@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
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import { a as RunRecord } from "./run-record-CaxbLq6Y.js";
|
|
2
2
|
import { a as PairedPromotionDecision, i as PairedMcNemarEvidence, l as PairedBootstrapOptions, r as PairedDecisionStatistic, t as PairedDecisionMethod, u as PairedBootstrapResult } from "./paired-promotion-decision-CQMJ5xGN.js";
|
|
3
3
|
import { r as LedgerHash } from "./canonical-CFpojCN5.js";
|
|
4
|
-
import { R as Scenario, a as CampaignResult, b as GenerationRecord, h as GateContext, j as MutableSurface, p as Gate, v as GateResult, w as JudgeScore } from "./types-
|
|
4
|
+
import { R as Scenario, a as CampaignResult, b as GenerationRecord, h as GateContext, j as MutableSurface, p as Gate, v as GateResult, w as JudgeScore } from "./types-DtkO1G9L.js";
|
|
5
5
|
import { z } from "zod";
|
|
6
6
|
//#region src/statistics/paired-binary.d.ts
|
|
7
7
|
/** A binomial proportion estimate with a confidence interval. */
|
|
@@ -1251,4 +1251,4 @@ declare function dimensionRegressions(candidate: Map<string, Record<string, Judg
|
|
|
1251
1251
|
}): DimensionRegression[];
|
|
1252
1252
|
//#endregion
|
|
1253
1253
|
export { ParetoResult as $, PairRunRecordsResult as A, EProcessStep as At, powerPreflight as B, pairedRiskDifferenceScore as Bt, signManifest as C, FinalEvidenceOutcome as Ct, MatchedRunRecordPair as D, EProcess as Dt, MatchedPair as E, openFinalEvidenceLedger as Et, comparePairedArms as F, RiskDifferenceResult as Ft, ObjectiveSource as G, AxisVerdict as H, wilson as Ht, pairArms as I, ScoreRiskDifferenceResult as It, PromotionPolicy as J, ParetoSignificanceGateOptions as K, pairRunRecords as L, mcnemar as Lt, PairedArmsComparison as M, ExactRiskDifferenceResult as Mt, PairedCorrectness as N, McNemarResult as Nt, PairArmsOptions as O, EProcessOptions as Ot, PairedMetricDelta as P, ProportionInterval as Pt, Objective as Q, PowerPreflight as R, pairedRiskDifference as Rt, manifestContentDigest as S, FinalEvidenceMeasurement as St, ComparePairedArmsOptions as T, FinalEvidenceReservation as Tt, BuildEvidenceVectorOptions as U, AxisEvidence as V, passAtK as Vt, EvidenceVector as W, paretoPolicy as X, buildEvidenceVector as Y, paretoSignificanceGate as Z, HypothesisManifest as _, attest as _t, dimensionRegressions as a, EvidenceAuthority as at, SignedManifestAlgo as b, FinalEvidenceError as bt, SequentialDecideFn as c, EvidenceReceipt as ct, SequentialObservation as d, isIndependentEvidence as dt, dominates as et, SequentialPairedGate as f, verifyEvidenceReceipt as ft, sequentialPairedGate as g, AttestedReport as gt, sequentialDecide as h, AttestationVerification as ht, PairedHoldout as i, CreateEvidenceReceiptInput as it, PairedArmRow as j, eProcess as jt, PairArmsResult as k, EProcessState as kt, SequentialDecideOptions as l, EvidenceReceiptVerification as lt, SequentialStreamState as m, AttestationProvenance as mt, HeldoutSignificance as n, CampaignEvidenceContext as nt, heldoutSignificance as o, EvidenceAuthorityKind as ot, SequentialPairedGateOptions as p, ATTESTATION_ALGORITHM as pt, PromotionObjective as q, HeldoutSignificanceOptions as r, createCampaignEvidenceReceipt as rt, pairHoldout as s, EvidenceBinding as st, DimensionRegression as t, paretoFrontier as tt, SequentialGateDecision as u, createEvidenceReceipt as ut, HypothesisResult as v, verifyAttestation as vt, verifyManifest as w, FinalEvidenceRecord as wt, hashJson as x, FinalEvidenceLedger as xt, SignedManifest as y, FinalEvidenceConflictError as yt, PowerPreflightOptions as z, pairedRiskDifferenceExact as zt };
|
|
1254
|
-
//# sourceMappingURL=statistical-heldout-
|
|
1254
|
+
//# sourceMappingURL=statistical-heldout-BHfLZHrm.d.ts.map
|
package/dist/{statistical-heldout-C9AbgREN.d.ts.map → statistical-heldout-BHfLZHrm.d.ts.map}
RENAMED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"statistical-heldout-
|
|
1
|
+
{"version":3,"file":"statistical-heldout-BHfLZHrm.d.ts","names":["_$__evidence_receipt0.EvidenceReceipt"],"sources":["../src/statistics/paired-binary.ts","../src/statistics/sequential-eprocess.ts","../src/experiment/final-evidence.ts","../src/attestation.ts","../src/experiment/evidence-receipt.ts","../src/experiment/campaign-evidence.ts","../src/pareto.ts","../src/campaign/gates/promotion-policy.ts","../src/campaign/gates/power-preflight.ts","../src/paired-arms.ts","../src/pre-registration.ts","../src/campaign/gates/sequential.ts","../src/campaign/gates/statistical-heldout.ts"],"mappings":";;;;;;;UAgBiB;;EAEf;;EAEA;;EAEA;;;;;;;;;;;;iBAac,OAAO,mBAAmB,WAAW,sBAAoB;;UA2DxD;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;;;;iBAec,QACd,SAAS,6BACT,WAAW,8BACV;;UAmBc;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