@tangle-network/agent-eval 0.133.0 → 0.133.2

package/dist/{index-BAvgST_9.d.ts → index-nhIYz9hn.d.ts}

@@ -1,3 +1,4 @@
+import { c as ValidationError } from "./errors-CEk209JS.js";
 //#region src/ledger-core/atomic-file-lock.d.ts
 /** Crash-safe filesystem lock shared by campaign persistence and run exclusion. */
 interface AtomicFileLockOwner {
@@ -38,6 +39,18 @@ declare function probeAtomicFileLock(options: AtomicFileLockOptions): AtomicFile
  */
 declare function tryAcquireAtomicFileLock(options: AtomicFileLockOptions): AtomicFileLockAcquisition;
 //#endregion
+//#region src/ledger-core/canonical.d.ts
+type LedgerHash = `sha256:${string}`;
+/** Shape of every ledger digest: `sha256:` followed by 64 lowercase hex chars. */
+declare const LEDGER_HASH_PATTERN: RegExp;
+/** A value has no faithful canonical-JSON form, so it cannot be stored or hashed. */
+declare class LedgerCanonicalizationError extends ValidationError {}
+/** Canonical JSON encoding (RFC 8785): object keys sorted by UTF-16 code unit,
+ * recursively. This is the byte form that is hashed and stored, so equal values
+ * always encode identically. */
+declare function canonicalString(value: unknown): string;
+declare function hashCanonical(value: unknown): LedgerHash;
+//#endregion
 //#region src/ledger-core/journal-file.d.ts
 /** Filesystem durability and cross-process exclusion for append-only journal files. */
 /** How a journal binding reports file-layer faults in its own error taxonomy. */
@@ -51,6 +64,10 @@ interface LedgerFileContext {
 /** Append one already-serialized line; the write is fsynced (file and directory)
  * before returning so an acknowledged append survives a crash. */
 declare function appendLedgerLine(path: string, line: string, context: LedgerFileContext): void;
+/** Replace a small sidecar file whole: write a temporary sibling, fsync it,
+ * then rename over the target. A reader therefore sees the previous contents
+ * or the new ones, never a partial file. */
+declare function writeLedgerFileAtomically(path: string, contents: string, context: LedgerFileContext): void;
 declare function withLedgerFileLock<T>(path: string, context: LedgerFileContext, run: () => T): T;
 type FileLockResult<T> = {
   acquired: true;
@@ -60,12 +77,26 @@ type FileLockResult<T> = {
 };
 declare function tryWithLedgerFileLock<T>(path: string, context: LedgerFileContext, run: () => T): FileLockResult<T>;
 //#endregion
+//#region src/ledger-core/trusted-head.d.ts
+/** A `(sequence, entryHash)` pair a writer pinned outside the journal. */
+interface LedgerTrustedHead {
+  sequence: number;
+  entryHash: LedgerHash;
+}
+/** Minimum entry shape the anchor check needs. */
+interface LedgerAnchoredEntry {
+  sequence: number;
+  entryHash: LedgerHash;
+}
+/** Sibling pin file for a journal — deliberately NOT the journal, so rewriting
+ * the log is not enough to rewrite the trust recorded about it. */
+declare function trustedHeadPathFor(journalPath: string): string;
+/** The entry at exactly `head.sequence` must still carry `head.entryHash`.
+ * Entries must already have passed chain verification, which is what makes
+ * index and sequence interchangeable here. */
+declare function verifyEntriesAgainstTrustedHead(entries: readonly LedgerAnchoredEntry[], head: LedgerTrustedHead, context: LedgerFileContext, subject: string): void;
+//#endregion
 //#region src/ledger-core/journal.d.ts
-type LedgerHash = `sha256:${string}`;
-/** Canonical JSON encoding: object keys sorted recursively. This is the byte
- * form that is hashed and stored, so identical values always encode identically. */
-declare function canonicalString(value: unknown): string;
-declare function hashCanonical(value: unknown): LedgerHash;
 /** Minimum shape of a journal event: the idempotency key. */
 interface LedgerEventBase {
   eventId: string;
@@ -113,19 +144,46 @@ interface LedgerAppendResult<Entry, Projection> {
   appended: boolean;
   projection: Projection;
 }
+interface LedgerAppendOptions {
+  /** Record the appended entry as this journal's trusted head, so a later read
+   * can prove nothing was deleted from the end. Ignored on the idempotent
+   * path: acknowledging an event that was already durable writes nothing, and
+   * moving the pin there would jump trust forward past entries this caller
+   * never saw. */
+  pinHead?: boolean;
+}
+interface FileLedgerJournalOptions {
+  /** Refuse to read a non-empty journal that has no trusted head.
+   *
+   * Off by default because journals written before pinning have no pin and
+   * must keep opening. Turn it on for a journal this process pins: without it,
+   * deleting the sibling pin file silently downgrades the journal back to a
+   * chain that cannot detect deletion. */
+  requireTrustedHead?: boolean;
+}
 /** Durable filesystem journal. Construction performs no I/O; `append` and
  * `replay` validate the complete existing file under the locks. */
 declare class FileLedgerJournal<Header extends object, Event extends LedgerEventBase, Projection> {
   readonly path: string;
+  /** Sibling file holding this journal's trusted head. */
+  readonly trustedHeadPath: string;
   private readonly codec;
   private readonly mutex;
-  constructor(path: string, codec: LedgerJournalCodec<Header, Event, Projection>);
+  private readonly requireTrustedHead;
+  constructor(path: string, codec: LedgerJournalCodec<Header, Event, Projection>, options?: FileLedgerJournalOptions);
   replay(): Promise<LedgerReplayResult<LedgerEntryOf<Header, Event>, Projection>>;
