@tangle-network/agent-eval 0.145.3 → 0.145.5
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 +22 -0
- package/README.md +5 -0
- package/dist/analyst/index.d.ts +1 -1
- package/dist/analyst/index.js +1 -1
- package/dist/{benchmark-command-Iaqlp3DU.js → benchmark-command-C44VE5rA.js} +2 -2
- package/dist/{benchmark-command-Iaqlp3DU.js.map → benchmark-command-C44VE5rA.js.map} +1 -1
- package/dist/benchmarks/index.d.ts +2 -2
- package/dist/benchmarks/index.js +2 -2
- package/dist/campaign/index.d.ts +4 -4
- package/dist/campaign/index.js +4 -4
- package/dist/{campaign-C5kujyqv.js → campaign-DGWSftLe.js} +503 -24
- package/dist/campaign-DGWSftLe.js.map +1 -0
- package/dist/cli.js +1 -1
- package/dist/contract/index.d.ts +3 -3
- package/dist/contract/index.js +3 -3
- package/dist/{define-agent-eval-sH24zBfM.d.ts → define-agent-eval-DcAcC1yQ.d.ts} +2 -2
- package/dist/{define-agent-eval-sH24zBfM.d.ts.map → define-agent-eval-DcAcC1yQ.d.ts.map} +1 -1
- package/dist/{define-agent-eval-CAL5HOz9.js → define-agent-eval-xaceIpPS.js} +2 -2
- package/dist/{define-agent-eval-CAL5HOz9.js.map → define-agent-eval-xaceIpPS.js.map} +1 -1
- package/dist/{index-CUczrQ7a.d.ts → index-DuKK7tJR.d.ts} +101 -4
- package/dist/index-DuKK7tJR.d.ts.map +1 -0
- package/dist/index.d.ts +2 -2
- package/dist/index.js +3 -3
- package/dist/{llm-judge-C2lQBv7h.js → llm-judge-BPKBdQTp.js} +7 -4
- package/dist/llm-judge-BPKBdQTp.js.map +1 -0
- package/dist/openapi.json +1 -1
- package/dist/{produced-state-Bgph_Qj-.js → produced-state-B_-TW9o0.js} +2 -2
- package/dist/{produced-state-Bgph_Qj-.js.map → produced-state-B_-TW9o0.js.map} +1 -1
- package/dist/{provenance-CCdxgLDT.d.ts → provenance-C_ln89b0.d.ts} +15 -1
- package/dist/provenance-C_ln89b0.d.ts.map +1 -0
- package/dist/{skillopt-optimization-method-C5cotF4E.d.ts → skillopt-optimization-method-Bb1OTHZW.d.ts} +9 -2
- package/dist/skillopt-optimization-method-Bb1OTHZW.d.ts.map +1 -0
- package/dist/{skillopt-optimization-method-B__Jnto4.js → skillopt-optimization-method-S7sdjzvh.js} +26 -3
- package/dist/skillopt-optimization-method-S7sdjzvh.js.map +1 -0
- package/docs/eval-surface-map.md +11 -0
- package/package.json +1 -1
- package/dist/campaign-C5kujyqv.js.map +0 -1
- package/dist/index-CUczrQ7a.d.ts.map +0 -1
- package/dist/llm-judge-C2lQBv7h.js.map +0 -1
- package/dist/provenance-CCdxgLDT.d.ts.map +0 -1
- package/dist/skillopt-optimization-method-B__Jnto4.js.map +0 -1
- package/dist/skillopt-optimization-method-C5cotF4E.d.ts.map +0 -1
|
@@ -8,8 +8,8 @@ import { A as LabeledScenarioWrite, D as LabeledScenarioSampleArgs, E as Labeled
|
|
|
8
8
|
import { d as CompletionVerdict, f as CorrectnessChecker, h as ProducedState, n as BackendIntegrityReport, s as RuntimeEventLike, u as CompletionRequirement } from "./backend-integrity-CsVin_Wb.js";
|
|
9
9
|
import { g as PairedArmsComparison } from "./pre-registration-DeRvl9sE.js";
|
|
10
10
|
import { d as PairedBootstrapResult } from "./paired-promotion-decision-CGzg0cI_.js";
|
|
11
|
-
import { lt as PlanCampaignRunOptions, st as CampaignRunPlan, xt as CampaignStorage } from "./provenance-
|
|
12
|
-
import "./skillopt-optimization-method-
|
|
11
|
+
import { lt as PlanCampaignRunOptions, st as CampaignRunPlan, xt as CampaignStorage } from "./provenance-C_ln89b0.js";
|
|
12
|
+
import "./skillopt-optimization-method-Bb1OTHZW.js";
|
|
13
13
|
import "./promotion-policy-D0nPhkSy.js";
|
|
14
14
|
import "./statistical-heldout-DC7cHtFD.js";
|
|
15
15
|
import { D as LedgerHash, m as LedgerTrustedHeadRemoval, p as LedgerTrustedHead } from "./index-B8Ui1mr1.js";
|
|
@@ -607,6 +607,20 @@ interface RunProfileMatrixOptions<TScenario extends Scenario, TArtifact> {
|
|
|
607
607
|
prompt: string;
|
|
608
608
|
completion: string;
|
|
609
609
|
} | undefined;
|
|
610
|
+
/**
|
|
611
|
+
* Optional explicit row selection. The matrix identity remains based on the
|
|
612
|
+
* complete profiles × scenarios × reps design; this invocation executes only
|
|
613
|
+
* rows accepted by the predicate.