;;;;;;;;iBAmBc,qBACd,SAAS,6BACT,WAAW,6BACX,sBACC;;;;UAkCc;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAgCc,0BACd,SAAS,6BACT,WAAW,6BACX,sBACC;;;UAyEc;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA8Dc,0BACd,SAAS,6BACT,WAAW,6BACX,sBACC;;;;;;;;;;;iBA6Ga,QAAQ,WAAW,WAAW;;;UChhB7B;;;EAGf;;;;EAIA;;;;EAIA;;;;;;;;EAQA,SAAS;;UAGM;;EAEf;;EAEA;;EAEA;;;;;;;;;;UAWe,sBAAsB;EACrC;EACA;EACA;;EAEA;;EAEA;;EAEA;;;EAGA;;UAGe;;;EAGf,OAAO,YAAY;EACnB,SAAS;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA8BK,SAAS,OAAM,kBAAuB;;;cCnFhD,mBAAiB,EAAA;;;;;;GAQZ,EAAA,KAAA;;KAGC,2BAA2B,EAAE,aAAa;cAEhD,mBAAiB,EAAA;;;GAKZ,EAAA,KAAA;KAEC,2BAA2B,EAAE,aAAa;UAgCrC;EACf,aAAa;EACb,iBAAiB;EACjB;IAAY,aAAa;IAA0B,WAAW;;;KAGpD,qBAAqB;EAC3B;EAAiB,OAAO;;EACxB;EAAkB;IAAS;IAA8C;;;UAE9D;EACf,QACE,OAAO,2BACN,QAAQ;IAAuB,QAAQ;IAAqB;;EAC/D,OACE,mBACA,aAAa,2BACZ,QAAQ;IAAuB,QAAQ;IAAqB;;EAC/D,QAAQ,QAAQ,qBAAqB;;;cAI1B,2BAA2B;WAE3B;EADX,YACW,8CACT;;;cAQS,mCAAmC;EAC9C,YAAY;;;iBA+HE,wBAAwB;EAAW;IAAiB;;;;;;;;;;;;;;;;;;;;;;;cCvMvD;UAEI;;EAEf,eAAe;;EAEf;;;EAGA;;EAEA;;EAEA;;;EAGA;;UAGe;;EAEf;EACA,YAAY;EACZ,kBAAkB;;EAElB;;UAGe;EACf;;EAEA;;;;;;;;iBAiBc,OAAO,iBAAiB,YAAY,wBAAwB;;;;;;;;;;iBAoB5D,kBACd,iBACA,UAAU,iBACT;;;cC1EG;;cAGA;KAQM,gCAAgC;UAS3B;WACN,MAAM;;WAEN;;UAGM;WACN,sBAAsB;;WAEtB;;WAEA;;WAEA;;WAEA;;WAEA;;WAEA;;WAEA;;WAEA;WACA,WAAW;;WAEX;;WAEA;;UAGM;WACN,SAAS;WACT,aAAa;;UAGP;WACN;WACA;;UAGM,mCAAmC,KAAK;;;;;;;iBAQzC,sBACd,OAAO,4BACP,YAAY,wBACX;;;;;iBA4Ba,sBAAsB,SAAS,kBAAkB;;;;;;iBAkCjD,sBAAsB,SAAS;;;;KCjJnC,0BAA0B,KACpC;EAOI,YAAY,KAAK;;;iBAGP,8BAA8B,UAAU,UAAU,GAAG;EACnE,UAAU,eAAe,GAAG;EAC5B,SAAS;EACT,SAAS;IAAAA;;;;;;;;;;;;;;;;;;KCPC;UAEK,UAAU;;EAEzB;EACA,WAAW;EACX,QAAQ,WAAW;;UAGJ,aAAa;EAC5B,UAAU;EACV,WAAW;;EAEX,cAAc;IAAQ,WAAW;IAAG,WAAW;;;;iBAIjC,UAAU,GAAG,GAAG,GAAG,GAAG,GAAG,YAAY,UAAU;;;;;;iBAmB/C,eAAe,GAAG,YAAY,KAAK,YAAY,UAAU,OAAO,aAAa;;;;;KCxBjF;EAAoB;;EAAwB;EAAmB;;UAE1D;;EAEf;EACA,QAAQ;;;EAGR,WAAW;;;;EAIX;;;;EAIA;;;;;EAKA;;;;KAKU;UAEK;EACf;EACA,QAAQ;EACR,WAAW;;;;;;;;EAQX,WAAW;;;;;EAKX;;;EAGA;IAAM;IAAa;;;EAEnB,mBAAmB;;EAEnB,SAAS;;;EAGT;;EAEA;EACA;EACA,gBAAgB;EAChB;EACA;EACA,SAAS;;UAGM;;EAEf,MAAM;;;EAGN;;;EAGA;IAAQ;IAAmB;;;;;;KAMjB,mBAAmB,IAAI,mBAAmB;UAErC;;;;EAIf;;EAEA;;EAEA;;EAEA;;;EAGA;;;;;;;;iBASc,oBAAoB,WAAW,kBAAkB,UAC/D,KAAK,YAAY,WAAW,YAC5B,YAAY,sBACZ,OAAM,6BACL;;;;;;cAkIU,cAAc;UA+EV,sCAAsC;;EAErD,YAAY;;;EAGZ,SAAS;;EAET;;;;;;;iBAQc,uBAAuB,qBAAqB,kBAAkB,WAAW,UACvF,SAAS,gCACR,KAAK,WAAW;;;;;;;;;;;;;;;;;UCzVF;;EAEf;;;EAGA;;EAEA;;EAEA;;;;EAIA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA;;;EAGA;EACA;EACA;;EAEA;;EAEA;;;iBAYc,eAAe,MAAM,wBAAwB;;;;;UCzB5C;;;EAGf;;;;;EAKA;;EAEA;;EAEA;;EAEA,UAAU;;UAGK;;EAEf;;EAEA;;;UAIe;EACf;;;;EAIA;EACA,UAAU;EACV,WAAW;;UAGI;;EAEf,OAAO;;;EAGP,kBAAkB;;EAElB,mBAAmB;;;;;;;;;;;;;;;;;;;;iBAqBL,SAAS,eAAe,gBAAgB,MAAM,kBAAkB;;UA8G/D;;EAEf;;EAEA;;EAEA,SAAS;;EAET,gBAAgB;;;UAID;EACf;;EAEA;;EAEA;;EAEA;;EAEA;;;;EAIA,aAAa;;EAEb;IAAY;IAAW;;;UAGR,iCAAiC;;;;EAIhD;;EAEA,YAAY;;UAGG;EACf;EACA;EACA;;;EAGA,aAAa;EACb,cAAc;;;;;;;;;;;;;;;iBAgBA,kBACd,eAAe,gBACf,MAAM,2BACL;UAmEc;EACf;EACA;EACA,UAAU;EACV,WAAW;;UAGI;EACf,OAAO;EACP,kBAAkB;EAClB,mBAAmB;;;;;;;;;iBAeL,eACd,uBAAuB,aACvB,wBAAwB,cACvB;;;;;;;;;;;;;;;;UCrWc;EACf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;EACA;;;KAIU;UAEK,uBAAuB;;EAEtC;;EAEA,MAAM;;UAGS;EACf,UAAU;EACV;EACA;EACA;;;EAGA;;EAEA,kBAAkB;EAGlB;;;;;;;;;;;;;;;;;iBAkBoB,SAAS,GAAG,KAAK,IAAI;;;;;iBAQ3B,sBAAsB,UAAU;;;;;;iBAc1B,aAAa,GAAG,qBAAqB,QAAQ;;;;;iBAS7C,eAAe,GAAG,iBAAiB;;;KCtE7C;UAEK;EACf,UAAU;;EAEV;;EAEA;;;EAGA;;UAGe;;;EAGf;;;;EAIA;;;EAGA;;EAEA;;;;EAIA;;;EAGA;;;;EAIA,8BAA8B;;;EAG9B,kBAAkB;;EAElB;;;;;;;EAOA,SAAS;;;;;;KAOC,wBAAwB;EAAkB,UAAU;;UAE/C,qBAAqB,qBAAqB,kBAAkB,WAAW,kBAC9E,KAAK,WAAW;;;;;;;;;;EAUxB,QAAQ,gBAAgB;;;EAGxB,SAAS;;;;;;;;;;;;;;;;;iBAuPK,qBAAqB,qBAAqB,kBAAkB,WAAW,UACrF,SAAS,8BACR,qBAAqB,WAAW;UA6GlB;;EAEf;;EAEA;;EAEA;;EAEA;;UAGe;GACd;IAAQ,SAAS;;IAAyB;IAAe;;;EAE1D,SAAS;;;;;;;;;;;;;;;;;;;;;;;;;iBA0BK,iBAAiB,UAAS,0BAA+B;;;UC9dxD;;EAEf;;EAEA;;EAEA;;;;;;;;;;;;;iBA4Ec,YACd,WAAW,YAAY,eAAe,cACtC,UAAU,YAAY,eAAe,cACrC,aAAa,aACb,SAAS,GAAG,oCACX;UAkEc;EACf,QAAQ;;;;;;;;;;;;EAYR,WAAW;;;;EAIX,iBAAiB;;;;;;;EAOjB,UAAU;;EAEV,mBAAmB;;EAEnB,SAAS;;;;EAIT;;EAEA;;EAEA;EACA;;EAEA;;EAEA;;EAEA,gBAAgB;;EAEhB;;;;EAIA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;;EAEA;EACA;;EAEA,8BAA8B;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA6BhB,oBACd,QAAQ,eACR,OAAM,6BACL;UA2Ec;EACf;;;EAGA,WAAW;;;;EAIX;;EAEA;IAAM;IAAa;;;EAEnB,mBAAmB;;EAEnB,SAAS;;EAET;;;;EAIA;EACA;EACA;EACA;EACA;;EAEA;EACA;;EAEA;;EAEA;;;;;;;;;;iBAkBc,qBACd,WAAW,YAAY,eAAe,cACtC,UAAU,YAAY,eAAe,cACrC,aAAa,aACb,8BACA;EACE;EACA;EACA;EACA;;;EAGA;EACA,8BAA8B;EAC9B;IAED"}
|
|
@@ -227,6 +227,21 @@ declare class HeldOutGate {
|
|
|
227
227
|
* Pairing is by `(experimentId, scenarioId, seed)`.