-  append(event: Event): Promise<LedgerAppendResult<LedgerEntryOf<Header, Event>, Projection>>;
+  append(event: Event, options?: LedgerAppendOptions): Promise<LedgerAppendResult<LedgerEntryOf<Header, Event>, Projection>>;
+  /** The pinned head, or null when this journal has never been pinned. */
+  trustedHead(): Promise<LedgerTrustedHead | null>;
+  /** Pin the current verified head. The chain and any existing pin are checked
+   * first, so the pin only ever moves forward along a journal that still
+   * carries the history it already recorded. */
+  pinTrustedHead(): Promise<LedgerTrustedHead>;
   private replayLocked;
+  private verifyTrustedHead;
   private readEntries;
   private project;
 }
 //#endregion
-export { AtomicFileLockUnavailable as C, AtomicFileLockOwner as S, tryAcquireAtomicFileLock as T, withLedgerFileLock as _, LedgerEventBase as a, AtomicFileLockError as b, LedgerLineContext as c, canonicalString as d, hashCanonical as f, tryWithLedgerFileLock as g, appendLedgerLine as h, LedgerEntryOf as i, LedgerProjector as l, LedgerFileContext as m, LedgerAppendResult as n, LedgerHash as o, FileLockResult as p, LedgerChainFields as r, LedgerJournalCodec as s, FileLedgerJournal as t, LedgerReplayResult as u, AtomicFileLock as v, probeAtomicFileLock as w, AtomicFileLockOptions as x, AtomicFileLockAcquisition as y };
-//# sourceMappingURL=index-BAvgST_9.d.ts.map
+export { AtomicFileLockOptions as A, LedgerCanonicalizationError as C, AtomicFileLock as D, hashCanonical as E, AtomicFileLockUnavailable as M, probeAtomicFileLock as N, AtomicFileLockAcquisition as O, tryAcquireAtomicFileLock as P, LEDGER_HASH_PATTERN as S, canonicalString as T, LedgerFileContext as _, LedgerChainFields as a, withLedgerFileLock as b, LedgerJournalCodec as c, LedgerReplayResult as d, LedgerAnchoredEntry as f, FileLockResult as g, verifyEntriesAgainstTrustedHead as h, LedgerAppendResult as i, AtomicFileLockOwner as j, AtomicFileLockError as k, LedgerLineContext as l, trustedHeadPathFor as m, FileLedgerJournalOptions as n, LedgerEntryOf as o, LedgerTrustedHead as p, LedgerAppendOptions as r, LedgerEventBase as s, FileLedgerJournal as t, LedgerProjector as u, appendLedgerLine as v, LedgerHash as w, writeLedgerFileAtomically as x, tryWithLedgerFileLock as y };
+//# sourceMappingURL=index-nhIYz9hn.d.ts.map

package/dist/index-nhIYz9hn.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index-nhIYz9hn.d.ts","names":[],"sources":["../src/ledger-core/atomic-file-lock.ts","../src/ledger-core/canonical.ts","../src/ledger-core/journal-file.ts","../src/ledger-core/trusted-head.ts","../src/ledger-core/journal.ts"],"mappings":";;;UAiBiB;WACN;WACA;WACA;;UAGM;WACN,OAAO;EAChB;;KAGU;WACG;WAA0B;WAAyB,QAAQ;;WAC3D;WAA0B;;KAE7B;WACG;WAAyB,MAAM;IAC1C;UAEa;WACN;WACA;;cAGE,4BAA4B;WACrB;;;iBASJ,oBACd,SAAS,wBACR;;;;;;;iBAea,yBACd,SAAS,wBACR;;;KCjDS;;cAGC,qBAAmB;;cAGnB,oCAAoC;;;;iBAKjC,gBAAgB;iBAQhB,cAAc,iBAAiB;;;;;UCzB9B;;EAEf;EACA,eAAe,iBAAiB;IAAY;MAAoB;;;;iBAKlD,iBAAiB,cAAc,cAAc,SAAS;;;;iBAetD,0BACd,cACA,kBACA,SAAS;iBAoBK,mBAAmB,GAAG,cAAc,SAAS,mBAAmB,WAAW,IAAI;KAQnF,eAAe;EAAO;EAAgB,OAAO;;EAAQ;;iBAEjD,sBAAsB,GACpC,cACA,SAAS,mBACT,WAAW,IACV,eAAe;;;;UC7BD;EACf;EACA,WAAW;;;UAII;EACf;EACA,WAAW;;;;iBAKG,mBAAmB;;;;iBA4DnB,gCACd,kBAAkB,uBAClB,MAAM,mBACN,SAAS,mBACT;;;;UCtFe;EACf;;;;;UAMe,kBAAkB,cAAc;EAC/C;EACA,cAAc;EACd,OAAO;EACP,WAAW;;KAGD,cAAc,uBAAuB,cAAc,mBAAmB,SAChF,kBAAkB;UAEH;EACf;EACA;;;;;UAMe,gBAAgB,OAAO;EACtC,MAAM,OAAO,OAAO;EACpB,OAAO,SAAS,UAAU;;UAGX,mBACf,uBACA,cAAc,iBACd,oBACQ;;EAER,QAAQ;EACR,cAAc,kBAAkB;;;;EAIhC,WAAW,cAAc,SAAS,oBAAoB,cAAc,QAAQ;;EAE5E,iBAAiB,OAAO,cAAc,QAAQ,QAAQ;EACtD,mBAAmB,gBAAgB,cAAc,QAAQ,QAAQ;;UAGlD,mBAAmB,OAAO;EACzC,SAAS;EACT,YAAY;;UAGG,mBAAmB,OAAO;EACzC,OAAO;;EAEP;EACA,YAAY;;UAGG;;;;;;EAMf;;UAGe;;;;;;;EAOf;;;;cAiBW,kBAAkB,uBAAuB,cAAc,iBAAiB;WAC1E;;WAEA;mBACQ;mBACA;mBACA;EAEjB,YACE,cACA,OAAO,mBAAmB,QAAQ,OAAO,aACzC,UAAS;EASL,UAAU,QAAQ,mBAAmB,cAAc,QAAQ,QAAQ;EAMnE,OACJ,OAAO,OACP,UAAS,sBACR,QAAQ,mBAAmB,cAAc,QAAQ,QAAQ;;EA2CtD,eAAe,QAAQ;;;;EAWvB,kBAAkB,QAAQ;UAqBxB;UAUA;UAkBA;UAgDA"}

package/dist/index.d.ts

@@ -19,26 +19,26 @@ import { A as ChatClient, B as SandboxSdkTransportOpts, C as ScenarioFile, D as
 import { _ as ViewTraceOversized, a as QueryTracesPage, c as SpanMatchRecord, d as TraceAnalystFilters, f as TraceAnalystSpan, g as ViewSpansResult, h as TraceAnalystTraceSummary, i as ErrorCluster, l as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, m as TraceAnalystSpanStatus, n as DEFAULT_TRACE_ANALYST_BUDGETS, o as SearchSpanResult, p as TraceAnalystSpanKind, r as DatasetOverview, s as SearchTraceResult, t as TraceAnalysisStore, u as TraceAnalystByteBudgets, v as ViewTraceResult } from "./store-CxJry_cs.js";
 import { A as AnalystRunResult, C as AnalystContext, D as AnalystRequirements, E as AnalystInputKind, F as computeFindingId, I as makeFinding, M as AnalystSeverity, N as AnalystUsageReceipt, O as AnalystRunEvent, P as EvidenceRef, S as Analyst, T as AnalystFinding, a as AnalystRegistryOptions, c as CreateTraceAnalystKindOpts, d as createTraceAnalystKind, f as renderPriorFindings, g as RawAnalystEvidence, i as AnalystRegistry, j as AnalystRunSummary, k as AnalystRunInputs, l as TraceAnalystGolden, n as buildDefaultAnalystRegistry, o as BudgetPolicy, p as renderUpstreamFindings, r as AnalystHooks, s as RegistryRunOpts, t as DefaultAnalystRegistryOptions, u as TraceAnalystKindSpec, v as RawAnalystFinding, w as AnalystCost } from "./default-registry-Cl3pHo4n.js";
 import { R as Scenario$1, S as JudgeConfig$1, w as JudgeScore$1 } from "./types-BokuXvOG.js";
