAAkB;CAC3E,IAAI,CAAC,OAAO,cAAc,kBAAkB,KAAK,qBAAqB,GACpE,MAAM,IAAI,UAAU,iEAAiE;CAEvF,MAAM,YAAY,QAAQ,aAAa;CACvC,iBAAiB,WAAW,WAAW;CACvC,MAAM,aAAa,QAAQ,cAAc;CACzC,IAAI,CAAC,OAAO,SAAS,UAAU,KAAK,cAAc,GAChD,MAAM,IAAI,UAAU,iDAAiD;CAEvE,MAAM,UAAU,QAAQ,WAAW,gBAAgB,QAAQ,KAAK;CAChE,MAAM,uBAAuB,QAAQ,wBAAwB;CAC7D,MAAM,wBAAwB,QAAQ,yBAAyB;CAC/D,MAAM,wBAAwB,QAAQ,yBAAyB;CAC/D,MAAM,qBAAqB,QAAQ,sBAAsB;CACzD,mCACE;EACE;EACA,aAAa,QAAQ,oBAAoB;EACzC,iBAAiB;EACjB,kBAAkB;EAClB,2BAA2B;EAC3B,8BAA8B;EAC9B;EACA,kBAAkB;CACpB,GACA,uBACF;CACA,uCAAuC,QAAQ,iBAAiB,4BAA4B;CAC5F,iBAAiB,oBAAoB,oBAAoB;CAEzD,MAAM,SAAS,gBAAgB,QAAQ,MAAM;CAC7C,MAAM,gBAAgB,sCAAsC,QAAQ,MAAM;CAC1E,OAAO;EACL,IAAI;EACJ,aAAa;EACb,OAAO,QAAQ;EACf,SAAS;EACT,iBAAiB;GACf,eAAe;GACf,UAAU,QAAQ;GAClB,OAAO,QAAQ;GACf,SAAS,EAAE,GAAG,QAAQ;GACtB,cAAc;GACd,mBAAmB;GACnB,sBAAsB;GACtB,iBAAiB;GACjB,YAAY;GACZ,oBAAoB,QAAQ,oBAAoB;GAChD,mBAAmB;GACnB,oBAAoB;GACpB,0BAA0B;GAC1B,mBAAmB,QAAQ,mBAAmB;GAC9C,uBAAuB;GACvB,gBAAgB;GAChB,QAAQ,SAAS,oBAAoB;GACrC,gBAAgB,QAAQ,WAAW;EACrC;EACA,MAAM,QAAQ,SAAS;GACrB,MAAM,WAAW,MAAM,uBAAuB;IAC5C,OAAO,QAAQ;IACf,UAAU,QAAQ,OAAO;IACzB,GAAI,QAAQ,kBAAkB,EAAE,QAAQ,QAAQ,gBAAgB,IAAI,CAAC;IACrE,GAAI,QAAQ,SAAS,EAAE,QAAQ,QAAQ,OAAO,IAAI,CAAC;GACrD,CAAC;GACD,IAAI;GACJ,MAAM,kBAAgE,CAAC;GACvE,MAAM,SAAS,MAAM,eAAe;IAClC,OAAO;IACP,KAAK,YAAY;KACf,aAAa,MAAM,iCAAiC;MAClD,MAAM,QAAQ;MACd,SAAS,QAAQ;MACjB,kBAAkB,gBAAgB;OAChC,gBAAgB,KAAK,gBAAgB,WAAW,CAAC;OACjD,QAAQ,gBAAgB,WAAW;MACrC;MACA,OAAO,QAAQ;MACf,QAAQ;OACN;OACA,aAAa,wBAAwB,QAAQ,kBAAkB,QAAQ,MAAM;OAC7E,iBAAiB;OACjB,kBAAkB;OAClB,2BAA2B;OAC3B,8BAA8B;OAC9B,kBAAkB;OAClB;MACF;MACA,YAAY,QAAQ;MACpB,SAAS;MACT,OAAO,QAAQ;MACf,OAAO,QAAQ;MACf,GAAI,QAAQ,WAAW,EAAE,MAAM,QAAQ,SAAS,IAAI,CAAC;MACrD,GAAI,QAAQ,SAAS,EAAE,QAAQ,QAAQ,OAAO,IAAI,CAAC;KACrD,CAAC;KACD,QAAQ,MAAM,gCAAgC;MAC5C,QAAQ;MACR,OAAO,QAAQ;MACf,OAAO,QAAQ,MAAM,KAAK,SAAS,KAAK,IAAI;MAC5C,QAAQ,QAAQ;KAClB,CAAC;KAwCD,MAAM,SAAS,kBAAkB,MAvCf,4BAAqC;MACrD,OAAO;MACP,YAAY;MACZ,QAAQ;MACR,OAAO;OACL,WAAW;OACX,UAAU,QAAQ;OAClB,cAAc,QAAQ;OACtB,YAAY;QACV,SAAS,WAAW;QACpB,QAAQ,WAAW;QACnB,OAAO,QAAQ;QACf;OACF;OACA,cAAc;QACZ,KAAK,SAAS;QACd,OAAO,SAAS;QAChB,WAAW;OACb;OACA,WAAW,QAAQ,MAAM,KAAK,EAAE,MAAM,aAAa,kBAAkB;QACnE;QACA;QACA;OACF,EAAE;OACF;OACA,QAAQ;QACN,eAAe,QAAQ,OAAO;QAC9B,aAAa,QAAQ,OAAO;QAC5B,gBAAgB,QAAQ,OAAO;OACjC;OAGA,GAAI,QAAQ,aAAa,EAAE,YAAY,QAAQ,WAAW,IAAI,CAAC;MACjE;MACA,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC;MAC3B,gBAAgB,CAAC,qBAAqB,OAAO,cAAc,aAAa,CAAC;MACzE;MACA,GAAI,QAAQ,SAAS,EAAE,QAAQ,QAAQ,OAAO,IAAI,CAAC;KACrD,CAAC,IACsC,OAAO,WAAW;MACvD,QAAQ,MAAM,yDAAyD;OACrE,QAAQ;OACR;OACA;MACF,CAAC;KACH,CAAC;KACD,MAAM,wBAAwB,WAAW,sBAAsB;KAC/D,MAAM,kBAAkB,WAAW,gBAAgB;KACnD,IAAI,OAAO,eAAe,uBACxB,MAAM,IAAI,MACR,qBAAqB,OAAO,WAAW,gDAAgD,uBACzF;KAEF,WAAW,wBAAwB;KACnC,OAAO;MACL,GAAG;MACH,WAAW,SAAS,MAAM;MAC1B,SAAS;OACP,GAAG,OAAO;OACV,sBAAsB;OACtB,4BAA4B;OAC5B;MACF;KACF;IACF;IACA,SAAS,YAAY;KAKnB,MAAM,UAAS,MAJO,QAAQ,WAAW,CACvC,GAAI,aAAa,CAAC,WAAW,MAAM,CAAC,IAAI,CAAC,GACzC,SAAS,MAAM,CACjB,CAAC,EAAA,CACsB,SAAS,UAC9B,MAAM,WAAW,aAAa,CAAC,MAAM,MAAM,IAAI,CAAC,CAClD;KACA,IAAI,OAAO,SAAS,GAClB,MAAM,IAAI,eAAe,QAAQ,kCAAkC;IAEvE;GACF,CAAC;GACD,QAAQ,MAAM,kCAAkC;IAC9C,QAAQ;IACR,aAAa,OAAO;IACpB,wBAAwB,OAAO,QAAQ;IACvC,YAAY,OAAO;IACnB,UAAU,OAAO,SAAS;GAC5B,CAAC;GACD,OAAO;EACT;CACF;AACF;AAEA,SAAS,kBACP,OACA,mBAC8C;CAC9C,IAAI,CAAC,SAAS,KAAK,GAAG,MAAM,IAAI,MAAM,0CAA0C;CAChF,IAAI,OAAO,MAAM,WAAW,YAAY,CAAC,MAAM,OAAO,KAAK,GACzD,MAAM,IAAI,MAAM,oCAAoC;CAMtD,MAAM,UAAU,sBAAsB,MAAM,QAAQ;CACpD,IAAI,QAAQ,kBAAkB,KAAA,GAC5B,MAAM,IAAI,MAAM,8CAA8C,QAAQ,eAAe;CAEvF,MAAM,WAAW,QAAQ;CACzB,MAAM,mBAAmB,QAAQ,SAAS;CAC1C,KAAK,MAAM,aAAa,QAAQ,UAC9B,kBACE,UAAU,OACV,GAAG,UAAU,OAAO,UAAU,OAAO,OAAO,UAAU,SAAS,GAAG,IAAI,UAAU,SAClF;CAEF,IAAI,CAAC,MAAM,QAAQ,MAAM,UAAU,GACjC,MAAM,IAAI,MAAM,6CAA6C;CAE/D,IAAI,CAAC,OAAO,cAAc,MAAM,UAAU,KAAM,MAAM,cAAyB,GAC7E,MAAM,IAAI,MAAM,4DAA4D;CAE9E,IAAI,CAAC,SAAS,MAAM,OAAO,GACzB,MAAM,IAAI,MAAM,2CAA2C;CAE7D,OAAO;EACL,QAAQ,MAAM;EACd;EACA,YAAY,MAAM;EAClB,YAAY,MAAM;EAClB,SAAS;GAAE,GAAG,MAAM;GAAS;EAAiB;CAChD;AACF;AAEA,SAAS,cAAc,SAA0C;CAC/D,KAAK,MAAM,CAAC,MAAM,UAAU,CAC1B,CAAC,WAAW,QAAQ,OAAO,GAC3B,CAAC,SAAS,QAAQ,KAAK,CACzB,GACE,IAAI,OAAO,UAAU,YAAY,CAAC,MAAM,KAAK,KAAK,UAAU,MAAM,KAAK,GACrE,MAAM,IAAI,UAAU,YAAY,KAAK,oCAAoC;CAG7E,IAAI,OAAO,QAAQ,SAAS,YAC1B,MAAM,IAAI,UAAU,kCAAkC;CAExD,IAAI,OAAO,QAAQ,oBAAoB,YACrC,MAAM,IAAI,UAAU,6CAA6C;CAEnE,IACE,QAAQ,qBAAqB,KAAA,MAC5B,CAAC,OAAO,cAAc,QAAQ,gBAAgB,KAAK,QAAQ,oBAAoB,IAEhF,MAAM,IAAI,UAAU,2DAA2D;AAEnF;AAEA,SAAS,wBACP,YACA,QACQ;CACR,IAAI,eAAe,KAAA,GAAW,OAAO;CACrC,MAAM,UAAU,OAAO,gBAAgB,OAAO,cAAc;CAC5D,IAAI,CAAC,OAAO,cAAc,OAAO,KAAK,WAAW,GAC/C,MAAM,IAAI,MAAM,iEAAiE;CAEnF,OAAO;AACT;AAEA,SAAS,iBAAiB,OAAe,OAAqB;CAC5D,IAAI,CAAC,OAAO,cAAc,KAAK,KAAK,SAAS,KAAK,QAAQ,oBACxD,MAAM,IAAI,UAAU,YAAY,MAAM,yBAAyB,oBAAoB;AAEvF;AAEA,SAAS,gBAAgB,OAAmC;CAC1D,MAAM,UAAU,oBAAoB,KAAK;CACzC,IAAI,CAAC,SACH,MAAM,IAAI,MACR,iCAAiC,MAAM,6CACzC;CAEF,OAAO;EACL,oBAAoB,QAAQ,QAAQ;EACpC,qBAAqB,QAAQ,SAAS;CACxC;AACF;AAEA,SAAS,gBACP,QAC4C;CAC5C,IAAI,CAAC,QAAQ,OAAO,KAAA;CACpB,OAAO;EACL,GAAI,OAAO,UAAU,EAAE,SAAS,OAAO,QAAQ,IAAI,CAAC;EACpD,GAAI,OAAO,OAAO,EAAE,MAAM,OAAO,KAAK,IAAI,CAAC;EAC3C,GAAI,OAAO,MAAM,EAAE,KAAK,4BAA4B,OAAO,GAAG,EAAE,IAAI,CAAC;EACrE,GAAI,OAAO,SAAS,EAAE,QAAQ,EAAE,GAAG,OAAO,OAAO,EAAE,IAAI,CAAC;CAC1D;AACF"}
|
|
1
|
+
{"version":3,"file":"dspy-rlm-engine-CSMwudLO.js","names":[],"sources":["../src/analyst/finding-codec.ts","../src/analyst/trace-tool-callback.ts","../src/analyst/dspy-rlm-engine.ts"],"sourcesContent":["/**\n * The raw-finding wire contract — one decoder for both sides of the process\n * boundary.\n *\n * A finding array crosses a process boundary twice: an optimizer bridge writes\n * it in Python, TypeScript reads it back. Two decoders written independently\n * accept different shapes, so Python could report a completed investigation\n * whose rows TypeScript then dropped — the paid work disappearing between two\n * \"successes\". This module is the TypeScript half; `finding_contract.json`\n * carries the same rules to the Python half, generated from the schema here.\n *\n * Decoding never invents an empty result. A value that is not an array of rows\n * fails with the exact type it was, and a malformed row is reported with its\n * index and path while its valid siblings survive.\n */\n\nimport { type RawAnalystFinding, RawAnalystFindingSchema } from './finding-signature'\nimport { parseFindingSubject } from './finding-subject'\nimport { coerceJson } from './parse-tolerant'\n\n/**\n * Version of the wire contract both languages implement. Bump when the\n * accepted row shape changes; the Python package pins the same value, and the\n * emitted contract carries it so a mismatched pair is visible.\n */\nexport const FINDING_WIRE_CONTRACT_VERSION = 1\n\n/** Why one row was refused. Stable across languages: Python reports the same codes. */\ntype FindingRejectionCode = 'not-an-object' | 'schema' | 'invalid-subject' | 'row-limit'\n\n/** A refused row, named precisely enough to repair or report. */\ninterface RejectedFindingRow {\n /** Position in the submitted array. */\n readonly index: number\n /** Dotted path to the offending field, `''` for the row itself. */\n readonly path: string\n readonly code: FindingRejectionCode\n readonly message: string\n}\n\ninterface DecodedFindingArray {\n readonly accepted: RawAnalystFinding[]\n readonly rejected: RejectedFindingRow[]\n /**\n * Set when the value was not a finding array at all — the array itself is\n * the failure, so `accepted` and `rejected` are both empty and a caller must\n * not read the result as \"no findings\".