|
|
614
|
+
*/
|
|
615
|
+
rowFilter?: (input: {
|
|
616
|
+
profile: AgentProfile$1;
|
|
617
|
+
scenario: TScenario;
|
|
618
|
+
rep: number;
|
|
619
|
+
}) => boolean;
|
|
620
|
+
/** Stable matrix identity supplied by a persisted profile-matrix plan. */
|
|
621
|
+
matrixId?: string;
|
|
622
|
+
/** Reuse cached failed cells. Normal matrix runs retry them by default. */
|
|
623
|
+
reuseFailedCells?: boolean;
|
|
610
624
|
}
|
|
611
625
|
interface ProfileSummary {
|
|
612
626
|
profileId: string;
|
|
@@ -752,6 +766,89 @@ interface ScoreboardRenderOptions {
|
|
|
752
766
|
*/
|
|
753
767
|
declare function renderScoreboardMarkdown(rows: readonly ScoreboardRow[], opts?: ScoreboardRenderOptions): string;
|
|
754
768
|
//#endregion
|
|
769
|
+
//#region src/campaign/presets/segmented-profile-matrix.d.ts
|
|
770
|
+
declare class SegmentedProfileMatrixError extends ProfileMatrixError {
|
|
771
|
+
constructor(message: string);
|
|
772
|
+
}
|
|
773
|
+
interface ProfileMatrixRow {
|
|
774
|
+
rowId: string;
|
|
775
|
+
ordinal: number;
|
|
776
|
+
profileId: string;
|
|
777
|
+
scenarioId: string;
|
|
778
|
+
rep: number;
|
|
779
|
+
}
|
|
780
|
+
interface ProfileMatrixPlan<TScenario extends Scenario, TArtifact> {
|
|
781
|
+
readonly schemaVersion: 1;
|
|
782
|
+
readonly matrixId: string;
|
|
783
|
+
readonly experimentId: string;
|
|
784
|
+
readonly planDigest: `sha256:${string}`;
|
|
785
|
+
readonly profiles: readonly AgentProfile$1[];
|
|
786
|
+
readonly scenarios: readonly TScenario[];
|
|
787
|
+
readonly judges: readonly JudgeConfig<TArtifact, TScenario>[];
|
|
788
|
+
readonly reps: number;
|
|
789
|
+
readonly seed: number;
|
|
790
|
+
readonly splitTag: NonNullable<RunProfileMatrixOptions<TScenario, TArtifact>['splitTag']>;
|
|
791
|
+
readonly commitSha: string;
|
|
792
|
+
/** Stable dispatch implementation/configuration identity used by cell caches. */
|
|
793
|
+
readonly dispatchRef: string;
|
|
794
|
+
readonly integrity: NonNullable<RunProfileMatrixOptions<TScenario, TArtifact>['integrity']>;
|
|
795
|
+
readonly allowMixed: boolean;
|
|
796
|
+
readonly validate: boolean;
|
|
797
|
+
readonly personaOf?: (scenario: TScenario) => string;
|
|
798
|
+
readonly corpusText?: (artifact: TArtifact, scenario: TScenario) => {
|
|
799
|
+
prompt: string;
|
|
800
|
+
completion: string;
|
|
801
|
+
} | undefined;
|
|
802
|
+
readonly rows: readonly ProfileMatrixRow[];
|
|
803
|
+
}
|
|
804
|
+
interface CreateProfileMatrixPlanOptions<TScenario extends Scenario, TArtifact> extends Pick<RunProfileMatrixOptions<TScenario, TArtifact>, 'profiles' | 'scenarios' | 'judges' | 'commitSha' | 'experimentId' | 'splitTag' | 'reps' | 'seed' | 'integrity' | 'allowMixed' | 'validate' | 'personaOf' | 'corpusText'> {
|
|
805
|
+
/** Stable dispatch implementation/configuration identity used by cell caches. */
|
|
806
|
+
dispatchRef: string;
|
|
807
|
+
}
|
|
808
|
+
interface ProfileMatrixCoverage {
|
|
809
|
+
expected: number;
|
|
810
|
+
present: number;
|
|
811
|
+
missing: string[];
|
|
812
|
+
failed: string[];
|
|
813
|
+
zeroScore: string[];
|
|
814
|
+
}
|
|
815
|
+
interface RunProfileMatrixSegmentOptions<TScenario extends Scenario, TArtifact> {
|
|
816
|
+
plan: ProfileMatrixPlan<TScenario, TArtifact>;
|
|
817
|
+
/** Stable external grant or attempt identity. Reuse it to resume. */
|
|
818
|
+
segmentId: string;
|
|
819
|
+
/** Explicit row ids from `plan.rows`; duplicate or unknown rows fail. */
|
|
820
|
+
rows: readonly (string | ProfileMatrixRow)[];
|
|
821
|
+
dispatch: ProfileDispatchFn<TScenario, TArtifact>;
|
|
822
|
+
runDir: string;
|
|
823
|
+
storage?: CampaignStorage;
|
|
824
|
+
maxConcurrency?: number;
|
|
825
|
+
maxProfileConcurrency?: number;
|
|
826
|
+
costCeiling?: number;
|
|
827
|
+
labeledStore?: RunProfileMatrixOptions<TScenario, TArtifact>['labeledStore'];
|
|
828
|
+
captureSource?: RunProfileMatrixOptions<TScenario, TArtifact>['captureSource'];
|
|
829
|
+
now?: () => Date;
|
|
830
|
+
}
|
|
831
|
+
interface ProfileMatrixSegmentResult<TArtifact, TScenario extends Scenario> {
|
|
832
|
+
segmentId: string;
|
|
833
|
+
rowIds: string[];
|
|
834
|
+
matrix: RunProfileMatrixResult<TArtifact, TScenario>;
|
|
835
|
+
coverage: ProfileMatrixCoverage;
|
|
836
|
+
}
|
|
837
|
+
interface FinalizeProfileMatrixOptions<TScenario extends Scenario, TArtifact> {
|
|
838
|
+
plan: ProfileMatrixPlan<TScenario, TArtifact>;
|
|
839
|
+
runDir: string;
|
|
840
|
+
storage?: CampaignStorage;
|
|
841
|
+
maxConcurrency?: number;
|
|
842
|
+
maxProfileConcurrency?: number;
|
|
843
|
+
now?: () => Date;
|
|
844
|
+
}
|
|