-import { B as BenchmarkSource, F as BenchmarkDatasetItem, H as deterministicSplit, I as BenchmarkEvaluation, L as BenchmarkFamily, N as BENCHMARK_SPLIT_SEED, P as BenchmarkAdapter, R as BenchmarkResponder, V as BenchmarkTaskKind, t as index_d_exports, z as BenchmarkScenario } from "./index-CiP2DRCj.js";
-import { $ as Objective, At as CanarySeverity, Cn as ReferenceEquivalenceJudgeResult, Ct as scoreRedTeamOutput, Dn as LlmJudgeDimension, Dt as CanaryKind, En as runReferenceEquivalenceJudge, Et as CanaryEvaluation, On as LlmJudgeOptions, Ot as CanaryOptions, Q as Direction, Sn as ReferenceEquivalenceJudgeOptions, St as redTeamReport, Tn as createReferenceEquivalenceJudge, Tt as CanaryAlert, _t as RedTeamCategory, at as scalarScore, bn as REFERENCE_EQUIVALENCE_JUDGE_VERSION, bt as RedTeamReport, et as ParetoResult, gt as RedTeamCase, ht as DEFAULT_RED_TEAM_CORPUS, it as paretoFrontierWithCrowding, jt as runCanaries, kn as llmJudge, kt as CanaryReport, nt as dominates, rt as paretoFrontier, tt as crowdingDistance, vt as RedTeamFinding, wn as ReferenceEquivalenceScenario, wt as toolNamesForRun, xn as ReferenceEquivalenceJudgeInput, xt as redTeamDataset, yn as REFERENCE_EQUIVALENCE_INPUT_LIMITS, yt as RedTeamPayload } from "./skillopt-optimization-method-CAASpcS3.js";
+import { B as BenchmarkSource, F as BenchmarkDatasetItem, H as deterministicSplit, I as BenchmarkEvaluation, L as BenchmarkFamily, N as BENCHMARK_SPLIT_SEED, P as BenchmarkAdapter, R as BenchmarkResponder, V as BenchmarkTaskKind, t as index_d_exports, z as BenchmarkScenario } from "./index-BREtv3ZZ.js";
+import { $ as Objective, At as CanarySeverity, Cn as ReferenceEquivalenceJudgeResult, Ct as scoreRedTeamOutput, Dn as LlmJudgeDimension, Dt as CanaryKind, En as runReferenceEquivalenceJudge, Et as CanaryEvaluation, On as LlmJudgeOptions, Ot as CanaryOptions, Q as Direction, Sn as ReferenceEquivalenceJudgeOptions, St as redTeamReport, Tn as createReferenceEquivalenceJudge, Tt as CanaryAlert, _t as RedTeamCategory, at as scalarScore, bn as REFERENCE_EQUIVALENCE_JUDGE_VERSION, bt as RedTeamReport, et as ParetoResult, gt as RedTeamCase, ht as DEFAULT_RED_TEAM_CORPUS, it as paretoFrontierWithCrowding, jt as runCanaries, kn as llmJudge, kt as CanaryReport, nt as dominates, rt as paretoFrontier, tt as crowdingDistance, vt as RedTeamFinding, wn as ReferenceEquivalenceScenario, wt as toolNamesForRun, xn as ReferenceEquivalenceJudgeInput, xt as redTeamDataset, yn as REFERENCE_EQUIVALENCE_INPUT_LIMITS, yt as RedTeamPayload } from "./skillopt-optimization-method-wHF5xsUv.js";
 import { a as analyzeTraces, i as AnalyzeTracesTurnSnapshot, n as AnalyzeTracesOptions, r as AnalyzeTracesResult, t as AnalyzeTracesInput } from "./analyst-BkTS3C58.js";
-import { $t as ArtifactEventLike, Bt as assertRealBackend, Cn as byteLengthRange, Dn as regexMatch, En as jsonHasKeys, Gt as HARNESS_NATIVE_MODEL, Ht as summarizeBackendIntegrity, Jt as agentProfileHash, Kt as HarnessType, Lt as BackendIntegrityError, Qt as harnessAxisOf, Rt as BackendIntegrityReport, Sn as ValidationResult, Tn as containsAll, Ut as AgentProfile, Vt as summarizeAgentReceiptIntegrity, Wt as CODING_HARNESSES, Xt as agentProfileModelId, Yt as agentProfileId, Zt as expandProfileAxes, _n as verifyCompletion, ai as MatchedPair, an as CompletionVerdict, ar as canonicalize, bn as ValidationContext, ci as PairArmsResult, cn as ProducedProposal, cr as signManifest, di as PairedArmsComparison, dn as SatisfiedBy, en as ProposalEventLike, fi as PairedCorrectness, fn as TaskGold, gi as pairRunRecords, gn as parseCorrectnessResponse, hi as pairArms, hn as createTokenRecallChecker, ii as ComparePairedArmsOptions, in as CompletionRequirement, ir as SignedManifestAlgo, li as PairRunRecordsResult, ln as ProducedState, lr as verifyManifest, mi as comparePairedArms, mn as createLlmCorrectnessChecker, nn as ToolCallEventLike, nr as HypothesisResult, oi as MatchedRunRecordPair, on as CorrectnessChecker, or as evaluateHypothesis, pi as PairedMetricDelta, pn as completionVerdict, qt as ProfileAxisSpec, rn as extractProducedState, rr as SignedManifest, si as PairArmsOptions, sn as LlmCorrectnessCheckerOpts, sr as hashJson, tn as RuntimeEventLike, tr as HypothesisManifest, ui as PairedArmRow, un as RequirementCheck, vn as Artifact$1, wn as composeValidators, xn as ValidationIssue, yn as ArtifactValidator, zt as assertRealAgentReceipts } from "./index-B_eUhU9K.js";
+import { $t as harnessAxisOf, Bt as assertRealAgentReceipts, Cn as ValidationResult, Dn as jsonHasKeys, En as containsAll, Gt as CODING_HARNESSES, Ht as summarizeAgentReceiptIntegrity, Jt as ProfileAxisSpec, Kt as HARNESS_NATIVE_MODEL, On as regexMatch, Qt as expandProfileAxes, Rt as BackendIntegrityError, Sn as ValidationIssue, Tn as composeValidators, Ut as summarizeBackendIntegrity, Vt as assertRealBackend, Wt as AgentProfile, Xt as agentProfileId, Yt as agentProfileHash, Zt as agentProfileModelId, _i as pairRunRecords, _n as parseCorrectnessResponse, ai as ComparePairedArmsOptions, an as CompletionRequirement, ar as SignedManifestAlgo, bn as ArtifactValidator, ci as PairArmsOptions, cn as LlmCorrectnessCheckerOpts, cr as hashJson, di as PairedArmRow, dn as RequirementCheck, en as ArtifactEventLike, fi as PairedArmsComparison, fn as SatisfiedBy, gi as pairArms, gn as createTokenRecallChecker, hi as comparePairedArms, hn as createLlmCorrectnessChecker, in as extractProducedState, ir as SignedManifest, li as PairArmsResult, ln as ProducedProposal, lr as signManifest, mi as PairedMetricDelta, mn as completionVerdict, nn as RuntimeEventLike, nr as HypothesisManifest, oi as MatchedPair, on as CompletionVerdict, or as canonicalize, pi as PairedCorrectness, pn as TaskGold, qt as HarnessType, rn as ToolCallEventLike, rr as HypothesisResult, si as MatchedRunRecordPair, sn as CorrectnessChecker, sr as evaluateHypothesis, tn as ProposalEventLike, ui as PairRunRecordsResult, un as ProducedState, ur as verifyManifest, vn as verifyCompletion, wn as byteLengthRange, xn as ValidationContext, yn as Artifact$1, zt as BackendIntegrityReport } from "./index-C7Wue8R6.js";