|
|
228
228
|
* Missing or duplicate identities throw instead of comparing by position. */
|
|
229
229
|
evaluate(candidate: RunRecord[], baseline: RunRecord[]): HeldOutGateDecision;
|
|
230
|
+
/** Refusals decided before any interval is computed: absent split scores,
|
|
231
|
+
* incomplete coverage, too few paired runs. */
|
|
232
|
+
private refuseOnMissingEvidence;
|
|
233
|
+
/** The paired holdout delta, decided by the one shared rule (see
|
|
234
|
+
* `paired-promotion-decision.ts`): the estimator comes from the outcome's
|
|
235
|
+
* shape, McNemar vetoes at a non-negative threshold, and a zero-width
|
|
236
|
+
* interval is refused. `medianPairedDelta` stays the literal median on
|
|
237
|
+
* EVERY path so the number that made the old gate blind remains visible. */
|
|
238
|
+
private decidePairedDelta;
|
|
239
|
+
/** Refusals read off the paired delta: a zero-width interval, too few pairs
|
|
240
|
+
* for the chosen estimator, or an interval that does not clear the threshold. */
|
|
241
|
+
private refuseOnDelta;
|
|
242
|
+
/** The overfit-gap and cost-ceiling refusals, run after the quality gates so
|
|
243
|
+
* a cost-driven rejection never masks a quality failure. */
|
|
244
|
+
private refuseOnOverfitOrCost;
|
|
230
245
|
/**
|
|
231
246
|
* Which paired statistic decides this data, and the shape facts behind it —
|
|
232
247
|
* for the early exits, where no interval is computed. Delegates to
|
|
@@ -523,4 +538,4 @@ interface ResearchReport {
|
|
|
523
538
|
declare function researchReport(runs: RunRecord[], opts?: ResearchReportOptions): Promise<ResearchReport>;
|
|
524
539
|
//#endregion
|
|
525
540
|
export { HeldOutGateRejectionCode as C, HeldOutGateDecision as S, gainHistogram as _, ResearchReportOptions as a, HeldOutGate as b, RESEARCH_REPORT_HARD_PAIR_FLOOR as c, GainDistributionOptions as d, ParetoFigureSpec as f, SummaryTableRow as g, SummaryTableOptions as h, ResearchReportMethodology as i, GainDistributionBin as l, SummaryTable as m, ResearchReportCandidate as n, ResearchReportRecommendation as o, ParetoPoint as p, ResearchReportDecision as r, researchReport as s, ResearchReport as t, GainDistributionFigureSpec as u, summaryTable as v, SplitCoverage as w, HeldOutGateConfig as x, GateEvidence as y };
|
|
526
|
-
//# sourceMappingURL=summary-report-
|
|
541
|
+
//# sourceMappingURL=summary-report-CkbRcSuu.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"summary-report-CkbRcSuu.d.ts","names":[],"sources":["../src/held-out-gate.ts","../src/summary-report-figures.ts","../src/research-report-analysis.ts","../src/research-report.ts"],"mappings":";;;;;KAuIY;;;;;;;;UAiBK;;EAEf;;;EAGA;;;;EAIA;;EAEA;;EAEA;;;EAGA;;UAGe;;;;EAIf;;;;;EAKA;;;;;;;;;;;EAWA;;;;EAIA;;;EAGA;;EAEA;;EAEA;;;;EAIA;;;;;;;;;;;;;;;;;EAiBA;;;;;;;;;;;;;;EAcA;;;;KAKG,iBAAiB;UAEL;;EAEf;;EAEA;;EAEA;;;;;;;;EAQA,iBAAiB;;;;;EAKjB,gBAAgB;;;;;EAKhB;;;;;;;;;EASA,gBAAgB;;;EAGhB;;;EAGA;IAAY;IAAa;;;;EAGzB;;;;;;;EAOA;IAAW;IAAW;IAAW;IAAqB;;;;;;EAKtD;;;;;EAKA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;EAKA;;;EAGA;;;;;;;EAOA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA,UAAU;;EAEV;;EAEA,eAAe;;;;;;;cAQJ;mBACM;mBACA;mBACA;mBACA;mBACA;mBACA;mBACA;mBACA;mBACA;mBACA;EAEjB,YAAY,QAAQ;;;;EAoCpB,SAAS,WAAW,aAAa,UAAU,cAAc;;;UAiCjD;;;;;;UAoEA;;;UAiCA;;;UAkEA;;;;;;;UAqDA;;;;UCxnBO;;;EAGf;;EAEA;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;EACf,MAAM;EACN;EACA;;EAEA;;;;;;;iBAQc,aAAa,MAAM,aAAa,OAAM,sBAA2B;UAkHhE;EACf;;EAEA;;EAEA;;EAEA;;;EAGA;;;EAGA;;UAGe;EACf;EACA;EACA,QAAQ;EACR;IAAQ;IAAc;;;UAoEP;;EAEf;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;EACA,MAAM;EACN;EACA;IAAM;IAAa;;;UAGJ;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;iBAQc,cACd,MAAM,aACN,qBACA,oBACA,OAAM,0BACL;;;;;;;;cCpSU;;;KCWD;UAOK;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;EASA;;;;;;;EAOA;IAAS;IAAa;;;;;;EAKtB;;;;;EAKA;;EAEA,gBAAgB,eAAe;;EAE/B,kBAAkB;;EAElB;;EAEA;;EAEA;;;;;;EAMA;;UAGe;EACf,UAAU;EACV;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;IAAU;IAAa;;;;;;;EAMvB;EACA;;;;;;;EAOA;EACA;EACA,OAAO;EACP,UAAU;EACV;;UAGe;;;;;EAKf;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;EACA;;;;;;EAMA;EACA;EACA;IAAQ;IAAa;;EACrB;EACA,gBAAgB;EAChB,YAAY;EACZ,SAAS;EACT;IACE,QAAQ;IACR,OAAO;;EAET,aAAa;EACb,kBAAkB;EAClB;EACA;;;;;;;;;;;;;;;;;;;;;iBAsBoB,eACpB,MAAM,aACN,OAAM,wBACL,QAAQ"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/supervisor-run/analyze.ts","../../src/supervisor-run/claude-code-reader.ts","../../src/supervisor-run/integrity-types.ts","../../src/supervisor-run/integrity.ts","../../src/supervisor-run/patch-stats.ts","../../src/supervisor-run/reader.ts","../../src/supervisor-run/render.ts","../../src/supervisor-run/report-command.ts","../../src/supervisor-run/source-facts.ts","../../src/supervisor-run/rollout-nodes.ts","../../src/supervisor-run/rollup.ts","../../src/supervisor-run/runtime-reader.ts","../../src/supervisor-run/terminal-record.ts"],"mappings":";;;;;;;
|
|
1
|
+
{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/supervisor-run/analyze.ts","../../src/supervisor-run/claude-code-reader.ts","../../src/supervisor-run/integrity-types.ts","../../src/supervisor-run/integrity.ts","../../src/supervisor-run/patch-stats.ts","../../src/supervisor-run/reader.ts","../../src/supervisor-run/render.ts","../../src/supervisor-run/report-command.ts","../../src/supervisor-run/source-facts.ts","../../src/supervisor-run/rollout-nodes.ts","../../src/supervisor-run/rollup.ts","../../src/supervisor-run/runtime-reader.ts","../../src/supervisor-run/terminal-record.ts"],"mappings":";;;;;;;wBAuCgB,4BACd,KAAK,sBACL,qBACC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;UCgCc;;WAEN;;;;;WAKA;WACA;WACA;WACA;WACA;WACA;WACA;;;;qBCrFE;KAED;KAEA;KAKA;UAqCK;WACN;WACA;;UAGM;WACN,MAAM;WACN;WACA,UAAU;WACV;WACA;WACA;WACA,mBAAmB;WACnB;WACA,WAAW,SAAS;;UAGd;WACN,eAAe;WACf;WACA;WACA,MAAM;WACN,iBAAiB;;UAGX;;WAEN;;;;;;;;wBC9BK,8BACd,OAAO,uBAAuB,mBAC9B,UAAS,gCACR;;;;wBC5Ca,WAAW,eAAe;;;;;;;;wBCgBpB,qBACpB,gBAAgB,sBAAsB,uBACrC,QAAQ;UAYM;;WAEN;;WAEA;;;;;;WAMA;;;;;;;;wBAeW,sBAAsB,eAAe;;;;;;;wBCtC3C,4BAA4B,GAAG;wBA8E/B,4BAA4B,GAAG;wBAsL/B,+BACd,QAAQ,qBACR;;;;;;;;;;;;;KC1QU;;UAQK;WACN,SAAS;WACT,SAAS;;;;UCIV;EACR;EACA;EACA;EACA;;EAEA;;;;;;EAMA;;UAGQ;WACC,YAAY,kBAAkB;EACvC,QAAQ;;;;;;EAMR;EACA;;;;;;EAMA;;UA6Ge;;EAEf;EACA;EACA;EACA;;EAEA,MAAM;;EAEN;;EAEA;EACA;;EAEA;EACA;;UAGe;EACf;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;EACA;EACA,OAAO;;EAEP;;UAGe;;EAEf;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;;EAEA;EACA;EACA;;EAEA;;EAEA;EACA;;EAEA;;EAEA;EACA;EACA;EACA;EACA,eAAe;EACf,uBAAuB;;UAGR;;EAEf;EACA;;UAGe;;EAEf;;EAEA;EACA;;;;;;UAOe;WACN,2BAA2B;WAC3B;WACA,iBAAiB;WACjB,iBAAiB;WACjB,uBAAuB;WACvB,uBAAuB;WACvB;;;;;;;IAOP;IACA;IACA;IACA;;IAEA;;IAEA;;IAEA;IACA;;IAEA;;IAEA;IACA;;IAEA;;WAEO,YAAY,oBAAoB;;WAEhC,wBAAwB;;WAExB;;;;;;;;;WASA;;WAEA;;;;;;WAMA,0BAA0B,SAAS;;;;;;;WAOnC;;WAEA,gBAAgB;;WAEhB,OAAO;WACP;WACA;;;;;;WAMA;WACA;;;;;;;;;;;;;;;;;;KAmDC;wBAkcI,oBAAoB,KAAK,uBAAuB;;;UClxB/C;;WAEN;;WAEA,QAAQ;;WAER;;WAEA;;WAEA;;WAEA;;WAEA;;WAEA;;WAEA;WACA;WACA;WACA;WACA;;WAEA;;;;;;;wBAwBK,0BACd,KAAK,sBACL,OAAM,2BACL;;;;;;;;wBCzFa,qBAAqB,kBAAkB,wBAAwB;;;UCwrB9D;;;;;;WAMN;;;;;;;;;;;;;;;;;;;wBA0EW,yBACpB,gBACA,OAAM,uBACL,QAAQ;;;;;;wBA0FW,0BAA0B,iBAAiB;;;;qBC11BpD;;qBAIA;UAEI;;WAEN,OAAO;;WAEP,QAAQ;;;;;WAKR,SAAS;;UAGH;WACN,WAAW;WACX,iBAAiB,SAAS;;;;;WAK1B,WAAW;;WAEX,SAAS,SAAS;;;;;WAKlB;;wBAmBK,mBAAmB,OAAO,sBAAsB"}