\n */\n readonly topLevelError?: string\n}\n\n/**\n * Rows past this count are refused rather than validated. A model that emits\n * thousands of rows has lost the plot, and the diagnostics for them would\n * dwarf the answer they came with.\n */\nexport const MAX_FINDING_ROWS = 500\n\n/**\n * Decode a submitted findings value into accepted rows plus per-row\n * diagnostics.\n *\n * Accepts an array of rows, or a string carrying one (a JSON array, possibly\n * fenced). Anything else — an object, a number, null, undefined — sets\n * `topLevelError` naming the type received: a caller that asked for findings\n * and got a number has a defect to report, not an empty result to record.\n */\nexport function decodeRawFindingArray(value: unknown): DecodedFindingArray {\n const rows = findingRows(value)\n if (typeof rows === 'string') return { accepted: [], rejected: [], topLevelError: rows }\n\n const accepted: RawAnalystFinding[] = []\n const rejected: RejectedFindingRow[] = []\n for (const [index, row] of rows.entries()) {\n if (index >= MAX_FINDING_ROWS) {\n rejected.push({\n index,\n path: '',\n code: 'row-limit',\n message: `findings array exceeds ${MAX_FINDING_ROWS} rows`,\n })\n continue\n }\n const decoded = decodeRow(row, index)\n if ('finding' in decoded) accepted.push(decoded.finding)\n else rejected.push(decoded.rejection)\n }\n return { accepted, rejected }\n}\n\nfunction decodeRow(\n row: unknown,\n index: number,\n): { finding: RawAnalystFinding } | { rejection: RejectedFindingRow } {\n if (row === null || typeof row !== 'object' || Array.isArray(row)) {\n return {\n rejection: {\n index,\n path: '',\n code: 'not-an-object',\n message: `finding row must be an object, received ${describe(row)}`,\n },\n }\n }\n const parsed = RawAnalystFindingSchema.safeParse(row)\n if (!parsed.success) {\n const issue = parsed.error.issues[0]\n const path = (issue?.path ?? []).join('.')\n // The schema refines `subject` against the grammar; name that refusal\n // separately so a repair turn can be told which rule the row broke.\n const code: FindingRejectionCode =\n path === 'subject' && parseFindingSubject((row as { subject?: string }).subject) === null\n ? 'invalid-subject'\n : 'schema'\n return {\n rejection: { index, path, code, message: issue?.message ?? 'row does not match the schema' },\n }\n }\n return { finding: parsed.data }\n}\n\n/**\n * Narrow a submitted value to candidate rows, or return the message naming why\n * it is not a findings array at all.\n */\nfunction findingRows(value: unknown): unknown[] | string {\n if (Array.isArray(value)) return value\n if (typeof value === 'string') {\n const parsed = coerceJson(value)\n if (parsed === undefined) {\n return 'findings must be a JSON array; the string received is not JSON'\n }\n if (Array.isArray(parsed)) return parsed\n const unwrapped = findingsProperty(parsed)\n if (unwrapped !== undefined) return unwrapped\n return `findings must be a JSON array, received ${describe(parsed)}`\n }\n const unwrapped = findingsProperty(value)\n if (unwrapped !== undefined) return unwrapped\n return `findings must be an array, received ${describe(value)}`\n}\n\n/** `{ findings: [...] }` is the one wrapper models emit often enough to unwrap. */\nfunction findingsProperty(value: unknown): unknown[] | undefined {\n if (value === null || typeof value !== 'object' || Array.isArray(value)) return undefined\n const inner = (value as Record<string, unknown>).findings\n return Array.isArray(inner) ? inner : undefined\n}\n\nfunction describe(value: unknown): string {\n if (value === null) return 'null'\n if (Array.isArray(value)) return 'an array'\n return `a ${typeof value}`\n}\n\n\n","import { randomBytes } from 'node:crypto'\nimport { createServer, type IncomingMessage, type ServerResponse } from 'node:http'\nimport {\n type ExternalOptimizerCallbackLimits,\n resolveExternalOptimizerCallbackLimits,\n} from '../campaign/external-optimizer-contracts'\nimport {\n closeServer,\n listenLocal,\n sendJsonIfOpen,\n waitForActiveHandlers,\n} from '../campaign/external-optimizer-http'\nimport type { TraceAnalysisToolDescriptor } from '../trace-analyst/tools'\n\nimport { isRecord } from '../llm-client/json'\ninterface TraceToolCallback {\n url: string\n token: string\n calls: () => number\n close: () => Promise<void>\n}\n\nexport type TraceToolCallbackLimits = ExternalOptimizerCallbackLimits\n\n/** Expose one bounded trace-tool set only on an authenticated loopback socket. */\nexport async function startTraceToolCallback(args: {\n tools: readonly TraceAnalysisToolDescriptor[]\n maxCalls: number\n /** Trace-tool request/response byte limits. Omitted fields use finite defaults. */\n limits?: Partial<TraceToolCallbackLimits>\n signal?: AbortSignal\n}): Promise<TraceToolCallback> {\n if (!Number.isSafeInteger(args.maxCalls) || args.maxCalls <= 0) {\n throw new TypeError('trace tool callback maxCalls must be a positive safe integer')\n }\n const limits = resolveExternalOptimizerCallbackLimits(args.limits, 'trace tool callback limits')\n args.signal?.throwIfAborted()\n const byName = new Map(args.tools.map((tool) => [tool.name, tool]))\n if (byName.size !== args.tools.length) {\n throw new Error('trace tool callback received duplicate tool names')\n }\n\n const token = randomBytes(32).toString('hex')\n let calls = 0\n let accepting = true\n let closePromise: Promise<void> | undefined\n const activeControllers = new Set<AbortController>()\n const activeHandlers = new Set<Promise<void>>()\n const server = createServer((request, response) => {\n if (!accepting) {\n sendJsonIfOpen(response, 503, { error: 'trace tool callback is closing' })\n return\n }\n const controller = new AbortController()\n const abortRequest = (): void => {\n request.destroy()\n response.destroy()\n }\n activeControllers.add(controller)\n controller.signal.addEventListener('abort', abortRequest, { once: true })\n\n let handler!: Promise<void>\n handler = handleRequest(request, response, controller.signal).finally(() => {\n controller.signal.removeEventListener('abort', abortRequest)\n activeControllers.delete(controller)\n activeHandlers.delete(handler)\n })\n activeHandlers.add(handler)\n void handler.catch(() => undefined)\n })\n const port = await listenLocal(server)\n const close = (): Promise<void> => {\n closePromise ??= closeCallback()\n return closePromise\n }\n const onAbort = (): void => {\n void close().catch(() => undefined)\n }\n args.signal?.addEventListener('abort', onAbort, { once: true })\n if (args.signal?.aborted) onAbort()\n\n return {\n url: `http://127.0.0.1:${port}/call`,\n token,\n calls: () => calls,\n close,\n }\n\n async function handleRequest(\n request: IncomingMessage,\n response: ServerResponse,\n signal: AbortSignal,\n ): Promise<void> {\n try {\n if (request.method !== 'POST' || request.url !== '/call') {\n sendJsonIfOpen(response, 404, { error: 'not found' })\n return\n }\n if (request.headers.authorization !== `Bearer ${token}`) {\n sendJsonIfOpen(response, 401, { error: 'unauthorized' })\n return\n }\n if (calls >= args.maxCalls) {\n sendJsonIfOpen(response, 429, { error: 'trace tool call limit reached' })\n return\n }\n const body = await readJson(request, limits.maxRequestBytes)\n if (!isRecord(body) || typeof body.name !== 'string' || !