 import { a as ContinuousCalibrationResult, c as SelfPreferenceResult, d as calibrateJudgeContinuous, f as continuousAgreement, h as verbosityBias, i as ContinuousAgreementOptions, l as VerbosityBiasResult, m as selfPreference, n as CandidateScore, o as GoldenItem, p as positionalBias, r as ContinuousAgreement, s as PositionalBiasResult, t as CalibrationResult, u as calibrateJudge } from "./judge-calibration-DFtEMlde.js";
 import { A as mannWhitneyU, B as pairedSignTest, C as confidenceInterval, D as holm, E as eProcess, F as normalizeScores, G as ranks, H as partialCredit, I as pairedBootstrap, J as spearmanR, K as requiredPairedSampleSize, L as pairedCohensDz, M as mcnemarPower, N as mcnemarRequiredN, O as interRaterReliability, P as mulberry32, Q as wilson, R as pairedMde, S as cohensD, T as corpusInterRaterAgreementFromJudgeScores, U as passAtK, V as pairedTTest, W as pearsonR, X as weightedMean, Y as weightedComposite, Z as wilcoxonSignedRank, _ as WeightedCompositeInput, a as CorpusScoreRecord, b as bonferroni, c as EProcessState, d as PairedBootstrapOptions, f as PairedBootstrapResult, g as SignTestAlternative, h as RiskDifferenceResult, i as CorpusAgreementReport, j as mcnemar, k as interpretCliffs, l as EProcessStep, m as ProportionInterval, n as CorpusAgreementOptions, o as EProcess, p as PairedSignTestResult, q as requiredSampleSize, r as CorpusAgreementPerDimension, s as EProcessOptions, t as CliffsMagnitude, u as McNemarResult, v as WeightedCompositeResult, w as corpusInterRaterAgreement, x as cliffsDelta, y as benjaminiHochberg, z as pairedRiskDifference } from "./statistics-DbvkkDPa.js";
 import { a as DatasetScenario, c as SliceOptions, i as DatasetProvenance, l as hashScenarios, n as DatasetDifficulty, o as DatasetSplit, r as DatasetManifest, s as HoldoutLockedError, t as Dataset } from "./dataset-BvtnC8Dc.js";
-import { C as HeldOutGateConfig, S as HeldOutGate, _ as paretoChart, a as ParetoPoint, b as GateDecision, c as ResearchReportCandidate, d as ResearchReportOptions, f as ResearchReportRecommendation, g as gainHistogram, h as SummaryTableRow, i as ParetoFigureSpec, l as ResearchReportDecision, m as SummaryTableOptions, n as GainDistributionFigureSpec, o as RESEARCH_REPORT_HARD_PAIR_FLOOR, p as SummaryTable, r as GainDistributionOptions, s as ResearchReport, t as GainDistributionBin, u as ResearchReportMethodology, v as researchReport, w as HeldOutGateRejectionCode, x as GateEvidence, y as summaryTable } from "./summary-report-DnUcjVpV.js";
+import { C as HeldOutGateConfig, S as HeldOutGate, _ as paretoChart, a as ParetoPoint, b as GateDecision, c as ResearchReportCandidate, d as ResearchReportOptions, f as ResearchReportRecommendation, g as gainHistogram, h as SummaryTableRow, i as ParetoFigureSpec, l as ResearchReportDecision, m as SummaryTableOptions, n as GainDistributionFigureSpec, o as RESEARCH_REPORT_HARD_PAIR_FLOOR, p as SummaryTable, r as GainDistributionOptions, s as ResearchReport, t as GainDistributionBin, u as ResearchReportMethodology, v as researchReport, w as HeldOutGateRejectionCode, x as GateEvidence, y as summaryTable } from "./summary-report-CFnQgNfg.js";
 import { a as FailureClassification, c as classifyFailure, i as DEFAULT_RULES, o as FailureContext, s as FailureRule } from "./failure-cluster-CqcvCcdR.js";
 import { _ as vitestTestParser, a as DockerSandboxDriver, c as SandboxHarness, d as SubprocessSandboxDriver, f as SubprocessSandboxDriverOptions, g as pytestTestParser, h as jestTestParser, i as runTestGradedScenario, l as SandboxHarnessResult, m as composeParsers, n as TestGradedRunResult, o as HarnessConfig, p as TestOutputParser, r as TestGradedScenario, s as SandboxDriver, t as TestGradedRunOptions, u as SandboxResult } from "./test-graded-scenario-D1TaI2va.js";
-import { t as AnalyzeRunsOptions } from "./analyze-runs-DZr7JW-m.js";
+import { t as AnalyzeRunsOptions } from "./analyze-runs-BmX-h_yn.js";
 import { A as replayFeedbackTrajectory, B as ControlRunResult, C as feedbackTrajectoriesToDatasetScenarios, D as parseFeedbackTrajectoriesJsonl, E as feedbackTrajectoryToOptimizerRow, F as ControlActionOutcome, G as ControlStopPolicies, H as ControlRuntimeError, I as ControlBudget, J as objectiveEval, K as StopDecision, L as ControlContext, M as summarizePreferenceMemory, N as withAssignedFeedbackSplit, O as renderPreferenceMemoryMarkdown, P as ControlActionFailureMode, Q as subjectiveEval, R as ControlDecision, S as createFeedbackTrajectory, T as feedbackTrajectoryToDatasetScenario, U as ControlSeverity, V as ControlRuntimeConfig, W as ControlStep, X as stopOnNoProgress, Y as runAgentControlLoop, Z as stopOnRepeatedAction, _ as InMemoryFeedbackTrajectoryStore, a as FeedbackLabelSource, b as assignFeedbackSplit, c as FeedbackReplayAdapter, d as FeedbackSplitPolicy, f as FeedbackTask, g as FileSystemFeedbackTrajectoryStore, h as FeedbackTrajectoryStore, i as FeedbackLabelKind, j as serializeFeedbackTrajectoriesJsonl, k as replayFeedbackTrajectories, l as FeedbackReplayResult, m as FeedbackTrajectoryFilter, n as FeedbackAttempt, o as FeedbackOptimizerRow, p as FeedbackTrajectory, q as allCriticalPassed, r as FeedbackLabel, s as FeedbackOutcome, t as FeedbackArtifactType, u as FeedbackSeverity, v as PreferenceMemoryEntry, w as feedbackTrajectoriesToOptimizerRows, x as controlRunToFeedbackTrajectory, y as ProposedSideEffect, z as ControlEvalResult } from "./feedback-trajectory-CVaeREXV.js";