|
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import { t as roundDecimal } from "../round-decimal-COrKglMh.js";
|
|
2
|
-
import { n as supervisorRunRolloutLines, r as SUPERVISOR_RUN_INTEGRITY_SCHEMA, t as analyzeSupervisorRunIntegrity } from "../integrity-
|
|
3
|
-
import { c as parseSupervisorTree, f as SUPERVISOR_RUN_ROLLUP_SCHEMA, g as unavailable, h as showMeasured, m as isUnavailable, n as RUNTIME_FAILED_STATUS, r as readTerminalRecord, t as NO_TERMINAL_RECORD } from "../terminal-record-
|
|
4
|
-
import { a as findSupervisorRunDirs, c as analyzeSupervisorRunSources, i as analyzeSupervisorRun, l as parsePatch, n as renderSupervisorRunHeadline, o as isRuntimeSupervisorRunDir, r as renderSupervisorRunMarkdown, s as readRuntimeSupervisorRun, t as renderSupervisorRollupMarkdown } from "../render-
|
|
2
|
+
import { n as supervisorRunRolloutLines, r as SUPERVISOR_RUN_INTEGRITY_SCHEMA, t as analyzeSupervisorRunIntegrity } from "../integrity-C8oUeB0g.js";
|
|
3
|
+
import { c as parseSupervisorTree, f as SUPERVISOR_RUN_ROLLUP_SCHEMA, g as unavailable, h as showMeasured, m as isUnavailable, n as RUNTIME_FAILED_STATUS, r as readTerminalRecord, t as NO_TERMINAL_RECORD } from "../terminal-record-D4Rv-j1W.js";
|
|
4
|
+
import { a as findSupervisorRunDirs, c as analyzeSupervisorRunSources, i as analyzeSupervisorRun, l as parsePatch, n as renderSupervisorRunHeadline, o as isRuntimeSupervisorRunDir, r as renderSupervisorRunMarkdown, s as readRuntimeSupervisorRun, t as renderSupervisorRollupMarkdown } from "../render-Cs34y0iX.js";
|
|
5
5
|
//#region src/supervisor-run/rollup.ts
|
|
6
6
|
/** Aggregate supervisor reports while retaining missing-measurement denominators. */
|
|
7
7
|
/**
|
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import { d as CostReceiptInput, p as CustomTokenPricing } from "./cost-ledger-Jk62OBfF.js";
|
|
2
2
|
import { i as AnalystFinding, n as AnalystContext, t as Analyst } from "./types-CW8ewATZ.js";
|
|
3
|
-
import { R as Scenario, S as JudgeConfig, w as JudgeScore } from "./types-
|
|
4
|
-
import { a as EvaluationResult, o as EvaluatorOptions, r as EvaluationContext } from "./evaluation-
|
|
3
|
+
import { R as Scenario, S as JudgeConfig, w as JudgeScore } from "./types-DtkO1G9L.js";
|
|
4
|
+
import { a as EvaluationResult, o as EvaluatorOptions, r as EvaluationContext } from "./evaluation-BOZ38vOH.js";
|
|
5
5
|
import { f as SystemOneResult, h as SystemOneState, i as SystemOneJson, o as SystemOneQuestions, u as SystemOneRequest } from "./systemone-protocol-DKoi1NdK.js";
|
|
6
6
|
import "./systemone-review-CwE6JGIp.js";
|
|
7
7
|
//#region src/systemone-domain-match.d.ts
|
|
@@ -132,4 +132,4 @@ interface SystemOneAnalystOptions<I, Q extends SystemOneQuestions = SystemOneQue
|
|
|
132
132
|
declare function systemOneAnalyst<I, Q extends SystemOneQuestions = SystemOneQuestions>(options: SystemOneAnalystOptions<I, Q>): Analyst<I>;
|
|
133
133
|
//#endregion
|
|
134
134
|
export { assessSystemOneDomainMatch as _, SystemOneOptions as a, systemOneEvaluator as c, SystemOneDomainCoverage as d, SystemOneDomainEvidence as f, SystemOneDomainMatchThresholds as g, SystemOneDomainMatchReport as h, SystemOneJudgeOptions as i, systemOneJudge as l, SystemOneDomainMatchReason as m, SystemOneEvaluate as n, normalizedSystemOneScore as o, SystemOneDomainMatchInput as p, SystemOneEvaluatorOptions as r, systemOneAnalyst as s, SystemOneAnalystOptions as t, PreparedSystemOneDomainMatch as u, prepareSystemOneDomainMatch as v };
|
|
135
|
-
//# sourceMappingURL=systemone-
|
|
135
|
+
//# sourceMappingURL=systemone-BMzDJb1i.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"systemone-
|
|
1
|
+
{"version":3,"file":"systemone-BMzDJb1i.d.ts","names":[],"sources":["../src/systemone-domain-match.ts","../src/systemone.ts"],"mappings":";;;;;;;KAYY;UAEK;GACd,cAAc;EACf;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA,YAAY;EACZ,UAAU;EACV,UAAU;EACV,YAAY;;KAGT;EACH;IACE;IACA;IACA;MAAY;MAAe;MAAkB;;;EAE/C;IACE;IACA;IACA,UAAU;;EAEZ;IACE;IACA;IACA,UAAU;;;UAIG,qCAAqC;EACpD,SAAS,iBAAiB;;EAE1B;;KAGU;UAUK;EACf;EACA;EACA;EACA;EACA;EACA,UAAU;EACV,QAAQ;EACR;IACE,OAAO;IACP,UAAU;IACV,QAAQ;;EAEV;IACE;IACA;IACA;;EAEF;;;iBAqHc,4BACd,OAAO,4BACN;;iBA+Ca,2BACd,UAAU,8BACV,eACC;;;KCxOS,oBAAoB,iBAAiB;UAChC;EACf,UAAU;EACV,aAAa,iBAAiB;EAC9B,gBAAgB,iBAAiB;EACjC,UAAU;EACV,WAAW,sBAAsB;EACjC,oBAAoB,OAAO,UAAU;;EAErC,eAAe,mBAAmB;;;iBAIpB,mBAAmB,SAAS,mCA2BtB,UAAU,oBAAkB,SACrC,iBAAiB,IAAE,UAClB,sBACT,QAAQ,iBAAiB,gBAAgB;KAUzC,MAAM,GAAG,UAAU;EAAc,UAAU;EAAG,UAAU;;KACxD,eAAe,GAAG,UAAU,sBAAsB,MAAM,OAAO,MAAM,IAAI,QAAQ;UACrE,yBAAyB;EACxC;EACA;EACA,WAAW;;UAEI,sBACf,GACA,UAAU,WAAW,UACrB,UAAU,qBAAqB,4BACvB;EACR;EACA;EACA,WAAW,eAAe,MAAM,GAAG,IAAI;EACvC,cACE,OAAO,MAAM,GAAG,IAChB,SAAS,sBACN,iBAAiB,QAAQ;EAC9B,aAAa,YAAY,GAAG;EAC5B,UAAU;EACV,OAAO,OAAO,gBAAgB,IAAI,OAAO,MAAM,GAAG,OAAO,aAAa,QAAQ;EAC9E,UACE,QAAQ,iBAAiB,gBAAgB,KACzC,OAAO,MAAM,GAAG,cACN;EACZ,aAAa,UAAU;;;iBA0BT,yBACd,QAAQ,iBACR,UAAU,yBACT;;iBAgCa,eACd,GACA,UAAU,WAAW,UACrB,UAAU,qBAAqB,oBAC/B,cAAc,SAAS,sBAAsB,GAAG,GAAG,KAAK,YAAY,GAAG;UAqDxD,wBAAwB,GAAG,UAAU,qBAAqB,4BACjE;EACR;EACA;EACA,WAAW,QAAQ;EACnB;EACA;EACA,WAAW,MAAM,OAAO,GAAG,SAAS,mBAAmB,IAAI,QAAQ;EACnE,cAAc,OAAO,GAAG,SAAS,mBAAmB,iBAAiB,QAAQ;EAC7E,WACE,OAAO,gBAAgB,IACvB,OAAO,GACP,SAAS,mBACN,mBAAmB,QAAQ;EAChC,UACE,QAAQ,iBAAiB,gBAAgB,KACzC,OAAO,GACP,SAAS,0BACC;;iBAGE,iBAAiB,GAAG,UAAU,qBAAqB,oBACjE,SAAS,wBAAwB,GAAG,KACnC,QAAQ"}
|
package/dist/systemone.d.ts
CHANGED
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { a as EvaluationResult, c as asJudge, i as EvaluationJudgeOptions, l as createEvaluator, n as EvaluationAnalystOptions, o as EvaluatorOptions, r as EvaluationContext, s as asAnalyst, t as CostedEvaluator } from "./evaluation-
|
|
1
|
+
import { a as EvaluationResult, c as asJudge, i as EvaluationJudgeOptions, l as createEvaluator, n as EvaluationAnalystOptions, o as EvaluatorOptions, r as EvaluationContext, s as asAnalyst, t as CostedEvaluator } from "./evaluation-BOZ38vOH.js";