('args' in body)) {\n sendJsonIfOpen(response, 400, { error: 'name and args are required' })\n return\n }\n const tool = byName.get(body.name)\n if (!tool) {\n sendJsonIfOpen(response, 404, { error: `unknown trace tool '${body.name}'` })\n return\n }\n calls += 1\n const result = await tool.handler(body.args, { signal })\n const encoded = JSON.stringify({ result })\n if (Buffer.byteLength(encoded) > limits.maxResponseBytes) {\n sendJsonIfOpen(response, 413, { error: 'trace tool response too large' })\n return\n }\n response.writeHead(200, {\n 'content-type': 'application/json; charset=utf-8',\n 'content-length': String(Buffer.byteLength(encoded)),\n })\n response.end(encoded)\n } catch (error) {\n sendJsonIfOpen(response, signal.aborted ? 499 : 400, {\n error: error instanceof Error ? error.message : String(error),\n })\n }\n }\n\n async function closeCallback(): Promise<void> {\n args.signal?.removeEventListener('abort', onAbort)\n accepting = false\n const closingServer = closeServer(server)\n server.closeIdleConnections?.()\n for (const controller of activeControllers) controller.abort()\n const [serverResult] = await Promise.allSettled([\n closingServer,\n waitForActiveHandlers(activeHandlers),\n ])\n if (activeControllers.size !== 0 || activeHandlers.size !== 0) {\n throw new Error('trace tool callback closed with active requests')\n }\n if (serverResult?.status === 'rejected') throw serverResult.reason\n }\n}\n\nfunction readJson(request: IncomingMessage, maxRequestBytes: number): Promise<unknown> {\n return new Promise((resolve, reject) => {\n let size = 0\n const chunks: Buffer[] = []\n request.on('data', (chunk: Buffer) => {\n size += chunk.length\n if (size > maxRequestBytes) {\n reject(new Error('trace tool request too large'))\n request.destroy()\n return\n }\n chunks.push(chunk)\n })\n request.on('error', reject)\n request.on('end', () => {\n try {\n resolve(JSON.parse(Buffer.concat(chunks).toString('utf8')))\n } catch (error) {\n reject(error)\n }\n })\n })\n}\n","import {\n assertExternalOptimizerModelBudget,\n type ExternalOptimizerModelCall,\n type ExternalOptimizerModelExecutionObservation,\n type ExternalOptimizerModelProxy,\n type ExternalOptimizerRunnerCommand,\n removeCredentialEnvironment,\n resolveExternalOptimizerCallbackLimits,\n resolveExternalOptimizerProcessLimits,\n} from '../campaign/external-optimizer-contracts'\nimport { startExternalOptimizerModelProxy } from '../campaign/external-optimizer-model-proxy'\nimport { runWithCleanup } from '../campaign/external-optimizer-resources'\nimport { runExternalOptimizerProcess } from '../campaign/external-optimizer-subprocess'\nimport type { CustomTokenPricing } from '../cost-ledger'\nimport { resolveModelPricing } from '../metrics'\nimport {\n DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS,\n type TraceAnalysisEngine,\n type TraceAnalysisEngineResult,\n} from './engine'\nimport { decodeRawFindingArray } from './finding-codec'\nimport { startTraceToolCallback, type TraceToolCallbackLimits } from './trace-tool-callback'\n\nimport { isRecord } from '../llm-client/json'\nconst DEFAULT_TIMEOUT_MS = 10 * 60_000\nconst DEFAULT_MAX_COST_USD = 1\nconst DEFAULT_MAX_MODEL_REQUEST_BYTES = 16 * 1024 * 1024\nconst DEFAULT_MAX_MODEL_RESPONSE_BYTES = 4 * 1024 * 1024\nconst DEFAULT_TRACE_TOOL_TIMEOUT_MS = 60_000\nconst MAX_TIMER_DELAY_MS = 2_147_483_647\nconst BRIDGE_MODULE = 'agent_eval_rpc.dspy_rlm_bridge'\n/** Bumped whenever this engine's execution behavior changes. */\nconst DSPY_RLM_ENGINE_VERSION = '1.0.0'\n\nexport interface DspyRlmTraceEngineOptions {\n /** Caller-owned execution path. Agent Eval never receives provider credentials. */\n call: ExternalOptimizerModelCall\n /** Stable public identity for the caller-owned path, such as an AgentProfile digest. */\n callRef: string\n /** Persist the finite execution record returned for every admitted call. */\n recordExecution: (observation: ExternalOptimizerModelExecutionObservation) => void\n model: string\n /** Exact provider rates. Required when the model is absent from the pricing table. */\n pricing?: CustomTokenPricing\n /** Maximum provider spend for one investigation. Default: 1 USD. */\n maxCostUsd?: number\n /** Controller response cap. Default: {@link DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS}. */\n maxOutputTokens?: number\n /**\n * Thinking tokens one controller turn may bill on top of its completion.\n * A reasoning model bills these beyond `maxOutputTokens`, so the cost\n * reservation must cover them. Default: four times the completion cap.\n */\n maxReasoningTokens?: number\n /** Maximum caller-owned model invocations. Default derives from the analysis limits. */\n maxModelRequests?: number\n /** Maximum model request bytes. Default: 16 MiB. */\n maxModelRequestBytes?: number\n /** Maximum model response bytes. Default: 4 MiB. */\n maxModelResponseBytes?: number\n /** Deadline for one caller-owned model invocation. Default: the whole analysis deadline. */\n modelRequestTimeoutMs?: number\n /** Trace-tool loopback request/response byte limits. */\n traceToolLimits?: Partial<TraceToolCallbackLimits>\n /** Deadline for one Python-to-Node trace-tool call. Default: 60 seconds. */\n traceToolTimeoutMs?: number\n /**\n * How the controller's reasoning and code fields are obtained.\n *\n * `tolerant` parses marker output strictly first, then recovers the fields\n * deterministically from prose plus a fenced code block — the shape coding\n * models naturally emit — at no extra model cost. `two-step` extracts with a\n * second call per turn. `chat` accepts marker output only. Default:\n * `tolerant`.\n */\n controlAdapter?: 'chat' | 'two-step' | 'tolerant'\n /** Python command used to load agent-eval-rpc[dspy]. Default: python. */\n runner?: ExternalOptimizerRunnerCommand\n /** Whole investigation deadline. Default: 10 minutes. */\n timeoutMs?: number\n}\n\n/** Use the official DSPy RLM as a bounded recursive trace-analysis engine. */\nexport function createDspyRlmTraceEngine(options: DspyRlmTraceEngineOptions): TraceAnalysisEngine {\n assertOptions(options)\n const maxOutputTokens = options.maxOutputTokens ?? DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS\n if (!Number.isSafeInteger(maxOutputTokens) || maxOutputTokens <= 0) {\n throw new TypeError('DSPy RLM maxOutputTokens must be a positive safe integer')\n }\n const controlAdapter = options.controlAdapter ?? 'tolerant'\n const maxReasoningTokens = options.maxReasoningTokens ?? maxOutputTokens * 4\n if (!Number.isSafeInteger(maxReasoningTokens) || maxReasoningTokens < 0) {\n throw new TypeError('DSPy RLM maxReasoningTokens must be a non-negative safe integer')\n }\n const timeoutMs = options.timeoutMs ?? DEFAULT_TIMEOUT_MS\n assertTimerDelay(timeoutMs, 'timeoutMs')\n const maxCostUsd = options.maxCostUsd ?? DEFAULT_MAX_COST_USD\n if (!Number.isFinite(maxCostUsd) || maxCostUsd <= 0) {\n throw new TypeError('DSPy RLM maxCostUsd must be positive and finite')\n }\n const pricing = options.pricing ?? pricingForModel(options.model)\n const maxModelRequestBytes = options.maxModelRequestBytes ?? DEFAULT_MAX_MODEL_REQUEST_BYTES\n const maxModelResponseBytes = options.maxModelResponseBytes ?? DEFAULT_MAX_MODEL_RESPONSE_BYTES\n const modelRequestTimeoutMs = options.modelRequestTimeoutMs ?? timeoutMs\n const traceToolTimeoutMs = options.traceToolTimeoutMs ?? DEFAULT_TRACE_TOOL_TIMEOUT_MS\n assertExternalOptimizerModelBudget(\n {\n maxCostUsd,\n maxRequests: options.maxModelRequests ?? 