 import { A as ActionExecutionPolicy, C as ReviewMemoryStore, D as inMemoryReviewStore, E as createLlmReviewer, M as evaluateActionPolicy, O as jsonlReviewStore, S as ReviewMemoryEntry, T as VerifyFn, _ as ProposeReviewReport, a as ProposeReviewControlAction, b as ReviewFn, c as ProposeReviewControlState, d as LlmJsonCall, f as LlmReviewerConfig, g as ProposeReviewConfig, h as ProposeOutput, i as scoreFromEvals, j as ActionPolicyDecision, k as runProposeReview, l as controlFailureClassFromVerification, m as ProposeInput, n as RunEvidenceMetadata, o as ProposeReviewControlConfig, p as ProposeFn, r as controlRunToRunRecord, s as ProposeReviewControlResult, t as ControlRunToRunRecordOptions, u as runProposeReviewAsControlLoop, v as ProposeReviewShot, w as Verification, x as ReviewInput, y as Review } from "./run-evidence-DokQtX0-.js";
-import { _ as ReleaseConfidenceStatus, a as JudgeReplayGateArgs, b as assertReleaseConfidence, c as judgeReplayGate, d as ReleaseConfidenceAxis, f as ReleaseConfidenceAxisName, g as ReleaseConfidenceScorecard, h as ReleaseConfidenceMetrics, i as BootstrapResult, l as ActionableSideInfo, m as ReleaseConfidenceIssue, n as renderReleaseReport, o as Verdict, p as ReleaseConfidenceInput, r as BootstrapOptions, s as bootstrapCi, t as RenderReleaseReportOptions, u as AsiSeverity, v as ReleaseConfidenceThresholds, x as evaluateReleaseConfidence, y as ReleaseTraceEvidence } from "./release-report-DfmKSIEE.js";
+import { _ as ReleaseConfidenceStatus, a as JudgeReplayGateArgs, b as assertReleaseConfidence, c as judgeReplayGate, d as ReleaseConfidenceAxis, f as ReleaseConfidenceAxisName, g as ReleaseConfidenceScorecard, h as ReleaseConfidenceMetrics, i as BootstrapResult, l as ActionableSideInfo, m as ReleaseConfidenceIssue, n as renderReleaseReport, o as Verdict, p as ReleaseConfidenceInput, r as BootstrapOptions, s as bootstrapCi, t as RenderReleaseReportOptions, u as AsiSeverity, v as ReleaseConfidenceThresholds, x as evaluateReleaseConfidence, y as ReleaseTraceEvidence } from "./release-report-CjHWa8Ia.js";
 import { A as isTrainableSplit, D as assertRolloutLine, E as assertMintedLines, T as assertMinted, b as RolloutSplit, h as RolloutLine, i as ChatToolCall, j as validateRolloutLine, k as isRolloutLine, n as ChatMessage, o as MintedRolloutLine, p as RolloutCapture, s as MintedRolloutOutcome, u as ROLLOUT_SCHEMA, w as ToolDef, x as RolloutStep, y as RolloutRole } from "./schema-Cef2cFmb.js";
 import { $ as ScorePreference, Ct as HarborToolCall, Dt as toHarborTrajectories, Et as relabelImportedSplit, J as MintRolloutOptions, Jt as toJsonl, Ot as toHarborTrajectory, Q as ScoreOrigin, Qt as toSftRows, St as HarborSubagentTrajectoryRef, Tt as fromHarborTrajectory, Ut as SftExportOptions, Vt as RewardRow, Wt as SftRow, X as RolloutScrubber, Y as MintRolloutResult, Yt as toRewardRows, Z as mintRolloutRows, _t as HarborMetrics, at as trainingScore, bt as HarborStep, dt as FromHarborOptions, et as isRealnessGated, ft as HARBOR_IMPORT_GAP, gt as HarborImageSource, ht as HarborFinalMetrics, it as trainingReward, mt as HarborContentPart, nt as observedSplitScore, pt as HarborAgent, rt as scoreOrigin, tt as observedScore, ut as ATIF_SCHEMA_VERSION, vt as HarborObservation, wt as HarborTrajectory, xt as HarborStepSource, yt as HarborObservationResult } from "./index-3cdlURSk.js";
 import { $ as isUnavailable, A as rollupSupervisorRuns, D as analyzeSupervisorRunSources, G as SupervisorRunReader, H as SUPERVISOR_RUN_SCHEMA, J as SupervisorRunSources, K as SupervisorRunReport, N as Measured, S as readClaudeCodeSupervisorRun, U as SourceLimits, X as Unavailable, Y as SupervisorRunTree, a as renderSupervisorRunMarkdown, c as analyzeSupervisorRun, et as showMeasured, h as writeSupervisorRunReport, i as renderSupervisorRunHeadline, n as supervisorRunRolloutLines, q as SupervisorRunRollup, x as claudeCodeSupervisorRunReader } from "./index-C61Wi7yg.js";
 import { a as compareToBaseline, c as Trajectory, d as ToolStats, f as ToolUseMetrics, i as MetricVerdict, l as TrajectoryStep, m as computeToolUseMetrics, n as BaselineReport, o as iqr, p as ToolUseOptions, r as MetricSamples, s as welchsTTest, t as BaselineOptions, u as buildTrajectory } from "./baseline-hG3K85h4.js";
 import { n as SeriesConvergenceResult, r as analyzeSeries, t as SeriesConvergenceOptions } from "./series-convergence-ofsqPWhs.js";
-import { _ as EvalCampaignResult, a as FailureMode, c as SteeringChange, d as CampaignRunContext, f as CampaignRunOutcome, g as EvalCampaignOptions, h as CampaignVariant, i as ExperimentResult, l as CampaignFactoryParams, m as CampaignScenario, n as CallbackResearcherOptions, o as NoopResearcher, p as CampaignRunner, r as ExperimentPlan, s as Researcher, t as CallbackResearcher, u as CampaignIntegrityPolicy, v as FailedRun, y as runEvalCampaign } from "./researcher-DMimgHtN.js";
+import { _ as EvalCampaignResult, a as FailureMode, c as SteeringChange, d as CampaignRunContext, f as CampaignRunOutcome, g as EvalCampaignOptions, h as CampaignVariant, i as ExperimentResult, l as CampaignFactoryParams, m as CampaignScenario, n as CallbackResearcherOptions, o as NoopResearcher, p as CampaignRunner, r as ExperimentPlan, s as Researcher, t as CallbackResearcher, u as CampaignIntegrityPolicy, v as FailedRun, y as runEvalCampaign } from "./researcher-CbSKhK8z.js";
