@tangle-network/agent-eval 0.190.1 → 0.192.0

package/dist/client-QwJpsNYU.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"client-QwJpsNYU.js","names":[],"sources":["../src/hosted/types.ts","../src/hosted/schemas.ts","../src/hosted/client.ts"],"sourcesContent":["/**\n * # Hosted-tier wire format — the schema that EVERY orchestrator (ours,\n * a partner's self-hosted one, a future open implementation) must accept.\n *\n * This package implements exactly one wire version. Servers reject every\n * other version instead of translating old payloads.\n *\n * The wire format is two event streams in one transport:\n *\n *   1. **Eval-run events** (`POST /v1/ingest/eval-runs`). Posted when a\n *      campaign / improvement-loop completes (or per-generation if\n *      streaming). Carries the structured result + per-cell scores +\n *      surface diffs the orchestrator stores for the dashboard.\n *\n *   2. **Trace spans** (`POST /v1/ingest/traces`). Standard OTLP-shaped\n *      spans with a few additional attributes so the orchestrator can\n *      pivot from eval-run → underlying execution. Compatible with any\n *      OTel collector.\n *\n * Both endpoints are authenticated with a bearer token + a tenant id\n * header. Tenants isolate everything downstream of ingest; no tenant\n * ever sees another tenant's data.\n */\n\nimport type { GateDecision, MutableSurface } from '../campaign/types'\nimport type { InsightReport } from '../contract/insight-report'\nimport type { RunTerminalOutcome } from '../run-record'\n\n// re-export so wire-format consumers can import the optional payload type\n// from `@tangle-network/agent-eval/hosted` without reaching into /contract.\nexport type { InsightReport } from '../contract/insight-report'\n\nexport const HOSTED_WIRE_VERSION = '2026-07-24.v1' as const\nexport type HostedWireVersion = typeof HOSTED_WIRE_VERSION\n\n// ── Transport headers ───────────────────────────────────────────────\n\n/** Every ingest request carries these. */\nexport interface HostedIngestHeaders {\n  /** Bearer token. The orchestrator validates against the tenant key. */\n  authorization: `Bearer ${string}`\n  /** Stable tenant id (the orchestrator-side primary key for the tenant). */\n  'x-tangle-tenant-id': string\n  /** Wire-version pin so the server can reject incompatible payloads. */\n  'x-tangle-wire-version': HostedWireVersion\n  /** Stable request key generated once and reused across retries. */\n  'idempotency-key': string\n}\n\n// ── Eval-run event ──────────────────────────────────────────────────\n\n/** Lifecycle stages of an eval-run as the substrate reports them. */\nexport type EvalRunStatus =\n  | 'started'\n  | 'baseline-complete'\n  | 'generation-complete'\n  | 'gate-decided'\n  | 'finished'\n  | 'errored'\n\nexport interface EvalRunCellScore {\n  /** Stable scenario id from the consumer's scenario set. */\n  scenarioId: string\n  /** Repetition index when reps > 1; 0 for the default. */\n  rep: number\n  /** Composite score across successful judges, or null when unscored. */\n  compositeMean: number | null\n  /** Per-judge and per-dimension scores; failed or missing judges are absent. */\n  dimensions: Record<string, Record<string, number>>\n  /** Root execution result, kept separate from task quality. */\n  terminalOutcome: RunTerminalOutcome\n  /** Canonical execution-error count, or null when the producer did not measure it. */\n  executionErrorCount: number | null\n  /** Per-cell dispatch or judge error. Missing on success. */\n  errorMessage?: string\n}\n\nexport interface EvalRunGenerationSnapshot {\n  /** Generation index. 0 is baseline. */\n  index: number\n  /** Candidate surface fingerprint (stable hash) — pivot key into the\n   *  trace stream to fetch the underlying execution. */\n  surfaceHash: string\n  /** The candidate surface itself. May be omitted to avoid PII when the\n   *  consumer prefers not to ship verbatim prompts. */\n  surface?: MutableSurface\n  /** Per-cell scores for this generation. */\n  cells: EvalRunCellScore[]\n  /** Mean across scored cells, or null when no cell has a task-quality label. */\n  compositeMean: number | null\n  /** Total $ spent across this generation. */\n  costUsd: number\n  /** Wall-clock duration of this generation. */\n  durationMs: number\n}\n\n/**\n * The top-level eval-run event. One ingest call per logical eval-run;\n * generations stream in incrementally via repeated calls with the same\n * `runId`. The orchestrator deduplicates by `(runId, generation.index)`.\n */\nexport interface EvalRunEvent {\n  /** Stable run id (the substrate's `runId`). UUID or substrate-generated. */\n  runId: string\n  /** Where this run was happening — derived from `RunCampaignOptions.runDir`. */\n  runDir: string\n  /** ISO-8601 timestamp the substrate recorded the event. */\n  timestamp: string\n  /** Lifecycle stage this event represents. */\n  status: EvalRunStatus\n  /** Free-form consumer tags (env, branch, model id, etc.). Searchable. */\n  labels: Record<string, string>\n  /** Baseline campaign snapshot. Present when status >= baseline-complete. */\n  baseline?: EvalRunGenerationSnapshot\n  /** Per-generation snapshots. Streams in; orchestrator appends. */\n  generations: EvalRunGenerationSnapshot[]\n  /** Final gate decision. Present when status >= gate-decided. */\n  gateDecision?: GateDecision\n  /** Held-out lift = winner-on-holdout - baseline-on-holdout. */\n  holdoutLift?: number\n  /** Total $ spent across baseline + every generation. */\n  totalCostUsd: number\n  /** Total wall-clock duration. */\n  totalDurationMs: number\n  /** Error message if status === 'errored'. */\n  errorMessage?: string\n  /** Rigor packet emitted alongside the run — distributional summary,\n   *  paired-bootstrap lift CI, judge stats, inter-rater agreement,\n   *  contamination check, failure clusters (when an analyst is wired),\n   *  outcome correlation (when downstream signal is supplied), and the\n   *  recommendations the dashboard surfaces verbatim. */\n  insightReport?: InsightReport\n}\n\n// ── Trace span event ────────────────────────────────────────────────\n\n/**\n * Canonical unsigned 64-bit integer encoded as a base-10 string.\n * JSON numbers cannot represent OTLP nanosecond timestamps exactly.\n */\nexport type UnixNanoTimestamp = string\n\n/**\n * OTel-shape span with a few additional attributes for eval-run pivoting.\n * Compatible with any OTLP collector — `name`, `traceId`, `spanId`,\n * `startTimeUnixNano`, `endTimeUnixNano`, `attributes` are stock OTel.\n */\nexport interface TraceSpanEvent {\n  traceId: string\n  spanId: string\n  parentSpanId?: string\n  name: string\n  startTimeUnixNano: UnixNanoTimestamp\n  endTimeUnixNano: UnixNanoTimestamp\n  attributes: Record<string, string | number | boolean>\n  events?: Array<{\n    timeUnixNano: UnixNanoTimestamp\n    name: string\n    attributes?: Record<string, string | number | boolean>\n  }>\n  status?: { code: 'OK' | 'ERROR' | 'UNSET'; message?: string }\n  /** Pivot back into the eval-run stream. */\n  'tangle.runId'?: string\n  /** Pivot to the specific generation. */\n  'tangle.generation'?: number\n  /** Pivot to the specific cell. */\n  'tangle.cellId'?: string\n  /** Pivot to the specific scenario. */\n  'tangle.scenarioId'?: string\n}\n\n// ── Ingest request bodies ───────────────────────────────────────────\n\nexport interface IngestEvalRunsRequest {\n  wireVersion: HostedWireVersion\n  events: EvalRunEvent[]\n}\n\nexport interface IngestTracesRequest {\n  wireVersion: HostedWireVersion\n  spans: TraceSpanEvent[]\n}\n\nexport interface IngestResponse {\n  /** Accepted events / spans count. */\n  accepted: number\n  /** Rejected events with reasons (validation failures, dup idempotency key, etc.). */\n  rejected: Array<{ index: number; reason: string }>\n}\n","import { z } from 'zod'\nimport type { MutableSurface } from '../campaign/types'\nimport type { InsightReport } from '../contract/insight-report'\nimport { FAILURE_CLASSES, type FailureClass } from '../trace/schema'\nimport type {\n  EvalRunCellScore,\n  EvalRunEvent,\n  EvalRunGenerationSnapshot,\n  IngestEvalRunsRequest,\n  IngestResponse,\n  IngestTracesRequest,\n  TraceSpanEvent,\n  UnixNanoTimestamp,\n} from './types'\nimport { HOSTED_WIRE_VERSION } from './types'\n\nconst finiteNumber = z.number().finite()\nconst nonNegativeNumber = finiteNumber.nonnegative()\nconst nonNegativeInteger = z.number().int().nonnegative()\nconst nonEmptyString = z\n  .string()\n  .min(1)\n  .refine((value) => value.trim().length > 0, 'must not be blank')\nconst attributeValue = z.union([z.string(), finiteNumber, z.boolean()])\nconst attributes = z.record(z.string(), attributeValue)\nconst FailureClassSchema = z.custom<FailureClass>(\n  (value) => typeof value === 'string' && FAILURE_CLASSES.includes(value as FailureClass),\n  `expected one of ${FAILURE_CLASSES.join(', ')}`,\n)\nconst UINT64_MAX = 18_446_744_073_709_551_615n\n\nexport const UnixNanoTimestampSchema: z.ZodType<UnixNanoTimestamp> = z\n  .string()\n  .regex(/^(0|[1-9][0-9]*)$/, 'expected an unsigned base-10 integer string')\n  .refine((value) => BigInt(value) <= UINT64_MAX, 'must fit in an unsigned 64-bit integer')\n\nconst GainDistributionBinSchema = z\n  .object({\n    lo: finiteNumber,\n    hi: finiteNumber,\n    count: nonNegativeInteger,\n  })\n  .strict()\n\nconst ScalarDistributionSchema = z\n  .object({\n    n: nonNegativeInteger,\n    mean: finiteNumber.nullable(),\n    p50: finiteNumber.nullable(),\n    p95: finiteNumber.nullable(),\n    stddev: nonNegativeNumber.nullable(),\n    min: finiteNumber.nullable(),\n    max: finiteNumber.nullable(),\n    histogram: z.array(GainDistributionBinSchema),\n    tailRuns: z\n      .array(\n        z\n          .object({\n            runId: nonEmptyString,\n            score: finiteNumber,\n          })\n          .strict(),\n      )\n      .optional(),\n  })\n  .strict()\n  .superRefine((distribution, ctx) => {\n    const values = [\n      distribution.mean,\n      distribution.p50,\n      distribution.p95,\n      distribution.stddev,\n      distribution.min,\n      distribution.max,\n    ]\n    if (distribution.n === 0 && values.some((value) => value !== null)) {\n      ctx.addIssue({\n        code: 'custom',\n        message: 'distribution values must be null when n is 0',\n      })\n    }\n    if (distribution.n > 0 && values.some((value) => value === null)) {\n      ctx.addIssue({\n        code: 'custom',\n        message: 'distribution values must be numbers when n is greater than 0',\n      })\n    }\n    if (\n      distribution.min !== null &&\n      distribution.max !== null &&\n      distribution.min > distribution.max\n    ) {\n      ctx.addIssue({ code: 'custom', path: ['min'], message: 'min must not exceed max' })\n    }\n  })\n\nconst TokenUsageInsightSchema = z\n  .object({\n    input: ScalarDistributionSchema,\n    output: ScalarDistributionSchema,\n    reasoning: ScalarDistributionSchema,\n    cached: ScalarDistributionSchema,\n    cacheWrite: ScalarDistributionSchema,\n    totals: z\n      .object({\n        input: nonNegativeNumber,\n        output: nonNegativeNumber,\n        reasoning: nonNegativeNumber,\n        cached: nonNegativeNumber,\n        cacheWrite: nonNegativeNumber,\n      })\n      .strict(),\n  })\n  .strict()\n\nconst ExecutionErrorOutcomeCellSchema = z\n  .object({\n    withErrors: nonNegativeInteger,\n    withoutErrors: nonNegativeInteger,\n    unreported: nonNegativeInteger,\n  })\n  .strict()\n\nconst ExecutionInsightSchema = z\n  .object({\n    durationMs: ScalarDistributionSchema,\n    queueMs: ScalarDistributionSchema,\n    tokenUsage: TokenUsageInsightSchema,\n    aggregateUsage: z\n      .object({\n        runs: nonNegativeInteger,\n        tokenUsage: TokenUsageInsightSchema,\n        costUsd: ScalarDistributionSchema,\n        totalCostUsd: nonNegativeNumber,\n      })\n      .strict(),\n    models: z.array(\n      z\n        .object({\n          model: nonEmptyString,\n          runs: nonNegativeInteger,\n        })\n        .strict(),\n    ),\n    modelCalls: z\n      .object({\n        runs: nonNegativeInteger,\n        events: nonNegativeInteger,\n        reportingRuns: nonNegativeInteger,\n      })\n      .strict(),\n    executionErrors: z\n      .object({\n        runs: nonNegativeInteger,\n        fraction: finiteNumber.min(0).max(1).nullable(),\n        events: nonNegativeInteger,\n        reportingRuns: nonNegativeInteger,\n        errorSpanEvents: nonNegativeInteger,\n        errorSpanReportingRuns: nonNegativeInteger,\n        byTerminalOutcome: z\n          .object({\n            succeeded: ExecutionErrorOutcomeCellSchema,\n            failed: ExecutionErrorOutcomeCellSchema,\n            cancelled: ExecutionErrorOutcomeCellSchema,\n            incomplete: ExecutionErrorOutcomeCellSchema,\n            unknown: ExecutionErrorOutcomeCellSchema,\n          })\n          .strict(),\n      })\n      .strict(),\n    terminalOutcomes: z\n      .object({\n        succeeded: nonNegativeInteger,\n        failed: nonNegativeInteger,\n        