845
|
+
interface FinalizedProfileMatrixResult<TArtifact, TScenario extends Scenario> extends RunProfileMatrixResult<TArtifact, TScenario> {
|
|
846
|
+
coverage: ProfileMatrixCoverage;
|
|
847
|
+
}
|
|
848
|
+
declare function createProfileMatrixPlan<TScenario extends Scenario, TArtifact>(opts: CreateProfileMatrixPlanOptions<TScenario, TArtifact>): ProfileMatrixPlan<TScenario, TArtifact>;
|
|
849
|
+
declare function runProfileMatrixSegment<TScenario extends Scenario, TArtifact>(opts: RunProfileMatrixSegmentOptions<TScenario, TArtifact>): Promise<ProfileMatrixSegmentResult<TArtifact, TScenario>>;
|
|
850
|
+
declare function finalizeProfileMatrix<TScenario extends Scenario, TArtifact>(opts: FinalizeProfileMatrixOptions<TScenario, TArtifact>): Promise<FinalizedProfileMatrixResult<TArtifact, TScenario>>;
|
|
851
|
+
//#endregion
|
|
755
852
|
//#region src/campaign/run-dir.d.ts
|
|
756
853
|
/** The shared, out-of-repo root for campaign/benchmark run bundles. Keeping run
|
|
757
854
|
* outputs here means they never land in a repo working tree (no per-repo
|
|
@@ -1413,5 +1510,5 @@ declare function verifyCodeSurface(surface: CodeSurface, worktreeDir?: string):
|
|
|
1413
1510
|
* identity against the checkout at `worktreeRef`. */
|
|
1414
1511
|
declare function resolveWorktreePath(surface: CodeSurface, worktreeDir?: string): string;
|
|
1415
1512
|
//#endregion
|
|
1416
|
-
export { SearchPlannedEvent as $,
|
|
1417
|
-
//# sourceMappingURL=index-
|
|
1513
|
+
export { SearchPlannedEvent as $, CrossSurfacePairEvidence as $n, RunProfileMatrixResult as $t, SearchAccountingAudit as A, CrossSurfaceAdditionRejectionReason as An, ProfileMatrixSegmentResult as At, SearchCostAccounting as B, CrossSurfaceComponentEvidence as Bn, ScoreboardRow as Bt, surfaceHash as C, discoverEvalFixtures as Cn, tangleTracesRoot as Ct, FileSearchLedger as D, analyzeCrossSurfaceInteractions as Dn, ProfileMatrixCoverage as Dt, acquireSingleRunLock as E, planEvalFixtureRun as En, FinalizedProfileMatrixResult as Et, SearchCandidateRegisteredEvent as F, CrossSurfaceCandidateComparison as Fn, runProfileMatrixSegment as Ft, SearchLedgerEvent as G, CrossSurfaceIneligibilityReason as Gn, renderScoreboardMarkdown as Gt, SearchLedger as H, CrossSurfaceDistribution as Hn, UserStory as Ht, SearchCandidateSlot as I, CrossSurfaceCandidateEvidence as In, PlaybackContext as It, SearchLedgerTrustedHeadMode as J, CrossSurfaceInteractionPath as Jn, userStoryScoreboard as Jt, SearchLedgerHash as K, CrossSurfaceInteractionAwareSelection as Kn, scoreUserStory as Kt, SearchCandidateSlotClosedEvent as L, CrossSurfaceCandidateOutcome as Ln, PlaybackDriver as Lt, SearchAttemptAccounting as M, CrossSurfaceBestSingleSelection as Mn, SegmentedProfileMatrixError as Mt, SearchCandidateDecidedEvent as N, CrossSurfaceBootstrapPolicy as Nn, createProfileMatrixPlan as Nt, OpenSearchLedgerOptions as O, AnalyzeCrossSurfaceInteractionsInput as On, ProfileMatrixPlan as Ot, SearchCandidateLineage as P, CrossSurfaceCandidate as Pn, finalizeProfileMatrix as Pt, SearchPlan as Q, CrossSurfacePairCompatibility as Qn, RunProfileMatrixOptions as Qt, SearchCandidateSurface as R, CrossSurfaceCandidateSummary as Rn, PlaybackStep as Rt, surfaceContentHash as S, PlanEvalFixtureRunOptions as Sn, resolveRunDir as St, SingleRunLockOptions as T, loadEvalFixtureScenarios as Tn, FinalizeProfileMatrixOptions as Tt, SearchLedgerAppendResult as U, CrossSurfaceEligibility as Un, UserStoryVerdict as Ut, SearchFailureReason as V, CrossSurfaceCompositionStep as Vn, ScoreboardSummary as Vt, SearchLedgerEntry as W, CrossSurfaceEvidenceBreakdown as Wn, makePlaybackDispatch as Wt, SearchOperationKind as X, CrossSurfaceInteractionTask as Xn, ProfileMatrixError as Xt, SearchModelIdentity as Y, CrossSurfaceInteractionReport as Yn, ProfileDispatchFn as Yt, SearchOperationRecordedEvent as Z, CrossSurfaceNaiveStackSelection as Zn, ProfileSummary as Zt, assertCodeSurfaceIdentity as _, EvalFixtureLoadOptions as _n, compareRankKeys as _t, WorktreeAdapterError as a, LabeledScenarioStoreError as an, CrossSurfaceSelections as ar, SearchSurfaceKind as at, componentSurfaceIdentityMaterial as b, EvalFixtureValidationMode as bn, scoreDiscrimination as bt, verifyCodeSurface as c, RolloutCall as cn, TraceAnalystArtifact as cr, SearchTokenAccounting as ct, PhoenixEvaluationResultLike as d, classifyUngroundedLiterals as dn, traceAnalystQualityJudge as dr, SearchLedgerConflictError as dt, ScenarioRollup as en, CrossSurfacePairIncompatibilityReason as er, SearchPlannedOperation as et, PhoenixEvaluatorLike as f, rolloutArgumentDiff as fn, SearchLedgerError