 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";
 //#region src/auto-pr.d.ts
 /**

package/dist/index.js

@@ -1,8 +1,7 @@
 import { t as __exportAll } from "./rolldown-runtime-8H4AJuhK.js";
-import { _ as observeAll, a as jsonlReviewStore, c as scoreFromEvals, d as runAgentControlLoop, f as stopOnNoProgress, g as noProgressDetector, h as errorStreakDetector, i as inMemoryReviewStore, l as allCriticalPassed, m as subjectiveEval, n as runProposeReviewAsControlLoop, o as runProposeReview, p as stopOnRepeatedAction, r as createLlmReviewer, s as controlRunToRunRecord, t as controlFailureClassFromVerification, u as objectiveEval, v as repeatedActionDetector, y as evaluateActionPolicy } from "./propose-review-control-Bqb7daEJ.js";
+import { _ as observeAll, a as jsonlReviewStore, c as scoreFromEvals, d as runAgentControlLoop, f as stopOnNoProgress, g as noProgressDetector, h as errorStreakDetector, i as inMemoryReviewStore, l as allCriticalPassed, m as subjectiveEval, n as runProposeReviewAsControlLoop, o as runProposeReview, p as stopOnRepeatedAction, r as createLlmReviewer, s as controlRunToRunRecord, t as controlFailureClassFromVerification, u as objectiveEval, v as repeatedActionDetector, y as evaluateActionPolicy } from "./propose-review-control-SQ-n9-We.js";
 import { a as NotFoundError, c as VerificationError, i as JudgeError, n as CaptureIntegrityError, o as ReplayError, r as ConfigError, s as ValidationError, t as AgentEvalError } from "./errors-8YnH8WlF.js";
-import { a as verifyManifest, i as signManifest, n as evaluateHypothesis, r as hashJson, t as canonicalize } from "./pre-registration-DakwTRXk.js";
-import { AGENT_PROFILE_KINDS, AgentProfileCellValidationError, agentProfileCellHashMaterial, agentProfileCellKey, assertRunAgentProfileCell, buildAgentInterfaceProfileCell, buildAgentProfileCell, groupRunsByAgentProfileCell, requireAgentProfileCell, toAgentProfileJson, validateAgentProfileCell, verifyAgentProfileCell } from "./profile-cell.js";
+import { _ as verifyManifest, a as assertRunAgentProfileCell, c as groupRunsByAgentProfileCell, d as validateAgentProfileCell, f as verifyAgentProfileCell, g as signManifest, h as hashJson, i as agentProfileCellKey, l as requireAgentProfileCell, m as evaluateHypothesis, n as AgentProfileCellValidationError, o as buildAgentInterfaceProfileCell, p as canonicalize, r as agentProfileCellHashMaterial, s as buildAgentProfileCell, t as AGENT_PROFILE_KINDS, u as toAgentProfileJson } from "./agent-profile-cell-OhuTee9n.js";
 import { F as makeFinding, I as computeTraceMetrics, L as createChatClient, P as computeFindingId, R as createAnalystAi, a as KNOWLEDGE_GAP_KIND_SPEC, d as renderUpstreamFindings, i as KNOWLEDGE_POISONING_KIND_SPEC, l as createTraceAnalystKind, n as AnalystRegistry, o as IMPROVEMENT_KIND_SPEC, r as DEFAULT_TRACE_ANALYST_KINDS, s as FAILURE_MODE_KIND_SPEC, t as buildDefaultAnalystRegistry, u as renderPriorFindings } from "./default-registry-D3T9XbuY.js";
 import { _ as resolveModelPricing, a as CostLedgerPersistenceError, c as costForTokenPricing, d as MODEL_PRICING, f as MetricsCollector, g as isModelPriced, h as estimateTokens, i as CostLedger, l as costForUsage, m as estimateCost, n as CostCallConflictError, o as CostReceiptCaptureError, p as TokenCounter, r as CostCeilingReachedError, s as CostReservationExceededError, t as CostAccountingIncompleteError, u as modelPriceKey } from "./cost-ledger-BrJxbrMy.js";
 import { a as providerFromBaseUrl, i as defaultProviderRedactor, n as InMemoryRawProviderSink, r as NoopRawProviderSink, t as FileSystemRawProviderSink } from "./raw-provider-sink-BQd7mzyT.js";
@@ -12,34 +11,34 @@ import { S as traceSpanKindToOpenInferenceKind, a as SpanNotFoundError, b as cla
 import { n as aggregateRunScore, r as clamp01, t as DEFAULT_RUN_SCORE_WEIGHTS } from "./run-score-iEEAWiBY.js";
 import { n as mapConcurrent, t as Mutex } from "./concurrency-MUjT7VjM.js";
 import { a as RunCritic, d as defaultIsMaterial, f as diffFindings, i as runSemanticConceptJudge, n as SEMANTIC_CONCEPT_JUDGE_VERSION, o as SKILL_USAGE_ANALYST, p as LockedJsonlAppender, r as createSemanticConceptJudge, s as SkillUsageAnalyst, t as DEFAULT_COMPLEXITY_WEIGHTS, u as FindingsStore } from "./semantic-concept-judge-BypLt6Fw.js";
-import { At as dominates, B as surfaceContentHash, Bt as JudgeParseError, Ct as llmJudge, D as runCanaries, Dt as fileVerdictCache, Et as contentHash, Ft as BackendIntegrityError, G as DEFAULT_MUTATION_PRIMITIVES, It as assertRealAgentReceipts, K as buildReflectionPrompt, Lt as assertRealBackend, Mt as paretoFrontierWithCrowding, Nt as scalarScore, Ot as inMemoryVerdictCache, Rt as summarizeAgentReceiptIntegrity, St as hashScenarios, Tt as canonicalJson, _t as redTeamReport, bt as Dataset, dt as REFERENCE_EQUIVALENCE_INPUT_LIMITS, ft as REFERENCE_EQUIVALENCE_JUDGE_VERSION, gt as redTeamDataset, ht as DEFAULT_RED_TEAM_CORPUS, jt as paretoFrontier, kt as crowdingDistance, mt as runReferenceEquivalenceJudge, pt as createReferenceEquivalenceJudge, q as parseReflectionResponse, vt as scoreRedTeamOutput, wt as cachedJudge, xt as HoldoutLockedError, yt as toolNamesForRun, zt as summarizeBackendIntegrity } from "./skillopt-optimization-method-DgN7U9iR.js";