cancelled: nonNegativeInteger,\n        incomplete: nonNegativeInteger,\n        unknown: nonNegativeInteger,\n      })\n      .strict(),\n  })\n  .strict()\n\nconst CostProvenanceSummarySchema = z\n  .object({\n    observed: z\n      .object({\n        n: nonNegativeInteger,\n        totalUsd: nonNegativeNumber,\n      })\n      .strict(),\n    estimated: z\n      .object({\n        n: nonNegativeInteger,\n        totalUsd: nonNegativeNumber,\n      })\n      .strict(),\n    lowerBound: z\n      .object({\n        n: nonNegativeInteger,\n        floorUsd: nonNegativeNumber,\n      })\n      .strict(),\n    uncaptured: z.object({ n: nonNegativeInteger }).strict(),\n    knownFraction: finiteNumber.min(0).max(1),\n  })\n  .strict()\n\nconst ParetoFigureSpecSchema = z\n  .object({\n    kind: z.literal('pareto-cost-quality'),\n    split: z.enum(['search', 'holdout']),\n    points: z.array(\n      z\n        .object({\n          candidateId: nonEmptyString,\n          cost: nonNegativeNumber,\n          quality: finiteNumber,\n          n: nonNegativeInteger,\n          onFrontier: z.boolean(),\n          gate: z.enum(['promote', 'reject']).optional(),\n        })\n        .strict(),\n    ),\n    axes: z.object({ x: z.literal('costUsd'), y: z.literal('score') }).strict(),\n  })\n  .strict()\n\nconst ContinuousAgreementSchema = z\n  .object({\n    weightedKappa: finiteNumber,\n    icc: finiteNumber,\n    pearson: finiteNumber,\n    spearman: finiteNumber,\n    ci: z\n      .object({\n        icc: z.tuple([finiteNumber, finiteNumber]),\n        weightedKappa: z.tuple([finiteNumber, finiteNumber]),\n      })\n      .strict(),\n    n: nonNegativeInteger,\n    raters: nonNegativeInteger,\n  })\n  .strict()\n\nconst JudgeInsightSchema = z\n  .object({\n    n: nonNegativeInteger,\n    meanScore: finiteNumber,\n    calibration: ContinuousAgreementSchema.optional(),\n    positionalBias: finiteNumber.optional(),\n    selfPreference: finiteNumber.optional(),\n    verbosityBias: finiteNumber.optional(),\n  })\n  .strict()\n\nconst InterRaterInsightSchema = z\n  .object({\n    raters: nonNegativeInteger,\n    jointlyRated: nonNegativeInteger,\n    kappa: finiteNumber,\n    icc: finiteNumber,\n    pearson: finiteNumber,\n    spearman: finiteNumber,\n    perPair: z.record(z.string(), finiteNumber),\n    disagreementCases: z.array(\n      z\n        .object({\n          runId: nonEmptyString,\n          ratings: z.array(\n            z\n              .object({\n                rater: nonEmptyString,\n                score: finiteNumber,\n              })\n              .strict(),\n          ),\n          range: nonNegativeNumber,\n        })\n        .strict(),\n    ),\n  })\n  .strict()\n\nconst LiftInsightSchema = z\n  .object({\n    baselineMean: finiteNumber,\n    candidateMean: finiteNumber,\n    delta: finiteNumber,\n    ci95: z.tuple([finiteNumber, finiteNumber]),\n    pValue: finiteNumber.min(0).max(1).nullable(),\n    n: nonNegativeInteger,\n    pairedRunN: nonNegativeInteger.optional(),\n    independentUnitIds: z.array(nonEmptyString).optional(),\n    minimumRequired: z.number().int().positive(),\n    decisionEligible: z.boolean(),\n    unpairedBaseline: nonNegativeInteger,\n    unpairedCandidate: nonNegativeInteger,\n    cohensD: finiteNumber.nullable(),\n    mde: nonNegativeNumber,\n    requiredN: nonNegativeInteger.nullable(),\n  })\n  .strict()\n\nconst FailureClusterInsightSchema = z\n  .object({\n    clusters: z.array(\n      z\n        .object({\n          id: nonEmptyString,\n          name: nonEmptyString,\n          share: finiteNumber.min(0).max(1),\n          exemplars: z.array(nonEmptyString).max(5),\n          suggestedFix: nonEmptyString.optional(),\n        })\n        .strict(),\n    ),\n    totalFailures: nonNegativeInteger,\n  })\n  .strict()\n\nconst ContaminationInsightSchema = z\n  .object({\n    leaks: nonNegativeInteger,\n    holdoutAuditPassed: z.boolean(),\n    details: z\n      .array(\n        z\n          .object({\n            runId: nonEmptyString,\n            canary: nonEmptyString,\n            matched: nonEmptyString,\n          })\n          .strict(),\n      )\n      .optional(),\n  })\n  .strict()\n\nconst OutcomeCorrelationInsightSchema = z\n  .object({\n    metric: nonEmptyString,\n    n: nonNegativeInteger,\n    pearson: finiteNumber,\n    spearman: finiteNumber,\n    rewardModel: z\n      .object({\n        intercept: finiteNumber,\n        slope: finiteNumber,\n        r2: finiteNumber,\n      })\n      .strict()\n      .optional(),\n  })\n  .strict()\n\nconst ReleaseSummarySchema = z\n  .object({\n    status: z.enum(['pass', 'warn', 'fail']),\n    axes: z.array(\n      z\n        .object({\n          name: z.enum(['quality-lift', 'contamination', 'composite-distribution']),\n          status: z.enum(['pass', 'warn', 'fail', 'not_evaluated']),\n          detail: nonEmptyString,\n        })\n        .strict(),\n    ),\n    issues: z.array(z.string()),\n  })\n  .strict()\n\nconst MetricDeltaBaseSchema = z\n  .object({\n    current: finiteNumber,\n    baseline: finiteNumber,\n    delta: finiteNumber,\n    baselineN: nonNegativeInteger,\n    currentN: nonNegativeInteger,\n  })\n  .strict()\n\nconst MetricDeltaSchema = z.discriminatedUnion('status', [\n  MetricDeltaBaseSchema.extend({\n    status: z.literal('ok'),\n    ci95: z.tuple([finiteNumber, finiteNumber]),\n    pValue: finiteNumber.min(0).max(1),\n    cohensD: finiteNumber,\n    significant: z.boolean(),\n  }),\n  MetricDeltaBaseSchema.extend({\n    status: z.enum(['insufficient-sample', 'zero-variance']),\n    ci95: z.null(),\n    pValue: z.null(),\n    cohensD: z.null(),\n    significant: z.literal(false),\n  }),\n])\n\nconst PriorPeriodComparisonSchema = z\n  .object({\n    baselineN: nonNegativeInteger,\n    currentN: nonNegativeInteger,\n    windowLabel: nonEmptyString.optional(),\n    metrics: z.record(z.string(), MetricDeltaSchema),\n    regressedMetrics: z.array(nonEmptyString),\n    improvedMetrics: z.array(nonEmptyString),\n    inconclusiveMetrics: z.array(nonEmptyString),\n  })\n  .strict()\n\nconst RecommendationSchema = z\n  .object({\n    priority: z.enum(['critical', 'high', 'medium', 'low']),\n    kind: z.enum(['ship', 'hold', 'investigate', 'fix', 'recalibrate', 'expand-corpus']),\n    title: nonEmptyString,\n    detail: nonEmptyString,\n    evidencePath: nonEmptyString.optional(),\n  })\n  .strict()\n\nexport const InsightReportSchema: z.ZodType<InsightReport> = z\n  .object({\n    n: nonNegativeInteger,\n    execution: ExecutionInsightSchema,\n    composite: ScalarDistributionSchema,\n    perDimension: z.record(z.string(), ScalarDistributionSchema),\n    costQuality: z\n      .object({\n        cost: ScalarDistributionSchema,\n        pareto: ParetoFigureSpecSchema,\n        provenance: CostProvenanceSummarySchema.optional(),\n        degraded: z\n          .object({\n            cost: nonEmptyString.optional(),\n            pareto: nonEmptyString.optional(),\n          })\n          .strict()\n          .optional(),\n      })\n      .strict(),\n    judges: z.record(z.string(), JudgeInsightSchema),\n    interRater: InterRaterInsightSchema.optional(),\n    lift: LiftInsightSchema.optional(),\n    failureClusters: FailureClusterInsightSchema.optional(),\n    contamination: ContaminationInsightSchema.optional(),\n    outcomeCorrelation: OutcomeCorrelationInsightSchema.optional(),\n    release: ReleaseSummarySchema,\n    priorPeriodComparison: PriorPeriodComparisonSchema.optional(),\n    failureClasses: z\n      .array(\n        z\n          .object({\n            failureClass: FailureClassSchema,\n            count: nonNegativeInteger,\n            share: finiteNumber.min(0).max(1),\n          })\n          .strict(),\n      )\n      .optional(),\n    recommendations: z.array(RecommendationSchema),\n  })\n  .strict()\n\nconst sha256Digest = z.custom<`sha256:${string}`>(\n  (value) => typeof value === 'string' && /^sha256:[0-9a-f]{64}$/.test(value),\n  'expected sha256:<64 lowercase hex characters>',\n)\n\nconst MutableSurfaceSchema: z.ZodType<MutableSurface> = z.union([\n  z.string(),\n  z\n    .object({\n      kind: z.literal('components'),\n      components: z.record(z.string(), z.string()),\n    })\n    .strict(),\n  z\n    .object({\n      kind: z.literal('code'),\n      worktreeRef: nonEmptyString,\n      baseRef: nonEmptyString,\n      baseCommit: nonEmptyString,\n      baseTree: nonEmptyString,\n      candidateCommit: nonEmptyString,\n      candidateTree: nonEmptyString,\n      patch: z\n        .object({\n          format: z.literal('git-diff-binary'),\n          sha256: sha256Digest,\n          byteLength: nonNegativeInteger,\n        })\n        .strict(),\n      summary: z.string().optional(),\n    })\n    .strict(),\n])\n\nconst RunTerminalOutcomeSchema = z.enum([\n  'succeeded',\n  'failed',\n  'cancelled',\n  'incomplete',\n  'unknown',\n])\n\nconst EvalRunCellScoreSchema: z.ZodType<EvalRunCellScore> = z\n  .object({\n    scenarioId: nonEmptyString,\n    rep: nonNegativeInteger,\n    compositeMean: finiteNumber.nullable(),\n    dimensions: z.record(z.string(), z.record(z.string(), finiteNumber)),\n    terminalOutcome: RunTerminalOutcomeSchema,\n    executionErrorCount: nonNegativeInteger.nullable(),\n    errorMessage: z.string().optional(),\n  })\n  .strict()\n\nconst EvalRunGenerationSnapshotSchema: z.ZodType<EvalRunGenerationSnapshot> = z\n  .object({\n    index: nonNegativeInteger,\n    surfaceHash: nonEmptyString,\n    surface: MutableSurfaceSchema.optional(),\n    cells: z.array(EvalRunCellScoreSchema),\n    compositeMean: finiteNumber.nullable(),\n    costUsd: nonNegativeNumber,\n    durationMs: nonNegativeNumber,\n  })\n  .strict()\n\nconst EvalRunStatusSchema = z.enum([\n  'started',\n  'baseline-complete',\n  'generation-complete',\n  'gate-decided',\n  'finished',\n  'errored',\n])\n\nconst GateDecisionSchema = z.enum([\n  'ship',\n  'hold',\n  'need_more_work',\n  'model_ceiling',\n  'arch_ceiling',\n])\n\nexport const EvalRunEventSchema: z.ZodType<EvalRunEvent> = z\n  .object({\n    runId: nonEmptyString,\n    runDir: nonEmptyString,\n    timestamp: z.string().datetime({ offset: true }),\n    status: EvalRunStatusSchema,\n    labels: z.record(z.string(), z.string()),\n    baseline: EvalRunGenerationSnapshotSchema.optional(),\n    generations: z.array(EvalRunGenerationSnapshotSchema),\n    gateDecision: GateDecisionSchema.optional(),\n    holdoutLift: finiteNumber.optional(),\n    totalCostUsd: nonNegativeNumber,\n    totalDurationMs: nonNegativeNumber,\n    errorMessage: z.string().optional(),\n    insightReport: InsightReportSchema.optional(),\n  })\n  .strict()\n  .superRefine((event, ctx) => {\n    if (event.baseline && event.baseline.index !== 0) {\n      ctx.addIssue({\n        code: 'custom',\n        path: ['baseline', 'index'],\n        message: 'baseline index must be 0',\n      })\n    }\n\n    const seen = new Set<number>()\n    for (let i = 0; i < event.generations.length; i++) {\n      const index = event.generations[i]!.index\n      if (seen.has(index)) {\n        ctx.addIssue({\n          code: 'custom',\n          path: ['generations', i, 'index'],\n          message: `duplicate generation index ${index}`,\n        })\n      }\n      seen.add(index)\n    }\n\n    if (event.status === 'errored' && !event.errorMessage?.trim()) {\n      ctx.addIssue({\n        code: 'custom',\n        path: ['errorMessage'],\n        message: 'errorMessage is required when status is errored',\n      })\n    }\n  })\n\nconst TraceSpanEventEntrySchema = z\n  .object({\n    timeUnixNano: UnixNanoTimestampSchema,\n    name: nonEmptyString,\n    attributes: attributes.optional(),\n  })\n  .strict()\n\nexport const TraceSpanEventSchema: z.ZodType<TraceSpanEvent> = z\n  .object({\n    traceId: nonEmptyString,\n    spanId: nonEmptyString,\n    parentSpanId: nonEmptyString.optional(),\n    name: nonEmptyString,\n    startTimeUnixNano: UnixNanoTimestampSchema,\n    endTimeUnixNano: UnixNanoTimestampSchema,\n    attributes,\n    events: z.array(TraceSpanEventEntrySchema).optional(),\n    status: z\n      .object({\n        code: z.enum(['OK', 'ERROR', 'UNSET']),\n        message: z.string().optional(),\n      })\n      .strict()\n      .optional(),\n    'tangle.runId': nonEmptyString.optional(),\n    'tangle.generation': nonNegativeInteger.optional(),\n    'tangle.cellId': nonEmptyString.optional(),\n    'tangle.scenarioId': nonEmptyString.optional(),\n  })\n  .strict()\n  .refine((span) => BigInt(span.endTimeUnixNano) >= BigInt(span.startTimeUnixNano), {\n    path: ['endTimeUnixNano'],\n    message: 'endTimeUnixNano must be greater than or equal to startTimeUnixNano',\n  })\n\nexport const IngestEvalRunsEnvelopeSchema = z\n  .object({\n    wireVersion: z.literal(HOSTED_WIRE_VERSION),\n    events: z.array(z.unknown()),\n  })\n  .strict()\n\nexport const IngestTracesEnvelopeSchema = z\n  .object({\n    wireVersion: z.literal(HOSTED_WIRE_VERSION),\n    spans: z.array(z.unknown()),\n  })\n  .strict()\n\nexport const IngestEvalRunsRequestSchema: z.ZodType<IngestEvalRunsRequest> = z\n  .object({\n    wireVersion: z.literal(HOSTED_WIRE_VERSION),\n    events: z.array(EvalRunEventSchema),\n  })\n  .strict()\n\nexport const IngestTracesRequestSchema: z.ZodType<IngestTracesRequest> = z\n  .object({\n    wireVersion: z.literal(HOSTED_WIRE_VERSION),\n    spans: z.array(TraceSpanEventSchema),\n  })\n  .strict()\n\nexport const IngestResponseSchema: z.ZodType<IngestResponse> = z\n  .object({\n    accepted: nonNegativeInteger,\n    rejected: z.array(\n      z\n        .object({\n          index: nonNegativeInteger,\n          reason: nonEmptyString,\n        })\n        .strict(),\n    ),\n  })\n  .strict()\n","/**\n * # Hosted-tier ingest client.