1,\n maxRequestBytes: maxModelRequestBytes,\n maxResponseBytes: maxModelResponseBytes,\n maxOutputTokensPerRequest: maxOutputTokens,\n maxReasoningTokensPerRequest: maxReasoningTokens,\n pricing,\n requestTimeoutMs: modelRequestTimeoutMs,\n },\n 'DSPy RLM model limits',\n )\n resolveExternalOptimizerCallbackLimits(options.traceToolLimits, 'DSPy RLM trace tool limits')\n assertTimerDelay(traceToolTimeoutMs, 'traceToolTimeoutMs')\n\n const runner = sanitizedRunner(options.runner)\n const processLimits = resolveExternalOptimizerProcessLimits(runner?.limits)\n return {\n id: 'dspy-rlm',\n description: 'Official DSPy RLM with bounded trace tools and metered model calls.',\n model: options.model,\n version: DSPY_RLM_ENGINE_VERSION,\n executionConfig: {\n bridge_module: BRIDGE_MODULE,\n call_ref: options.callRef,\n model: options.model,\n pricing: { ...pricing },\n max_cost_usd: maxCostUsd,\n max_output_tokens: maxOutputTokens,\n max_reasoning_tokens: maxReasoningTokens,\n control_adapter: controlAdapter,\n timeout_ms: timeoutMs,\n max_model_requests: options.maxModelRequests ?? null,\n max_request_bytes: maxModelRequestBytes,\n max_response_bytes: maxModelResponseBytes,\n model_request_timeout_ms: modelRequestTimeoutMs,\n trace_tool_limits: options.traceToolLimits ?? null,\n trace_tool_timeout_ms: traceToolTimeoutMs,\n process_limits: processLimits,\n runner: runner ? 'caller-supplied' : 'default',\n runner_command: runner?.command ?? null,\n },\n async analyze(request) {\n const callback = await startTraceToolCallback({\n tools: request.tools,\n maxCalls: request.limits.maxToolCalls,\n ...(options.traceToolLimits ? { limits: options.traceToolLimits } : {}),\n ...(request.signal ? { signal: request.signal } : {}),\n })\n let modelProxy: ExternalOptimizerModelProxy | undefined\n const modelExecutions: ExternalOptimizerModelExecutionObservation[] = []\n const result = await runWithCleanup({\n label: 'DSPy RLM trace-analysis resources',\n run: async () => {\n modelProxy = await startExternalOptimizerModelProxy({\n call: options.call,\n callRef: options.callRef,\n recordExecution: (observation) => {\n modelExecutions.push(structuredClone(observation))\n options.recordExecution(observation)\n },\n model: options.model,\n budget: {\n maxCostUsd,\n maxRequests: resolveMaxModelRequests(options.maxModelRequests, request.limits),\n maxRequestBytes: maxModelRequestBytes,\n maxResponseBytes: maxModelResponseBytes,\n maxOutputTokensPerRequest: maxOutputTokens,\n maxReasoningTokensPerRequest: maxReasoningTokens,\n requestTimeoutMs: modelRequestTimeoutMs,\n pricing,\n },\n costLedger: request.costLedger,\n channel: 'analyst',\n phase: request.costPhase,\n actor: request.analystId,\n ...(request.costTags ? { tags: request.costTags } : {}),\n ...(request.signal ? { signal: request.signal } : {}),\n })\n request.log?.('trace analyst engine started', {\n engine: 'dspy-rlm',\n model: options.model,\n tools: request.tools.map((tool) => tool.name),\n limits: request.limits,\n })\n const raw = await runExternalOptimizerProcess<unknown>({\n label: 'DSPy RLM trace analysis',\n tempPrefix: 'agent-eval-dspy-rlm-',\n module: BRIDGE_MODULE,\n input: {\n operation: 'analyze',\n question: request.question,\n instructions: request.instructions,\n modelProxy: {\n baseUrl: modelProxy.baseUrl,\n apiKey: modelProxy.apiKey,\n model: options.model,\n maxOutputTokens,\n },\n toolCallback: {\n url: callback.url,\n token: callback.token,\n timeoutMs: traceToolTimeoutMs,\n },\n toolSpecs: request.tools.map(({ name, description, parameters }) => ({\n name,\n description,\n parameters,\n })),\n controlAdapter,\n limits: {\n maxIterations: request.limits.maxIterations,\n maxLlmCalls: request.limits.maxLlmCalls,\n maxOutputChars: request.limits.maxOutputChars,\n },\n // Omitted entirely when the caller supplied none, so a request\n // without structured inputs sends the payload it always sent.\n ...(request.taskInputs ? { taskInputs: request.taskInputs } : {}),\n },\n ...(runner ? { runner } : {}),\n additionalArgs: ['--max-input-bytes', String(processLimits.maxInputBytes)],\n timeoutMs,\n ...(request.signal ? { signal: request.signal } : {}),\n })\n const parsed = parseBridgeOutput(raw, (index, reason) => {\n request.log?.('finding rejected: bridge row failed schema validation', {\n engine: 'dspy-rlm',\n index,\n reason,\n })\n })\n const successfulCompletions = modelProxy.successfulCompletions()\n const requestAttempts = modelProxy.requestAttempts()\n if (parsed.modelCalls !== successfulCompletions) {\n throw new Error(\n `DSPy RLM reported ${parsed.modelCalls} model calls, but the provider proxy recorded ${successfulCompletions}`,\n )\n }\n modelProxy.assertExecutionComplete()\n return {\n ...parsed,\n toolCalls: callback.calls(),\n runtime: {\n ...parsed.runtime,\n modelRequestAttempts: requestAttempts,\n modelSuccessfulCompletions: successfulCompletions,\n modelExecutions,\n },\n } satisfies TraceAnalysisEngineResult\n },\n cleanup: async () => {\n const results = await Promise.allSettled([\n ...(modelProxy ? [modelProxy.close()] : []),\n callback.close(),\n ])\n const errors = results.flatMap((entry) =>\n entry.status === 'rejected' ? [entry.reason] : [],\n )\n if (errors.length > 0) {\n throw new AggregateError(errors, 'DSPy RLM resource cleanup failed')\n }\n },\n })\n request.log?.('trace analyst engine completed', {\n engine: 'dspy-rlm',\n model_calls: result.modelCalls,\n model_request_attempts: result.runtime.modelRequestAttempts,\n tool_calls: result.toolCalls,\n findings: result.findings.length,\n })\n return result\n },\n }\n}\n\nfunction parseBridgeOutput(\n value: unknown,\n onRejectedFinding: (index: number, reason: string) => void,\n): Omit<TraceAnalysisEngineResult, 'toolCalls'> {\n if (!isRecord(value)) throw new Error('DSPy RLM bridge output must be an object')\n if (typeof value.answer !== 'string' || !value.answer.trim()) {\n throw new Error('DSPy RLM bridge returned no answer')\n }\n // Findings are model output: one malformed row is model noise, not a bridge\n // fault, and the rest of the paid investigation must survive it. The codec\n // is the same decoder the Python side runs, so a row this bridge accepts is\n // a row the optimizer could report, and vice versa.\n const decoded = decodeRawFindingArray(value.findings)\n if (decoded.topLevelError !== undefined) {\n throw new Error(`DSPy RLM bridge findings must be an array: ${decoded.topLevelError}`)\n }\n const findings = decoded.accepted\n const rejectedFindings = decoded.rejected.length\n for (const rejection of decoded.rejected) {\n onRejectedFinding(\n rejection.index,\n `${rejection.code}${rejection.path ? ` at ${rejection.path}` : ''}: ${rejection.message}`,\n )\n }\n if (!Array.isArray(value.trajectory)) {\n throw new Error('DSPy RLM bridge trajectory must be an array')\n }\n if (!Number.isSafeInteger(value.modelCalls) || (value.modelCalls as number) <= 0) {\n throw new Error('DSPy RLM bridge modelCalls must be a positive safe integer')\n }\n if (!isRecord(value.runtime)) {\n throw new Error('DSPy RLM bridge runtime must be an object')\n }\n return {\n answer: value.answer,\n findings,\n trajectory: value.trajectory,\n modelCalls: value.modelCalls as number,\n runtime: { ...value.runtime, rejectedFindings },\n }\n}\n\nfunction assertOptions(options: DspyRlmTraceEngineOptions): void {\n for (const [name, value] of [\n ['callRef', options.callRef],\n ['model', options.model],\n ] as const) {\n if (typeof value !