-import { A as spearmanR, B as verbosityBias, C as pairedTTest, D as ranks, E as pearsonR, F as calibrateJudge, I as calibrateJudgeContinuous, L as continuousAgreement, M as weightedMean, N as wilcoxonSignedRank, O as requiredPairedSampleSize, P as wilson, R as positionalBias, S as pairedSignTest, T as passAtK, _ as normalizeScores, a as confidenceInterval, b as pairedMde, c as eProcess, d as interpretCliffs, f as mannWhitneyU, g as mulberry32, h as mcnemarRequiredN, i as cohensD, j as weightedComposite, k as requiredSampleSize, l as holm, m as mcnemarPower, n as bonferroni, o as corpusInterRaterAgreement, p as mcnemar, r as cliffsDelta, s as corpusInterRaterAgreementFromJudgeScores, t as benjaminiHochberg, u as interRaterReliability, v as pairedBootstrap, w as partialCredit, x as pairedRiskDifference, y as pairedCohensDz, z as selfPreference } from "./statistics-CnnxdpOg.js";
-import { n as pairArms, r as pairRunRecords, t as comparePairedArms } from "./paired-arms-D9D0wXj2.js";
+import { At as dominates, B as surfaceContentHash, Bt as JudgeParseError, Ct as llmJudge, D as runCanaries, Dt as fileVerdictCache, Et as contentHash, Ft as BackendIntegrityError, G as DEFAULT_MUTATION_PRIMITIVES, It as assertRealAgentReceipts, K as buildReflectionPrompt, Lt as assertRealBackend, Mt as paretoFrontierWithCrowding, Nt as scalarScore, Ot as inMemoryVerdictCache, Rt as summarizeAgentReceiptIntegrity, St as hashScenarios, Tt as canonicalJson, _t as redTeamReport, bt as Dataset, dt as REFERENCE_EQUIVALENCE_INPUT_LIMITS, ft as REFERENCE_EQUIVALENCE_JUDGE_VERSION, gt as redTeamDataset, ht as DEFAULT_RED_TEAM_CORPUS, jt as paretoFrontier, kt as crowdingDistance, mt as runReferenceEquivalenceJudge, pt as createReferenceEquivalenceJudge, q as parseReflectionResponse, vt as scoreRedTeamOutput, wt as cachedJudge, xt as HoldoutLockedError, yt as toolNamesForRun, zt as summarizeBackendIntegrity } from "./skillopt-optimization-method-CF6a327Q.js";
+import { A as spearmanR, B as positionalBias, C as pairedTTest, D as ranks, E as pearsonR, H as verbosityBias, L as calibrateJudge, M as weightedMean, N as wilcoxonSignedRank, O as requiredPairedSampleSize, P as wilson, R as calibrateJudgeContinuous, S as pairedSignTest, T as passAtK, V as selfPreference, _ as normalizeScores, a as confidenceInterval, b as pairedMde, c as eProcess, d as interpretCliffs, f as mannWhitneyU, g as mulberry32, h as mcnemarRequiredN, i as cohensD, j as weightedComposite, k as requiredSampleSize, l as holm, m as mcnemarPower, n as bonferroni, o as corpusInterRaterAgreement, p as mcnemar, r as cliffsDelta, s as corpusInterRaterAgreementFromJudgeScores, t as benjaminiHochberg, u as interRaterReliability, v as pairedBootstrap, w as partialCredit, x as pairedRiskDifference, y as pairedCohensDz, z as continuousAgreement } from "./statistics-DWM_AyLe.js";
+import { n as pairArms, r as pairRunRecords, t as comparePairedArms } from "./paired-arms-6XItKzd1.js";
 import { n as llmSpanFromProvider, t as TraceEmitter } from "./emitter-CPBAhxum.js";
 import { a as scoreOrigin, n as observedScore, o as trainingReward, r as observedSplitScore, s as trainingScore, t as isRealnessGated } from "./reward-nw2xZGZG.js";
 import { a as isRetrievalSpan, i as isLlmSpan, n as TRACE_SCHEMA_VERSION, o as isSandboxSpan, r as isJudgeSpan, s as isToolSpan, t as FAILURE_CLASSES } from "./schema-CRhEY1SO.js";
-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-CN8Zd21B.js";
-import { a as evaluateReleaseConfidence, i as assertReleaseConfidence, n as bootstrapCi, r as judgeReplayGate, t as renderReleaseReport } from "./release-report-oWt9f2k-.js";
+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-BIwU2wdV.js";
+import { a as evaluateReleaseConfidence, i as assertReleaseConfidence, n as bootstrapCi, r as judgeReplayGate, t as renderReleaseReport } from "./release-report-wuilQkvK.js";
 import { c as assertMintedLines, f as isRolloutLine, i as ROLLOUT_SCHEMA, m as validateRolloutLine, p as isTrainableSplit, s as assertMinted, u as assertRolloutLine } from "./schema-C6DW4ZHR.js";
 import { i as toRewardRows, r as toJsonl, s as toSftRows } from "./exporters-q9iL-2Jf.js";
-import { a as toHarborTrajectories, i as relabelImportedSplit, n as HARBOR_IMPORT_GAP, o as toHarborTrajectory, r as fromHarborTrajectory, t as ATIF_SCHEMA_VERSION } from "./rollout-VEo41J0N.js";
+import { a as toHarborTrajectories, i as relabelImportedSplit, n as HARBOR_IMPORT_GAP, o as toHarborTrajectory, r as fromHarborTrajectory, t as ATIF_SCHEMA_VERSION } from "./rollout-CreDz__7.js";
 import { t as buildTrajectory } from "./trajectory-D_7rLrvE.js";
-import { t as mintRolloutRows } from "./mint-D5_87M5L.js";
-import { D as showMeasured, E as isUnavailable, S as rollupSupervisorRuns, T as SUPERVISOR_RUN_SCHEMA, _ as claudeCodeSupervisorRunReader, f as renderSupervisorRunHeadline, l as writeSupervisorRunReport, n as analyzeSupervisorRun, p as renderSupervisorRunMarkdown, t as supervisorRunRolloutLines, v as readClaudeCodeSupervisorRun, y as analyzeSupervisorRunSources } from "./supervisor-run-_lnTLM3z.js";
+import { t as mintRolloutRows } from "./mint-BvkwcYZU.js";
+import { D as showMeasured, E as isUnavailable, S as rollupSupervisorRuns, T as SUPERVISOR_RUN_SCHEMA, _ as claudeCodeSupervisorRunReader, f as renderSupervisorRunHeadline, l as writeSupervisorRunReport, n as analyzeSupervisorRun, p as renderSupervisorRunMarkdown, t as supervisorRunRolloutLines, v as readClaudeCodeSupervisorRun, y as analyzeSupervisorRunSources } from "./supervisor-run-B7lUGoyZ.js";