as ft, isTransientTransportFailure as g, EvalFixtureFile as gn, campaignMeanComposite as gt, TransientFailureOptions as h, EvalFixture as hn, campaignBreakdown as ht, WorktreeAdapter as i, FsLabeledScenarioStoreOptions as in, CrossSurfaceSelectionPolicy as ir, SearchSurfaceEvidence as it, SearchArtifactRef as j, CrossSurfaceAttemptCompleteness as jn, RunProfileMatrixSegmentOptions as jt, SEARCH_LEDGER_SCHEMA as k, CrossSurfaceAdditionDecision as kn, ProfileMatrixRow as kt, AutoevalsScoreLike as l, ScoredRollout as ln, TraceAnalystScenario as lr, openSearchLedger as lt, phoenixEvaluatorJudge as m, neutralizationGate as mn, CampaignBreakdown as mt, GitWorktreeAdapterOptions as n, neutralizeText as nn, CrossSurfaceRankedSingle as nr, SearchSourceRef as nt, gitWorktreeAdapter as o, RolloutArgumentDiff as on, CrossSurfaceTaskRow as or, SearchTaskAttemptedEvent as ot, autoevalsScorerJudge as p, NeutralizationGateOptions as pn, SearchLedgerIntegrityError as pt, SearchLedgerReplay as q, CrossSurfaceInteractionEffect as qn, scoreboardSummary as qt, Worktree as r, FsLabeledScenarioStore as rn, CrossSurfaceRelativeCost as rr, SearchSurfaceEffect as rt, resolveWorktreePath as s, RolloutArgumentDiffOptions as sn, BuildTraceAnalystSurfaceDispatchOptions as sr, SearchTaskOutcome as st, CodeSurfaceVerification as t, runProfileMatrix as tn, CrossSurfacePairwiseEntry as tr, SearchPlannedTask as tt, AutoevalsScorerLike as u, UngroundedLiteralReport as un, buildTraceAnalystSurfaceDispatch as ur, validateSearchLedgerEvent as ut, assertComponentSurface as v, EvalFixtureRunPlan as vn, DiscriminationScore as vt, SingleRunLock as w, loadEvalFixture as wn, CreateProfileMatrixPlanOptions as wt, renderSurfaceDiff as x, LoadEvalFixtureScenariosOptions as xn, selectDiscriminative as xt, codeSurfaceIdentityMaterial as y, EvalFixtureScenario as yn, ScenarioSignal as yt, SearchCompletedEvent as z, CrossSurfaceComponent as zn, ScoreboardRenderOptions as zt };
|
|
1514
|
+
//# sourceMappingURL=index-DuKK7tJR.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"index-DuKK7tJR.d.ts","names":[],"sources":["../src/campaign/analyst-surface.ts","../src/campaign/cross-surface-types.ts","../src/campaign/cross-surface-interaction.ts","../src/campaign/fixtures.ts","../src/campaign/gates/neutralization-gate.ts","../src/campaign/grounded-reflection.ts","../src/campaign/labeled-store/fs-adapter.ts","../src/campaign/neutralize.ts","../src/campaign/presets/run-profile-matrix.ts","../src/campaign/presets/playback.ts","../src/campaign/presets/segmented-profile-matrix.ts","../src/campaign/run-dir.ts","../src/campaign/scenario-selection.ts","../src/campaign/score-utils.ts","../src/campaign/search-ledger-errors.ts","../src/campaign/search-ledger.ts","../src/campaign/single-run-lock.ts","../src/campaign/surface-identity.ts","../src/campaign/transient-failure.ts","../src/campaign/upstream-evaluators.ts","../src/campaign/worktree/index.ts"],"mappings":";;;;;;;;;;;;;;;;;UAWiB,6BAA6B;EAC5C;EACA,YAAY;EACZ,YAAY;EACZ,yBAAyB;EACzB,kBAAkB;;UAGH;EACf,mBAAmB;EACnB,QAAQ;EACR,MAAM;;UAGS;EACf,QAAQ;IACN;IACA,YAAY;IACZ;IACA,QAAQ;MACN,QAAQ;;iBAGE,iCACd,SAAS,2CAET,SAAS,gBACT,UAAU,sBACV,SAAS,oBACN,QAAQ;iBAgBG,4BAA4B,YAC1C,sBACA;;;;KCtDU;;UAGK;EACf;EACA;;EAEA;;;UAIe;EACf;EACA;EACA;EACA;;;UAIe;EACf;;EAEA;;EAEA;;;;;;;UAQe;EACf;EACA;;EAEA;EACA,cAAc;EACd;EACA;;;;;;EAMA,MAAM;EACN,mBAAmB;;EAEnB;;UAGe;EACf;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;;EAEA,kCAAkC;;EAElC;;UAGe,qCACf,aAAa,sBAAsB;EAEnC,qBAAqB;EACrB,qBAAqB;EACrB,eAAe;EACf;;EAEA;;EAEA;;EAEA;;EAEA;EACA,WAAW;EACX,WAAW;;UAGI;EACf;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;;UAGe;EACf,aAAa;EACb;EACA;EACA;EACA;;KAGU;UAWK;EACf;EACA,SAAS;;UAGM;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf,WAAW;EACX,SAAS;EACT,OAAO;EACP,OAAO,eAAe;EACtB,QAAQ;EACR,QAAQ;;EAER,sBAAsB;;EAEtB,aAAa;;UAGE;EACf;EACA;EACA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA,QAAQ;EACR,cAAc,eAAe;;UAGd;EACf;EACA;EACA;EACA;EACA;;UAGe;EACf;;EAEA;EACA;;UAGe;EACf,SAAS;EACT;EACA;EACA;EACA;EACA,eAAe;EACf,gBAAgB;;KAGN;UAYK;EACf;EACA,SAAS;EACT;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA,4BAA4B,iCAAiC;EAC7D,wBAAwB,eAAe;EACvC,QAAQ;EACR,QAAQ;EACR,aAAa;EACb,eAAe;;UAGA;EACf;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;;UAGM;;EAEf;EACA;;KAGU;UAaK;EACf;EACA;EACA;EACA;EACA;EACA,uBAAuB;EACvB;EACA;EACA,SAAS;;UAGM;EACf;EACA;EACA,YAAY;EACZ;;;UAIe;EACf;EACA;EACA;EACA;EACA,OAAO;;UAGQ;;EAEf;;EAEA;EACA;;EAEA;EACA;;EAEA,gBAAgB;;EAEhB,OAAO;;UAGQ;EACf,YAAY;EACZ,YAAY;EACZ,kBAAkB;;UAGH,8BACf,aAAa,sBAAsB;EAEnC;EACA;EACA;EACA;;EAEA,MAAM;EACN,iBAAiB;EACjB,iBAAiB;EACjB,YAAY;EACZ,UAAU;EACV,YAAY;;;;;;;;;iBCnRE,gCAAgC,aAAa,qBAC3D,OAAO,qCAAqC,QAC3C,8BAA8B;;;KC7CrB;UAEK;EACf;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA,OAAO;EACP;;UAGe,4BAA4B;EAC3C;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;;EAEf,aAAa;;EAEb;;UAGe,wCAAwC;EACvD;;UAGe,0BAA0B,6BACjC,KACN,uBAAuB,qBAAqB;EAG9C;EACA,aAAa;EACb;EACA;EACA,UAAU;;KAGA,qBAAqB;EAC/B,UAAU,MAAM,KAAK;;;iBAIP,qBAAqB;;;;;iBA4BrB,gBACd,kBACA,cACA,UAAS,yBACR;;iBA4Ca,yBACd,kBACA,UAAS,kCACR;;;;;iBAuBa,mBAAmB,qBACjC,SAAS,0BAA0B,aAClC;;;UChJc,0BAA0B,kBAAkB,WAAW;EACtE,WAAW;;;;;EAKX;;;;;;iBAuBc,mBAAmB,WAAW,kBAAkB,UAC9D,SAAS,0BAA0B,aAClC,KAAK,WAAW;;;;;;;;;;;;;;;;;;;;;;;;;;;;;UC9BF;WACN;WACA,MAAM,SAAS;;;UAIT;;WAEN;;WAEA;WACA,gBAAgB;;UAGV;;WAEN;;WAEA;;UAGM;;WAEN;;WAEA,eAAe;;WAEf,eAAe;;;;;;;;iBASV,oBACd,mBAAmB,iBACnB,OAAM,6BACL;UAsCc;;WAEN;;WAEA;;;;;;;;;iBAUK,2BACd,cACA,MAAM,KAAK,0DACV;;;UCjFc;;EAEf;;;EAGA;;EAEA;;;cAIW,kCAAkC;WAE3B;EADlB,YACkB,cAChB;;;;;;cAiBS,kCAAkC;mBAIhB;mBAHZ;mBACA;EAEjB,YAA6B,SAAS;EAKhC,QAAQ,OAAO,uBAAuB;EAYtC,OAAO,MAAM,4BAA4B,QAAQ;EA8CjD,QAAQ;IACZ;IACA;IACA,UAAU;IACV,SAAS,OAAO;;UAkCV;UAiCA;UAmBA;UAkBA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBCtNM,eAAe;;;;;;cC6ClB,2BAA2B;EACtC,YAAY;;;;;KAQF,kBAAkB,kBAAkB,UAAU,cACxD,SAAS,gBACT,UAAU,WACV,KAAK,oBACF,QAAQ;UAEI,wBAAwB,kBAAkB,UAAU;;EAEnE,UAAU;;EAEV,WAAW;;EAEX,UAAU,kBAAkB,WAAW;;EAEvC,SAAS,YAAY,WAAW;;;EAGhC;;;EAGA;;;EAGA;;;;EAIA;;EAEA,WAAW;;EAEX;;EAEA;;;;;;;;EAQA;;;EAGA;;EAEA;;;EAGA;;EAEA;;EAEA,eAAe;EACf,gBAAgB;;;EAGhB,UAAU;;EAEV,YAAY;;;EAGZ,aAAa,UAAU;;;;EAIvB;;;;;;;EAOA,cACE,UAAU,WACV,UAAU;IACL;IAAgB;;;;;;;EAMvB,aAAa;IAAS,SAAS;IAAc,UAAU;IAAW;;;EAElE;;EAEA;;UAGe;EACf;EACA;EACA;;EAEA;;EAEA;;EAEA;EACA,gBAAgB;;;EAGhB,WAAW;;UAGI;EACf;EACA;;UAGe,uBAAuB,WAAW,kBAAkB;EACnE;EACA;;;;EAIA,SAAS;EACT,WAAW,eAAe;EAC1B,YAAY,eAAe;;EAE3B,YAAY,eAAe;;EAE3B,WAAW;;EAEX,WAAW,eAAe,eAAe,WAAW;;;;;iBA6LhC,iBAAiB,kBAAkB,UAAU,WACjE,MAAM,wBAAwB,WAAW,aACxC,QAAQ,uBAAuB,WAAW;;;;;UCtX5B;;EAEf;;EAEA,UAAU;;;;;;;UAQK,kBAAkB;;EAEjC;;EAEA,OAAO;;EAEP,cAAc;;;UAIC,wBAAwB;EACvC,SAAS;;;;;;;;;;;UAYM,eAAe,eAAe,YAAY;EACzD,IAAI,OAAO,QAAQ,KAAK,kBAAkB,iBAAiB;;;;;;;iBAQ7C,qBAAqB,eAAe,WAClD,QAAQ,eAAe,UACtB,kBAAkB,QAAQ;;UAQZ,yBAAyB;EACxC;;;;;;;;iBASoB,eACpB,OAAO,WACP,OAAO,eACP,kBAAkB,qBACjB,QAAQ;;UAUM;EACf;EACA;EACA;EACA;EACA;EACA;;;;;;;iBAQc,oBAAoB,mBAAmB,qBAAqB;;UAkB3D;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;iBAIc,kBAAkB,eAAe,kBAAkB;UAwBlD;;EAEf;;EAEA,OAAO;;EAEP;;;;;;;;;iBAkBc,yBACd,eAAe,iBACf,OAAM;;;cC1KK,oCAAoC;EAC/C,YAAY;;UAKG;EACf;EACA;EACA;EACA;EACA;;UAGe,kBAAkB,kBAAkB,UAAU;WACpD;WACA;WACA;WACA;WACA,mBAAmB;WACnB,oBAAoB;WACpB,iBAAiB,YAAY,WAAW;WACxC;WACA;WACA,UAAU,YAAY,wBAAwB,WAAW;WACzD;;WAEA;WACA,WAAW,YAAY,wBAAwB,WAAW;WAC1D;WACA;WACA,aAAa,UAAU;WACvB,cACP,UAAU,WACV,UAAU;IACL;IAAgB;;WACd,eAAe;;UAGT,+BAA+B,kBAAkB,UAAU,mBAClE,KACN,wBAAwB,WAAW;;EAgBrC;;UAGe;EACf;EACA;EACA;EACA;EACA;;UAGe,+BAA+B,kBAAkB,UAAU;EAC1E,MAAM,kBAAkB,WAAW;;EAEnC;;EAEA,yBAAyB;EACzB,UAAU,kBAAkB,WAAW;EACvC;EACA,UAAU;EACV;EACA;EACA;EACA,eAAe,wBAAwB,WAAW;EAClD,gBAAgB,wBAAwB,WAAW;EACnD,YAAY;;UAGG,2BAA2B,WAAW,kBAAkB;EACvE;EACA;EACA,QAAQ,uBAAuB,WAAW;EAC1C,UAAU;;UAGK,6BAA6B,kBAAkB,UAAU;EACxE,MAAM,kBAAkB,WAAW;EACnC;EACA,UAAU;EACV;EACA;EACA,YAAY;;UAGG,6BAA6B,WAAW,kBAAkB,kBACjE,uBAAuB,WAAW;EAC1C,UAAU;;iBAGI,wBAAwB,kBAAkB,UAAU,WAClE,MAAM,+BAA+B,WAAW,aAC/C,kBAAkB,WAAW;iBA0HV,wBAAwB,kBAAkB,UAAU,WACxE,MAAM,+BAA+B,WAAW,aAC/C,QAAQ,2BAA2B,WAAW;iBAkD3B,sBAAsB,kBAAkB,UAAU,WACtE,MAAM,6BAA6B,WAAW,aAC7C,QAAQ,6BAA6B,WAAW;;;;;;;;iBCrTnC;;;;;iBAQA,cAAc,gBAAgB;;;;;;;;;;;;;;;;;UCD7B;EACf;;EAEA;;UAGe;EACf;;EAEA;;EAEA;EACA;;EAEA;;;;;;;;;;iBA4Cc,oBACd,SAAS,kBACT;EAAS;IACR;;;;;;;;;;;iBA0Ba,qBACd,SAAS,kBACT,WACA;EAAS;;;;;;;iBC3FK,sBAAsB,WAAW,kBAAkB,UACjE,UAAU,eAAe,WAAW;;;;iBA0CtB,gBAAgB,sBAAsB;UAWrC;;EAEf,YAAY;;;;;EAKZ,WAAW;IAAQ;IAAoB;IAAmB;IAAgB;;;;;iBAK5D,kBAAkB,WAAW,kBAAkB,UAC7D,UAAU,eAAe,WAAW,aACnC;;;;cC/EU,0BAA0B;;cAG1B,mCAAmC;;cAGnC,kCAAkC;;;cCgClC;KAED,mBAAmB;KAEnB;;;UAYK;EACf;EACA;EACA,QAAQ;EACR;;;;UAKe;EACf;EACA;;UAGe;EACf;EACA;;UAGe;EACf;EACA,MAAM;EACN,UAAU;;UAGK;;EAEf;EACA;EACA;EACA;EACA,gBAAgB;;KAGN;UAOK;EACf;EACA,QAAQ;EACR,WAAW;;;EAGX;;UAGe;EACf;EACA,MAAM;;UAGS;EACf;;;EAGA;;UAGe;;EAEf,gBAAgB;;EAEhB,OAAO;;EAEP,YAAY;;KAGF;EAEN;EACA;EACA;EACA;;EAGA;EACA;;KAGM;EAEN;EACA;EACA;;EAGA;;EAEA;EACA;;UAGW;EACf,QAAQ;EACR,MAAM;;UAGS;EACf;EACA;;KAGU;EAEN;EACA;EACA,SAAS;;EAGT;EACA;EACA,SAAS;EACT,SAAS;;EAGT;EACA,SAAS;EACT,OAAO;IAAwB;;;KAGzB;EAEN;EACA;EACA;EACA;EACA;;EAGA;EACA;;;;UAKW;EACf;EACA;EACA;EACA,QAAQ;EACR,UAAU;;UAGF;EACR;EACA;EACA,WAAW;;UAGI,2BAA2B;EAC1C;EACA,MAAM;;UAGS,uCAAuC;EACtD;EACA;EACA;EACA;EACA,SAAS;EACT,UAAU;;UAGK,uCAAuC;EACtD;EACA;EACA;EACA,QAAQ;;UAGO,iCAAiC;EAChD;EACA;EACA;