== 'string' || !value.trim() || value !== value.trim()) {\n throw new TypeError(`DSPy RLM ${name} must be a trimmed non-empty string`)\n }\n }\n if (typeof options.call !== 'function') {\n throw new TypeError('DSPy RLM call must be a function')\n }\n if (typeof options.recordExecution !== 'function') {\n throw new TypeError('DSPy RLM recordExecution must be a function')\n }\n if (\n options.maxModelRequests !== undefined &&\n (!Number.isSafeInteger(options.maxModelRequests) || options.maxModelRequests <= 0)\n ) {\n throw new TypeError('DSPy RLM maxModelRequests must be a positive safe integer')\n }\n}\n\nfunction resolveMaxModelRequests(\n configured: number | undefined,\n limits: { maxIterations: number; maxLlmCalls: number },\n): number {\n if (configured !== undefined) return configured\n const derived = limits.maxIterations + limits.maxLlmCalls + 1\n if (!Number.isSafeInteger(derived) || derived <= 0) {\n throw new Error('DSPy RLM analysis limits produce an invalid model request limit')\n }\n return derived\n}\n\nfunction assertTimerDelay(value: number, field: string): void {\n if (!Number.isSafeInteger(value) || value <= 0 || value > MAX_TIMER_DELAY_MS) {\n throw new TypeError(`DSPy RLM ${field} must be between 1 and ${MAX_TIMER_DELAY_MS}`)\n }\n}\n\nfunction pricingForModel(model: string): CustomTokenPricing {\n const pricing = resolveModelPricing(model)\n if (!pricing) {\n throw new Error(\n `no pricing is configured for '${model}'; provide DspyRlmTraceEngineOptions.pricing`,\n )\n }\n return {\n inputUsdPerMillion: pricing.input * 1_000,\n outputUsdPerMillion: pricing.output * 1_000,\n }\n}\n\nfunction sanitizedRunner(\n runner: ExternalOptimizerRunnerCommand | undefined,\n): ExternalOptimizerRunnerCommand | undefined {\n if (!runner) return undefined\n return {\n ...(runner.command ? { command: runner.command } : {}),\n ...(runner.args ? { args: runner.args } : {}),\n ...(runner.env ? { env: removeCredentialEnvironment(runner.env) } : {}),\n ...(runner.limits ? { limits: { ...runner.limits } } : {}),\n }\n}\n\n"],"mappings":";;;;;;;;;;;;;;;AAmEA,SAAgB,sBAAsB,OAAqC;CACzE,MAAM,OAAO,YAAY,KAAK;CAC9B,IAAI,OAAO,SAAS,UAAU,OAAO;EAAE,UAAU,CAAC;EAAG,UAAU,CAAC;EAAG,eAAe;CAAK;CAEvF,MAAM,WAAgC,CAAC;CACvC,MAAM,WAAiC,CAAC;CACxC,KAAK,MAAM,CAAC,OAAO,QAAQ,KAAK,QAAQ,GAAG;EACzC,IAAI,SAAA,KAA2B;GAC7B,SAAS,KAAK;IACZ;IACA,MAAM;IACN,MAAM;IACN,SAAS;GACX,CAAC;GACD;EACF;EACA,MAAM,UAAU,UAAU,KAAK,KAAK;EACpC,IAAI,aAAa,SAAS,SAAS,KAAK,QAAQ,OAAO;OAClD,SAAS,KAAK,QAAQ,SAAS;CACtC;CACA,OAAO;EAAE;EAAU;CAAS;AAC9B;AAEA,SAAS,UACP,KACA,OACoE;CACpE,IAAI,QAAQ,QAAQ,OAAO,QAAQ,YAAY,MAAM,QAAQ,GAAG,GAC9D,OAAO,EACL,WAAW;EACT;EACA,MAAM;EACN,MAAM;EACN,SAAS,2CAA2C,SAAS,GAAG;CAClE,EACF;CAEF,MAAM,SAAS,wBAAwB,UAAU,GAAG;CACpD,IAAI,CAAC,OAAO,SAAS;EACnB,MAAM,QAAQ,OAAO,MAAM,OAAO;EAClC,MAAM,QAAQ,OAAO,QAAQ,CAAC,EAAA,CAAG,KAAK,GAAG;EAOzC,OAAO,EACL,WAAW;GAAE;GAAO;GAAM,MAJ1B,SAAS,aAAa,oBAAqB,IAA6B,OAAO,MAAM,OACjF,oBACA;GAE4B,SAAS,OAAO,WAAW;EAAgC,EAC7F;CACF;CACA,OAAO,EAAE,SAAS,OAAO,KAAK;AAChC;;;;;AAMA,SAAS,YAAY,OAAoC;CACvD,IAAI,MAAM,QAAQ,KAAK,GAAG,OAAO;CACjC,IAAI,OAAO,UAAU,UAAU;EAC7B,MAAM,SAAS,WAAW,KAAK;EAC/B,IAAI,WAAW,KAAA,GACb,OAAO;EAET,IAAI,MAAM,QAAQ,MAAM,GAAG,OAAO;EAClC,MAAM,YAAY,iBAAiB,MAAM;EACzC,IAAI,cAAc,KAAA,GAAW,OAAO;EACpC,OAAO,2CAA2C,SAAS,MAAM;CACnE;CACA,MAAM,YAAY,iBAAiB,KAAK;CACxC,IAAI,cAAc,KAAA,GAAW,OAAO;CACpC,OAAO,uCAAuC,SAAS,KAAK;AAC9D;;AAGA,SAAS,iBAAiB,OAAuC;CAC/D,IAAI,UAAU,QAAQ,OAAO,UAAU,YAAY,MAAM,QAAQ,KAAK,GAAG,OAAO,KAAA;CAChF,MAAM,QAAS,MAAkC;CACjD,OAAO,MAAM,QAAQ,KAAK,IAAI,QAAQ,KAAA;AACxC;AAEA,SAAS,SAAS,OAAwB;CACxC,IAAI,UAAU,MAAM,OAAO;CAC3B,IAAI,MAAM,QAAQ,KAAK,GAAG,OAAO;CACjC,OAAO,KAAK,OAAO;AACrB;;;;AChIA,eAAsB,uBAAuB,MAMd;CAC7B,IAAI,CAAC,OAAO,cAAc,KAAK,QAAQ,KAAK,KAAK,YAAY,GAC3D,MAAM,IAAI,UAAU,8DAA8D;CAEpF,MAAM,SAAS,uCAAuC,KAAK,QAAQ,4BAA4B;CAC/F,KAAK,QAAQ,eAAe;CAC5B,MAAM,SAAS,IAAI,IAAI,KAAK,MAAM,KAAK,SAAS,CAAC,KAAK,MAAM,IAAI,CAAC,CAAC;CAClE,IAAI,OAAO,SAAS,KAAK,MAAM,QAC7B,MAAM,IAAI,MAAM,mDAAmD;CAGrE,MAAM,QAAQ,YAAY,EAAE,CAAC,CAAC,SAAS,KAAK;CAC5C,IAAI,QAAQ;CACZ,IAAI,YAAY;CAChB,IAAI;CACJ,MAAM,oCAAoB,IAAI,IAAqB;CACnD,MAAM,iCAAiB,IAAI,IAAmB;CAC9C,MAAM,SAAS,cAAc,SAAS,aAAa;EACjD,IAAI,CAAC,WAAW;GACd,eAAe,UAAU,KAAK,EAAE,OAAO,iCAAiC,CAAC;GACzE;EACF;EACA,MAAM,aAAa,IAAI,gBAAgB;EACvC,MAAM,qBAA2B;GAC/B,QAAQ,QAAQ;GAChB,SAAS,QAAQ;EACnB;EACA,kBAAkB,IAAI,UAAU;EAChC,WAAW,OAAO,iBAAiB,SAAS,cAAc,EAAE,MAAM,KAAK,CAAC;EAExE,IAAI;EACJ,UAAU,cAAc,SAAS,UAAU,WAAW,MAAM,CAAC,CAAC,cAAc;GAC1E,WAAW,OAAO,oBAAoB,SAAS,YAAY;GAC3D,kBAAkB,OAAO,UAAU;GACnC,eAAe,OAAO,OAAO;EAC/B,CAAC;EACD,eAAe,IAAI,OAAO;EAC1B,QAAa,YAAY,KAAA,CAAS;CACpC,CAAC;CACD,MAAM,OAAO,MAAM,YAAY,MAAM;CACrC,MAAM,cAA6B;EACjC,iBAAiB,cAAc;EAC/B,OAAO;CACT;CACA,MAAM,gBAAsB;EAC1B,MAAW,CAAC,CAAC,YAAY,KAAA,CAAS;CACpC;CACA,KAAK,QAAQ,iBAAiB,SAAS,SAAS,EAAE,MAAM,KAAK,CAAC;CAC9D,IAAI,KAAK,QAAQ,SAAS,QAAQ;CAElC,OAAO;EACL,KAAK,oBAAoB,KAAK;EAC9B;EACA,aAAa;EACb;CACF;CAEA,eAAe,cACb,SACA,UACA,QACe;EACf,IAAI;GACF,IAAI,QAAQ,WAAW,UAAU,QAAQ,QAAQ,SAAS;IACxD,eAAe,UAAU,KAAK,EAAE,OAAO,YAAY,CAAC;IACpD;GACF;GACA,IAAI,QAAQ,QAAQ,kBAAkB,UAAU,SAAS;IACvD,eAAe,UAAU,KAAK,EAAE,OAAO,eAAe,CAAC;IACvD;GACF;GACA,IAAI,SAAS,KAAK,UAAU;IAC1B,eAAe,UAAU,KAAK,EAAE,OAAO,gCAAgC,CAAC;IACxE;GACF;GACA,MAAM,OAAO,MAAM,SAAS,SAAS,OAAO,eAAe;GAC3D,IAAI,CAAC,SAAS,IAAI,KAAK,OAAO,KAAK,SAAS,YAAY,EAAE,UAAU,OAAO;IACzE,eAAe,UAAU,KAAK,EAAE,OAAO,6BAA6B,CAAC;IACrE;GACF;GACA,MAAM,OAAO,OAAO,IAAI,KAAK,IAAI;GACjC,IAAI,CAAC,MAAM;IACT,eAAe,UAAU,KAAK,EAAE,OAAO,uBAAuB,KAAK,KAAK,GAAG,CAAC;IAC5E;GACF;GACA,SAAS;GACT,MAAM,SAAS,MAAM,KAAK,QAAQ,KAAK,MAAM,EAAE,OAAO,CAAC;GACvD,MAAM,UAAU,KAAK,UAAU,EAAE,OAAO,CAAC;GACzC,IAAI,OAAO,WAAW,OAAO,IAAI,OAAO,kBAAkB;IACxD,eAAe,UAAU,KAAK,EAAE,OAAO,gCAAgC,CAAC;IACxE;GACF;GACA,SAAS,UAAU,KAAK;IACtB,gBAAgB;IAChB,kBAAkB,OAAO,OAAO,WAAW,OAAO,CAAC;GACrD,CAAC;GACD,SAAS,IAAI,OAAO;EACtB,SAAS,OAAO;GACd,eAAe,UAAU,OAAO,UAAU,MAAM,KAAK,EACnD,OAAO,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK,EAC9D,CAAC;EACH;CACF;CAEA,eAAe,gBAA+B;EAC5C,KAAK,QAAQ,oBAAoB,SAAS,OAAO;EACjD,YAAY;EACZ,MAAM,gBAAgB,YAAY,MAAM;EACxC,OAAO,uBAAuB;EAC9B,KAAK,MAAM,cAAc,mBAAmB,WAAW,MAAM;EAC7D,MAAM,CAAC,gBAAgB,MAAM,QAAQ,WAAW,CAC9C,eACA,sBAAsB,cAAc,CACtC,CAAC;EACD,IAAI,kBAAkB,SAAS,KAAK,eAAe,SAAS,GAC1D,MAAM,IAAI,MAAM,iDAAiD;EAEnE,IAAI,cAAc,WAAW,YAAY,MAAM,aAAa;CAC9D;AACF;AAEA,SAAS,SAAS,SAA0B,iBAA2C;CACrF,OAAO,IAAI,SAAS,SAAS,WAAW;EACtC,IAAI,OAAO;EACX,MAAM,SAAmB,CAAC;EAC1B,QAAQ,GAAG,SAAS,UAAkB;GACpC,QAAQ,MAAM;GACd,IAAI,OAAO,iBAAiB;IAC1B,uBAAO,IAAI,MAAM,8BAA8B,CAAC;IAChD,QAAQ,QAAQ;IAChB;GACF;GACA,OAAO,KAAK,KAAK;EACnB,CAAC;EACD,QAAQ,GAAG,SAAS,MAAM;EAC1B,QAAQ,GAAG,aAAa;GACtB,IAAI;IACF,QAAQ,KAAK,MAAM,OAAO,OAAO,MAAM,CAAC,CAAC,SAAS,MAAM,CAAC,CAAC;GAC5D,SAAS,OAAO;IACd,OAAO,KAAK;GACd;EACF,CAAC;CACH,CAAC;AACH;;;ACtJA,MAAM,qBAAqB;AAC3B,MAAM,uBAAuB;AAC7B,MAAM,kCAAkC;AACxC,MAAM,mCAAmC;AACzC,MAAM,gCAAgC;AACtC,MAAM,qBAAqB;AAC3B,MAAM,gBAAgB;;AAEtB,MAAM,0BAA0B;;AAmDhC,SAAgB,yBAAyB,SAAyD;CAChG,cAAc,OAAO;CACrB,MAAM,kBAAkB,QAAQ,mBAAA;CAChC,IAAI,CAAC,OAAO,cAAc,eAAe,KAAK,mBAAmB,GAC/D,MAAM,IAAI,UAAU,0DAA0D;CAEhF,MAAM,iBAAiB,QAAQ,kBAAkB;CACjD,MAAM,qBAAqB,QAAQ,sBAAsB,kBAAkB;CAC3E,IAAI,CAAC,OAAO,cAAc,kBAAkB,KAAK,qBAAqB,GACpE,MAAM,IAAI,UAAU,iEAAiE;CAEvF,MAAM,YAAY,QAAQ,aAAa;CACvC,iBAAiB,WAAW,WAAW;CACvC,MAAM,aAAa,QAAQ,cAAc;CACzC,IAAI,CAAC,OAAO,SAAS,UAAU,KAAK,cAAc,GAChD,MAAM,IAAI,UAAU,iDAAiD;CAEvE,MAAM,UAAU,QAAQ,WAAW,gBAAgB,QAAQ,KAAK;CAChE,MAAM,uBAAuB,