|
|
2
2
|
import { S as systemOneUsage, a as SystemOneQuestion, b as parseSystemOneResult, c as SystemOneRecordedQuestions, d as SystemOneResponseError, f as SystemOneResult, g as normalizeSystemOneResult, h as SystemOneState, i as SystemOneJson, l as SystemOneRecordedRequest, m as SystemOneScoreCriteria, n as SystemOneAnswerFor, o as SystemOneQuestions, p as SystemOneResultProfile, r as SystemOneDescription, s as SystemOneRecordedQuestion, t as SystemOneAnswer, u as SystemOneRequest, v as parseSystemOneRecordedRequest, x as systemOneConfidence, y as parseSystemOneRequest } from "./systemone-protocol-DKoi1NdK.js";
|
|
3
|
-
import { _ as assessSystemOneDomainMatch, a as SystemOneOptions, c as systemOneEvaluator, d as SystemOneDomainCoverage, f as SystemOneDomainEvidence, g as SystemOneDomainMatchThresholds, h as SystemOneDomainMatchReport, i as SystemOneJudgeOptions, l as systemOneJudge, m as SystemOneDomainMatchReason, n as SystemOneEvaluate, o as normalizedSystemOneScore, p as SystemOneDomainMatchInput, r as SystemOneEvaluatorOptions, s as systemOneAnalyst, t as SystemOneAnalystOptions, u as PreparedSystemOneDomainMatch, v as prepareSystemOneDomainMatch } from "./systemone-
|
|
3
|
+
import { _ as assessSystemOneDomainMatch, a as SystemOneOptions, c as systemOneEvaluator, d as SystemOneDomainCoverage, f as SystemOneDomainEvidence, g as SystemOneDomainMatchThresholds, h as SystemOneDomainMatchReport, i as SystemOneJudgeOptions, l as systemOneJudge, m as SystemOneDomainMatchReason, n as SystemOneEvaluate, o as normalizedSystemOneScore, p as SystemOneDomainMatchInput, r as SystemOneEvaluatorOptions, s as systemOneAnalyst, t as SystemOneAnalystOptions, u as PreparedSystemOneDomainMatch, v as prepareSystemOneDomainMatch } from "./systemone-BMzDJb1i.js";
|
|
4
4
|
import { a as SystemOneReviewReport, c as systemOneReviewFindings, i as SystemOneReviewInput, n as SystemOneReviewAssessment, o as assessSystemOneReview, r as SystemOneReviewCheck, s as prepareSystemOneReview, t as PreparedSystemOneReview } from "./systemone-review-CwE6JGIp.js";
|
|
5
5
|
export { CostedEvaluator, EvaluationAnalystOptions, EvaluationContext, EvaluationJudgeOptions, EvaluationResult, EvaluatorOptions, PreparedSystemOneDomainMatch, PreparedSystemOneReview, SystemOneAnalystOptions, type SystemOneAnswer, type SystemOneAnswerFor, type SystemOneDescription, SystemOneDomainCoverage, SystemOneDomainEvidence, SystemOneDomainMatchInput, SystemOneDomainMatchReason, SystemOneDomainMatchReport, SystemOneDomainMatchThresholds, SystemOneEvaluate, SystemOneEvaluatorOptions, type SystemOneJson, SystemOneJudgeOptions, SystemOneOptions, type SystemOneQuestion, type SystemOneQuestions, type SystemOneRecordedQuestion, type SystemOneRecordedQuestions, type SystemOneRecordedRequest, type SystemOneRequest, SystemOneResponseError, type SystemOneResult, type SystemOneResultProfile, SystemOneReviewAssessment, SystemOneReviewCheck, SystemOneReviewInput, SystemOneReviewReport, type SystemOneScoreCriteria, type SystemOneState, asAnalyst, asJudge, assessSystemOneDomainMatch, assessSystemOneReview, createEvaluator, normalizeSystemOneResult, normalizedSystemOneScore, parseSystemOneRecordedRequest, parseSystemOneRequest, parseSystemOneResult, prepareSystemOneDomainMatch, prepareSystemOneReview, systemOneAnalyst, systemOneConfidence, systemOneEvaluator, systemOneJudge, systemOneReviewFindings, systemOneUsage };
|
|
@@ -193,24 +193,84 @@ function summarizeDialect(seen) {
|
|
|
193
193
|
if (seen.size > 1) return "mixed";
|
|
194
194
|
return seen.has("runtime-envelope") ? "runtime-envelope" : "flat";
|
|
195
195
|
}
|
|
196
|
-
|
|
197
|
-
|
|
198
|
-
const
|
|
199
|
-
const
|
|
196
|
+
/** One spawned event as a row; a field that is present but unusable marks the row invalid. */
|
|
197
|
+
function spawnRowOf(ev, id, sourceRow, rootId) {
|
|
198
|
+
const parent = typeof ev.parent === "string" && ev.parent.length > 0 ? ev.parent : null;
|
|
199
|
+
const label = typeof ev.label === "string" ? ev.label : "";
|
|
200
|
+
const invalidFields = [];
|
|
201
|
+
if (id.length === 0) invalidFields.push("id");
|
|
202
|
+
const parentOmittedForFirstRoot = ev.parent === void 0 && rootId === null;
|
|
203
|
+
if (ev.parent !== null && !parentOmittedForFirstRoot && !(typeof ev.parent === "string" && ev.parent.length > 0)) invalidFields.push("parent");
|
|
204
|
+
if (label.length === 0) invalidFields.push("label");
|
|
205
|
+
if (ev.role !== void 0 && ev.role !== "supervisor" && ev.role !== "worker") invalidFields.push("role");
|
|
206
|
+
if (ev.profileDigest !== void 0 && !(typeof ev.profileDigest === "string" && ev.profileDigest.length > 0)) invalidFields.push("profileDigest");
|
|
207
|
+
if (ev.runtime !== void 0 && !(typeof ev.runtime === "string" && ev.runtime.length > 0)) invalidFields.push("runtime");
|
|
208
|
+
const role = ev.role === "supervisor" || ev.role === "worker" ? ev.role : null;
|
|
209
|
+
const valid = invalidFields.length === 0;
|
|
210
|
+
if (valid && parent === null && rootId === null) rootId = id;
|
|
211
|
+
return {
|
|
212
|
+
sourceRow,
|
|
213
|
+
id,
|
|
214
|
+
parent,
|
|
215
|
+
label,
|
|
216
|
+
role,
|
|
217
|
+
profileDigest: typeof ev.profileDigest === "string" && ev.profileDigest.length > 0 ? ev.profileDigest : null,
|
|
218
|
+
runtime: typeof ev.runtime === "string" && ev.runtime.length > 0 ? ev.runtime : null,
|
|
219
|
+
at: ms(ev.at),
|
|
220
|
+
valid,
|
|
221
|
+
invalidFields
|
|
222
|
+
};
|
|
223
|
+
}
|
|
224
|
+
/** A settled event as a close row, with its verdict and spend. */
|
|
225
|
+
function settledCloseOf(ev, id) {
|
|
226
|
+
const verdict = readVerdict(ev.verdict);
|
|
227
|
+
return {
|
|
228
|
+
id,
|
|
229
|
+
kind: "settled",
|
|
230
|
+
status: typeof ev.status === "string" ? ev.status : null,
|
|
231
|
+
verdict: verdict.label,
|
|
232
|
+
valid: verdict.valid,
|
|
233
|
+
score: verdict.score,
|
|
234
|
+
rawVerdict: verdict.raw,
|
|
235
|
+
reason: typeof ev.reason === "string" ? ev.reason : null,
|
|
236
|
+
infra: typeof ev.infra === "boolean" ? ev.infra : null,
|
|
237
|
+
at: ms(ev.at),
|
|
238
|
+
spend: readSpend(ev.spent),
|
|
239
|
+
hasSpend: asRecord(ev.spent).tokens !== void 0
|
|
240
|
+
};
|
|
241
|
+
}
|
|
242
|
+
/** A cancelled event as a close row: it carries no verdict and no spend. */
|
|
243
|
+
function cancelledCloseOf(ev, id) {
|
|
244
|
+
return {
|
|
245
|
+
id,
|
|
246
|
+
kind: "cancelled",
|
|
247
|
+
status: "cancelled",
|
|
248
|
+
verdict: typeof ev.reason === "string" ? ev.reason : null,
|
|
249
|
+
valid: null,
|
|
250
|
+
score: null,
|
|
251
|
+
rawVerdict: typeof ev.reason === "string" ? ev.reason : null,