\n *\n * Ships eval-run events + trace spans to any orchestrator (ours, a\n * partner's self-hosted one, or a future open implementation) that\n * speaks the wire format in `./types.ts`.\n *\n * Three modes:\n *   - **Ours:** point at `https://orchestrator.tangle.tools` (the host root —\n *     the client appends the versioned `/v1/ingest/...` path itself; a trailing\n *     `/v1` on the endpoint is tolerated and normalized away). We handle ingest\n *     + storage + dashboard.\n *   - **Self-hosted:** point at whatever URL runs the reference receiver\n *     from `examples/hosted-ingest-server/`.\n *   - **Off (default):** when `hostedTenant` is unset, nothing is sent.\n *     Everything stays local.\n */\n\nimport {\n  IngestEvalRunsRequestSchema,\n  IngestResponseSchema,\n  IngestTracesRequestSchema,\n} from './schemas'\nimport {\n  type EvalRunEvent,\n  HOSTED_WIRE_VERSION,\n  type HostedWireVersion,\n  type IngestEvalRunsRequest,\n  type IngestResponse,\n  type IngestTracesRequest,\n  type TraceSpanEvent,\n} from './types'\n\nexport interface HostedTenant {\n  /** Orchestrator endpoint base URL (no trailing slash). Required. */\n  endpoint: string\n  /** Bearer token issued by the orchestrator. Required. */\n  apiKey: string\n  /** Tenant id — the orchestrator's primary key for this consumer. Required. */\n  tenantId: string\n  /** Optional `fetch` override (auth wrappers, custom agent, test mocks). */\n  fetchImpl?: typeof fetch\n  /** Per-call timeout in ms. Default 30s. */\n  timeoutMs?: number\n  /** Retries on 5xx / network errors. Default 2. */\n  retries?: number\n}\n\nexport interface HostedClient {\n  ingestEvalRun(event: EvalRunEvent, idempotencyKey?: string): Promise<IngestResponse>\n  ingestEvalRuns(events: EvalRunEvent[], idempotencyKey?: string): Promise<IngestResponse>\n  ingestTraces(spans: TraceSpanEvent[], idempotencyKey?: string): Promise<IngestResponse>\n  readonly tenant: HostedTenant\n  readonly wireVersion: HostedWireVersion\n}\n\ninterface RequestOptions {\n  idempotencyKey?: string\n  signal?: AbortSignal\n}\n\nconst MAX_IDEMPOTENCY_KEY_LENGTH = 256\n\nfunction sleep(ms: number): Promise<void> {\n  return new Promise((resolve) => setTimeout(resolve, ms))\n}\n\nfunction normalizeHostedBase(endpoint: string): string {\n  return endpoint.trim().replace(/\\/+$/, '').replace(/\\/v1$/, '')\n}\n\nfunction resolveIdempotencyKey(key: string | undefined): string {\n  const resolved = key ?? globalThis.crypto.randomUUID()\n  if (resolved.trim().length === 0) throw new Error('idempotency key must not be blank')\n  if (resolved.length > MAX_IDEMPOTENCY_KEY_LENGTH) {\n    throw new Error(`idempotency key must be at most ${MAX_IDEMPOTENCY_KEY_LENGTH} characters`)\n  }\n  return resolved\n}\n\nfunction responseValidationReason(error: {\n  issues: Array<{ path: PropertyKey[]; message: string }>\n}) {\n  return error.issues\n    .map(\n      (issue) =>\n        `${issue.path.length > 0 ? issue.path.map(String).join('.') : 'value'}: ${issue.message}`,\n    )\n    .join('; ')\n}\n\nasync function post<TReq>(\n  tenant: HostedTenant,\n  path: string,\n  body: TReq,\n  opts: RequestOptions = {},\n): Promise<IngestResponse> {\n  const timeoutMs = tenant.timeoutMs ?? 30_000\n  const maxRetries = tenant.retries ?? 2\n  const f: typeof fetch = tenant.fetchImpl ?? ((...args) => fetch(...args))\n  const base = normalizeHostedBase(tenant.endpoint)\n  const url = `${base}${path}`\n  const idempotencyKey = resolveIdempotencyKey(opts.idempotencyKey)\n  const headers: Record<string, string> = {\n    'content-type': 'application/json',\n    authorization: `Bearer ${tenant.apiKey}`,\n    'x-tangle-tenant-id': tenant.tenantId,\n    'x-tangle-wire-version': HOSTED_WIRE_VERSION,\n    'idempotency-key': idempotencyKey,\n  }\n\n  let lastError: unknown\n  for (let attempt = 0; attempt <= maxRetries; attempt++) {\n    const ourTimeout = AbortSignal.timeout(timeoutMs)\n    const combinedSignal = opts.signal ? AbortSignal.any([opts.signal, ourTimeout]) : ourTimeout\n    let res: Response\n    try {\n      res = await f(url, {\n        method: 'POST',\n        headers,\n        body: JSON.stringify(body),\n        signal: combinedSignal,\n      })\n    } catch (err) {\n      if (opts.signal?.aborted) throw err\n      lastError = err\n      if (attempt === maxRetries) throw err\n      await sleep(2 ** attempt * 200 + Math.random() * 200)\n      continue\n    }\n\n    if (!res.ok) {\n      const text = await res.text().catch(() => '')\n      const error = new Error(`hosted ingest ${url} failed (${res.status}): ${text.slice(0, 500)}`)\n      const retryable = res.status >= 500 || res.status === 408 || res.status === 429\n      if (!retryable || attempt === maxRetries) throw error\n      lastError = error\n      await sleep(2 ** attempt * 200 + Math.random() * 200)\n      continue\n    }\n\n    let rawResponse: unknown\n    try {\n      rawResponse = await res.json()\n    } catch (error) {\n      throw new Error(`hosted ingest ${url} returned invalid JSON`, { cause: error })\n    }\n    const parsed = IngestResponseSchema.safeParse(rawResponse)\n    if (!parsed.success) {\n      throw new Error(\n        `hosted ingest ${url} returned an invalid response: ${responseValidationReason(parsed.error)}`,\n      )\n    }\n    return parsed.data\n  }\n  throw lastError ?? new Error('hosted ingest exhausted retries')\n}\n\nexport function createHostedClient(tenant: HostedTenant): HostedClient {\n  if (normalizeHostedBase(tenant.endpoint).length === 0) throw new Error('endpoint is required')\n  if (tenant.apiKey.trim().length === 0) throw new Error('apiKey is required')\n  if (tenant.tenantId.trim().length === 0) throw new Error('tenantId is required')\n  if (\n    tenant.timeoutMs !== undefined &&\n    (!Number.isFinite(tenant.timeoutMs) || tenant.timeoutMs <= 0)\n  ) {\n    throw new Error('timeoutMs must be greater than 0')\n  }\n  if (tenant.retries !== undefined && (!Number.isInteger(tenant.retries) || tenant.retries < 0)) {\n    throw new Error('retries must be a non-negative integer')\n  }\n\n  return {\n    tenant,\n    wireVersion: HOSTED_WIRE_VERSION,\n\n    async ingestEvalRun(event, idempotencyKey) {\n      return this.ingestEvalRuns([event], idempotencyKey)\n    },\n\n    async ingestEvalRuns(events, idempotencyKey) {\n      const body: IngestEvalRunsRequest = IngestEvalRunsRequestSchema.parse({\n        wireVersion: HOSTED_WIRE_VERSION,\n        events,\n      })\n      return post<IngestEvalRunsRequest>(tenant, '/v1/ingest/eval-runs', body, {\n        idempotencyKey,\n      })\n    },\n\n    async ingestTraces(spans, idempotencyKey) {\n      const body: IngestTracesRequest = IngestTracesRequestSchema.parse({\n        wireVersion: HOSTED_WIRE_VERSION,\n        spans,\n      })\n      return post<IngestTracesRequest>(tenant, '/v1/ingest/traces', body, {\n        idempotencyKey,\n      })\n    },\n  }\n}\n\n/**\n * Build a `HostedClient` from environment, or `undefined` when ingest is not\n * configured — the canonical, fail-soft wiring every product uses so eval-run +\n * trace provenance lands in the Intelligence dashboard with ONE call:\n *\n *   const hosted = hostedClientFromEnv()\n *   // ...run the loop...\n *   await emitLoopProvenance({ ..., hostedClient: hosted })  // no-op if undefined\n *\n * Returns `undefined` (NOT an error) when any of endpoint / apiKey / tenantId is\n * missing — so a product wires the ship call unconditionally and it stays a\n * no-op until the env is set. Env precedence:\n *   - endpoint:  `TANGLE_INGEST_URL` → `TANGLE_ORCHESTRATOR_URL`\n *   - apiKey:    `TANGLE_INGEST_API_KEY` → `TANGLE_API_KEY`\n *   - tenantId:  `TANGLE_TENANT_ID`\n * A trailing slash on the endpoint is stripped. Pass `overrides` to supply any\n * field directly (e.g. a fixed `tenantId` per product) — overrides win over env.\n */\n/**\n * Build a {@link HostedTenant} config from env — the input `selfImprove`'s\n * `hostedTenant` and `emitLoopProvenance` take. Same env precedence + overrides\n * as {@link hostedClientFromEnv}; returns `undefined` (not an error) when any of\n * endpoint / apiKey / tenantId is missing, so a product wires\n * `hostedTenant: hostedTenantFromEnv({ tenantId: 'my-agent' })` unconditionally\n * and it stays off until the env is set.\n */\nexport function hostedTenantFromEnv(\n  overrides: Partial<HostedTenant> & { env?: Record<string, string | undefined> } = {},\n): HostedTenant | undefined {\n  const env = overrides.env ?? process.env\n  const endpoint = (\n    overrides.endpoint ??\n    env.TANGLE_INGEST_URL ??\n    env.TANGLE_ORCHESTRATOR_URL\n  )?.trim()\n  const apiKey = (overrides.apiKey ?? env.TANGLE_INGEST_API_KEY ?? env.TANGLE_API_KEY)?.trim()\n  const tenantId = (overrides.tenantId ?? env.TANGLE_TENANT_ID)?.trim()\n  if (!endpoint || !apiKey || !tenantId) return undefined\n  const tenant: HostedTenant = { endpoint: endpoint.replace(/\\/+$/, ''), apiKey, tenantId }\n  if (overrides.fetchImpl) tenant.fetchImpl = overrides.fetchImpl\n  if (overrides.timeoutMs !== undefined) tenant.timeoutMs = overrides.timeoutMs\n  if (overrides.retries !== undefined) tenant.retries = overrides.retries\n  return tenant\n}\n\nexport function hostedClientFromEnv(\n  overrides: Partial<HostedTenant> & { env?: Record<string, string | undefined> } = {},\n): HostedClient | undefined {\n  const tenant = hostedTenantFromEnv(overrides)\n  return tenant ? createHostedClient(tenant) : undefined\n}\n"],"mappings":";;;AAgCA,MAAa,sBAAsB;;;AChBnC,MAAM,eAAe,EAAE,OAAO,CAAC,CAAC,OAAO;AACvC,MAAM,oBAAoB,aAAa,YAAY;AACnD,MAAM,qBAAqB,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,YAAY;AACxD,MAAM,iBAAiB,EACpB,OAAO,CAAC,CACR,IAAI,CAAC,CAAC,CACN,QAAQ,UAAU,MAAM,KAAK,CAAC,CAAC,SAAS,GAAG,mBAAmB;AACjE,MAAM,iBAAiB,EAAE,MAAM;CAAC,EAAE,OAAO;CAAG;CAAc,EAAE,QAAQ;AAAC,CAAC;AACtE,MAAM,aAAa,EAAE,OAAO,EAAE,OAAO,GAAG,cAAc;AACtD,MAAM,qBAAqB,EAAE,QAC1B,UAAU,OAAO,UAAU,YAAY,gBAAgB,SAAS,KAAqB,GACtF,mBAAmB,gBAAgB,KAAK,IAAI,GAC9C;AACA,MAAM,aAAa;AAEnB,MAAa,0BAAwD,EAClE,OAAO,CAAC,CACR,MAAM,qBAAqB,6CAA6C,CAAC,CACzE,QAAQ,UAAU,OAAO,KAAK,KAAK,YAAY,wCAAwC;AAE1F,MAAM,4BAA4B,EAC/B,OAAO;CACN,IAAI;CACJ,IAAI;CACJ,OAAO;AACT,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,2BAA2B,EAC9B,OAAO;CACN,GAAG;CACH,MAAM,aAAa,SAAS;CAC5B,KAAK,aAAa,SAAS;CAC3B,KAAK,aAAa,SAAS;CAC3B,QAAQ,kBAAkB,SAAS;CACnC,KAAK,aAAa,SAAS;CAC3B,KAAK,aAAa,SAAS;CAC3B,WAAW,EAAE,MAAM,yBAAyB;CAC5C,UAAU,EACP,MACC,EACG,OAAO;EACN,OAAO;EACP,OAAO;CACT,CAAC,CAAC,CACD,OAAO,CACZ,CAAC,CACA,SAAS;AACd,CAAC,CAAC,CACD,OAAO,CAAC,CACR,aAAa,cAAc,QAAQ;CAClC,MAAM,SAAS;EACb,aAAa;EACb,aAAa;EACb,aAAa;EACb,aAAa;EACb,aAAa;EACb,aAAa;CACf;CACA,IAAI,aAAa,MAAM,KAAK,OAAO,MAAM,UAAU,UAAU,IAAI,GAC/D,IAAI,SAAS;EACX,MAAM;EACN,SAAS;CACX,CAAC;CAEH,IAAI,aAAa,IAAI,KAAK,OAAO,MAAM,UAAU,UAAU,IAAI,GAC7D,IAAI,SAAS;EACX,MAAM;EACN,SAAS;CACX,CAAC;CAEH,IACE,aAAa,QAAQ,QACrB,aAAa,QAAQ,QACrB,aAAa,MAAM,aAAa,KAEhC,IAAI,SAAS;EAAE,MAAM;EAAU,MAAM,CAAC,KAAK;EAAG,SAAS;CAA0B,CAAC;AAEtF,CAAC;AAEH,MAAM,0BAA0B,EAC7B,OAAO;CACN,OAAO;CACP,QAAQ;CACR,WAAW;CACX,QAAQ;CACR,YAAY;CACZ,QAAQ,EACL,OAAO;EACN,OAAO;EACP,QAAQ;EACR,WAAW;EACX,QAAQ;EACR,YAAY;CACd,CAAC,CAAC,CACD,OAAO;AACZ,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,kCAAkC,EACrC,OAAO;CACN,YAAY;CACZ,eAAe;CACf,YAAY;AACd,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,yBAAyB,EAC5B,OAAO;CACN,YAAY;CACZ,SAAS;CACT,YAAY;CACZ,gBAAgB,EACb,OAAO;EACN,MAAM;EACN,YAAY;EACZ,SAAS;EACT,cAAc;CAChB,CAAC,CAAC,CACD,OAAO;CACV,QAAQ,EAAE,MACR,EACG,OAAO;EACN,OAAO;EACP,MAAM;CACR,CAAC,CAAC,CACD,OAAO,CACZ;CACA,YAAY,EACT,OAAO;EACN,MAAM;EACN,QAAQ;EACR,eAAe;CACjB,CAAC,CAAC,CACD,OAAO;CACV,iBAAiB,EACd,OAAO;EACN,MAAM;EACN,UAAU,aAAa,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS;EAC9C,QAAQ;EACR,eAAe;EACf,iBAAiB;EACjB,wBAAwB;EACxB,mBAAmB,EAChB,OAAO;GACN,WAAW;GACX,QAAQ;GACR,WAAW;GACX,YAAY;GACZ,SAAS;EACX,CAAC,CAAC,CACD,OAAO;CACZ,CAAC,CAAC,CACD,OAAO;CACV,kBAAkB,EACf,OAAO;EACN,WAAW;EACX,QAAQ;EACR,WAAW;EACX,YAAY;EACZ,SAAS;CACX,CAAC,CAAC,CACD,OAAO;AACZ,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,8BAA8B,EACjC,OAAO;CACN,UAAU,EACP,OAAO;EACN,GAAG;EACH,UAAU;CACZ,CAAC,CAAC,CACD,OAAO;CACV,WAAW,EACR,OAAO;EACN,GAAG;EACH,UAAU;CACZ,CAAC,CAAC,CACD,OAAO;CACV,YAAY,EACT,OAAO;EACN,GAAG;EACH,UAAU;CACZ,CAAC,CAAC,CACD,OAAO;CACV,YAAY,EAAE,OAAO,EAAE,GAAG,mBAAmB,CAAC,CAAC,CAAC,OAAO;CACvD,eAAe,aAAa,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC;AAC1C,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