QAAQ,wBAAwB;CAC7D,MAAM,wBAAwB,QAAQ,yBAAyB;CAC/D,MAAM,wBAAwB,QAAQ,yBAAyB;CAC/D,MAAM,qBAAqB,QAAQ,sBAAsB;CACzD,mCACE;EACE;EACA,aAAa,QAAQ,oBAAoB;EACzC,iBAAiB;EACjB,kBAAkB;EAClB,2BAA2B;EAC3B,8BAA8B;EAC9B;EACA,kBAAkB;CACpB,GACA,uBACF;CACA,uCAAuC,QAAQ,iBAAiB,4BAA4B;CAC5F,iBAAiB,oBAAoB,oBAAoB;CAEzD,MAAM,SAAS,gBAAgB,QAAQ,MAAM;CAC7C,MAAM,gBAAgB,sCAAsC,QAAQ,MAAM;CAC1E,OAAO;EACL,IAAI;EACJ,aAAa;EACb,OAAO,QAAQ;EACf,SAAS;EACT,iBAAiB;GACf,eAAe;GACf,UAAU,QAAQ;GAClB,OAAO,QAAQ;GACf,SAAS,EAAE,GAAG,QAAQ;GACtB,cAAc;GACd,mBAAmB;GACnB,sBAAsB;GACtB,iBAAiB;GACjB,YAAY;GACZ,oBAAoB,QAAQ,oBAAoB;GAChD,mBAAmB;GACnB,oBAAoB;GACpB,0BAA0B;GAC1B,mBAAmB,QAAQ,mBAAmB;GAC9C,uBAAuB;GACvB,gBAAgB;GAChB,QAAQ,SAAS,oBAAoB;GACrC,gBAAgB,QAAQ,WAAW;EACrC;EACA,MAAM,QAAQ,SAAS;GACrB,MAAM,WAAW,MAAM,uBAAuB;IAC5C,OAAO,QAAQ;IACf,UAAU,QAAQ,OAAO;IACzB,GAAI,QAAQ,kBAAkB,EAAE,QAAQ,QAAQ,gBAAgB,IAAI,CAAC;IACrE,GAAI,QAAQ,SAAS,EAAE,QAAQ,QAAQ,OAAO,IAAI,CAAC;GACrD,CAAC;GACD,IAAI;GACJ,MAAM,kBAAgE,CAAC;GACvE,MAAM,SAAS,MAAM,eAAe;IAClC,OAAO;IACP,KAAK,YAAY;KACf,aAAa,MAAM,iCAAiC;MAClD,MAAM,QAAQ;MACd,SAAS,QAAQ;MACjB,kBAAkB,gBAAgB;OAChC,gBAAgB,KAAK,gBAAgB,WAAW,CAAC;OACjD,QAAQ,gBAAgB,WAAW;MACrC;MACA,OAAO,QAAQ;MACf,QAAQ;OACN;OACA,aAAa,wBAAwB,QAAQ,kBAAkB,QAAQ,MAAM;OAC7E,iBAAiB;OACjB,kBAAkB;OAClB,2BAA2B;OAC3B,8BAA8B;OAC9B,kBAAkB;OAClB;MACF;MACA,YAAY,QAAQ;MACpB,SAAS;MACT,OAAO,QAAQ;MACf,OAAO,QAAQ;MACf,GAAI,QAAQ,WAAW,EAAE,MAAM,QAAQ,SAAS,IAAI,CAAC;MACrD,GAAI,QAAQ,SAAS,EAAE,QAAQ,QAAQ,OAAO,IAAI,CAAC;KACrD,CAAC;KACD,QAAQ,MAAM,gCAAgC;MAC5C,QAAQ;MACR,OAAO,QAAQ;MACf,OAAO,QAAQ,MAAM,KAAK,SAAS,KAAK,IAAI;MAC5C,QAAQ,QAAQ;KAClB,CAAC;KAwCD,MAAM,SAAS,kBAAkB,MAvCf,4BAAqC;MACrD,OAAO;MACP,YAAY;MACZ,QAAQ;MACR,OAAO;OACL,WAAW;OACX,UAAU,QAAQ;OAClB,cAAc,QAAQ;OACtB,YAAY;QACV,SAAS,WAAW;QACpB,QAAQ,WAAW;QACnB,OAAO,QAAQ;QACf;OACF;OACA,cAAc;QACZ,KAAK,SAAS;QACd,OAAO,SAAS;QAChB,WAAW;OACb;OACA,WAAW,QAAQ,MAAM,KAAK,EAAE,MAAM,aAAa,kBAAkB;QACnE;QACA;QACA;OACF,EAAE;OACF;OACA,QAAQ;QACN,eAAe,QAAQ,OAAO;QAC9B,aAAa,QAAQ,OAAO;QAC5B,gBAAgB,QAAQ,OAAO;OACjC;OAGA,GAAI,QAAQ,aAAa,EAAE,YAAY,QAAQ,WAAW,IAAI,CAAC;MACjE;MACA,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC;MAC3B,gBAAgB,CAAC,qBAAqB,OAAO,cAAc,aAAa,CAAC;MACzE;MACA,GAAI,QAAQ,SAAS,EAAE,QAAQ,QAAQ,OAAO,IAAI,CAAC;KACrD,CAAC,IACsC,OAAO,WAAW;MACvD,QAAQ,MAAM,yDAAyD;OACrE,QAAQ;OACR;OACA;MACF,CAAC;KACH,CAAC;KACD,MAAM,wBAAwB,WAAW,sBAAsB;KAC/D,MAAM,kBAAkB,WAAW,gBAAgB;KACnD,IAAI,OAAO,eAAe,uBACxB,MAAM,IAAI,MACR,qBAAqB,OAAO,WAAW,gDAAgD,uBACzF;KAEF,WAAW,wBAAwB;KACnC,OAAO;MACL,GAAG;MACH,WAAW,SAAS,MAAM;MAC1B,SAAS;OACP,GAAG,OAAO;OACV,sBAAsB;OACtB,4BAA4B;OAC5B;MACF;KACF;IACF;IACA,SAAS,YAAY;KAKnB,MAAM,UAAS,MAJO,QAAQ,WAAW,CACvC,GAAI,aAAa,CAAC,WAAW,MAAM,CAAC,IAAI,CAAC,GACzC,SAAS,MAAM,CACjB,CAAC,EAAA,CACsB,SAAS,UAC9B,MAAM,WAAW,aAAa,CAAC,MAAM,MAAM,IAAI,CAAC,CAClD;KACA,IAAI,OAAO,SAAS,GAClB,MAAM,IAAI,eAAe,QAAQ,kCAAkC;IAEvE;GACF,CAAC;GACD,QAAQ,MAAM,kCAAkC;IAC9C,QAAQ;IACR,aAAa,OAAO;IACpB,wBAAwB,OAAO,QAAQ;IACvC,YAAY,OAAO;IACnB,UAAU,OAAO,SAAS;GAC5B,CAAC;GACD,OAAO;EACT;CACF;AACF;AAEA,SAAS,kBACP,OACA,mBAC8C;CAC9C,IAAI,CAAC,SAAS,KAAK,GAAG,MAAM,IAAI,MAAM,0CAA0C;CAChF,IAAI,OAAO,MAAM,WAAW,YAAY,CAAC,MAAM,OAAO,KAAK,GACzD,MAAM,IAAI,MAAM,oCAAoC;CAMtD,MAAM,UAAU,sBAAsB,MAAM,QAAQ;CACpD,IAAI,QAAQ,kBAAkB,KAAA,GAC5B,MAAM,IAAI,MAAM,8CAA8C,QAAQ,eAAe;CAEvF,MAAM,WAAW,QAAQ;CACzB,MAAM,mBAAmB,QAAQ,SAAS;CAC1C,KAAK,MAAM,aAAa,QAAQ,UAC9B,kBACE,UAAU,OACV,GAAG,UAAU,OAAO,UAAU,OAAO,OAAO,UAAU,SAAS,GAAG,IAAI,UAAU,SAClF;CAEF,IAAI,CAAC,MAAM,QAAQ,MAAM,UAAU,GACjC,MAAM,IAAI,MAAM,6CAA6C;CAE/D,IAAI,CAAC,OAAO,cAAc,MAAM,UAAU,KAAM,MAAM,cAAyB,GAC7E,MAAM,IAAI,MAAM,4DAA4D;CAE9E,IAAI,CAAC,SAAS,MAAM,OAAO,GACzB,MAAM,IAAI,MAAM,2CAA2C;CAE7D,OAAO;EACL,QAAQ,MAAM;EACd;EACA,YAAY,MAAM;EAClB,YAAY,MAAM;EAClB,SAAS;GAAE,GAAG,MAAM;GAAS;EAAiB;CAChD;AACF;AAEA,SAAS,cAAc,SAA0C;CAC/D,KAAK,MAAM,CAAC,MAAM,UAAU,CAC1B,CAAC,WAAW,QAAQ,OAAO,GAC3B,CAAC,SAAS,QAAQ,KAAK,CACzB,GACE,IAAI,OAAO,UAAU,YAAY,CAAC,MAAM,KAAK,KAAK,UAAU,MAAM,KAAK,GACrE,MAAM,IAAI,UAAU,YAAY,KAAK,oCAAoC;CAG7E,IAAI,OAAO,QAAQ,SAAS,YAC1B,MAAM,IAAI,UAAU,kCAAkC;CAExD,IAAI,OAAO,QAAQ,oBAAoB,YACrC,MAAM,IAAI,UAAU,6CAA6C;CAEnE,IACE,QAAQ,qBAAqB,KAAA,MAC5B,CAAC,OAAO,cAAc,QAAQ,gBAAgB,KAAK,QAAQ,oBAAoB,IAEhF,MAAM,IAAI,UAAU,2DAA2D;AAEnF;AAEA,SAAS,wBACP,YACA,QACQ;CACR,IAAI,eAAe,KAAA,GAAW,OAAO;CACrC,MAAM,UAAU,OAAO,gBAAgB,OAAO,cAAc;CAC5D,IAAI,CAAC,OAAO,cAAc,OAAO,KAAK,WAAW,GAC/C,MAAM,IAAI,MAAM,iEAAiE;CAEnF,OAAO;AACT;AAEA,SAAS,iBAAiB,OAAe,OAAqB;CAC5D,IAAI,CAAC,OAAO,cAAc,KAAK,KAAK,SAAS,KAAK,QAAQ,oBACxD,MAAM,IAAI,UAAU,YAAY,MAAM,yBAAyB,oBAAoB;AAEvF;AAEA,SAAS,gBAAgB,OAAmC;CAC1D,MAAM,UAAU,oBAAoB,KAAK;CACzC,IAAI,CAAC,SACH,MAAM,IAAI,MACR,iCAAiC,MAAM,6CACzC;CAEF,OAAO;EACL,oBAAoB,QAAQ,QAAQ;EACpC,qBAAqB,QAAQ,SAAS;CACxC;AACF;AAEA,SAAS,gBACP,QAC4C;CAC5C,IAAI,CAAC,QAAQ,OAAO,KAAA;CACpB,OAAO;EACL,GAAI,OAAO,UAAU,EAAE,SAAS,OAAO,QAAQ,IAAI,CAAC;EACpD,GAAI,OAAO,OAAO,EAAE,MAAM,OAAO,KAAK,IAAI,CAAC;EAC3C,GAAI,OAAO,MAAM,EAAE,KAAK,4BAA4B,OAAO,GAAG,EAAE,IAAI,CAAC;EACrE,GAAI,OAAO,SAAS,EAAE,QAAQ,EAAE,GAAG,OAAO,OAAO,EAAE,IAAI,CAAC;CAC1D;AACF"}
|
|
@@ -0,0 +1,393 @@
|
|
|
1
|
+
import { c as verifyAgentProfileCell, i as buildAgentProfileCell, l as hashJson } from "./agent-profile-cell-B1V1gUZs.js";
|
|
2
|
+
import { c as validateRunRecord } from "./run-record-DhEJgoPz.js";
|
|
3
|
+
import { t as researchReport } from "./summary-report-BsXJoYX5.js";
|
|
4
|
+
import { t as FileSystemRawProviderSink } from "./raw-provider-sink-75w4FWBj.js";
|
|
5
|
+
import { t as TraceEmitter } from "./emitter-DbMXUIVP.js";
|
|
6
|
+
import { n as assertRunCaptured, t as RunIntegrityError } from "./integrity-W4CADcFc.js";
|
|
7
|
+
//#region src/eval-campaign.ts
|
|
8
|
+
/**
|
|
9
|
+
* EvalCampaign — opinionated matrix runner that wires the four
|
|
10
|
+
* capture-integrity directives by construction.
|
|
11
|
+
*
|
|
12
|
+
* The canonical benchmark shape — matrix runner → for each
|
|
13
|
+
* (variant, scenario, seed) → start a TraceEmitter → call LLMs → end the
|
|
14
|
+
* run → analyze — has a bug class at the integration boundary: raw
|
|
15
|
+
* events not captured, route silently wrong, integrity not asserted,
|
|
16
|
+
* analyst never run. The directives in `SKILL.md § Capture integrity`
|
|
17
|
+
* are the mitigations.
|
|
18
|
+
*
|
|
19
|
+
* `EvalCampaign` is the structural fix — consumers don't wire the
|
|
20
|
+
* integrity surface themselves; the campaign owns it. Specifically:
|
|
21
|
+
*
|
|
22
|
+
* - constructs a per-run `TraceStore` and `RawProviderSink` via factories
|
|
23
|
+
* - builds the run's `ChatClient` through `chatFactory`, handing it the
|
|
24
|
+
* run's raw sink and trace context — a transport built any other way has
|
|
25
|
+
* no raw HTTP envelope, and `assertRunCaptured` says so
|
|
26
|
+
* - calls `assertRunCaptured` after every `endRun` and routes failures
|
|
27
|
+
* through a configurable policy (`throw` / `mark_failed` / `log`)
|
|
28
|
+
* - assembles per-run `RunRecord`s and runs `researchReport` at the end
|
|
29
|
+
* so the campaign artifact is launch-decision-grade by default
|
|
30
|
+
* - embeds the campaign fingerprint (a SHA-256 over the canonicalised
|
|
31
|
+
* run set) and optional `preregistrationHash` in the report
|
|
32
|
+
*
|
|
33
|
+
* The runner contract is intentionally narrow: produce a `CampaignRunOutcome`
|
|
34
|
+
* given a fully-wired `CampaignRunContext`. Everything orchestration-shaped
|
|
35
|
+
* lives in the campaign. This is the inversion-of-control point — consumers
|
|
36
|
+
* stop writing matrix runners and start writing scenario-runners.