|
|
252
|
+
reason: typeof ev.reason === "string" ? ev.reason : null,
|
|
253
|
+
infra: null,
|
|
254
|
+
at: ms(ev.at),
|
|
255
|
+
spend: {
|
|
256
|
+
tokens: {
|
|
257
|
+
input: 0,
|
|
258
|
+
output: 0,
|
|
259
|
+
cacheRead: 0,
|
|
260
|
+
cacheWrite: 0,
|
|
261
|
+
hasCache: false,
|
|
262
|
+
breakdownKnown: true
|
|
263
|
+
},
|
|
264
|
+
tokensKnown: true,
|
|
265
|
+
usd: 0,
|
|
266
|
+
usdKnown: true
|
|
267
|
+
},
|
|
268
|
+
hasSpend: false
|
|
269
|
+
};
|
|
270
|
+
}
|
|
271
|
+
function readJournal(events) {
|
|
200
272
|
const spawns = [];
|
|
201
273
|
const closes = [];
|
|
202
|
-
let brainIn = 0;
|
|
203
|
-
let brainOut = 0;
|
|
204
|
-
let brainCacheRead = 0;
|
|
205
|
-
let brainCacheWrite = 0;
|
|
206
|
-
let brainHasCache = false;
|
|
207
|
-
let brainCacheBreakdownKnown = true;
|
|
208
|
-
let brainTokensKnownCount = 0;
|
|
209
|
-
let brainTokensUnknownCount = 0;
|
|
210
|
-
let brainUsd = 0;
|
|
211
|
-
let brainUsdKnownCount = 0;
|
|
212
|
-
let brainUsdUnknownCount = 0;
|
|
213
|
-
let meteredCount = 0;
|
|
214
274
|
let rootId = null;
|
|
215
275
|
let unreadableRows = 0;
|
|
216
276
|
const ignoredByKind = /* @__PURE__ */ new Map();
|
|
@@ -240,73 +300,11 @@ function parseSupervisorTree(src) {
|
|
|
240
300
|
});
|
|
241
301
|
}
|
|
242
302
|
if (kind === "spawned") {
|
|
243
|
-
const
|
|
244
|
-
|
|
245
|
-
|
|
246
|
-
|
|
247
|
-
|
|
248
|
-
if (ev.parent !== null && !parentOmittedForFirstRoot && !(typeof ev.parent === "string" && ev.parent.length > 0)) invalidFields.push("parent");
|
|
249
|
-
if (label.length === 0) invalidFields.push("label");
|
|
250
|
-
if (ev.role !== void 0 && ev.role !== "supervisor" && ev.role !== "worker") invalidFields.push("role");
|
|
251
|
-
if (ev.profileDigest !== void 0 && !(typeof ev.profileDigest === "string" && ev.profileDigest.length > 0)) invalidFields.push("profileDigest");
|
|
252
|
-
if (ev.runtime !== void 0 && !(typeof ev.runtime === "string" && ev.runtime.length > 0)) invalidFields.push("runtime");
|
|
253
|
-
const role = ev.role === "supervisor" || ev.role === "worker" ? ev.role : null;
|
|
254
|
-
const valid = invalidFields.length === 0;
|
|
255
|
-
if (valid && parent === null && rootId === null) rootId = id;
|
|
256
|
-
spawns.push({
|
|
257
|
-
sourceRow,
|
|
258
|
-
id,
|
|
259
|
-
parent,
|
|
260
|
-
label,
|
|
261
|
-
role,
|
|
262
|
-
profileDigest: typeof ev.profileDigest === "string" && ev.profileDigest.length > 0 ? ev.profileDigest : null,
|
|
263
|
-
runtime: typeof ev.runtime === "string" && ev.runtime.length > 0 ? ev.runtime : null,
|
|
264
|
-
at: ms(ev.at),
|
|
265
|
-
valid,
|
|
266
|
-
invalidFields
|
|
267
|
-
});
|
|
268
|
-
} else if (kind === "settled") {
|
|
269
|
-
const verdict = readVerdict(ev.verdict);
|
|
270
|
-
closes.push({
|
|
271
|
-
id,
|
|
272
|
-
kind: "settled",
|
|
273
|
-
status: typeof ev.status === "string" ? ev.status : null,
|
|
274
|
-
verdict: verdict.label,
|
|
275
|
-
valid: verdict.valid,
|
|
276
|
-
score: verdict.score,
|
|
277
|
-
rawVerdict: verdict.raw,
|
|
278
|
-
reason: typeof ev.reason === "string" ? ev.reason : null,
|
|
279
|
-
infra: typeof ev.infra === "boolean" ? ev.infra : null,
|
|
280
|
-
at: ms(ev.at),
|
|
281
|
-
spend: readSpend(ev.spent),
|
|
282
|
-
hasSpend: asRecord(ev.spent).tokens !== void 0
|
|
283
|
-
});
|
|
284
|
-
} else if (kind === "cancelled") closes.push({
|
|
285
|
-
id,
|
|
286
|
-
kind: "cancelled",
|
|
287
|
-
status: "cancelled",
|
|
288
|
-
verdict: typeof ev.reason === "string" ? ev.reason : null,
|
|
289
|
-
valid: null,
|
|
290
|
-
score: null,
|
|
291
|
-
rawVerdict: typeof ev.reason === "string" ? ev.reason : null,
|
|
292
|
-
reason: typeof ev.reason === "string" ? ev.reason : null,
|
|
293
|
-
infra: null,
|
|
294
|
-
at: ms(ev.at),
|
|
295
|
-
spend: {
|
|
296
|
-
tokens: {
|
|
297
|
-
input: 0,
|
|
298
|
-
output: 0,
|
|
299
|
-
cacheRead: 0,
|
|
300
|
-
cacheWrite: 0,
|
|
301
|
-
hasCache: false,
|
|
302
|
-
breakdownKnown: true
|
|
303
|
-
},
|
|
304
|
-
tokensKnown: true,
|
|
305
|
-
usd: 0,
|
|
306
|
-
usdKnown: true
|
|
307
|
-
},
|
|
308
|
-
hasSpend: false
|
|
309
|
-
});
|
|
303
|
+
const spawn = spawnRowOf(ev, id, sourceRow, rootId);
|
|
304
|
+
if (spawn.valid && spawn.parent === null && rootId === null) rootId = id;
|
|
305
|
+
spawns.push(spawn);
|
|
306
|
+
} else if (kind === "settled") closes.push(settledCloseOf(ev, id));
|
|
307
|
+
else if (kind === "cancelled") closes.push(cancelledCloseOf(ev, id));
|
|
310
308
|
else if (kind === "metered") meteredRows.push({
|
|
311
309
|
id,
|
|
312
310
|
spend: readSpend(ev.spend),
|
|
@@ -314,6 +312,31 @@ function parseSupervisorTree(src) {
|
|
|
314
312
|
});
|
|
315
313
|
else throw new Error(`unhandled journal event kind ${JSON.stringify(kind)}`);
|
|
316
314
|
}
|
|
315
|
+
return {
|
|
316
|
+
spawns,
|
|
317
|
+
closes,
|
|
318
|
+
rootId,
|
|
319
|
+
unreadableRows,
|
|
320
|
+
ignoredByKind,
|
|
321
|
+
dialectsSeen,
|
|
322
|
+
resourceRecords,
|
|
323
|
+
meteredRows
|
|
324
|
+
};
|
|
325
|
+
}
|
|
326
|
+
/** Only root-attributed metered events are the supervisor brain's own spend. A nested driver meters under its own id; its spend reaches this tree through its settled close, so folding it into the brain here would count the same tokens twice. */
|
|
327
|
+
function foldBrainSpend(meteredRows, rootId) {
|
|
328
|
+
let brainIn = 0;
|
|
329
|
+
let brainOut = 0;
|
|
330
|
+
let brainCacheRead = 0;
|
|
331
|
+
let brainCacheWrite = 0;
|
|
332
|
+
let brainHasCache = false;
|
|
333
|
+
let brainCacheBreakdownKnown = true;
|
|
334
|
+
let brainTokensKnownCount = 0;
|
|
335
|
+
let brainTokensUnknownCount = 0;
|
|
336
|
+
let brainUsd = 0;
|
|
337
|
+
let brainUsdKnownCount = 0;
|
|
338
|
+
let brainUsdUnknownCount = 0;
|
|
339
|
+
let meteredCount = 0;
|
|
317
340
|
for (const row of meteredRows) {
|
|
318
341
|
if (rootId === null || row.id !== rootId) continue;
|
|
319
342
|
if (row.spend.tokensKnown) {
|
|
@@ -331,105 +354,118 @@ function parseSupervisorTree(src) {
|
|
|
331
354
|
} else brainUsdUnknownCount += 1;
|
|
332
355
|
meteredCount += 1;
|
|
333
356
|
}
|
|
334
|
-
|
|
335
|
-
|
|
336
|
-
|
|
337
|
-
|
|
338
|
-
|
|
339
|
-
|
|
340
|
-
|
|
341
|
-
|
|
342
|
-
|
|
343
|
-
|
|
344
|
-
|
|
345
|
-
|
|
346
|
-
|
|
347
|
-
|
|
348
|
-
|
|
349
|
-
|
|
350
|
-
|
|
351
|
-
|
|
352
|
-
|
|
353
|
-
|
|
354
|
-
|
|
355
|
-
|
|
356
|
-
|
|
357
|
-
|
|
358
|
-
|
|
359
|
-
|
|
360
|
-
|
|
361
|
-
|
|
362
|
-
|
|
363
|
-
|
|
364
|
-
|
|
365
|
-
|
|
366
|
-
|
|
367
|
-
|
|
368
|
-
|
|
369
|
-
|
|
370
|
-
|
|
357
|
+
return {
|
|
358
|
+
tokensIn: brainIn,
|
|
359
|
+
tokensOut: brainOut,
|
|
360
|
+