,yBAAyB,EAC5B,OAAO;CACN,MAAM,EAAE,QAAQ,qBAAqB;CACrC,OAAO,EAAE,KAAK,CAAC,UAAU,SAAS,CAAC;CACnC,QAAQ,EAAE,MACR,EACG,OAAO;EACN,aAAa;EACb,MAAM;EACN,SAAS;EACT,GAAG;EACH,YAAY,EAAE,QAAQ;EACtB,MAAM,EAAE,KAAK,CAAC,WAAW,QAAQ,CAAC,CAAC,CAAC,SAAS;CAC/C,CAAC,CAAC,CACD,OAAO,CACZ;CACA,MAAM,EAAE,OAAO;EAAE,GAAG,EAAE,QAAQ,SAAS;EAAG,GAAG,EAAE,QAAQ,OAAO;CAAE,CAAC,CAAC,CAAC,OAAO;AAC5E,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,4BAA4B,EAC/B,OAAO;CACN,eAAe;CACf,KAAK;CACL,SAAS;CACT,UAAU;CACV,IAAI,EACD,OAAO;EACN,KAAK,EAAE,MAAM,CAAC,cAAc,YAAY,CAAC;EACzC,eAAe,EAAE,MAAM,CAAC,cAAc,YAAY,CAAC;CACrD,CAAC,CAAC,CACD,OAAO;CACV,GAAG;CACH,QAAQ;AACV,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,qBAAqB,EACxB,OAAO;CACN,GAAG;CACH,WAAW;CACX,aAAa,0BAA0B,SAAS;CAChD,gBAAgB,aAAa,SAAS;CACtC,gBAAgB,aAAa,SAAS;CACtC,eAAe,aAAa,SAAS;AACvC,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,0BAA0B,EAC7B,OAAO;CACN,QAAQ;CACR,cAAc;CACd,OAAO;CACP,KAAK;CACL,SAAS;CACT,UAAU;CACV,SAAS,EAAE,OAAO,EAAE,OAAO,GAAG,YAAY;CAC1C,mBAAmB,EAAE,MACnB,EACG,OAAO;EACN,OAAO;EACP,SAAS,EAAE,MACT,EACG,OAAO;GACN,OAAO;GACP,OAAO;EACT,CAAC,CAAC,CACD,OAAO,CACZ;EACA,OAAO;CACT,CAAC,CAAC,CACD,OAAO,CACZ;AACF,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,oBAAoB,EACvB,OAAO;CACN,cAAc;CACd,eAAe;CACf,OAAO;CACP,MAAM,EAAE,MAAM,CAAC,cAAc,YAAY,CAAC;CAC1C,QAAQ,aAAa,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS;CAC5C,GAAG;CACH,YAAY,mBAAmB,SAAS;CACxC,oBAAoB,EAAE,MAAM,cAAc,CAAC,CAAC,SAAS;CACrD,iBAAiB,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,SAAS;CAC3C,kBAAkB,EAAE,QAAQ;CAC5B,kBAAkB;CAClB,mBAAmB;CACnB,SAAS,aAAa,SAAS;CAC/B,KAAK;CACL,WAAW,mBAAmB,SAAS;AACzC,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,8BAA8B,EACjC,OAAO;CACN,UAAU,EAAE,MACV,EACG,OAAO;EACN,IAAI;EACJ,MAAM;EACN,OAAO,aAAa,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC;EAChC,WAAW,EAAE,MAAM,cAAc,CAAC,CAAC,IAAI,CAAC;EACxC,cAAc,eAAe,SAAS;CACxC,CAAC,CAAC,CACD,OAAO,CACZ;CACA,eAAe;AACjB,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,6BAA6B,EAChC,OAAO;CACN,OAAO;CACP,oBAAoB,EAAE,QAAQ;CAC9B,SAAS,EACN,MACC,EACG,OAAO;EACN,OAAO;EACP,QAAQ;EACR,SAAS;CACX,CAAC,CAAC,CACD,OAAO,CACZ,CAAC,CACA,SAAS;AACd,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,kCAAkC,EACrC,OAAO;CACN,QAAQ;CACR,GAAG;CACH,SAAS;CACT,UAAU;CACV,aAAa,EACV,OAAO;EACN,WAAW;EACX,OAAO;EACP,IAAI;CACN,CAAC,CAAC,CACD,OAAO,CAAC,CACR,SAAS;AACd,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,uBAAuB,EAC1B,OAAO;CACN,QAAQ,EAAE,KAAK;EAAC;EAAQ;EAAQ;CAAM,CAAC;CACvC,MAAM,EAAE,MACN,EACG,OAAO;EACN,MAAM,EAAE,KAAK;GAAC;GAAgB;GAAiB;EAAwB,CAAC;EACxE,QAAQ,EAAE,KAAK;GAAC;GAAQ;GAAQ;GAAQ;EAAe,CAAC;EACxD,QAAQ;CACV,CAAC,CAAC,CACD,OAAO,CACZ;CACA,QAAQ,EAAE,MAAM,EAAE,OAAO,CAAC;AAC5B,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,wBAAwB,EAC3B,OAAO;CACN,SAAS;CACT,UAAU;CACV,OAAO;CACP,WAAW;CACX,UAAU;AACZ,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,oBAAoB,EAAE,mBAAmB,UAAU,CACvD,sBAAsB,OAAO;CAC3B,QAAQ,EAAE,QAAQ,IAAI;CACtB,MAAM,EAAE,MAAM,CAAC,cAAc,YAAY,CAAC;CAC1C,QAAQ,aAAa,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC;CACjC,SAAS;CACT,aAAa,EAAE,QAAQ;AACzB,CAAC,GACD,sBAAsB,OAAO;CAC3B,QAAQ,EAAE,KAAK,CAAC,uBAAuB,eAAe,CAAC;CACvD,MAAM,EAAE,KAAK;CACb,QAAQ,EAAE,KAAK;CACf,SAAS,EAAE,KAAK;CAChB,aAAa,EAAE,QAAQ,KAAK;AAC9B,CAAC,CACH,CAAC;AAED,MAAM,8BAA8B,EACjC,OAAO;CACN,WAAW;CACX,UAAU;CACV,aAAa,eAAe,SAAS;CACrC,SAAS,EAAE,OAAO,EAAE,OAAO,GAAG,iBAAiB;CAC/C,kBAAkB,EAAE,MAAM,cAAc;CACxC,iBAAiB,EAAE,MAAM,cAAc;CACvC,qBAAqB,EAAE,MAAM,cAAc;AAC7C,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,uBAAuB,EAC1B,OAAO;CACN,UAAU,EAAE,KAAK;EAAC;EAAY;EAAQ;EAAU;CAAK,CAAC;CACtD,MAAM,EAAE,KAAK;EAAC;EAAQ;EAAQ;EAAe;EAAO;EAAe;CAAe,CAAC;CACnF,OAAO;CACP,QAAQ;CACR,cAAc,eAAe,SAAS;AACxC,CAAC,CAAC,CACD,OAAO;AAEV,MAAa,sBAAgD,EAC1D,OAAO;CACN,GAAG;CACH,WAAW;CACX,WAAW;CACX,cAAc,EAAE,OAAO,EAAE,OAAO,GAAG,wBAAwB;CAC3D,aAAa,EACV,OAAO;EACN,MAAM;EACN,QAAQ;EACR,YAAY,4BAA4B,SAAS;EACjD,UAAU,EACP,OAAO;GACN,MAAM,eAAe,SAAS;GAC9B,QAAQ,eAAe,SAAS;EAClC,CAAC,CAAC,CACD,OAAO,CAAC,CACR,SAAS;CACd,CAAC,CAAC,CACD,OAAO;CACV,QAAQ,EAAE,OAAO,EAAE,OAAO,GAAG,kBAAkB;CAC/C,YAAY,wBAAwB,SAAS;CAC7C,MAAM,kBAAkB,SAAS;CACjC,iBAAiB,4BAA4B,SAAS;CACtD,eAAe,2BAA2B,SAAS;CACnD,oBAAoB,gCAAgC,SAAS;CAC7D,SAAS;CACT,uBAAuB,4BAA4B,SAAS;CAC5D,gBAAgB,EACb,MACC,EACG,OAAO;EACN,cAAc;EACd,OAAO;EACP,OAAO,aAAa,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC;CAClC,CAAC,CAAC,CACD,OAAO,CACZ,CAAC,CACA,SAAS;CACZ,iBAAiB,EAAE,MAAM,oBAAoB;AAC/C,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,eAAe,EAAE,QACpB,UAAU,OAAO,UAAU,YAAY,wBAAwB,KAAK,KAAK,GAC1E,+CACF;AAEA,MAAM,uBAAkD,EAAE,MAAM;CAC9D,EAAE,OAAO;CACT,EACG,OAAO;EACN,MAAM,EAAE,QAAQ,YAAY;EAC5B,YAAY,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,OAAO,CAAC;CAC7C,CAAC,CAAC,CACD,OAAO;CACV,EACG,OAAO;EACN,MAAM,EAAE,QAAQ,MAAM;EACtB,aAAa;EACb,SAAS;EACT,YAAY;EACZ,UAAU;EACV,iBAAiB;EACjB,eAAe;EACf,OAAO,EACJ,OAAO;GACN,QAAQ,EAAE,QAAQ,iBAAiB;GACnC,QAAQ;GACR,YAAY;EACd,CAAC,CAAC,CACD,OAAO;EACV,SAAS,EAAE,OAAO,CAAC,CAAC,SAAS;CAC/B,CAAC,CAAC,CACD,OAAO;AACZ,CAAC;AAED,MAAM,2BAA2B,EAAE,KAAK;CACtC;CACA;CACA;CACA;CACA;AACF,CAAC;AAED,MAAM,yBAAsD,EACzD,OAAO;CACN,YAAY;CACZ,KAAK;CACL,eAAe,aAAa,SAAS;CACrC,YAAY,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,OAAO,EAAE,OAAO,GAAG,YAAY,CAAC;CACnE,iBAAiB;CACjB,qBAAqB,mBAAmB,SAAS;CACjD,cAAc,EAAE,OAAO,CAAC,CAAC,SAAS;AACpC,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,kCAAwE,EAC3E,OAAO;CACN,OAAO;CACP,aAAa;CACb,SAAS,qBAAqB,SAAS;CACvC,OAAO,EAAE,MAAM,sBAAsB;CACrC,eAAe,aAAa,SAAS;CACrC,SAAS;CACT,YAAY;AACd,CAAC,CAAC,CACD,OAAO;AAEV,MAAM,sBAAsB,EAAE,KAAK;CACjC;CACA;CACA;CACA;CACA;CACA;AACF,CAAC;AAED,MAAM,qBAAqB,EAAE,KAAK;CAChC;CACA;CACA;CACA;CACA;AACF,CAAC;AAED,MAAa,qBAA8C,EACxD,OAAO;CACN,OAAO;CACP,QAAQ;CACR,WAAW,EAAE,OAAO,CAAC,CAAC,SAAS,EAAE,QAAQ,KAAK,CAAC;CAC/C,QAAQ;CACR,QAAQ,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,OAAO,CAAC;CACvC,UAAU,gCAAgC,SAAS;CACnD,aAAa,EAAE,MAAM,+BAA+B;CACpD,cAAc,mBAAmB,SAAS;CAC1C,aAAa,aAAa,SAAS;CACnC,cAAc;CACd,iBAAiB;CACjB,cAAc,EAAE,OAAO,CAAC,CAAC,SAAS;CAClC,eAAe,oBAAoB,SAAS;AAC9C,CAAC,CAAC,CACD,OAAO,CAAC,CACR,aAAa,OAAO,QAAQ;CAC3B,IAAI,MAAM,YAAY,MAAM,SAAS,UAAU,GAC7C,IAAI,SAAS;EACX,MAAM;EACN,MAAM,CAAC,YAAY,OAAO;EAC1B,SAAS;CACX,CAAC;CAGH,MAAM,uBAAO,IAAI,IAAY;CAC7B,KAAK,IAAI,IAAI,GAAG,IAAI,MAAM,YAAY,QAAQ,KAAK;EACjD,MAAM,QAAQ,MAAM,YAAY,EAAE,CAAE;EACpC,IAAI,KAAK,IAAI,KAAK,GAChB,IAAI,SAAS;GACX,MAAM;GACN,MAAM;IAAC;IAAe;IAAG;GAAO;GAChC,SAAS,8BAA8B;EACzC,CAAC;EAEH,KAAK,IAAI,KAAK;CAChB;CAEA,IAAI,MAAM,WAAW,aAAa,CAAC,MAAM,cAAc,KAAK,GAC1D,IAAI,SAAS;EACX,MAAM;EACN,MAAM,CAAC,cAAc;EACrB,SAAS;CACX,CAAC;AAEL,CAAC;AAEH,MAAM,4BAA4B,EAC/B,OAAO;CACN,cAAc;CACd,MAAM;CACN,YAAY,WAAW,SAAS;AAClC,CAAC,CAAC,CACD,OAAO;AAEV,MAAa,uBAAkD,EAC5D,OAAO;CACN,SAAS;CACT,QAAQ;CACR,cAAc,eAAe,SAAS;CACtC,MAAM;CACN,mBAAmB;CACnB,iBAAiB;CACjB;CACA,QAAQ,EAAE,MAAM,yBAAyB,CAAC,CAAC,SAAS;CACpD,QAAQ,EACL,OAAO;EACN,MAAM,EAAE,KAAK;GAAC;GAAM;GAAS;EAAO,CAAC;EACrC,SAAS,EAAE,OAAO,CAAC,CAAC,SAAS;CAC/B,CAAC,CAAC,CACD,OAAO,CAAC,CACR,SAAS;CACZ,gBAAgB,eAAe,SAAS;CACxC,qBAAqB,mBAAmB,SAAS;CACjD,iBAAiB,eAAe,SAAS;CACzC,qBAAqB,eAAe,SAAS;AAC/C,CAAC,CAAC,CACD,OAAO,CAAC,CACR,QAAQ,SAAS,OAAO,KAAK,eAAe,KAAK,OAAO,KAAK,iBAAiB,GAAG;CAChF,MAAM,CAAC,iBAAiB;CACxB,SAAS;AACX,CAAC;AAEyC,EACzC,OAAO;CACN,aAAa,EAAE,QAAQ,mBAAmB;CAC1C,QAAQ,EAAE,MAAM,EAAE,QAAQ,CAAC;AAC7B,CAAC,CAAC,CACD,OAAO;AAEgC,EACvC,OAAO;CACN,aAAa,EAAE,QAAQ,mBAAmB;CAC1C,OAAO,EAAE,MAAM,EAAE,QAAQ,CAAC;AAC5B,CAAC,CAAC,CACD,OAAO;AAEV,MAAa,8BAAgE,EAC1E,OAAO;CACN,aAAa,EAAE,QAAQ,mBAAmB;CAC1C,QAAQ,EAAE,MAAM,kBAAkB;AACpC,CAAC,CAAC,CACD,OAAO;AAEV,MAAa,4BAA4D,EACtE,OAAO;CACN,aAAa,EAAE,QAAQ,mBAAmB;CAC1C,OAAO,EAAE,MAAM,oBAAoB;AACrC,CAAC,CAAC,CACD,OAAO;AAEV,MAAa,uBAAkD,EAC5D,OAAO;CACN,UAAU;CACV,UAAU,EAAE,MACV,EACG,OAAO;EACN,OAAO;EACP,QAAQ;CACV,CAAC,CAAC,CACD,OAAO,CACZ;AACF,CAAC,CAAC,CACD,OAAO;;;;;;;;;;;;;;;;;;;;ACjmBV,MAAM,6BAA6B;AAEnC,SAAS,MAAM,IAA2B;CACxC,OAAO,IAAI,SAAS,YAAY,WAAW,SAAS,EAAE,CAAC;AACzD;AAEA,SAAS,oBAAoB,UAA0B;CACrD,OAAO,SAAS,KAAK,CAAC,CAAC,QAAQ,QAAQ,EAAE,CAAC,CAAC,QAAQ,SAAS,EAAE;AAChE;AAEA,SAAS,sBAAsB,KAAiC;CAC9D,MAAM,WAAW,OAAO,WAAW,OAAO,WAAW;CACrD,IAAI,SAAS,KAAK,CAAC,CAAC,WAAW,GAAG,MAAM,IAAI,MAAM,mCAAmC;CACrF,IAAI,SAAS,SAAS,4BACpB,MAAM,IAAI,MAAM,mCAAmC,2BAA2B,YAAY;CAE5F,OAAO;AACT;AAEA,SAAS,yBAAyB,OAE/B;CACD,OAAO,MAAM,OACV,KACE,UACC,GAAG,MAAM,KAAK,SAAS,IAAI,MAAM,KAAK,IAAI,MAAM,CAAC,CAAC,KAAK,GAAG,IAAI,QAAQ,IAAI,MAAM,SACpF,CAAC,CACA,KAAK,IAAI;AACd;AAEA,eAAe,KACb,QACA,MACA,MACA,OAAuB,CAAC,GACC;CACzB,MAAM,YAAY,OAAO,aAAa;CACtC,MAAM,aAAa,OAAO,WAAW;CACrC,MAAM,IAAkB,OAAO,eAAe,GAAG,SAAS,MAAM,GAAG,IAAI;CAEvE,MAAM,MAAM,GADC,oBAAoB,OAAO,QACtB,IAAI;CACtB,MAAM,iBAAiB,sBAAsB,KAAK,cAAc;CAChE,MAAM,UAAkC;EACtC,gBAAgB;EAChB,eAAe,UAAU,OAAO;EAChC,sBAAsB,OAAO;EAC7B,yBAAyB;EACzB,mBAAmB;CACrB;CAEA,IAAI;CACJ,KAAK,IAAI,UAAU,GAAG,WAAW,YAAY,WAAW;EACtD,MAAM,aAAa,YAAY,QAAQ,SAAS;EAChD,MAAM,iBAAiB,KAAK,SAAS,YAAY,IAAI,CAAC,KAAK,QAAQ,UAAU,CAAC,IAAI;EAClF,IAAI;EACJ,IAAI;GACF,MAAM,MAAM,EAAE,KAAK;IACjB,QAAQ;IACR;IACA,MAAM,KAAK,UAAU,IAAI;IACzB,QAAQ;GACV,CAAC;EACH,SAAS,KAAK;GACZ,IAAI,KAAK,QAAQ,SAAS,MAAM;GAChC,YAAY;GACZ,IAAI,YAAY,YAAY,MAAM;GAClC,MAAM,MAAM,KAAK,UAAU,MAAM,KAAK,OAAO,IAAI,GAAG;GACpD;EACF;EAEA,IAAI,CAAC,IAAI,IAAI;GACX,MAAM,OAAO,MAAM,IAAI,KAAK,CAAC,CAAC,YAAY,EAAE;GAC5C,MAAM,wBAAQ,IAAI,MAAM,iBAAiB,IAAI,WAAW,IAAI,OAAO,KAAK,KAAK,MAAM,GAAG,GAAG,GAAG;GAE5F,IAAI,EADc,IAAI,UAAU,OAAO,IAAI,WAAW,OAAO,IAAI,WAAW,QAC1D,YAAY,YAAY,MAAM;GAChD,YAAY;GACZ,MAAM,MAAM,KAAK,UAAU,MAAM,KAAK,OAAO,IAAI,GAAG;GACpD;EACF;EAEA,IAAI;EACJ,IAAI;GACF,cAAc,MAAM,IAAI,KAAK;EAC/B,SAAS,OAAO;GACd,MAAM,IAAI,MAAM,iBAAiB,IAAI,yBAAyB,EAAE,OAAO,MAAM,CAAC;EAChF;EACA,MAAM,SAAS,qBAAqB,UAAU,WAAW;EACzD,IAAI,CAAC,OAAO,SACV,MAAM,IAAI,MACR,iBAAiB,IAAI,iCAAiC,yBAAyB,OAAO,KAAK,GAC7F;EAEF,OAAO,OAAO;CAChB;CACA,MAAM,6BAAa,IAAI,MAAM,iCAAiC;AAChE;AAEA,SAAgB,mBAAmB,QAAoC;CACrE,IAAI,oBAAoB,OAAO,QAAQ,CAAC,CAAC,WAAW,GAAG,MAAM,IAAI,MAAM,sBAAsB;CAC7F,IAAI,OAAO,OAAO,KAAK,CAAC,CAAC,WAAW,GAAG,MAAM,IAAI,MAAM,oBAAoB;CAC3E,IAAI,OAAO,SAAS,KAAK,CAAC,CAAC,WAAW,GAAG,MAAM,IAAI,MAAM,sBAAsB;CAC/E,IACE,OAAO,cAAc,KAAA,MACpB,CAAC,OAAO,SAAS,OAAO,SAAS,KAAK,OAAO,aAAa,IAE3D,MAAM,IAAI,MAAM,kCAAkC;CAEpD,IAAI,OAAO,YAAY,KAAA,MAAc,CAAC,OAAO,UAAU,OAAO,OAAO,KAAK,OAAO,UAAU,IACzF,MAAM,IAAI,MAAM,wCAAwC;CAG1D,OAAO;EACL;EACA,aAAa;EAEb,MAAM,cAAc,OAAO,gBAAgB;GACzC,OAAO,KAAK,eAAe,CAAC,KAAK,GAAG,cAAc;EACpD;EAEA,MAAM,eAAe,QAAQ,gBAAgB;GAK3C,OAAO,KAA4B,QAAQ,wBAJP,4BAA4B,MAAM;IACpE,aAAa;IACb;GACF,CACsE,GAAG,EACvE,eACF,CAAC;EACH;EAEA,MAAM,aAAa,OAAO,gBAAgB;GAKxC,OAAO,KAA0B,QAAQ,qBAJP,0BAA0B,MAAM;IAChE,aAAa;IACb;GACF,CACiE,GAAG,EAClE,eACF,CAAC;EACH;CACF;AACF;;;;;;;;;;;;;;;;;;;;;;;;;;;AA4BA,SAAgB,oBACd,YAAkF,CAAC,GACzD;CAC1B,MAAM,MAAM,UAAU,OAAO,QAAQ;CACrC,MAAM,YACJ,UAAU,YACV,IAAI,qBACJ,IAAI,wBAAA,EACH,KAAK;CACR,MAAM,UAAU,UAAU,UAAU,IAAI,yBAAyB,IAAI,eAAA,EAAiB,KAAK;CAC3F,MAAM,YAAY,UAAU,YAAY,IAAI,iBAAA,EAAmB,KAAK;CACpE,IAAI,CAAC,YAAY,CAAC,UAAU,CAAC,UAAU,OAAO,KAAA;CAC9C,MAAM,SAAuB;EAAE,UAAU,SAAS,QAAQ,QAAQ,EAAE;EAAG;EAAQ;CAAS;CACxF,IAAI,UAAU,WAAW,OAAO,YAAY,UAAU;CACtD,IAAI,UAAU,cAAc,KAAA,GAAW,OAAO,YAAY,UAAU;CACpE,IAAI,UAAU,YAAY,KAAA,GAAW,OAAO,UAAU,UAAU;CAChE,OAAO;AACT;AAEA,SAAgB,oBACd,YAAkF,CAAC,GACzD;CAC1B,MAAM,SAAS,oBAAoB,SAAS;CAC5C,OAAO,SAAS,mBAAmB,MAAM,IAAI,KAAA;AAC/C"}

package/dist/define-agent-eval-DPizhHfy.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"define-agent-eval-DPizhHfy.js","names":[],"sources":["../src/contract/self-improve-reporting.ts","../src/contract/self-improve-method.ts","../src/contract/self-improve.ts","../src/contract/define-agent-eval.ts"],"sourcesContent":["import { campaignCellToRunRecord } from '../campaign/run-record'\nimport { pairedCampaignComposites } from '../campaign/score-utils'\nimport { surfaceContentHash } from '../campaign/surface-identity'\nimport type {\n  CampaignCellResult,\n  CampaignResult,\n  MutableSurface,\n  Scenario,\n} from '../campaign/types'\nimport { ValidationError } from '../errors'\nimport { modelHasSnapshot, type RunRecord, type RunSplitTag } from '../run-record'\n\nexport function pairedCompositeSummary<TArtifact, TScenario extends Scenario>(\n  baseline: CampaignResult<TArtifact, TScenario>,\n  winner: CampaignResult<TArtifact, TScenario>,\n  independentUnitByScenarioId?: ReadonlyMap<string, string>,\n) {\n  const scores = pairedCampaignComposites(baseline, winner, independentUnitByScenarioId)\n  const perScenario = (campaign: CampaignResult<TArtifact, TScenario>) =>\n    Object.fromEntries(\n      Object.entries(campaign.aggregates.byScenario).map(([id, aggregate]) => [\n        id,\n        aggregate.meanComposite,\n      ]),\n    )\n  return {\n    baseline: { compositeMean: scores.beforeMean, perScenario: perScenario(baseline) },\n    winner: { compositeMean: scores.afterMean, perScenario: perScenario(winner) },\n    baselineComposites: scores.before,\n    observations: scores.observations,\n  }\n}\n\n/** 32-bit FNV-1a over raw UTF-16 code units, rendered as hex for a cell key.\n *\n *  Frozen: the key names a persisted cell, so a change orphans every cell\n *  already written. This is a cell-key function, not a general-purpose hash. */\nfunction hashString(s: string): string {\n  let h = 2166136261 >>> 0\n  for (let i = 0; i < s.length; i++) {\n    h ^= s.charCodeAt(i)\n    h = Math.imul(h, 16777619) >>> 0\n  }\n  return h.toString(16).padStart(8, '0')\n}\n\n/**\n * Adapt campaign cells into the `RunRecord` shape `analyzeRuns()` consumes.