|
|
37
|
+
*
|
|
38
|
+
* Out of scope for v1 (tracked in `docs/research-report-methodology.md`):
|
|
39
|
+
*
|
|
40
|
+
* - Distributed/cluster execution (concurrency is local async)
|
|
41
|
+
* - Adaptive sampling / sequential interim looks
|
|
42
|
+
* - Resume from partial state across crashes
|
|
43
|
+
* - LLM-call retry beyond what the caller's transport already does
|
|
44
|
+
*/
|
|
45
|
+
const DEFAULT_INTEGRITY = {
|
|
46
|
+
llmSpansMin: 1,
|
|
47
|
+
requireRawCoverageOfLlmSpans: true,
|
|
48
|
+
requireOutcome: true
|
|
49
|
+
};
|
|
50
|
+
async function runEvalCampaign(opts) {
|
|
51
|
+
assertCampaignPreflight(opts);
|
|
52
|
+
const settings = resolveCampaignSettings(opts);
|
|
53
|
+
const campaignFingerprint = await fingerprintCampaign(opts, settings);
|
|
54
|
+
const startedAt = new Date(settings.now()).toISOString();
|
|
55
|
+
const pool = {
|
|
56
|
+
opts,
|
|
57
|
+
settings,
|
|
58
|
+
cells: planCampaignCells(opts, settings.seeds),
|
|
59
|
+
cursor: 0,
|
|
60
|
+
aborting: false,
|
|
61
|
+
genuineErrors: [],
|
|
62
|
+
openRuns: /* @__PURE__ */ new Map(),
|
|
63
|
+
runs: [],
|
|
64
|
+
integrityReports: [],
|
|
65
|
+
failedRuns: []
|
|
66
|
+
};
|
|
67
|
+
await drainCampaignPool(pool);
|
|
68
|
+
const report = await buildCampaignReport(opts, settings, pool.runs);
|
|
69
|
+
const endedAt = new Date(settings.now()).toISOString();
|
|
70
|
+
return {
|
|
71
|
+
campaignId: opts.campaignId,
|
|
72
|
+
campaignFingerprint,
|
|
73
|
+
preregistrationHash: settings.preregistrationHash,
|
|
74
|
+
runs: pool.runs,
|
|
75
|
+
integrityReports: pool.integrityReports,
|
|
76
|
+
failedRuns: pool.failedRuns,
|
|
77
|
+
report,
|
|
78
|
+
startedAt,
|
|
79
|
+
endedAt
|
|
80
|
+
};
|
|
81
|
+
}
|
|
82
|
+
/** Reject structurally invalid campaigns before any work starts. */
|
|
83
|
+
function assertCampaignPreflight(opts) {
|
|
84
|
+
if (opts.variants.length === 0) throw new Error("runEvalCampaign: variants must be non-empty.");
|
|
85
|
+
if (opts.scenarios.length === 0) throw new Error("runEvalCampaign: scenarios must be non-empty.");
|
|
86
|
+
const variantIds = /* @__PURE__ */ new Set();
|
|
87
|
+
for (const v of opts.variants) {
|
|
88
|
+
if (variantIds.has(v.id)) throw new Error(`runEvalCampaign: duplicate variant id "${v.id}".`);
|
|
89
|
+
variantIds.add(v.id);
|
|
90
|
+
}
|
|
91
|
+
const scenarioIds = /* @__PURE__ */ new Set();
|
|
92
|
+
for (const s of opts.scenarios) {
|
|
93
|
+
if (scenarioIds.has(s.scenarioId)) throw new Error(`runEvalCampaign: duplicate scenarioId "${s.scenarioId}".`);
|
|
94
|
+
scenarioIds.add(s.scenarioId);
|
|
95
|
+
}
|
|
96
|
+
if (opts.report?.comparator && !variantIds.has(opts.report.comparator)) throw new Error(`runEvalCampaign: report.comparator "${opts.report.comparator}" is not a configured variantId.`);
|
|
97
|
+
if (!opts.commitSha) throw new Error("runEvalCampaign: commitSha is required (every RunRecord needs it).");
|
|
98
|
+
}
|
|
99
|
+
/** Apply defaults to the optional campaign options. */
|
|
100
|
+
function resolveCampaignSettings(opts) {
|
|
101
|
+
return {
|
|
102
|
+
seeds: opts.seeds ?? [
|
|
103
|
+
0,
|
|
104
|
+
1,
|
|
105
|
+
2
|
|
106
|
+
],
|
|
107
|
+
splitTag: opts.splitTag ?? "holdout",
|
|
108
|
+
concurrency: Math.max(1, opts.concurrency ?? 1),
|
|
109
|
+
integrity: {
|
|
110
|
+
...DEFAULT_INTEGRITY,
|
|
111
|
+
...opts.integrity ?? {}
|
|
112
|
+
},
|
|
113
|
+
onIntegrityFailure: opts.onIntegrityFailure ?? "mark_failed",
|
|
114
|
+
now: opts.now ?? (() => Date.now()),
|
|
115
|
+
executionRef: opts.executionRef ?? null,
|
|
116
|
+
preregistrationHash: opts.preregistrationHash ?? null,
|
|
117
|
+
rawSinkFactory: opts.rawSinkFactory ?? defaultRawSinkFactory(opts.workDir)
|
|
118
|
+
};
|
|
119
|
+
}
|
|
120
|
+
/** SHA-256 over the canonicalised matrix definition. */
|
|
121
|
+
function fingerprintCampaign(opts, settings) {
|
|
122
|
+
return hashJson({
|
|
123
|
+
campaignId: opts.campaignId,
|
|
124
|
+
variants: opts.variants.map((v) => v.id).sort(),
|
|
125
|
+
scenarios: opts.scenarios.map((s) => s.scenarioId).sort(),
|
|
126
|
+
seeds: [...settings.seeds].sort((a, b) => a - b),
|
|
127
|
+
splitTag: settings.splitTag,
|
|
128
|
+
comparator: opts.report?.comparator ?? null,
|
|
129
|
+
executionRef: settings.executionRef,
|
|
130
|
+
preregistrationHash: settings.preregistrationHash
|
|
131
|
+
});
|
|
132
|
+
}
|
|
133
|
+
/** Expand variants x scenarios x seeds, variant-major. */
|
|
134
|
+
function planCampaignCells(opts, seeds) {
|
|
135
|
+
const cells = [];
|
|
136
|
+
for (const variant of opts.variants) for (const scenario of opts.scenarios) for (const seed of seeds) cells.push({
|
|
137
|
+
variant,
|
|
138
|
+
scenario,
|
|
139
|
+
seed
|
|
140
|
+
});
|
|
141
|
+
return cells;
|
|
142
|
+
}
|
|
143
|
+
/**
|
|
144
|
+
* Run the bounded-concurrency worker pool to completion, finalize any run left
|
|
145
|
+
* open, and re-throw the genuine errors the workers captured.
|
|
146
|
+
*/
|
|
147
|
+
async function drainCampaignPool(pool) {
|
|
148
|
+
const workers = Array.from({ length: Math.min(pool.settings.concurrency, pool.cells.length) }, () => campaignWorker(pool));
|
|
149
|
+
await Promise.allSettled(workers);
|
|
150
|
+
for (const [runId, emitter] of pool.openRuns) await finalizeAbort(emitter, runId, "campaign aborted: genuine error in a sibling run");
|
|
151
|
+
pool.openRuns.clear();
|
|
152
|
+
if (pool.genuineErrors.length > 0) throw pool.genuineErrors.length === 1 ? pool.genuineErrors[0] : new AggregateError(pool.genuineErrors, `runEvalCampaign: ${pool.genuineErrors.length} runs failed with genuine (non-run-level) errors`);
|
|
153
|
+
}
|
|
154
|
+
/** Pull cells off the shared cursor until none remain or the pool is aborting. */
|
|
155
|
+
async function campaignWorker(pool) {
|
|
156
|
+
while (!pool.aborting) {
|
|
157
|
+
const i = pool.cursor++;
|
|
158
|
+
if (i >= pool.cells.length) return;
|
|
159
|
+
const cell = pool.cells[i];
|
|
160
|
+
try {
|
|
161
|
+
const result = await runCampaignCell(pool, cell);
|
|
162
|
+
pool.runs.push(result.record);
|
|
163
|
+
pool.integrityReports.push(result.integrity);
|
|
164
|
+
} catch (err) {
|
|
165
|
+
if (err instanceof CellExecutionError) {
|
|
166
|
+
pool.failedRuns.push(err.failed);
|
|
167
|
+
if (err.integrity) pool.integrityReports.push(err.integrity);
|
|
168
|
+
} else {
|
|
169
|
+
pool.genuineErrors.push(err);
|
|
170
|
+
pool.aborting = true;
|
|
171
|
+
return;
|
|
172
|
+
}
|
|
173
|
+
}
|
|
174
|
+
}
|
|
175
|
+
}
|
|
176
|
+
/** Run one matrix cell: runner, integrity check, then the validated RunRecord. */
|
|
177
|
+
async function runCampaignCell(pool, cell) {
|
|
178
|
+
const { opts, settings } = pool;
|
|
179
|
+
const runId = (opts.runId ?? defaultRunId)({
|
|
180
|
+
campaignId: opts.campaignId,
|
|
181
|
+
runId: "",
|
|
182
|
+
variantId: cell.variant.id,
|
|
183
|
+
scenarioId: cell.scenario.scenarioId,
|
|
184
|
+
seed: cell.seed
|
|
185
|
+
});
|
|
186
|
+
const factoryParams = {
|
|
187
|
+
campaignId: opts.campaignId,
|
|
188
|
+
runId,
|
|
189
|
+
variantId: cell.variant.id,
|
|
190
|
+
scenarioId: cell.scenario.scenarioId,
|
|
191
|
+
seed: cell.seed
|
|
192
|
+
};
|
|
193
|
+
const store = opts.storeFactory(factoryParams);
|
|
194
|
+
const rawSink = settings.rawSinkFactory(factoryParams);
|
|
195
|
+
const emitter = new TraceEmitter(store, {
|
|
196
|
+
runId,
|
|
197
|
+
now: opts.now,
|
|
198
|
+
onRunComplete: opts.onRunComplete
|
|
199
|
+
});
|
|
200
|
+
pool.openRuns.set(runId, emitter);
|
|
201
|
+
const chat = opts.chatFactory({
|
|
202
|
+
rawSink,
|
|
203
|
+
runId
|
|
204
|
+
});
|
|
205
|
+
const ctx = {
|
|
206
|
+
runId,
|
|
207
|
+
experimentId: opts.campaignId,
|
|
208
|
+
variant: cell.variant.payload,
|
|
209
|
+
variantId: cell.variant.id,
|
|
210
|
+
scenarioId: cell.scenario.scenarioId,
|
|
211
|
+
scenarioTags: cell.scenario.tags ?? {},
|
|
212
|
+
seed: cell.seed,
|
|
213
|
+
splitTag: settings.splitTag,
|
|
214
|
+
emitter,
|
|
215
|
+
store,
|
|
216
|
+
rawSink,
|
|
217
|
+
chat
|
|
218
|
+
};
|
|
219
|
+
try {
|
|
220
|
+
const wallStart = settings.now();
|
|
221
|
+
const outcome = await invokeCampaignRunner(opts, cell, ctx);
|
|
222
|
+
const wallMs = settings.now() - wallStart;
|
|
223
|
+
const integrityReport = await assertRunCaptured(store, runId, {
|
|
224
|
+
...settings.integrity,
|
|
225
|
+
rawSink
|
|
226
|
+
});
|
|
227
|
+
if (!integrityReport.ok) admitOrRejectIntegrity(settings, cell, runId, integrityReport);
|
|
228
|
+