EACA;EACA;IACE;IACA,QAAQ;;EAEV;IACE,OAAO;IACP,OAAO;IACP,WAAW;;EAEb,SAAS;EACT,YAAY;EACZ,iBAAiB;;UAGF,qCAAqC;EACpD;EACA;EACA,eAAe;EACf;IAEM;IACA,OAAO;IACP,QAAQ;;IAGR;IACA,QAAQ;;EAEd;IACM;;IACA;IAAmB,SAAS;;IAC5B;IAAkB,SAAS;;EACjC,YAAY;;UAGG,oCAAoC;EACnD;EACA;EACA;IACM;;IAEA;IACA,QAAQ;;;UAIC,6BAA6B;EAC5C;EACA;IAEM;IACA;;IAGA;IACA,QAAQ;;;KAIJ,oBACR,qBACA,iCACA,iCACA,2BACA,+BACA,8BACA;UAEa;EACf,eAAe;EACf;EACA;EACA,cAAc;EACd,OAAO;EACP,WAAW;;KAGD;EAEN;EACA;EACA;EACA;EACA;;EAGA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGW;EACf;EACA;EACA;EACA;EACA;EACA;EACA;IAAY;IAAgB;IAAgB;;EAC5C;IAAqB;IAAmB;IAAiB;;EACzD;IAAa;IAAkB;IAAkB;;EACjD;IACE;IACA;IACA;IACA;IACA;IACA;;EAEF;EACA;EACA,YAAY;EACZ,UAAU;;UAGK;EACf,SAAS;EACT,MAAM;EACN,YAAY;EACZ,sBAAsB;EACtB,UAAU;EACV,YAAY;EACZ,WAAW;EACX,YAAY;EACZ,OAAO;;UAGQ;EACf,OAAO;;EAEP;EACA,QAAQ;;;;iBA+ZM,0BAA0B,iBAAiB;;;;;;;;;;;;;;;KAsB/C;UAEK;EACf;EACA;EACA,cAAc;;UAGC;WACN;WACA;;WAEA;EACT,OAAO,OAAO,oBAAoB,QAAQ;EAC1C,UAAU,QAAQ;;EAElB,eAAe,QAAQ;;;EAGvB,kBAAkB,QAAQ;;;;;;;EAO1B,oBAAoB,QAAQ;;;;iBAKd,iBAAiB,SAAS,0BAA0B;;cA8BvD,4BAA4B;WAC9B;WACA;WACA;mBACQ;mBACA;EAMjB,YAAY,cAAc,oBAAoB,cAAa;EAerD,UAAU,QAAQ;EAIlB,OAAO,OAAO,oBAAoB,QAAQ;EAU1C,eAAe,QAAQ;EAIvB,kBAAkB,QAAQ;EAI1B,oBAAoB,QAAQ;;;;;;;;;;;;;;;;;;;UC/3BnB;;WAEN;;WAEA;;WAEA;;WAEA;;UAGM;;EAEf;;;;;;;iBAwBc,qBAAqB,MAAM,uBAAuB;;;;iBCpDlD,0BAA0B,2BAA2B,WAAW;;iBAmChE,uBAAuB,2BAA2B,WAAW;;iBA8B7D,iCAAiC,SAAS;;;;iBAa1C,4BAA4B,SAAS;;iBAgBrC,mBAAmB,SAAS;;iBAW5B,YAAY,SAAS;;iBAKrB,kBACd,eAAe,gBACf,iBAAiB;;;;;;;;;;;;;;;;;;;;;UCpGF;;;;;;;WAON;;WAEA,yBAAyB;;;;;;iBAYpB,4BACd,oCACA,OAAM;;;UCnCS;EACf;EACA;EACA;;UAGe,qBAAqB,gBAAgB;EACpD;EACA;EACA;EACA,SACE,QAAQ,SACR,SAAS,4BACR,QAAQ;;UAGI;EACf;EACA;EACA,WAAW;;KAGD,oBAAoB,eAAe,4BAC7C,OAAO,QACP,SAAS,8BACN,qBAAqB,QAAQ;UAEjB;WACN,QAAQ;WACR;;KAGN,sBAAsB,WAAW,KACpC,iBAAiB;EAGjB;;UAGQ,qBAAqB,kBAAkB;EAC/C;EACA;EACA;EACA,aAAa,UAAU;;EAEvB,eAAe;;iBAGD,sBACd,gBAAgB,yBAChB,WACA,kBAAkB,WAAW,UAE7B,WAAW,qBAAqB,UAChC,SAAS,qBAAqB;EAC5B,SAAS;IAAS,UAAU;IAAW,UAAU;MAAc;EAC/D,WAAW,sBAAsB;IAElC,YAAY,WAAW;iBA8CV,qBACd,eAAe,yBACf,WACA,kBAAkB,WAAW,UAE7B,QAAQ,oBAAoB,SAC5B,SAAS,qBAAqB;EAC5B;EACA,SAAS;IAAS,UAAU;IAAW,UAAU;MAAc;;EAEzD;EAAc;;EACd;EAAa,UAAU,sBAAsB;KAEpD,YAAY,WAAW;;;KCxFrB,qBAAqB;KACrB,iBAAiB,SAAS;KAC1B,aAAa,gBAAgB,aAAa,MAAM,mBAAmB;UAcvD;;WAEN;;WAEA;;WAEA;;WAEA;;WAEA;;UAGM;;EAEf,OAAO;IAAQ;IAAiB;MAAkB,QAAQ;;;EAG1D,SAAS,UAAU,UAAU,kBAAkB,QAAQ;;EAEvD,QAAQ,UAAU,WAAW;;;cAIlB,6BAA6B;WAG7B;EAFX,YACE,iBACS;;UAOI;;EAEf;;EAEA;;EAEA;;;;EAIA,MAAM;;UAobS;;EAEf;;EAEA;;EAEA;;;EAGA,YAAY;;;;;iBAoJE,mBAAmB,MAAM,4BAA4B;;;;iBA0FrD,kBACd,SAAS,aACT,uBACC;;;iBAOa,oBAAoB,SAAS,aAAa"}
|
package/dist/index.d.ts
CHANGED
|
@@ -35,10 +35,10 @@ import { o as InsightReport } from "./insight-report-CRi-Ufrj.js";
|
|
|
35
35
|
import { C as analyzeRuns, b as AnalyzeRunsOptions, d as RawAnalystFinding, i as TraceAnalysisEngineResult, n as TraceAnalysisEngine } from "./engine-otFpE2gF.js";
|
|
36
36
|
import { o as MintedRolloutLine } from "./schema-Cef2cFmb.js";
|
|
37
37
|
import { i as BenchmarkEvaluation } from "./types-BEPZc6eo.js";
|
|
38
|
-
import { A as RedTeamCategory, At as llmJudge, B as CanaryReport, F as scoreRedTeamOutput, I as CanaryAlert, L as CanaryEvaluation, M as RedTeamReport, N as redTeamDataset, O as DEFAULT_RED_TEAM_CORPUS, P as redTeamReport, R as CanaryKind, V as runCanaries, at as RunCampaignOptions, it as CampaignCellFailureReceipt, j as RedTeamFinding, k as RedTeamCase, kt as LlmJudgeOptions, ot as runCampaign, z as CanaryOptions } from "./provenance-
|
|
38
|
+
import { A as RedTeamCategory, At as llmJudge, B as CanaryReport, F as scoreRedTeamOutput, I as CanaryAlert, L as CanaryEvaluation, M as RedTeamReport, N as redTeamDataset, O as DEFAULT_RED_TEAM_CORPUS, P as redTeamReport, R as CanaryKind, V as runCanaries, at as RunCampaignOptions, it as CampaignCellFailureReceipt, j as RedTeamFinding, k as RedTeamCase, kt as LlmJudgeOptions, ot as runCampaign, z as CanaryOptions } from "./provenance-C_ln89b0.js";
|
|
39
39
|
import { a as paretoFrontier, i as dominates, n as Objective, r as ParetoResult } from "./pareto-BqNW3LJR.js";
|
|
40
40
|
import { n as SandboxDriver } from "./sandbox-harness-BlSOu4LX.js";
|
|
41
|
-
import { a as DefinedAgentEval, c as SelfImproveOptions, f as selfImprove, i as DefineAgentEvalOptions, o as defineAgentEval, u as SelfImproveResult } from "./define-agent-eval-
|
|
41
|
+
import { a as DefinedAgentEval, c as SelfImproveOptions, f as selfImprove, i as DefineAgentEvalOptions, o as defineAgentEval, u as SelfImproveResult } from "./define-agent-eval-DcAcC1yQ.js";
|
|
42
42
|