\n * Each cell becomes one run; `candidateId` is the caller-supplied label so\n * baseline + winner pair cleanly on `(experimentId, scenarioId, seed)`.\n *\n * `promptHash` identifies the executed surface; `configHash` identifies the candidate label.\n */\nexport function cellsToRunRecords<TArtifact>(\n  cells: ReadonlyArray<CampaignCellResult<TArtifact>>,\n  candidateId: 'baseline' | 'winner',\n  runId: string,\n  surface: MutableSurface,\n  splitTag: RunSplitTag,\n  fallbackModel?: string,\n): RunRecord[] {\n  const promptHash = surfaceContentHash(surface)\n  const configHash = surfaceContentHash(candidateId)\n  return cells.map((cell) => {\n    const receiptModels = cell.resolvedModels ?? (cell.resolvedModel ? [cell.resolvedModel] : [])\n    if (receiptModels.length > 1) {\n      throw new ValidationError(\n        `selfImprove cell ${cell.cellId} used multiple agent models: ${receiptModels.join(', ')}`,\n      )\n    }\n    const model = receiptModels[0] ?? fallbackModel\n    if (!model) {\n      throw new ValidationError(\n        `selfImprove.model is required when cell ${cell.cellId} has no paid-call model receipt`,\n      )\n    }\n    if (!modelHasSnapshot(model)) {\n      throw new ValidationError(\n        `selfImprove model \"${model}\" lacks a snapshot version for cell ${cell.cellId}`,\n      )\n    }\n    return campaignCellToRunRecord(cell, {\n      runId: `${runId}::${candidateId}::${cell.cellId}`,\n      experimentId: runId,\n      candidateId,\n      // scenarioId is explicit; seed keeps repeated runs distinct.\n      seed:\n        cell.rep * 1_000_000 +\n        hashString(cell.scenarioId)\n          .slice(0, 6)\n          .split('')\n          .reduce((a, c) => (a * 31 + c.charCodeAt(0)) >>> 0, 0),\n      model,\n      promptHash,\n      configHash,\n      commitSha: 'cell',\n      splitTag,\n    })\n  })\n}\n","import { openAutoPr } from '../campaign/auto-pr'\nimport { campaignSplitDigest } from '../campaign/coverage'\nimport { evaluationUnitMap } from '../campaign/final-evidence'\nimport { defaultProductionGate } from '../campaign/gates/default-production-gate'\nimport { executeOptimizationMethod } from '../campaign/optimization-method'\nimport {\n  type ComparisonCost,\n  combineComparisonCosts,\n  costFromLedgerSummary,\n  type OptimizationMethodResult,\n} from '../campaign/presets/compare-optimization-methods'\nimport { runFinalComparison } from '../campaign/presets/run-final-comparison'\nimport { campaignMeasurementDigest, canonicalDigest } from '../campaign/provenance'\nimport {\n  campaignCellExecutionEvidence,\n  campaignCellJudgeDimensions,\n  campaignCellTaskScore,\n} from '../campaign/run-record'\nimport type { CampaignComparisonUnits } from '../campaign/score-utils'\nimport type { SearchHistoryCoverageRow } from '../campaign/search-history-receipt'\nimport type { CampaignStorage } from '../campaign/storage'\nimport { surfaceContentHash, surfaceHash } from '../campaign/surface-identity'\nimport type { Scenario } from '../campaign/types'\nimport type { CostLedgerHandle, CostLedgerSummary } from '../cost-ledger'\nimport { createCampaignEvidenceReceipt } from '../experiment/campaign-evidence'\nimport type { EvaluationClaim } from '../experiment/claim'\nimport type { EvidenceReceipt } from '../experiment/evidence-receipt'\nimport { createHostedClient } from '../hosted/client'\nimport type { EvalRunGenerationSnapshot } from '../hosted/types'\nimport { analyzeRuns } from './analyze-runs'\nimport type {\n  SelfImproveMethodOptions,\n  SelfImproveOptions,\n  SelfImproveProposerResult,\n} from './self-improve'\nimport { cellsToRunRecords, pairedCompositeSummary } from './self-improve-reporting'\n\nexport interface SelfImproveMethodProvenance {\n  schema: 'tangle.method-improvement'\n  recordDigest: `sha256:${string}`\n  runId: string\n  runDir: string\n  timestamp: string\n  baselineContentHash: string\n  winnerContentHash: string\n  diff: string\n  optimizationMethod: NonNullable<SelfImproveProposerResult<Scenario, unknown>['optimization']>\n  claim?: EvaluationClaim\n  evidence: {\n    trainSplitDigest: `sha256:${string}`\n    selectionSplitDigest: `sha256:${string}`\n    holdoutSplitDigest: `sha256:${string}`\n    baselineCampaignDigest: `sha256:${string}`\n    winnerCampaignDigest: `sha256:${string}`\n    costReceiptsDigest: `sha256:${string}`\n    neutralizedCampaignDigest?: `sha256:${string}`\n    holdoutObservations?: CampaignComparisonUnits\n  }\n  gate: Awaited<ReturnType<typeof runFinalComparison>>['gateResult']\n  holdout?: 'deferred'\n  baselineHoldoutComposite?: number\n  winnerHoldoutComposite?: number\n  heldOutLift?: number\n  cost: ComparisonCost\n  totalCostUsd: number\n  totalDurationMs: number\n}\n\nexport interface SelfImproveMethodResult<TScenario extends Scenario, TArtifact>\n  extends Omit<\n    SelfImproveProposerResult<TScenario, TArtifact>,\n    | 'mode'\n    | 'baseline'\n    | 'winner'\n    | 'provenance'\n    | 'generationsExplored'\n    | 'cost'\n    | 'raw'\n    | 'optimization'\n    | 'insight'\n  > {\n  evidence?: { baseline: EvidenceReceipt; winner: EvidenceReceipt }\n  searchHistoryCoverage: SearchHistoryCoverageRow\n  mode: 'method'\n  /** No final measurement exists when holdout is deferred. */\n  baseline: SelfImproveProposerResult<TScenario, TArtifact>['baseline'] | null\n  winner: Omit<SelfImproveProposerResult<TScenario, TArtifact>['winner'], 'compositeMean'> & {\n    compositeMean: number | null\n  }\n  provenance: SelfImproveMethodProvenance\n  optimization: NonNullable<SelfImproveProposerResult<TScenario, TArtifact>['optimization']>\n  /** Includes the method's reported cost and final measurements without double counting receipts. */\n  cost: ComparisonCost\n  /** Only calls actually recorded by the shared ledger have token and channel breakdowns. */\n  ledgerCost: CostLedgerSummary\n  insight?: SelfImproveProposerResult<TScenario, TArtifact>['insight']\n  raw: Awaited<ReturnType<typeof runFinalComparison<TScenario, TArtifact>>> & {\n    kind: 'method'\n    baselineSurface: SelfImproveProposerResult<TScenario, TArtifact>['raw']['baselineSurface']\n    winnerSurface: SelfImproveProposerResult<TScenario, TArtifact>['raw']['winnerSurface']\n    winnerSurfaceHash: string\n    method: OptimizationMethodResult\n    cost: ComparisonCost\n    prResult?: ReturnType<typeof openAutoPr>\n  }\n}\n\n/** Complete methods own search and selection; this path only measures their final choice. */\nexport async function runSelfImproveMethod<TScenario extends Scenario, TArtifact>(args: {\n  opts: SelfImproveOptions<TScenario, TArtifact> &\n    Pick<SelfImproveMethodOptions<TScenario, TArtifact>, 'method'>\n  train: TScenario[]\n  selection: TScenario[]\n  holdout: TScenario[]\n  costLedger: CostLedgerHandle\n  storage: CampaignStorage\n  runDir: string\n  startedAt: number\n}): Promise<SelfImproveMethodResult<TScenario, TArtifact>> {\n  const { opts, train, selection, holdout, costLedger, storage, runDir, startedAt } = args\n  const budget = opts.budget ?? {}\n  if (opts.autoOnPromote === 'pr' && (!opts.ghOwner || !opts.ghRepo)) {\n    throw new Error(\"selfImprove: autoOnPromote='pr' requires ghOwner + ghRepo\")\n  }\n  const evidenceContext = opts.evidence === undefined ? undefined : structuredClone(opts.evidence)\n  if (evidenceContext && budget.holdout === 'deferred')\n    throw new Error('selfImprove: evidence receipts require measured holdout')\n  const baselineSurface = structuredClone(opts.baselineSurface)\n  const judge = opts.judge\n  const {\n    selected,\n    cost: methodCost,\n    history,\n  } = await executeOptimizationMethod({\n    method: opts.method,\n    storage,\n    searchHistoryPolicy: opts.searchHistoryPolicy,\n    searchHistoryVerification: opts.searchHistoryVerification,\n    input: {\n      baselineSurface,\n      trainScenarios: train,\n      selectionScenarios: selection,\n      dispatchWithSurface: opts.agent,\n      judges: [judge],\n      runDir: `${runDir}/optimization/${opts.method.name.replace(/[^a-zA-Z0-9._-]/g, '_')}`,\n      seed: 42,\n      runOptions: {\n        storage,\n        maxConcurrency: budget.maxConcurrency ?? 2,\n        reps: budget.reps,\n        dispatchRef: opts.dispatchRef,\n        dispatchTimeoutMs: opts.dispatchTimeoutMs,\n        cellRetry: opts.cellRetry,\n        cellPlacement: opts.cellPlacement,\n        labeledStore: opts.labeledStore,\n        captureSource: opts.captureSource,\n        expectUsage: opts.expectUsage ?? 'assert',\n      },\n      costLedger,\n    },\n  })\n  const finalPhase = 'selfImprove.method-final'\n  const finalCost = (): ComparisonCost =>\n    combineComparisonCosts(\n      ['holdout.baseline', 'holdout.winner', 'holdout.neutralized', 'promotion.gate'].map(\n        (phase) => ({\n          label: phase,\n          cost: costFromLedgerSummary(costLedger.summary({ phase: `${finalPhase}.${phase}` })),\n        }),\n      ),\n    )\n  const totalCost = (): ComparisonCost =>\n    combineComparisonCosts([\n      { label: opts.method.name, cost: methodCost },\n      { label: 'final comparison', cost: finalCost() },\n    ])\n  const gate =\n    opts.gate ??\n    defaultProductionGate<TArtifact, TScenario>({\n      holdoutScenarios: holdout,\n      deltaThreshold: opts.claim?.minimumEffect ?? 0.05,\n      independentUnitByScenarioId: opts.claim ? evaluationUnitMap(opts.claim, holdout) : undefined,\n    })\n  const comparison = await runFinalComparison({\n    claim: opts.claim,\n    finalEvidence: opts.finalEvidence,\n    baselineSurface,\n    winnerSurface: selected.winnerSurface,\n    scenarios: holdout,\n    dispatchWithSurface: opts.agent,\n    dispatchRef: opts.dispatchRef,\n    judges: [judge],\n    holdout: budget.holdout,\n    neutralize: opts.neutralize,\n    gate: {\n      ...gate,\n      async decide(context) {\n        const cost = totalCost()\n        if (\n          budget.dollars !== undefined &&\n          (!cost.accountingComplete || cost.totalCostUsd > budget.dollars)\n        ) {\n          return {\n            decision: 'hold' as const,\n            reasons: ['complete method spend cannot satisfy the declared run budget'],\n            contributingGates: [\n              {\n                name: 'budget',\n                status: 'fail' as const,\n                detail: { cost, budgetUsd: budget.dollars },\n              },\n            ],\n          }\n        }\n        return gate.decide(context)\n      },\n    },\n    runDir,\n    storage,\n    costLedger,\n    costPhase: finalPhase,\n    reps: budget.reps,\n    maxConcurrency: budget.maxConcurrency ?? 2,\n    dispatchTimeoutMs: opts.dispatchTimeoutMs,\n    cellRetry: opts.cellRetry,\n    cellPlacement: opts.cellPlacement,\n    expectUsage: opts.expectUsage ?? 'assert',\n    label: 'selfImprove',\n  })\n  const evidence = evidenceContext\n    ? {\n        baseline: createCampaignEvidenceReceipt({\n          campaign: comparison.baselineOnHoldout,\n          surface: baselineSurface,\n          context: evidenceContext,\n        }),\n        winner: createCampaignEvidenceReceipt({\n          campaign: comparison.winnerOnHoldout,\n          surface: selected.winnerSurface,\n          context: evidenceContext,\n        }),\n      }\n    : undefined\n  const deferred = comparison.holdout === 'deferred'\n  const report = deferred\n    ? null\n    : pairedCompositeSummary(\n        comparison.baselineOnHoldout,\n        comparison.winnerOnHoldout,\n        opts.claim ? evaluationUnitMap(opts.claim, holdout) : undefined,\n      )\n  const baseline = report?.baseline ?? null\n  const winner = report?.winner ?? null\n  const lift = baseline && winner ? winner.compositeMean - baseline.compositeMean : undefined\n  const insight = deferred\n    ? undefined\n    : await analyzeRuns({\n        runs: [\n          ...cellsToRunRecords(\n            comparison.baselineOnHoldout.cells,\n            'baseline',\n            runDir,\n            baselineSurface,\n            'holdout',\n            opts.model,\n          ),\n          ...(comparison.winnerOnHoldout === comparison.baselineOnHoldout\n            ? []\n            : cellsToRunRecords(\n                comparison.winnerOnHoldout.cells,\n                'winner',\n                runDir,\n                selected.winnerSurface,\n                'holdout',\n                opts.model,\n              )),\n        ],\n        baselineCandidateId: 'baseline',\n        independentUnitByScenarioId: opts.claim\n          ? evaluationUnitMap(opts.claim, holdout)\n          : undefined,\n        decisionThreshold: opts.claim?.minimumEffect ?? 0.05,\n        ...(comparison.winnerOnHoldout === comparison.baselineOnHoldout\n          ? {}\n          : { candidateCandidateId: 'winner' }),\n      })\n  const cost = totalCost()\n  const optimization = {\n    name: opts.method.name,\n    cost: methodCost,\n    ...(selected.durationMs === undefined ? {} : { durationMs: selected.durationMs }),\n    ...