cacheRead: brainCacheRead,
|
|
361
|
+
cacheWrite: brainCacheWrite,
|
|
362
|
+
hasCache: brainHasCache,
|
|
363
|
+
cacheBreakdownKnown: brainCacheBreakdownKnown,
|
|
364
|
+
tokensKnownCount: brainTokensKnownCount,
|
|
365
|
+
tokensUnknownCount: brainTokensUnknownCount,
|
|
366
|
+
usd: brainUsd,
|
|
367
|
+
usdKnownCount: brainUsdKnownCount,
|
|
368
|
+
usdUnknownCount: brainUsdUnknownCount,
|
|
369
|
+
meteredCount
|
|
370
|
+
};
|
|
371
|
+
}
|
|
372
|
+
/** Steer counts from a worker's inbox and acknowledgements, or the reason they cannot be stated exactly. */
|
|
373
|
+
function accountSteers(facts) {
|
|
374
|
+
const requestsById = /* @__PURE__ */ new Map();
|
|
375
|
+
for (const request of facts.steerRequests) {
|
|
376
|
+
if (request.requestId === null) continue;
|
|
377
|
+
requestsById.set(request.requestId, (requestsById.get(request.requestId) ?? 0) + 1);
|
|
378
|
+
}
|
|
379
|
+
const acknowledgementsById = /* @__PURE__ */ new Map();
|
|
380
|
+
for (const acknowledgement of facts.steerAcknowledgements) {
|
|
381
|
+
if (acknowledgement.requestId === null) continue;
|
|
382
|
+
const matches = acknowledgementsById.get(acknowledgement.requestId) ?? [];
|
|
383
|
+
matches.push(acknowledgement);
|
|
384
|
+
acknowledgementsById.set(acknowledgement.requestId, matches);
|
|
385
|
+
}
|
|
386
|
+
const queueProblems = [];
|
|
387
|
+
if (!facts.inboxCaptured) queueProblems.push("inbox absent");
|
|
388
|
+
if (!facts.eventsCaptured) queueProblems.push("events absent");
|
|
389
|
+
if (facts.inboxInvalidRows > 0) queueProblems.push("inbox contains malformed rows");
|
|
390
|
+
if (facts.eventsInvalidRows > 0) queueProblems.push("events contain malformed rows");
|
|
391
|
+
if (facts.steerRequests.some((request) => request.requestId === null)) queueProblems.push("queued request id missing");
|
|
392
|
+
if ([...requestsById.values()].some((count) => count > 1)) queueProblems.push("queued request id duplicated");
|
|
393
|
+
if (facts.steerAcknowledgements.some((acknowledgement) => acknowledgement.requestId === null)) queueProblems.push("acknowledgement request id missing");
|
|
394
|
+
if ([...acknowledgementsById.keys()].some((id) => !requestsById.has(id))) queueProblems.push("acknowledgement has no queued request");
|
|
395
|
+
if (queueProblems.length === 0) facts.steersQueued = requestsById.size;
|
|
396
|
+
else facts.steersQueuedUnavailable = [...new Set(queueProblems)].join("; ");
|
|
397
|
+
const deliveryProblems = [...queueProblems];
|
|
398
|
+
if ([...acknowledgementsById.values()].some((rows) => rows.length > 1)) deliveryProblems.push("acknowledgement request id duplicated");
|
|
399
|
+
if (facts.steerAcknowledgements.some((acknowledgement) => acknowledgement.delivered === null)) deliveryProblems.push("acknowledgement delivery status missing");
|
|
400
|
+
if (deliveryProblems.length === 0) facts.steersDelivered = [...requestsById.keys()].filter((id) => {
|
|
401
|
+
const acknowledgement = acknowledgementsById.get(id);
|
|
402
|
+
return acknowledgement?.length === 1 && acknowledgement[0]?.delivered === true;
|
|
403
|
+
}).length;
|
|
404
|
+
else facts.steersDeliveredUnavailable = [...new Set(deliveryProblems)].join("; ");
|
|
405
|
+
}
|
|
406
|
+
/** Everything one worker's inbox and event log say: timing, verdict, questions and steer accounting. */
|
|
407
|
+
function readWorkerLog(w, sourceIndex) {
|
|
408
|
+
const facts = {
|
|
409
|
+
sourceIndex,
|
|
410
|
+
workerId: w.workerId ?? null,
|
|
411
|
+
label: w.label,
|
|
412
|
+
inboxCaptured: w.inbox !== null,
|
|
413
|
+
eventsCaptured: w.events !== null,
|
|
414
|
+
inboxInvalidRows: 0,
|
|
415
|
+
eventsInvalidRows: 0,
|
|
416
|
+
started: null,
|
|
417
|
+
finished: false,
|
|
418
|
+
finishedAt: null,
|
|
419
|
+
passed: null,
|
|
420
|
+
score: null,
|
|
421
|
+
finishedPatchBytes: null,
|
|
422
|
+
evidenceBytes: 0,
|
|
423
|
+
steersQueued: null,
|
|
424
|
+
steersDelivered: null,
|
|
425
|
+
steersQueuedUnavailable: null,
|
|
426
|
+
steersDeliveredUnavailable: null,
|
|
427
|
+
questions: 0,
|
|
428
|
+
steerRequests: [],
|
|
429
|
+
steerAcknowledgements: []
|
|
430
|
+
};
|
|
431
|
+
const parsedInbox = parseJsonlWithDiagnostics(w.inbox);
|
|
432
|
+
const parsedEvents = parseJsonlWithDiagnostics(w.events);
|
|
433
|
+
facts.inboxInvalidRows = parsedInbox.invalidRows;
|
|
434
|
+
facts.eventsInvalidRows = parsedEvents.invalidRows;
|
|
435
|
+
for (const [row, req] of parsedInbox.rows.entries()) {
|
|
436
|
+
if (typeof req.message !== "string" || req.message.trim().length === 0) continue;
|
|
437
|
+
facts.steerRequests.push({
|
|
438
|
+
requestId: typeof req.id === "string" && req.id.length > 0 ? req.id : null,
|
|
439
|
+
row
|
|
440
|
+
});
|
|
441
|
+
}
|
|
442
|
+
for (const [row, ev] of parsedEvents.rows.entries()) {
|
|
443
|
+
const kind = ev.kind;
|
|
444
|
+
if (kind === "message") {
|
|
445
|
+
if (ev.direction === "up") {
|
|
446
|
+
facts.questions += 1;
|
|
447
|
+
continue;
|
|
448
|
+
}
|
|
449
|
+
facts.steerAcknowledgements.push({
|
|
450
|
+
requestId: typeof ev.requestId === "string" && ev.requestId.length > 0 ? ev.requestId : null,
|
|
451
|
+
delivered: typeof ev.delivered === "boolean" ? ev.delivered : null,
|
|
371
452
|
row
|
|
372
453
|
});
|
|
454
|
+
} else if (kind === "started") facts.started = ms(ev.at);
|
|
455
|
+
else if (kind === "finished") {
|
|
456
|
+
facts.finished = true;
|
|
457
|
+
facts.finishedAt = ms(ev.at);
|
|
458
|
+
facts.passed = typeof ev.passed === "boolean" ? ev.passed : null;
|
|
459
|
+
facts.score = typeof ev.score === "number" && Number.isFinite(ev.score) ? ev.score : null;
|
|
460
|
+
facts.finishedPatchBytes = typeof ev.patchBytes === "number" ? ev.patchBytes : null;
|
|
461
|
+
facts.evidenceBytes = typeof ev.evidence === "string" ? ev.evidence.length : 0;
|
|
373
462
|
}
|
|
374
|
-
for (const [row, ev] of parsedEvents.rows.entries()) {
|
|
375
|
-
const kind = ev.kind;
|
|
376
|
-
if (kind === "message") {
|
|
377
|
-
if (ev.direction === "up") {
|
|
378
|
-
facts.questions += 1;
|
|
379
|
-
continue;
|
|
380
|
-
}
|
|
381
|
-
facts.steerAcknowledgements.push({
|
|
382
|
-
requestId: typeof ev.requestId === "string" && ev.requestId.length > 0 ? ev.requestId : null,
|
|
383
|
-
delivered: typeof ev.delivered === "boolean" ? ev.delivered : null,
|
|
384
|
-
row
|
|
385
|
-
});
|
|
386
|
-
} else if (kind === "started") facts.started = ms(ev.at);
|
|
387
|
-
else if (kind === "finished") {
|
|
388
|
-
facts.finished = true;
|
|
389
|
-
facts.finishedAt = ms(ev.at);
|
|
390
|
-
facts.passed = typeof ev.passed === "boolean" ? ev.passed : null;
|
|
391
|
-