 import { n as TRACE_ANALYST_ACTOR_DESCRIPTION, r as TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, t as analyzeTraces } from "./analyst-LsnNpSkm.js";
-import { C as planTraceInsightQuestions, E as traceAnalystOnRunComplete, S as inferDomainKeywords, T as tokenizeDomainWords, _ as buildTraceInsightContext, a as convertTraceStoresToOtlp, b as describeTraceInsightScope, c as otelRunCompleteHook, d as captureFetchToRawSink, f as otlpRowsToRunRecords, g as flattenOtlpExportToNdjson, h as otlpToTraceRunRecords, i as iterateRawCalls, l as OTEL_AGENT_EVAL_SCOPE, m as otlpToRunRecords, n as ReplayCacheMissError, o as createOtelExporter, p as otlpRowsToTraceRunRecords, r as createReplayFetch, s as createOtelTracingStore, t as ReplayCache, u as exportRunAsOtlp, v as buildTraceInsightPrompt, w as scoreTraceInsightReadiness, x as domainEvidencePattern, y as defaultTraceInsightPanel } from "./replay-BMR4TEYY.js";
+import { C as planTraceInsightQuestions, E as traceAnalystOnRunComplete, S as inferDomainKeywords, T as tokenizeDomainWords, _ as buildTraceInsightContext, a as convertTraceStoresToOtlp, b as describeTraceInsightScope, c as otelRunCompleteHook, d as captureFetchToRawSink, f as otlpRowsToRunRecords, g as flattenOtlpExportToNdjson, h as otlpToTraceRunRecords, i as iterateRawCalls, l as OTEL_AGENT_EVAL_SCOPE, m as otlpToRunRecords, n as ReplayCacheMissError, o as createOtelExporter, p as otlpRowsToTraceRunRecords, r as createReplayFetch, s as createOtelTracingStore, t as ReplayCache, u as exportRunAsOtlp, v as buildTraceInsightPrompt, w as scoreTraceInsightReadiness, x as domainEvidencePattern, y as defaultTraceInsightPanel } from "./replay-CJfGLdx4.js";
 import { n as extractUsageFromResponse, r as extractUsageFromSse, t as extractUsage } from "./extract-usage-2j25whHw.js";
-import { B as harnessAxisOf, F as HARNESS_NATIVE_MODEL, G as parseCorrectnessResponse, H as completionVerdict, I as agentProfileHash, K as verifyCompletion, L as agentProfileId, P as CODING_HARNESSES, R as agentProfileModelId, U as createLlmCorrectnessChecker, V as extractProducedState, W as createTokenRecallChecker, z as expandProfileAxes } from "./campaign-CyQMCCEU.js";
+import { B as harnessAxisOf, F as HARNESS_NATIVE_MODEL, G as parseCorrectnessResponse, H as completionVerdict, I as agentProfileHash, K as verifyCompletion, L as agentProfileId, P as CODING_HARNESSES, R as agentProfileModelId, U as createLlmCorrectnessChecker, V as extractProducedState, W as createTokenRecallChecker, z as expandProfileAxes } from "./campaign-BJjn1rhw.js";
 import { a as judgeSpans, c as runsForScenario, i as hasCapturedToolArgs, l as toolSpans, n as argHash, o as llmSpans, r as groupBy, s as runFailureClass, t as aggregateLlm } from "./query-Di7eEQ79.js";
-import { a as checkBehavioralCanary, i as canaryLeakView, o as checkCanaries, r as HoldoutAuditor, s as runBehavioralCanaries, t as analyzeRuns } from "./analyze-runs-BjPn_fOS.js";
-import { a as summaryTable, i as researchReport, n as gainHistogram, r as paretoChart, t as RESEARCH_REPORT_HARD_PAIR_FLOOR } from "./summary-report-BNs5nmXI.js";
+import { a as checkBehavioralCanary, i as canaryLeakView, o as checkCanaries, r as HoldoutAuditor, s as runBehavioralCanaries, t as analyzeRuns } from "./analyze-runs-B-afTpCv.js";
+import { a as summaryTable, i as researchReport, n as gainHistogram, r as paretoChart, t as RESEARCH_REPORT_HARD_PAIR_FLOOR } from "./summary-report-Ci17nIdU.js";
 import { a as composeParsers, c as vitestTestParser, i as SubprocessSandboxDriver, n as DockerSandboxDriver, o as jestTestParser, r as SandboxHarness, s as pytestTestParser, t as runTestGradedScenario } from "./test-graded-scenario-BsqWLmPt.js";
 import { a as InMemoryTraceStore, i as FileSystemTraceStore, n as assertRunCaptured, r as throwIfRunIncomplete, t as RunIntegrityError } from "./integrity-BzRbCHzi.js";
 import { i as redactValue, n as REDACTION_VERSION, r as redactString, t as DEFAULT_REDACTION_RULES } from "./redact-7Aq1ukl-.js";
-import { a as DEFAULT_RULES, i as computeToolUseMetrics, n as iqr, o as classifyFailure, r as welchsTTest, t as compareToBaseline } from "./baseline-HsBvw_dk.js";
+import { a as DEFAULT_RULES, i as computeToolUseMetrics, n as iqr, o as classifyFailure, r as welchsTTest, t as compareToBaseline } from "./baseline-DcX5hQDv.js";
 import { t as analyzeSeries } from "./series-convergence-CjO2QdRW.js";
-import { _ as deterministicSplit, g as BENCHMARK_SPLIT_SEED, t as benchmarks_exports } from "./benchmarks-DncrVrSr.js";
-import { t as runEvalCampaign } from "./eval-campaign-CYkUME2T.js";
+import { _ as deterministicSplit, g as BENCHMARK_SPLIT_SEED, t as benchmarks_exports } from "./benchmarks-CJr1H1_a.js";
+import { t as runEvalCampaign } from "./eval-campaign-DXhpZghy.js";
 import { n as pairedEvalueSequence, t as evaluateInterimReleaseConfidence } from "./sequential-Br0mAPHA.js";
 import { AxGEPA, AxSignature, ax } from "@ax-llm/ax";
 import { accessSync, appendFileSync, constants, cpSync, existsSync, mkdirSync, promises, readFileSync, readdirSync, statSync, writeFileSync } from "node:fs";
@@ -10226,6 +10225,14 @@ var HeldOutGate = class {
 			},
 			pairedPValue: wilcoxon.p
 		};
+		if (!Number.isFinite(ci.low) || !Number.isFinite(ci.high) || ci.low === 0 && ci.high === 0) return {
+			promote: false,
+			candidateId,
+			baselineId,
+			evidence,
+			reason: `indeterminate_delta: paired holdout median CI=[${fmt(ci.low)}, ${fmt(ci.high)}] carries no direction`,
+			rejectionCode: "indeterminate_delta"
+		};
 		if (!(ci.low > this.pairedDeltaThreshold)) return {
 			promote: false,
 			candidateId,