(selected.provenance === undefined ? {} : { provenance: selected.provenance }),\n  }\n  const durationMs = Date.now() - startedAt\n  const receipts = costLedger.list()\n  const record = {\n    schema: 'tangle.method-improvement' as const,\n    runId: `${runDir}#${startedAt}`,\n    runDir,\n    timestamp: new Date(startedAt).toISOString(),\n    baselineContentHash: surfaceContentHash(baselineSurface),\n    winnerContentHash: surfaceContentHash(selected.winnerSurface),\n    diff: comparison.promotedDiff,\n    optimizationMethod: optimization,\n    ...(opts.claim ? { claim: opts.claim } : {}),\n    evidence: {\n      trainSplitDigest: campaignSplitDigest(train, budget.reps ?? 1),\n      selectionSplitDigest: campaignSplitDigest(selection, budget.reps ?? 1),\n      holdoutSplitDigest: campaignSplitDigest(holdout, budget.reps ?? 1),\n      baselineCampaignDigest: campaignMeasurementDigest(comparison.baselineOnHoldout),\n      winnerCampaignDigest: campaignMeasurementDigest(comparison.winnerOnHoldout),\n      costReceiptsDigest: canonicalDigest(receipts),\n      ...(report ? { holdoutObservations: report.observations } : {}),\n      ...(comparison.neutralizedOnHoldout\n        ? { neutralizedCampaignDigest: campaignMeasurementDigest(comparison.neutralizedOnHoldout) }\n        : {}),\n    },\n    gate: comparison.gateResult,\n    ...(deferred ? { holdout: 'deferred' as const } : {}),\n    ...(baseline && winner\n      ? {\n          baselineHoldoutComposite: baseline.compositeMean,\n          winnerHoldoutComposite: winner.compositeMean,\n          heldOutLift: lift,\n        }\n      : {}),\n    cost,\n    totalCostUsd: cost.totalCostUsd,\n    totalDurationMs: durationMs,\n  }\n  // The persisted JSON and returned record must have the same canonical content.\n  const serialized: Omit<SelfImproveMethodProvenance, 'recordDigest'> = JSON.parse(\n    JSON.stringify(record),\n  )\n  const provenance: SelfImproveMethodProvenance = {\n    ...serialized,\n    recordDigest: canonicalDigest(serialized),\n  }\n  storage.ensureDir(runDir)\n  storage.write(`${runDir}/method-provenance.json`, JSON.stringify(provenance, null, 2))\n  opts.onProvenance?.(provenance)\n  opts.onProgress?.({\n    kind: 'gate.decided',\n    decision: comparison.gateResult.decision,\n    ...(lift === undefined ? {} : { lift }),\n  })\n  const prResult =\n    opts.autoOnPromote === 'pr' && comparison.gateResult.decision === 'ship'\n      ? openAutoPr({\n          result: comparison.winnerOnHoldout,\n          gate: comparison.gateResult,\n          promotedDiff: comparison.promotedDiff,\n          ghOwner: opts.ghOwner!,\n          ghRepo: opts.ghRepo!,\n        })\n      : undefined\n  const result: SelfImproveMethodResult<TScenario, TArtifact> = {\n    ...(evidence ? { evidence } : {}),\n    mode: 'method',\n    ...(opts.claim ? { claim: opts.claim } : {}),\n    ...(comparison.finalEvidence ? { finalEvidence: comparison.finalEvidence } : {}),\n    searchHistoryCoverage: history,\n    baseline,\n    winner: {\n      surface: selected.winnerSurface,\n      compositeMean: winner?.compositeMean ?? null,\n      perScenario: winner?.perScenario ?? {},\n      label: opts.method.name,\n    },\n    ...(lift === undefined ? {} : { lift }),\n    diff: comparison.promotedDiff,\n    gateDecision: comparison.gateResult.decision,\n    provenance,\n    optimization,\n    cost,\n    ledgerCost: costLedger.summary(),\n    totalCostUsd: cost.totalCostUsd,\n    durationMs,\n    receipts,\n    ...(insight ? { insight } : {}),\n    ...(selected.searchHistory ? { searchHistory: selected.searchHistory } : {}),\n    raw: {\n      ...comparison,\n      kind: 'method',\n      baselineSurface,\n      winnerSurface: selected.winnerSurface,\n      winnerSurfaceHash: surfaceHash(selected.winnerSurface),\n      method: selected,\n      cost,\n      ...(prResult ? { prResult } : {}),\n    },\n  }\n  if (opts.hostedTenant) {\n    const snapshot = (\n      index: number,\n      surface: typeof baselineSurface,\n      campaign: typeof comparison.baselineOnHoldout,\n      compositeMean: number | null,\n    ): EvalRunGenerationSnapshot => ({\n      index,\n      surfaceHash: surfaceHash(surface),\n      surface,\n      cells: campaign.cells.map((cell) => {\n        const execution = campaignCellExecutionEvidence(cell)\n        return {\n          scenarioId: cell.scenarioId,\n          rep: cell.rep,\n          compositeMean: campaignCellTaskScore(cell) ?? null,\n          dimensions: campaignCellJudgeDimensions(cell),\n          terminalOutcome: execution.terminalOutcome,\n          executionErrorCount: execution.executionErrorCount ?? null,\n          ...(cell.error ? { errorMessage: cell.error } : {}),\n        }\n      }),\n      compositeMean,\n      costUsd: campaign.aggregates.cost.totalCostUsd,\n      durationMs: campaign.durationMs,\n    })\n    try {\n      await createHostedClient(opts.hostedTenant).ingestEvalRun({\n        runId: provenance.runId,\n        runDir,\n        timestamp: provenance.timestamp,\n        status: 'finished',\n        labels: { ...opts.hostedLabels, mode: 'method' },\n        baseline: snapshot(\n          0,\n          baselineSurface,\n          comparison.baselineOnHoldout,\n          baseline?.compositeMean ?? null,\n        ),\n        generations: [\n          snapshot(\n            1,\n            selected.winnerSurface,\n            comparison.winnerOnHoldout,\n            winner?.compositeMean ?? null,\n          ),\n        ],\n        gateDecision: result.gateDecision,\n        ...(lift === undefined ? {} : { holdoutLift: lift }),\n        totalCostUsd: result.totalCostUsd,\n        totalDurationMs: durationMs,\n        ...(insight ? { insightReport: insight } : {}),\n      })\n    } catch (error) {\n      console.warn(\n        `[agent-eval] hosted ingest failed (continuing): ${error instanceof Error ? error.message : String(error)}`,\n      )\n    }\n  }\n  return result\n}\n","import { defineEvaluationClaim, type EvaluationClaim } from '../experiment/claim'\n/**\n * Run one complete improvement job.\n *\n * A caller-owned `proposer` can generate candidates across local generations.\n * An external `method`, such as official GEPA or SkillOpt, owns its complete\n * search and returns one candidate. Both paths remeasure the selected candidate\n * against cases that candidate generation never receives.\n */\n\nimport type { ProposalFinding } from '../analyst/types'\nimport {\n  assertIndependentEvaluationSplit,\n  captureFinalEvidencePolicy,\n  evaluationUnitMap,\n  type FinalEvidencePolicy,\n  type FinalEvidenceUse,\n  reserveFinalEvidence,\n} from '../campaign/final-evidence'\nimport { defaultProductionGate } from '../campaign/gates/default-production-gate'\nimport { type PowerPreflight, powerPreflight } from '../campaign/gates/power-preflight'\nimport { captureJudge } from '../campaign/judge-snapshot'\nimport type {\n  OptimizationMethod,\n  OptimizationMethodProvenance,\n  OptimizationMethodResult,\n} from '../campaign/presets/compare-optimization-methods'\nimport {\n  type RunImprovementLoopResult,\n  runImprovementLoop,\n} from '../campaign/presets/run-improvement-loop'\nimport type {\n  PremeasuredOptimizationBaseline,\n  RunOptimizationOptions,\n} from '../campaign/presets/run-optimization'\nimport {\n  emitLoopProvenance,\n  type LoopProvenanceRecord,\n  loopProvenanceArgsFromResult,\n} from '../campaign/provenance'\nimport type { CampaignCellRetryPolicy } from '../campaign/run-campaign'\nimport { resolveRunDir } from '../campaign/run-dir'\nimport {\n  campaignCellExecutionEvidence,\n  campaignCellJudgeDimensions,\n  campaignCellTaskScore,\n} from '../campaign/run-record'\nimport type { SearchHistoryReceipt } from '../campaign/search-history-receipt'\nimport {\n  assertSearchHistoryAdmissionOptions,\n  type SearchHistoryAdmissionOptions,\n} from '../campaign/search-history-receipt'\nimport {\n  type CampaignStorage,\n  createRunCostLedger,\n  fsCampaignStorage,\n  inMemoryCampaignStorage,\n} from '../campaign/storage'\nimport { surfaceHash } from '../campaign/surface-identity'\nimport type {\n  DispatchContext,\n  Gate,\n  JudgeConfig,\n  LabeledScenarioStore,\n  MutableSurface,\n  Scenario,\n  SurfaceProposer,\n} from '../campaign/types'\nimport type { CostLedgerHandle, CostLedgerSummary, CostReceipt } from '../cost-ledger'\nimport type { CampaignEvidenceContext } from '../experiment/campaign-evidence'\nimport { createHostedClient, type HostedTenant } from '../hosted/client'\nimport type { EvalRunCellScore, EvalRunEvent, EvalRunGenerationSnapshot } from '../hosted/types'\nimport type { RunSplitTag } from '../run-record'\nimport { analyzeRuns } from './analyze-runs'\nimport type { InsightReport } from './insight-report'\nimport {\n  runSelfImproveMethod,\n  type SelfImproveMethodProvenance,\n  type SelfImproveMethodResult,\n} from './self-improve-method'\nimport { cellsToRunRecords, pairedCompositeSummary } from './self-improve-reporting'\n\nexport type { SelfImproveMethodProvenance, SelfImproveMethodResult } from './self-improve-method'\n\nexport interface SelfImproveBudget {\n  /** Hard spend cap across the full run. Each paid call reserves its enforced\n   *  maximum before dispatch, so completed spend cannot cross this amount. */\n  dollars?: number\n  /** Proposer generations. Default: 3. External methods own their rounds and\n   *  require this value to be omitted or set to 1. Set 0 only for a\n   *  proposer-free baseline run. */\n  generations?: number\n  /** Candidates the proposer emits per generation. Default 2. */\n  populationSize?: number\n  /** Max concurrent cells across the loop. Default 2. */\n  maxConcurrency?: number\n  /** Candidate campaigns scored in parallel. Default 1. Total concurrent\n   *  cells are bounded by `candidateConcurrency * maxConcurrency`. */\n  candidateConcurrency?: number\n  /** Fraction of `scenarios` held out from training, used for the gate.\n   *  Default 0.25. Ignored when `holdoutScenarios` is set explicitly. */\n  holdoutFraction?: number\n  /** Fraction of the non-final cases reserved for method selection.\n   *  Default 0.25. Used only with `method` and ignored when\n   *  `selectionScenarios` is supplied explicitly. */\n  selectionFraction?: number\n  /** Explicit held-out scenarios; overrides `holdoutFraction`. */\n  holdoutScenarios?: Scenario[]\n  /** Holdout policy. Default `'measured'`: split, re-score baseline vs winner\n   *  on the held-out set, gate on that comparison. `'deferred'`: run the\n   *  improvement-set campaigns + search promotion, dispatch ZERO holdout cells,\n   *  force the gate to `'hold'`, return `lift: undefined`, and record\n   *  `holdout: 'deferred'` in the provenance record — for callers that measure\n   *  the held-out comparison in a separate later run instead of faking a\n   *  static holdout scenario and recording a meaningless lift. Unless\n   *  `holdoutScenarios` reserves an explicit set, ALL scenarios train. */\n  holdout?: 'measured' | 'deferred'\n  /** Repeated executions per scenario. A claim's independent-unit count stays unchanged. Default 1. */\n  reps?: number\n  /** DEPTH dial forwarded to the proposer's `propose()` as\n   *  `ctx.maxImprovementShots` — max iterations an agentic candidate generator\n   *  may take per candidate (verify-in-session retries). Unset ⇒ the\n   *  proposer's own default. */\n  maxImprovementShots?: number\n}\n\nexport type SelfImproveProgressEvent =\n  | { kind: 'baseline.started'; scenarios: number }\n  | { kind: 'baseline.completed'; compositeMean: number; durationMs: number }\n  | { kind: 'generation.started'; index: number; populationSize: number }\n  | { kind: 'generation.completed'; index: number; bestComposite: number; durationMs: number }\n  // `lift` is absent when `budget.holdout === 'deferred'` — no held-out\n  // measurement ran, and the search-split delta must not masquerade as one.\n  | { kind: 'gate.decided'; decision: string; lift?: number }\n  | { kind: 'power.estimated'; n: number; sd: number; mde: number; underpowered: boolean }\n\nexport interface SelfImproveOptions<TScenario extends Scenario, TArtifact>\n  extends SearchHistoryAdmissionOptions {\n  /**\n   * Your agent — a function that takes the current `MutableSurface`\n   * (typically a system prompt the loop is optimizing) plus the\n   * scenario + cell ctx, and returns the artifact your judge scores.\n   *\n   * Same shape as `RunOptimizationOptions.dispatchWithSurface`. Wrap a\n   * plain `Dispatch` if you don't have a surface seam:\n   *\n   *   agent: (_surface, scenario, ctx) => yourPlainDispatch(scenario, ctx)\n   *\n   * That mode evaluates without mutating any surface — useful as a\n   * baseline-only run (set `budget.generations = 0`).\n   */\n  agent: (surface: MutableSurface, scenario: TScenario, ctx: DispatchContext) => Promise<TArtifact>\n\n  /**\n   * Snapshot-bearing model identity for agents that do not report a paid-call\n   * receipt through `ctx.cost.runPaidCall()`.\n   *\n   * Omit this when every cell reports its concrete model in a receipt.