facts.score = typeof ev.score === "number" && Number.isFinite(ev.score) ? ev.score : null;
|
|
392
|
-
facts.finishedPatchBytes = typeof ev.patchBytes === "number" ? ev.patchBytes : null;
|
|
393
|
-
facts.evidenceBytes = typeof ev.evidence === "string" ? ev.evidence.length : 0;
|
|
394
|
-
}
|
|
395
|
-
}
|
|
396
|
-
const requestsById = /* @__PURE__ */ new Map();
|
|
397
|
-
for (const request of facts.steerRequests) {
|
|
398
|
-
if (request.requestId === null) continue;
|
|
399
|
-
requestsById.set(request.requestId, (requestsById.get(request.requestId) ?? 0) + 1);
|
|
400
|
-
}
|
|
401
|
-
const acknowledgementsById = /* @__PURE__ */ new Map();
|
|
402
|
-
for (const acknowledgement of facts.steerAcknowledgements) {
|
|
403
|
-
if (acknowledgement.requestId === null) continue;
|
|
404
|
-
const matches = acknowledgementsById.get(acknowledgement.requestId) ?? [];
|
|
405
|
-
matches.push(acknowledgement);
|
|
406
|
-
acknowledgementsById.set(acknowledgement.requestId, matches);
|
|
407
|
-
}
|
|
408
|
-
const queueProblems = [];
|
|
409
|
-
if (!facts.inboxCaptured) queueProblems.push("inbox absent");
|
|
410
|
-
if (!facts.eventsCaptured) queueProblems.push("events absent");
|
|
411
|
-
if (facts.inboxInvalidRows > 0) queueProblems.push("inbox contains malformed rows");
|
|
412
|
-
if (facts.eventsInvalidRows > 0) queueProblems.push("events contain malformed rows");
|
|
413
|
-
if (facts.steerRequests.some((request) => request.requestId === null)) queueProblems.push("queued request id missing");
|
|
414
|
-
if ([...requestsById.values()].some((count) => count > 1)) queueProblems.push("queued request id duplicated");
|
|
415
|
-
if (facts.steerAcknowledgements.some((acknowledgement) => acknowledgement.requestId === null)) queueProblems.push("acknowledgement request id missing");
|
|
416
|
-
if ([...acknowledgementsById.keys()].some((id) => !requestsById.has(id))) queueProblems.push("acknowledgement has no queued request");
|
|
417
|
-
if (queueProblems.length === 0) facts.steersQueued = requestsById.size;
|
|
418
|
-
else facts.steersQueuedUnavailable = [...new Set(queueProblems)].join("; ");
|
|
419
|
-
const deliveryProblems = [...queueProblems];
|
|
420
|
-
if ([...acknowledgementsById.values()].some((rows) => rows.length > 1)) deliveryProblems.push("acknowledgement request id duplicated");
|
|
421
|
-
if (facts.steerAcknowledgements.some((acknowledgement) => acknowledgement.delivered === null)) deliveryProblems.push("acknowledgement delivery status missing");
|
|
422
|
-
if (deliveryProblems.length === 0) facts.steersDelivered = [...requestsById.keys()].filter((id) => {
|
|
423
|
-
const acknowledgement = acknowledgementsById.get(id);
|
|
424
|
-
return acknowledgement?.length === 1 && acknowledgement[0]?.delivered === true;
|
|
425
|
-
}).length;
|
|
426
|
-
else facts.steersDeliveredUnavailable = [...new Set(deliveryProblems)].join("; ");
|
|
427
|
-
workerLogRows.push(facts);
|
|
428
|
-
workerLogs.set(workerSourceKey(w), facts);
|
|
429
463
|
}
|
|
430
|
-
|
|
431
|
-
|
|
432
|
-
|
|
464
|
+
accountSteers(facts);
|
|
465
|
+
return facts;
|
|
466
|
+
}
|
|
467
|
+
/** The earliest and latest stamped event across spawns, closes and metered rows. */
|
|
468
|
+
function eventSpan(spawns, closes, meteredRows) {
|
|
433
469
|
let firstEventAt = null;
|
|
434
470
|
let lastEventAt = null;
|
|
435
471
|
const widenEventSpan = (at) => {
|
|
@@ -440,6 +476,31 @@ function parseSupervisorTree(src) {
|
|
|
440
476
|
for (const spawn of spawns) widenEventSpan(spawn.at);
|
|
441
477
|
for (const close of closes) widenEventSpan(close.at);
|
|
442
478
|
for (const metered of meteredRows) widenEventSpan(metered.at);
|
|
479
|
+
return {
|
|
480
|
+
firstEventAt,
|
|
481
|
+
lastEventAt
|
|
482
|
+
};
|
|
483
|
+
}
|
|
484
|
+
function parseSupervisorTree(src) {
|
|
485
|
+
const parsedJournal = parseJsonlWithDiagnostics(src.journal);
|
|
486
|
+
const events = parsedJournal.rows;
|
|
487
|
+
const state = parseJsonRecord(src.state);
|
|
488
|
+
const { spawns, closes, rootId, unreadableRows, ignoredByKind, dialectsSeen, resourceRecords, meteredRows } = readJournal(events);
|
|
489
|
+
const brain = foldBrainSpend(meteredRows, rootId);
|
|
490
|
+
const workerSpawns = spawns.filter((s) => s.valid && s.id !== rootId);
|
|
491
|
+
const workerIds = new Set(workerSpawns.map((s) => s.id));
|
|
492
|
+
const workerCloses = closes.filter((c) => workerIds.has(c.id));
|
|
493
|
+
const workerLogs = /* @__PURE__ */ new Map();
|
|
494
|
+
const workerLogRows = [];
|
|
495
|
+
for (const [sourceIndex, w] of (src.workers ?? []).entries()) {
|
|
496
|
+
const facts = readWorkerLog(w, sourceIndex);
|
|
497
|
+
workerLogRows.push(facts);
|
|
498
|
+
workerLogs.set(workerSourceKey(w), facts);
|
|
499
|
+
}
|
|
500
|
+
const startedAt = ms(state?.startedAt) ?? spawns[0]?.at ?? null;
|
|
501
|
+
const rootClose = rootId === null ? null : closes.find((close) => close.id === rootId);
|
|
502
|
+
const completedAt = ms(state?.completedAt) ?? rootClose?.at ?? null;
|
|
503
|
+
const { firstEventAt, lastEventAt } = eventSpan(spawns, closes, meteredRows);
|
|
443
504
|
const result = parseJsonRecord(src.result);
|
|
444
505
|
if (result && Object.hasOwn(result, "spentTotal")) resourceRecords.push({
|
|
445
506
|
nodeId: rootId,
|
|
@@ -454,20 +515,7 @@ function parseSupervisorTree(src) {
|
|
|
454
515
|
closes,
|
|
455
516
|
workerSpawns,
|
|
456
517
|
workerCloses,
|
|
457
|
-
brain
|
|
458
|
-
tokensIn: brainIn,
|
|
459
|
-
tokensOut: brainOut,
|
|
460
|
-
cacheRead: brainCacheRead,
|
|
461
|
-
cacheWrite: brainCacheWrite,
|
|
462
|
-
hasCache: brainHasCache,
|
|
463
|
-
cacheBreakdownKnown: brainCacheBreakdownKnown,
|
|
464
|
-
tokensKnownCount: brainTokensKnownCount,
|
|
465
|
-
tokensUnknownCount: brainTokensUnknownCount,
|
|
466
|
-
usd: brainUsd,
|
|
467
|
-
usdKnownCount: brainUsdKnownCount,
|
|
468
|
-
usdUnknownCount: brainUsdUnknownCount,
|
|
469
|
-
meteredCount
|
|
470
|
-
},
|
|
518
|
+
brain,
|
|
471
519
|
workerLogs,
|
|
472
520
|
workerLogRows,
|
|
473
521
|
journalRows: events.length + parsedJournal.invalidRows,
|
|
@@ -555,4 +603,4 @@ function readTerminalRecord(input) {
|
|
|
555
603
|
//#endregion
|
|
556
604
|
export { nonEmptyString as a, parseSupervisorTree as c, NO_SOURCE_LIMITS as d, SUPERVISOR_RUN_ROLLUP_SCHEMA as f, unavailable as g, showMeasured as h, asRecord as i, recordOrNull as l, isUnavailable as m, RUNTIME_FAILED_STATUS as n, parseJsonRecord as o, SUPERVISOR_RUN_SCHEMA as p, readTerminalRecord as r, parseJsonl as s, NO_TERMINAL_RECORD as t, workerSourceKey as u };
|
|
557
605
|
|
|
558
|
-
//# sourceMappingURL=terminal-record-
|
|
606
|
+
//# sourceMappingURL=terminal-record-D4Rv-j1W.js.map
|