const record = buildCampaignRunRecord(opts, settings, cell, runId, outcome, wallMs);
|
|
229
|
+
await stampAgentProfile(opts, cell, record, runId, outcome);
|
|
230
|
+
return {
|
|
231
|
+
record: validateRunRecord(record),
|
|
232
|
+
integrity: integrityReport
|
|
233
|
+
};
|
|
234
|
+
} finally {
|
|
235
|
+
pool.openRuns.delete(runId);
|
|
236
|
+
}
|
|
237
|
+
}
|
|
238
|
+
/** Invoke the runner; a throw aborts the run and becomes a run-level failure. */
|
|
239
|
+
async function invokeCampaignRunner(opts, cell, ctx) {
|
|
240
|
+
try {
|
|
241
|
+
return await opts.runner(ctx);
|
|
242
|
+
} catch (err) {
|
|
243
|
+
const message = err instanceof Error ? err.message : String(err);
|
|
244
|
+
await finalizeAbort(ctx.emitter, ctx.runId, message);
|
|
245
|
+
throw new CellExecutionError({
|
|
246
|
+
runId: ctx.runId,
|
|
247
|
+
variantId: cell.variant.id,
|
|
248
|
+
scenarioId: cell.scenario.scenarioId,
|
|
249
|
+
seed: cell.seed,
|
|
250
|
+
reason: "runner_threw",
|
|
251
|
+
error: message
|
|
252
|
+
});
|
|
253
|
+
}
|
|
254
|
+
}
|
|
255
|
+
/** Apply the integrity-failure policy to a report that is not ok. */
|
|
256
|
+
function admitOrRejectIntegrity(settings, cell, runId, integrityReport) {
|
|
257
|
+
switch (settings.onIntegrityFailure) {
|
|
258
|
+
case "throw": throw new RunIntegrityError(integrityReport);
|
|
259
|
+
case "mark_failed": throw new CellExecutionError({
|
|
260
|
+
runId,
|
|
261
|
+
variantId: cell.variant.id,
|
|
262
|
+
scenarioId: cell.scenario.scenarioId,
|
|
263
|
+
seed: cell.seed,
|
|
264
|
+
reason: "integrity_failed",
|
|
265
|
+
error: integrityReport.issues.map((i) => i.code).join(", ")
|
|
266
|
+
}, integrityReport);
|
|
267
|
+
}
|
|
268
|
+
}
|
|
269
|
+
/** Assemble the RunRecord for a successful run, before profile stamping. */
|
|
270
|
+
function buildCampaignRunRecord(opts, settings, cell, runId, outcome, wallMs) {
|
|
271
|
+
const recordOutcome = { raw: outcome.raw ?? {} };
|
|
272
|
+
if (settings.splitTag === "holdout") recordOutcome.holdoutScore = outcome.score;
|
|
273
|
+
else recordOutcome.searchScore = outcome.score;
|
|
274
|
+
if (outcome.judgeScores !== void 0) recordOutcome.judgeScores = outcome.judgeScores;
|
|
275
|
+
return {
|
|
276
|
+
runId,
|
|
277
|
+
experimentId: opts.campaignId,
|
|
278
|
+
candidateId: cell.variant.id,
|
|
279
|
+
seed: cell.seed,
|
|
280
|
+
model: outcome.model,
|
|
281
|
+
promptHash: outcome.promptHash,
|
|
282
|
+
configHash: outcome.configHash,
|
|
283
|
+
commitSha: opts.commitSha,
|
|
284
|
+
wallMs,
|
|
285
|
+
costUsd: outcome.costUsd,
|
|
286
|
+
costProvenance: outcome.costProvenance,
|
|
287
|
+
tokenUsage: outcome.tokenUsage,
|
|
288
|
+
terminalOutcome: "succeeded",
|
|
289
|
+
judgeMetadata: outcome.judgeMetadata,
|
|
290
|
+
outcome: recordOutcome,
|
|
291
|
+
...outcome.failureClass ? { failureClass: outcome.failureClass } : {},
|
|
292
|
+
failureMode: outcome.failureMode,
|
|
293
|
+
splitTag: settings.splitTag,
|
|
294
|
+
scenarioId: cell.scenario.scenarioId
|
|
295
|
+
};
|
|
296
|
+
}
|
|
297
|
+
/** Resolve the run's agent profile (outcome first, then options) and stamp it on the record. */
|
|
298
|
+
async function stampAgentProfile(opts, cell, record, runId, outcome) {
|
|
299
|
+
const profileSource = outcome.agentProfile ?? (typeof opts.agentProfile === "function" ? await opts.agentProfile({
|
|
300
|
+
campaignId: opts.campaignId,
|
|
301
|
+
runId,
|
|
302
|
+
variantId: cell.variant.id,
|
|
303
|
+
scenarioId: cell.scenario.scenarioId,
|
|
304
|
+
seed: cell.seed,
|
|
305
|
+
variant: cell.variant.payload,
|
|
306
|
+
scenarioTags: cell.scenario.tags ?? {}
|
|
307
|
+
}) : opts.agentProfile);
|
|
308
|
+
if (profileSource !== void 0) {
|
|
309
|
+
const agentProfile = await resolveAgentProfileCell(profileSource);
|
|
310
|
+
assertAgentProfileMatchesRun(agentProfile, outcome.model, outcome.promptHash);
|
|
311
|
+
record.agentProfile = agentProfile;
|
|
312
|
+
}
|
|
313
|
+
}
|
|
314
|
+
/** Build the optional research report over the successful runs. */
|
|
315
|
+
async function buildCampaignReport(opts, settings, runs) {
|
|
316
|
+
if (!opts.report) return void 0;
|
|
317
|
+
const reportOpts = {
|
|
318
|
+
...opts.report,
|
|
319
|
+
comparator: opts.report.comparator,
|
|
320
|
+
split: settings.splitTag === "dev" ? "search" : settings.splitTag,
|
|
321
|
+
generatedAt: new Date(settings.now()).toISOString(),
|
|
322
|
+
preregistrationHash: settings.preregistrationHash ?? void 0
|
|
323
|
+
};
|
|
324
|
+
return researchReport(runs, reportOpts);
|
|
325
|
+
}
|
|
326
|
+
var CellExecutionError = class extends Error {
|
|
327
|
+
failed;
|
|
328
|
+
integrity;
|
|
329
|
+
constructor(failed, integrity) {
|
|
330
|
+
super(`cell ${failed.variantId}/${failed.scenarioId}@${failed.seed} failed: ${failed.reason}`);
|
|
331
|
+
this.failed = failed;
|
|
332
|
+
this.integrity = integrity;
|
|
333
|
+
}
|
|
334
|
+
};
|
|
335
|
+
/**
|
|
336
|
+
* Abort a run whose owning work threw or was orphaned by a sibling's genuine
|
|
337
|
+
* error, finalizing the emitter so hooks fire and the store records a terminal
|
|
338
|
+
* status. Safe to call unconditionally: it only aborts runs that are still
|
|
339
|
+
* `running`. Two no-op cases are intentional and benign:
|
|
340
|
+
*
|
|
341
|
+
* - the run was never started (absent from the store) — nothing to finalize
|
|
342
|
+
* - the run is already terminal (`completed` / `failed` / `aborted`) —
|
|
343
|
+
* finalizing again would overwrite the real outcome (e.g. flip a passed run
|
|
344
|
+
* to `aborted` with `{ pass: false, notes: reason }`), so we leave it alone
|
|
345
|
+
*
|
|
346
|
+
* Any OTHER failure of the store read or the abort write (disk full, FS fault,
|
|
347
|
+
* backend down) is a genuine diagnostic and propagates rather than being
|
|
348
|
+
* swallowed.
|
|
349
|
+
*/
|
|
350
|
+
async function finalizeAbort(emitter, runId, reason) {
|
|
351
|
+
const existing = await emitter.traceStore.getRun(runId);
|
|
352
|
+
if (existing === void 0) return;
|
|
353
|
+
if (existing.status !== "running") return;
|
|
354
|
+
const droppedBefore = emitter.captureStats().dropped;
|
|
355
|
+
await emitter.abortRun(reason);
|
|
356
|
+
const after = emitter.captureStats();
|
|
357
|
+
if (after.dropped > droppedBefore) throw new Error(`finalizeAbort: run ${runId} could not be finalized: ${after.lastError}`);
|
|
358
|
+
}
|
|
359
|
+
function defaultRawSinkFactory(workDir) {
|
|
360
|
+
return (params) => {
|
|
361
|
+
if (!workDir) throw new Error("runEvalCampaign: rawSinkFactory not supplied and workDir not set. Pass either to enable raw provider capture, or pass `new NoopRawProviderSink()` via rawSinkFactory to opt out explicitly.");
|
|
362
|
+
return new FileSystemRawProviderSink({ dir: `${workDir}/raw-events/${params.runId}` });
|
|
363
|
+
};
|
|
364
|
+
}
|
|
365
|
+
async function resolveAgentProfileCell(input) {
|
|
366
|
+
if (isAgentProfileCell(input)) {
|
|
367
|
+
if (!await verifyAgentProfileCell(input)) throw new Error(`runEvalCampaign: agentProfile.cellId does not match its content`);
|
|
368
|
+
return input;
|
|
369
|
+
}
|
|
370
|
+
return buildAgentProfileCell(input);
|
|
371
|
+
}
|
|
372
|
+
function isAgentProfileCell(input) {
|
|
373
|
+
return "schemaVersion" in input && "cellId" in input;
|
|
374
|
+
}
|
|
375
|
+
function assertAgentProfileMatchesRun(profile, model, promptHash) {
|
|
376
|
+
if (profile.model !== void 0 && profile.model !== model) throw new Error(`runEvalCampaign: agentProfile.model "${profile.model}" does not match outcome.model "${model}"`);
|
|
377
|
+
if (profile.promptHash !== void 0 && profile.promptHash !== promptHash) throw new Error(`runEvalCampaign: agentProfile.promptHash "${profile.promptHash}" does not match outcome.promptHash "${promptHash}"`);
|
|
378
|
+
}
|
|
379
|
+
function defaultRunId(params) {
|
|
380
|
+
const base = `${params.campaignId}::${params.variantId}::${params.scenarioId}::${params.seed}`;
|
|
381
|
+
let h1 = 2166136261;
|
|
382
|
+
let h2 = 305419896;
|
|
383
|
+
for (let i = 0; i < base.length; i++) {
|
|
384
|
+
const c = base.charCodeAt(i);
|
|
385
|
+
h1 = Math.imul(h1 ^ c, 16777619) >>> 0;
|
|
386
|
+
h2 = Math.imul(h2 ^ c, 2654435761) >>> 0;
|
|
387
|
+
}
|
|
388
|
+
return `run-${h1.toString(16).padStart(8, "0")}${h2.toString(16).padStart(8, "0")}`;
|
|
389
|
+
}
|
|
390
|
+
//#endregion
|
|
391
|
+
export { runEvalCampaign as t };
|
|
392
|
+
|
|
393
|
+
//# sourceMappingURL=eval-campaign-BUMIxKTE.js.map
|