\n   */\n  model?: string\n  /** Version of execution behavior outside the candidate surface, used by measurement caches. */\n  dispatchRef?: string\n\n  /** Scenarios to evaluate against. Train/holdout split is computed from\n   *  these unless `budget.holdoutScenarios` is set explicitly. */\n  scenarios: TScenario[]\n\n  /** Judge that scores artifacts. Bring your own, or wrap `llmJudge`. */\n  judge: JudgeConfig<TArtifact, TScenario>\n\n  /** Starting surface — system prompt, JSON config, anything `MutableSurface`\n   *  accepts. The proposer mutates this each generation. */\n  baselineSurface: MutableSurface\n\n  /** Budget + loop shape. All fields optional. */\n  budget?: SelfImproveBudget\n\n  /**\n   * Complete prior measurement of `baselineSurface` over the TRAIN split.\n   * Forwarded to the loop body, which validates its surface hash, scenario\n   * split, seed (42), reps, evaluator manifest, and coverage, then skips the baseline search\n   * campaign entirely — no baseline dispatch, no resumability lookup. The\n   * train split is `scenarios` minus the holdout split, so premeasure with\n   * exactly that scenario set (explicit `budget.holdoutScenarios`, or\n   * `budget.holdout: 'deferred'` with no reserved set, makes the train split\n   * deterministic). Prior spend stays in the imported campaign aggregates and\n   * is not re-added to this run's cost ledger. Premeasure with\n   * `dispatchRef: surfaceDispatchRef(baselineSurface, dispatchRef)` and the same judge revision.\n   */\n  premeasuredBaseline?: PremeasuredOptimizationBaseline<TArtifact, TScenario>\n\n  /**\n   * Candidate generator for this local generation loop.\n   * Required when `budget.generations` is greater than zero.\n   */\n  proposer?: SurfaceProposer<ProposalFinding>\n\n  /**\n   * Complete optimization method, such as official GEPA or SkillOpt.\n   * The method receives disjoint train and selection cases and never receives\n   * the final comparison cases. Mutually exclusive with `proposer`.\n   */\n  method?: OptimizationMethod<TScenario, TArtifact>\n\n  /** Explicit method-selection cases. They must also appear in `scenarios`\n   *  and must not overlap the final comparison cases. */\n  selectionScenarios?: TScenario[]\n\n  /** Custom gate. Default is `defaultProductionGate` with\n   *  `deltaThreshold: 0.05` on the held-out split. */\n  gate?: Gate<TArtifact, TScenario>\n\n  /** Placebo control. When supplied AND the winner differs from baseline, the\n   *  loop scores a THIRD held-out arm: the winner surface with its content\n   *  footprint-matched-blanked by this fn (typically via `neutralizeText`). Its\n   *  scores reach the gate as `ctx.neutralizedJudgeScores`, letting a\n   *  `neutralizationGate` reject a win whose lift survives blanking the content\n   *  (decorative — driven by footprint, not content). Costs one extra held-out\n   *  campaign; omit to skip. Compose `neutralizationGate` into `gate` to act on it. */\n  neutralize?: (winnerSurface: MutableSurface, baselineSurface: MutableSurface) => MutableSurface\n\n  /** Storage backend. A filesystem run directory uses `fsCampaignStorage()`;\n   *  a `mem://` directory uses in-memory storage. External methods default to\n   *  a filesystem directory because their official state must survive. */\n  storage?: CampaignStorage\n\n  /** Run directory. Proposer mode defaults to\n   *  `mem://selfImprove-<timestamp>`. External method mode defaults to\n   *  `.agent-eval/runs/self-improve-<timestamp>`. */\n  runDir?: string\n\n  /** Fires once the durable provenance record is written.\n   *  Receives the structured record for inline assertions / custom routing. */\n  onProvenance?: (record: LoopProvenanceRecord | SelfImproveMethodProvenance) => void\n\n  /** Distributed execution seam — same as `RunCampaignOptions.cellPlacement`.\n   *  Returns an opaque placement key the substrate forwards to your agent\n   *  as `ctx.placement`. Combined with `httpDispatch` from\n   *  `/adapters/http`, fans cells across regions. */\n  cellPlacement?: (input: {\n    scenario: TScenario\n    rep: number\n    generation?: number\n  }) => string | undefined\n\n  /** Per-cell agent dispatch deadline, applied to baseline, candidate, and\n   *  held-out campaigns. Default 600_000 ms. Set 0 to disable. */\n  dispatchTimeoutMs?: number\n\n  /** Bounded in-run retry of failed cells — same as\n   *  `RunCampaignOptions.cellRetry`, applied to baseline, candidate, and\n   *  held-out campaigns. Pair with `transientDispatchFailure()` so a\n   *  transport hiccup (a router 503, a dropped stream) is re-dispatched in\n   *  the same slot instead of leaving holdout coverage incomplete. Absent by\n   *  default: a failed cell is final. */\n  cellRetry?: CampaignCellRetryPolicy\n\n  /** Streaming hook — fires on baseline + each generation + gate decision.\n   *  Consumer routes events wherever (UI, dashboard, logs). */\n  onProgress?: (event: SelfImproveProgressEvent) => void\n\n  /** Auto-promotion behavior on a ship decision. Default `'none'` — we\n   *  return the winner; you ship it however you ship. `'pr'` opens a\n   *  GitHub PR via `openAutoPr`; requires `ghOwner` + `ghRepo`. */\n  autoOnPromote?: 'pr' | 'none'\n  ghOwner?: string\n  ghRepo?: string\n\n  /**\n   * Opt-in: ship eval-run events to a hosted orchestrator (ours, your\n   * self-hosted one, or any compatible implementation of the\n   * `docs/hosted-ingest-spec.md` wire format). When set, the substrate\n   * POSTs the final `EvalRunEvent` to `${endpoint}/v1/ingest/eval-runs`\n   * after the loop completes. Failures are logged but do not fail the\n   * loop — local result is always returned.\n   *\n   * For our orchestrator: `{ endpoint: 'https://orchestrator.tangle.tools/v1', apiKey, tenantId }`.\n   *\n   * For your self-hosted: any URL serving the wire format. See\n   * `examples/hosted-ingest-server/` for the reference receiver.\n   */\n  hostedTenant?: HostedTenant\n\n  /** Free-form labels attached to the hosted event (env, branch, model id,\n   *  etc.). Ignored when `hostedTenant` is unset. */\n  hostedLabels?: Record<string, string>\n\n  /** Capture every search artifact and judge score to this store.\n   *  The store is output only and is never exposed to candidate generation.\n   *  Pass `'off'` to disable. Default: off. */\n  labeledStore?: LabeledScenarioStore | 'off'\n\n  /** Capture-source tag for `labeledStore`. Default `'eval-run'`. */\n  captureSource?: 'production-trace' | 'eval-run' | 'manual' | 'red-team' | 'synthetic'\n\n  /**\n   * Per-cell backend-integrity expectation — the fail-loud guard. A cell that\n   * produced an artifact but reported `costUsd === 0` AND zero tokens is a\n   * stub. Modes: `'assert'` throws on the first such cell, `'warn'` logs it,\n   * `'off'` skips the check (offline/replay). Default `'assert'` — `selfImprove`\n   * is the real-run path, so a stub fails loud rather than scoring a clean 0.\n   */\n  expectUsage?: 'assert' | 'warn' | 'off'\n\n  /**\n   * Per-generation findings producer. Runs once on the baseline campaign (as\n   * `generation: -1`) before generation 0 proposes — so single-generation runs\n   * propose with trace context — and again after each generation is scored;\n   * whatever it returns REPLACES the proposer's `findings` for the next\n   * `propose()`. Plug a trace-analyst registry / HALO here. When absent,\n   * findings stay `opts.findings`.\n   */\n  analyzeGeneration?: RunOptimizationOptions<TScenario, TArtifact>['analyzeGeneration']\n\n  /** Static findings forwarded to the proposer's `propose()` as `ctx.findings`\n   *  (a findings-grounded proposer consumes them). Default: none. */\n  findings?: ProposalFinding[]\n\n  /** Override how the WINNER is selected among coverage-complete candidates.\n   *  Defaults to the scalar mean composite (historical behavior). A binary-with-\n   *  replicates consumer whose ship-gate counts an instance resolved only when\n   *  every replicate resolved passes a fail-closed lexicographic key here so that\n   *  winner-selection and the ship-gate rank on the identical metric and cannot\n   *  invert. See `RunOptimizationOptions.selectionRankKey`. */\n  selectionRankKey?: RunOptimizationOptions<TScenario, TArtifact>['selectionRankKey']\n\n  /** Override which scored surface each generation MUTATES. Defaults to the\n   *  global incumbent (hill-climb; the recorded lineage is a chain). Pass\n   *  `crowdedFrontierParent({ seed })` or a custom policy to draw the parent\n   *  from the Pareto frontier; promotion still compares against the incumbent.\n   *  Proposer mode only. See `RunOptimizationOptions.selectParent`. */\n  selectParent?: RunOptimizationOptions<TScenario, TArtifact>['selectParent']\n\n  /** Record this run's candidate search into a durable `SearchLedger` and\n   *  return the bounded receipt on `searchHistory`. See\n   *  `RunOptimizationOptions.searchLedger`. */\n  searchLedger?: RunOptimizationOptions<TScenario, TArtifact>['searchLedger']\n  /** Complete-method final measurement receipts; authority and environment remain caller-owned. */\n  evidence?: CampaignEvidenceContext\n  /** Declare the population and independent units without requiring fresh data. */\n  claim?: EvaluationClaim\n  /** Opt into durable fresh-evidence consumption for this final comparison. */\n  finalEvidence?: FinalEvidencePolicy\n}\n\nexport interface SelfImproveProposerResult<TScenario extends Scenario, TArtifact> {\n  mode: 'proposer'\n  claim?: EvaluationClaim\n  finalEvidence?: FinalEvidenceUse\n  /** Composite mean across all scenarios, baseline run. When\n   *  `budget.holdout === 'deferred'` this is measured on the improvement\n   *  (search) split — no holdout campaign ran. */\n  baseline: {\n    compositeMean: number\n    perScenario: Record<string, number>\n  }\n  /** Composite mean on the held-out set, winner run. When\n   *  `budget.holdout === 'deferred'` this is the winner's improvement-set\n   *  (search) measurement — no holdout campaign ran. */\n  winner: {\n    compositeMean: number\n    perScenario: Record<string, number>\n    surface: MutableSurface\n    /** Proposer label for the promoted change. Absent ⇒ winner == baseline or\n     *  a bare-surface mutator. */\n    label?: string\n    /** Proposer rationale — the \"because Z\" that motivated the promoted change.\n     *  Threaded from the proposer's `ProposedCandidate` through the loop.\n     *  Absent ⇒ winner == baseline. */\n    rationale?: string\n  }\n  /** `winner.compositeMean - baselineOnHoldout.compositeMean`. Positive\n   *  means the gate observed improvement. Absent iff\n   *  `budget.holdout === 'deferred'` — no held-out measurement ran, so there\n   *  is no lift to report (never a fabricated 0). */\n  lift?: number\n  /** The explicit baseline→winner unified diff. Always present (empty string\n   *  when winner == baseline). */\n  diff: string\n  /** Durable, queryable provenance record: candidate→cell→gate→promote chain +\n   *  rationale + diff + backend provenance. The artifact the hosted ingest\n   *  path stores; the +lift RECOMPUTES from `record.heldOutLift`. */\n  provenance: LoopProvenanceRecord\n  /** `defaultProductionGate.decide()` result. */\n  gateDecision: 'ship' | 'hold' | 'need_more_work' | 'model_ceiling' | 'arch_ceiling'\n  /** Number of generations actually explored (may be less than the\n   *  budget if the proposer gave up early). */\n  generationsExplored: number\n  /** Wall-clock total. */\n  durationMs: number\n  /** Total newly observed cost across the full run. */\n  totalCostUsd: number\n  /** Canonical run-wide spend summary. */\n  cost: CostLedgerSummary\n  /** Run-wide receipts across proposal, search, holdout, judging, analysis,\n   *  and promotion work, with phase and actor attribution. */\n  receipts: CostReceipt[]\n  /** Bounded proof envelope over this run's canonical search ledger. Present\n   *  only when `searchLedger` was supplied. */\n  searchHistory?: SearchHistoryReceipt\n  /** Exact external method and source identity, when `method` was used. */\n  optimization?: {\n    name: string\n    cost: OptimizationMethodResult['cost']\n    durationMs?: number\n    provenance?: OptimizationMethodProvenance\n  }\n  /**\n   * Rigor packet: distributional summary, paired-bootstrap lift CI,\n   * judge stats, contamination check, recommendations. Wired through\n   * `analyzeRuns()` on the baseline + winner cells of the campaign.\n   * Hosted-tier dashboards render this as the v3-vs-v4 decision view.\n   */\n  insight: InsightReport\n  /** Approximate detectable lift from baseline holdout observations.\n   *  Declared independent units determine n and baseline variance.\n   *  Absent with fewer than three observations or a deferred holdout. */\n  power?: PowerPreflight\n  /**\n   * Raw substrate result for advanced inspection — full per-generation\n   * candidates, full campaign artifacts, all judge scores. Useful for\n   * debugging or reporting beyond the summary.