import { a as PairedEvalueStep, c as pairedEvalueSequence, i as PairedEvalueSequence, n as InterimReleaseConfidenceInput, o as SequentialDecision, r as PairedEvalueOptions, s as evaluateInterimReleaseConfidence, t as InterimReleaseConfidence } from "./sequential-CYwq6Ff_.js";
|
|
43
43
|
import { n as TrajectoryStep, r as buildTrajectory, t as Trajectory } from "./trajectory-YC15QDYQ.js";
|
|
44
44
|
import { a as runCounterfactual, i as CounterfactualRunner, n as CounterfactualMutation, r as CounterfactualResult, t as CounterfactualContext } from "./counterfactual--bpysZF0.js";
|
package/dist/index.js
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import { t as __exportAll } from "./rolldown-runtime-8H4AJuhK.js";
|
|
2
2
|
import { i as JudgeError, o as NotFoundError, r as ConfigError, s as ValidationError, t as AgentEvalError } from "./errors-Dngq5h35.js";
|
|
3
3
|
import { r as hashJson, t as canonicalize } from "./pre-registration-DakwTRXk.js";
|
|
4
|
-
import { a as CODING_HARNESSES, c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, n as completionVerdict, o as HARNESS_NATIVE_MODEL, r as createLlmCorrectnessChecker, s as agentProfileHash, t as extractProducedState, u as expandProfileAxes } from "./produced-state-
|
|
4
|
+
import { a as CODING_HARNESSES, c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, n as completionVerdict, o as HARNESS_NATIVE_MODEL, r as createLlmCorrectnessChecker, s as agentProfileHash, t as extractProducedState, u as expandProfileAxes } from "./produced-state-B_-TW9o0.js";
|
|
5
5
|
import { AGENT_PROFILE_KINDS, agentProfileCellHashMaterial, agentProfileCellKey, buildAgentProfileCell, groupRunsByAgentProfileCell, toAgentProfileJson, verifyAgentProfileCell } from "./profile-cell.js";
|
|
6
6
|
import { c as mulberry32 } from "./internal-BDHPCnjk.js";
|
|
7
7
|
import { a as spearmanR, i as ranks, n as partialCredit, o as weightedComposite, r as pearsonR, s as weightedMean, t as confidenceInterval } from "./descriptive-B5MwKfbf.js";
|
|
@@ -15,12 +15,12 @@ import { t as eProcess } from "./sequential-eprocess-CbUt2htw.js";
|
|
|
15
15
|
import { n as iqr, r as welchsTTest } from "./baseline-BhPRQBVn.js";
|
|
16
16
|
import { i as isLlmSpan, r as isJudgeSpan, s as isToolSpan, t as FAILURE_CLASSES } from "./schema-CRhEY1SO.js";
|
|
17
17
|
import { a as judgeSpans, c as runsForScenario, n as argHash } from "./query-Di7eEQ79.js";
|
|
18
|
-
import { i as analyzeRuns, o as checkCanaries, r as selfImprove, t as defineAgentEval } from "./define-agent-eval-
|
|
18
|
+
import { i as analyzeRuns, o as checkCanaries, r as selfImprove, t as defineAgentEval } from "./define-agent-eval-xaceIpPS.js";
|
|
19
19
|
import { r as observedSplitScore, t as isRealnessGated } from "./reward-nw2xZGZG.js";
|
|
20
20
|
import { a as roundTripRunRecord, i as parseRunRecordSafe, n as isRunRecord, o as runTaskScore, r as modelHasSnapshot, s as validateRunRecord, t as RunRecordValidationError } from "./run-record-BvHPVS-i.js";
|
|
21
21
|
import { a as summaryTable, n as gainHistogram, r as paretoChart } from "./summary-report-Blysd6Z2.js";
|
|
22
22
|
import { n as contentHash, r as fileVerdictCache, t as canonicalJson } from "./verdict-cache-mZf5FEiY.js";
|
|
23
|
-
import { B as DEFAULT_RED_TEAM_CORPUS, F as parseReflectionResponse, H as redTeamReport, K as runCampaign, P as buildReflectionPrompt, Q as summarizeBackendIntegrity, U as scoreRedTeamOutput, V as redTeamDataset, W as runCanaries, X as BackendIntegrityError, Z as assertRealBackend, _ as paretoFrontier, g as dominates, k as surfaceContentHash, t as llmJudge } from "./llm-judge-
|
|
23
|
+
import { B as DEFAULT_RED_TEAM_CORPUS, F as parseReflectionResponse, H as redTeamReport, K as runCampaign, P as buildReflectionPrompt, Q as summarizeBackendIntegrity, U as scoreRedTeamOutput, V as redTeamDataset, W as runCanaries, X as BackendIntegrityError, Z as assertRealBackend, _ as paretoFrontier, g as dominates, k as surfaceContentHash, t as llmJudge } from "./llm-judge-BPKBdQTp.js";
|
|
24
24
|
import { a as resolveModelPricing, i as isModelPriced, n as estimateCost, r as estimateTokens, t as MODEL_PRICING } from "./metrics-Cl0L1KUy.js";
|
|
25
25
|
import { a as CostLedgerPersistenceError, c as costForTokenPricing, i as CostLedger, l as costForUsage, n as CostCallConflictError, o as CostReceiptCaptureError, r as CostCeilingReachedError, s as CostReservationExceededError, t as CostAccountingIncompleteError, u as modelPriceKey } from "./cost-ledger-BSe92yAV.js";
|
|
26
26
|
import { n as REDACTION_VERSION, r as redactString, t as DEFAULT_REDACTION_RULES } from "./redact-7Aq1ukl-.js";
|
|
@@ -580,7 +580,7 @@ async function executeCell(args) {
|
|
|
580
580
|
cellId: args.slot.cellId,
|
|
581
581
|
reason: cached.reason
|
|
582
582
|
}]);
|
|
583
|
-
if (cached.status === "hit") {
|
|
583
|
+
if (cached.status === "hit" && (!cached.cell.error || args.opts.reuseFailedCells)) {
|
|
584
584
|
const receiptProblem = cachedCellReceiptProblem(cached.cell, args.costLedger, stableCostTags);
|
|
585
585
|
if (receiptProblem === void 0) {
|
|
586
586
|
const cell = withCurrentAgentModelEvidence(cached.cell, args.costLedger, stableCostTags);
|
|
@@ -1067,7 +1067,10 @@ async function runCampaign(opts) {
|
|
|
1067
1067
|
const runAttemptId = globalThis.crypto.randomUUID();
|
|
1068
1068
|
const cells = [];
|
|
1069
1069
|
const artifactsByPath = {};
|
|
1070
|
-
const schedule = buildCellSchedule(opts.scenarios, seed, reps)
|
|
1070
|
+
const schedule = buildCellSchedule(opts.scenarios, seed, reps).filter((slot) => opts.cellFilter ? opts.cellFilter({
|
|
1071
|
+
scenario: slot.scenario,
|
|
1072
|
+
rep: slot.rep
|
|
1073
|
+
}) : true);
|
|
1071
1074
|
if (resumable) assertScheduleCachesReusable({
|
|
1072
1075
|
schedule,
|
|
1073
1076
|
runDir: opts.runDir,
|
|
@@ -4855,6 +4858,6 @@ function firstString(value) {
|
|
|
4855
4858
|
return typeof value === "string" && value.trim() ? value : void 0;
|
|
4856
4859
|
}
|
|
4857
4860
|
//#endregion
|
|
4858
|
-
export { assertCampaignDesign as $, surfaceHash as A, DEFAULT_RED_TEAM_CORPUS as B, optimizationTokenUsageFromSummary as C, componentSurfaceIdentityMaterial as D, codeSurfaceIdentityMaterial as E, parseReflectionResponse as F, runEval as G, redTeamReport as H, recoverTruncatedJson as I, resolveRunDir as J, runCampaign as K, assertGepaCandidatePopulationSummary as L, campaignMeanComposite as M, compareRankKeys as N, renderSurfaceDiff as O, buildReflectionPrompt as P, summarizeBackendIntegrity as Q, readGepaCandidatePopulationArtifact as R, costFromLedgerSummary as S, assertComponentSurface as T, scoreRedTeamOutput as U, redTeamDataset as V, runCanaries as W, BackendIntegrityError as X, tangleTracesRoot as Y, assertRealBackend as Z, paretoFrontier as _, canonicalDigest as a, combineComparisonCosts as b, loopProvenanceSpans as c, verifyLoopProvenanceRecord as d, assertCampaignSplitIdentity as et, runImprovementLoop as f, dominates as g, labelTrustRank as h, campaignMeasurementDigest as i, campaignBreakdown as j, surfaceContentHash as k, provenanceRecordPath as l, isProposedCandidate as m, JudgeParseError as n, campaignSplitDigest as nt, emitLoopProvenance as o, runOptimization as p, planCampaignRun as q, buildLoopProvenanceRecord as r, campaignSplitDigestFromIdentities as rt, loopProvenanceArgsFromResult as s, llmJudge as t, campaignScenarioIdentity as tt, provenanceSpansPath as u, openAutoPr as v, assertCodeSurfaceIdentity as w, compareOptimizationMethods as x, assertOptimizationResult as y, defaultProductionGate as z };
|
|
4861
|
+
export { assertCampaignDesign as $, surfaceHash as A, DEFAULT_RED_TEAM_CORPUS as B, optimizationTokenUsageFromSummary as C, componentSurfaceIdentityMaterial as D, codeSurfaceIdentityMaterial as E, parseReflectionResponse as F, runEval as G, redTeamReport as H, recoverTruncatedJson as I, resolveRunDir as J, runCampaign as K, assertGepaCandidatePopulationSummary as L, campaignMeanComposite as M, compareRankKeys as N, renderSurfaceDiff as O, buildReflectionPrompt as P, summarizeBackendIntegrity as Q, readGepaCandidatePopulationArtifact as R, costFromLedgerSummary as S, assertComponentSurface as T, scoreRedTeamOutput as U, redTeamDataset as V, runCanaries as W, BackendIntegrityError as X, tangleTracesRoot as Y, assertRealBackend as Z, paretoFrontier as _, canonicalDigest as a, computeManifestHash as at, combineComparisonCosts as b, loopProvenanceSpans as c, verifyLoopProvenanceRecord as d, assertCampaignSplitIdentity as et, runImprovementLoop as f, dominates as g, labelTrustRank as h, campaignMeasurementDigest as i, cellCachePath as it, campaignBreakdown as j, surfaceContentHash as k, provenanceRecordPath as l, isProposedCandidate as m, JudgeParseError as n, campaignSplitDigest as nt, emitLoopProvenance as o, runOptimization as p, planCampaignRun as q, buildLoopProvenanceRecord as r, campaignSplitDigestFromIdentities as rt, loopProvenanceArgsFromResult as s, llmJudge as t, campaignScenarioIdentity as tt, provenanceSpansPath as u, openAutoPr as v, assertCodeSurfaceIdentity as w, compareOptimizationMethods as x, assertOptimizationResult as y, defaultProductionGate as z };
|
|
4859
4862
|
|
|
4860
|
-
//# sourceMappingURL=llm-judge-
|
|
4863
|
+
//# sourceMappingURL=llm-judge-BPKBdQTp.js.map
|