\n   */\n  raw: RunImprovementLoopResult<TArtifact, TScenario>\n}\n\nexport type SelfImproveResult<TScenario extends Scenario, TArtifact> =\n  | SelfImproveProposerResult<TScenario, TArtifact>\n  | SelfImproveMethodResult<TScenario, TArtifact>\n\nexport type SelfImproveMethodOptions<TScenario extends Scenario, TArtifact> = Omit<\n  SelfImproveOptions<TScenario, TArtifact>,\n  'proposer' | 'onProvenance'\n> & {\n  method: OptimizationMethod<TScenario, TArtifact>\n  proposer?: never\n  onProvenance?: (record: SelfImproveMethodProvenance) => void\n}\n\nexport type SelfImproveProposerOptions<TScenario extends Scenario, TArtifact> = Omit<\n  SelfImproveOptions<TScenario, TArtifact>,\n  'method' | 'onProvenance'\n> & {\n  method?: never\n  onProvenance?: (record: LoopProvenanceRecord) => void\n}\n\n/** Failed self-improvement run with an immutable receipt snapshot. */\nexport class SelfImproveRunError extends Error {\n  readonly cost: CostLedgerSummary\n  readonly receipts: CostReceipt[]\n\n  constructor(cause: unknown, ledger: CostLedgerHandle) {\n    const original = cause instanceof Error ? cause : new Error(String(cause))\n    super(original.message, { cause: original })\n    this.name = 'SelfImproveRunError'\n    this.cost = ledger.summary()\n    this.receipts = ledger.list()\n  }\n}\n\nfunction assertSelfImproveSearchMode<TScenario extends Scenario, TArtifact>(\n  opts: SelfImproveOptions<TScenario, TArtifact>,\n): void {\n  assertSearchHistoryAdmissionOptions(opts)\n  if (\n    !opts.method &&\n    (opts.searchHistoryPolicy !== undefined ||\n      opts.searchHistoryVerification !== undefined ||\n      opts.evidence !== undefined)\n  )\n    throw new Error('selfImprove: search history admission requires method')\n  if (opts.method && opts.proposer) {\n    throw new Error('selfImprove: method and proposer are mutually exclusive')\n  }\n  if (!opts.method) {\n    if (opts.selectionScenarios !== undefined) {\n      throw new Error('selfImprove: selectionScenarios requires method')\n    }\n    return\n  }\n  if (\n    typeof opts.method.name !== 'string' ||\n    !opts.method.name.trim() ||\n    opts.method.name.trim() !== opts.method.name ||\n    typeof opts.method.optimize !== 'function'\n  ) {\n    throw new Error('selfImprove: method must have a trimmed name and optimize(input)')\n  }\n  const budget = opts.budget\n  if (budget?.generations !== undefined) {\n    throw new Error(\n      'selfImprove: method owns its rounds; budget.generations applies only to proposer mode',\n    )\n  }\n  if (budget?.populationSize !== undefined) {\n    throw new Error(\n      'selfImprove: method owns its candidates; budget.populationSize applies only to proposer mode',\n    )\n  }\n  if (\n    budget?.candidateConcurrency !== undefined ||\n    budget?.maxImprovementShots !== undefined ||\n    opts.analyzeGeneration !== undefined ||\n    opts.findings !== undefined ||\n    opts.selectParent !== undefined ||\n    opts.premeasuredBaseline !== undefined ||\n    opts.selectionRankKey !== undefined ||\n    opts.searchLedger !== undefined\n  ) {\n    throw new Error(\n      'selfImprove: candidateConcurrency, maxImprovementShots, analyzeGeneration, findings, selectParent, premeasuredBaseline, selectionRankKey, and searchLedger apply only to proposer mode',\n    )\n  }\n}\n\nfunction splitMethodPartitions<TScenario extends Scenario>(\n  searchScenarios: TScenario[],\n  explicitSelection: TScenario[] | undefined,\n  fraction: number,\n  unitByScenario?: ReadonlyMap<string, string>,\n): { train: TScenario[]; selection: TScenario[] } {\n  if (!Number.isFinite(fraction) || fraction <= 0 || fraction >= 1) {\n    throw new Error('selfImprove: budget.selectionFraction must be in (0, 1)')\n  }\n  const byId = new Map<string, TScenario>()\n  for (const scenario of searchScenarios) {\n    if (byId.has(scenario.id)) {\n      throw new Error(`selfImprove: duplicate scenario id '${scenario.id}'`)\n    }\n    byId.set(scenario.id, scenario)\n  }\n  if (explicitSelection) {\n    if (explicitSelection.length === 0) {\n      throw new Error('selfImprove: selectionScenarios must not be empty')\n    }\n    const selectionIds = new Set<string>()\n    for (const scenario of explicitSelection) {\n      if (!byId.has(scenario.id)) {\n        throw new Error(\n          `selfImprove: selection scenario '${scenario.id}' is absent from the non-final cases`,\n        )\n      }\n      if (selectionIds.has(scenario.id)) {\n        throw new Error(`selfImprove: duplicate selection scenario id '${scenario.id}'`)\n      }\n      selectionIds.add(scenario.id)\n    }\n    const train = searchScenarios.filter((scenario) => !selectionIds.has(scenario.id))\n    if (train.length === 0) {\n      throw new Error('selfImprove: method train split is empty')\n    }\n    if (unitByScenario) {\n      const selectedUnits = new Set([...selectionIds].map((id) => unitByScenario.get(id)))\n      if (train.some((scenario) => selectedUnits.has(unitByScenario.get(scenario.id)))) {\n        throw new Error('selfImprove: training and selection share independent units')\n      }\n    }\n    return {\n      train,\n      selection: explicitSelection.map((scenario) => byId.get(scenario.id)!),\n    }\n  }\n  if (searchScenarios.length < 2) {\n    throw new Error('selfImprove: method requires at least two non-final scenarios')\n  }\n  if (unitByScenario) {\n    const split = splitTrainHoldout(searchScenarios, fraction, unitByScenario)\n    return { train: split.train, selection: split.holdout }\n  }\n  const sorted = [...searchScenarios].sort(\n    (a, b) => stableScenarioHash(a.id) - stableScenarioHash(b.id),\n  )\n  const count = Math.max(1, Math.min(sorted.length - 1, Math.round(sorted.length * fraction)))\n  return {\n    selection: sorted.slice(0, count),\n    train: sorted.slice(count),\n  }\n}\n\n/** 32-bit FNV-1a over raw UTF-16 code units, read as an unsigned int and used\n *  only to order scenarios deterministically.\n *\n *  Frozen: the order it produces assigns scenarios to partitions, so a change\n *  re-partitions decisions already recorded. It masks no byte, unlike\n *  `fnv1a32` in `src/partition-held-out.ts`, so the two disagree on any\n *  non-ASCII id and cannot be shared. */\nfunction stableScenarioHash(value: string): number {\n  let hash = 2166136261 >>> 0\n  for (let index = 0; index < value.length; index++) {\n    hash ^= value.charCodeAt(index)\n    hash = Math.imul(hash, 16777619) >>> 0\n  }\n  return hash\n}\n\n/** Deterministic split by scenario identity, or by source identity for new-unit claims. */\nfunction splitTrainHoldout<TScenario extends Scenario>(\n  scenarios: TScenario[],\n  fraction: number,\n  unitByScenario?: ReadonlyMap<string, string>,\n): { train: TScenario[]; holdout: TScenario[] } {\n  if (unitByScenario) {\n    const units = [...new Set(scenarios.map((scenario) => unitByScenario.get(scenario.id)!))].sort(\n      (a, b) => stableScenarioHash(a) - stableScenarioHash(b) || (a < b ? -1 : a > b ? 1 : 0),\n    )\n    if (units.length < 2)\n      throw new Error('selfImprove: splitting needs at least two independent units')\n    const count = Math.max(1, Math.min(units.length - 1, Math.round(units.length * fraction)))\n    const finalUnits = new Set(units.slice(0, count))\n    return {\n      holdout: scenarios.filter((scenario) => finalUnits.has(unitByScenario.get(scenario.id)!)),\n      train: scenarios.filter((scenario) => !finalUnits.has(unitByScenario.get(scenario.id)!)),\n    }\n  }\n  const sorted = [...scenarios].sort((a, b) => stableScenarioHash(a.id) - stableScenarioHash(b.id))\n  const nHoldout = Math.max(1, Math.min(sorted.length - 1, Math.round(sorted.length * fraction)))\n  return {\n    holdout: sorted.slice(0, nHoldout),\n    train: sorted.slice(nHoldout),\n  }\n}\n\n/**\n * Latest search campaign measured for the winner surface; the baseline search\n * campaign when the winner IS the baseline. Used by the deferred-holdout\n * summary, where no holdout campaign exists to summarize.\n */\nfunction winnerSearchCampaign<TScenario extends Scenario, TArtifact>(\n  result: RunImprovementLoopResult<TArtifact, TScenario>,\n): RunImprovementLoopResult<TArtifact, TScenario>['baselineCampaign'] {\n  for (let i = result.generations.length - 1; i >= 0; i--) {\n    const measured = result.generations[i]?.surfaces.find(\n      (s) => s.surfaceHash === result.winnerSurfaceHash,\n    )\n    if (measured) return measured.campaign\n  }\n  return result.baselineCampaign\n}\n\n/**\n * One-shot self-improvement loop. See module docstring for defaults +\n * extension points.\n *\n * @example Minimum:\n *\n *   const result = await selfImprove({\n *     agent: (surface, scenario, ctx) => myAgent(surface, scenario, ctx.signal),\n *     scenarios,\n *     judge,\n *     baselineSurface: DEFAULT_PROMPT,\n *     proposer,\n *   })\n *   console.log(`lift: ${result.lift.toFixed(3)} (${result.gateDecision})`)\n *\n * @example Distributed (workers in three regions):\n *\n *   await selfImprove({\n *     agent: httpDispatch({ resolveUrl: ({ placement }) => REGION_URLS[placement!] }),\n *     scenarios,\n *     judge,\n *     baselineSurface: DEFAULT_PROMPT,\n *     cellPlacement: ({ scenario }) => scenario.region,\n *     budget: { maxConcurrency: 12 },\n *   })\n */\nexport function selfImprove<TScenario extends Scenario, TArtifact>(\n  opts: SelfImproveMethodOptions<TScenario, TArtifact>,\n): Promise<SelfImproveMethodResult<TScenario, TArtifact>>\nexport function selfImprove<TScenario extends Scenario, TArtifact>(\n  opts: SelfImproveProposerOptions<TScenario, TArtifact>,\n): Promise<SelfImproveProposerResult<TScenario, TArtifact>>\nexport function selfImprove<TScenario extends Scenario, TArtifact>(\n  opts: SelfImproveOptions<TScenario, TArtifact>,\n): Promise<SelfImproveResult<TScenario, TArtifact>>\nexport async function selfImprove<TScenario extends Scenario, TArtifact>(\n  opts:\n    | SelfImproveOptions<TScenario, TArtifact>\n    | SelfImproveMethodOptions<TScenario, TArtifact>\n    | SelfImproveProposerOptions<TScenario, TArtifact>,\n): Promise<SelfImproveResult<TScenario, TArtifact>> {\n  opts = {\n    ...opts,\n    judge: captureJudge(opts.judge),\n    ...(opts.claim || opts.finalEvidence\n      ? {\n          claim: opts.claim && defineEvaluationClaim(opts.claim),\n          finalEvidence: opts.finalEvidence && captureFinalEvidencePolicy(opts.finalEvidence),\n          scenarios: structuredClone(opts.scenarios),\n          selectionScenarios: opts.selectionScenarios && structuredClone(opts.selectionScenarios),\n          budget: opts.budget && structuredClone(opts.budget),\n        }\n      : {}),\n  }\n  const startedAt = Date.now()\n  const requestedRunDir =\n    opts.runDir ??\n    (opts.method ? `.agent-eval/runs/self-improve-${startedAt}` : `mem://selfImprove-${startedAt}`)\n  const runDir = resolveRunDir(requestedRunDir)\n  const storage =\n    opts.storage ?? (runDir.startsWith('mem://') ? inMemoryCampaignStorage() : fsCampaignStorage())\n  const costLedger = createRunCostLedger({\n    storage,\n    runDir,\n    costCeilingUsd: opts.budget?.dollars,\n  })\n  try {\n    return await runSelfImprove(\n      opts as SelfImproveOptions<TScenario, TArtifact>,\n      costLedger,\n      startedAt,\n      runDir,\n      storage,\n    )\n  } catch (error) {\n    throw new SelfImproveRunError(error, costLedger)\n  }\n}\n\nasync function runSelfImprove<TScenario extends Scenario, TArtifact>(\n  opts: SelfImproveOptions<TScenario, TArtifact>,\n  costLedger: CostLedgerHandle,\n  startedAt: number,\n  runDir: string,\n  storage: CampaignStorage,\n): Promise<SelfImproveResult<TScenario, TArtifact>> {\n  const budget = opts.budget ?? {}\n  assertSelfImproveSearchMode(opts)\n  const maxConcurrency = budget.maxConcurrency ?? 2\n  const holdoutFraction = budget.holdoutFraction ?? 0.25\n  const holdoutMode = budget.holdout ?? 'measured'\n  const holdoutDeferred = holdoutMode === 'deferred'\n  const expectUsage = opts.expectUsage ?? 'assert'\n  const unitByScenario = opts.claim ? evaluationUnitMap(opts.claim, opts.scenarios) : undefined\n  const splitUnitByScenario =\n    opts.claim?.generalization === 'new-units' ? unitByScenario : undefined\n\n  // Deferred holdout without an explicitly reserved set trains on EVERYTHING:\n  // there is no held-out measurement in this run, so carving out a fraction\n  // would waste scenarios. An explicit `holdoutScenarios` set stays reserved\n  // (excluded from training) even when deferred, for the later measured run.\n  const explicitHoldout = budget.holdoutScenarios\n  const { train, holdout } = explicitHoldout\n    ? {\n        train: opts.scenarios.filter((s) => !explicitHoldout.some((h) => h.id === s.id)),\n        holdout: explicitHoldout as TScenario[],\n      }\n    : holdoutDeferred\n      ? { train: opts.scenarios, holdout: [] as TScenario[] }\n      : splitTrainHoldout(opts.scenarios, holdoutFraction, splitUnitByScenario)\n\n  if (train.length === 0) {\n    throw new Error(\n      'selfImprove: train split is empty. Reduce holdoutFraction or pass more scenarios.',\n    )\n  }\n  if (holdout.length === 0 && !holdoutDeferred) {\n    throw new Error('selfImprove: holdout split is empty. Pass more scenarios.')\n  }\n\n  if (opts.claim?.generalization === 'new-units') {\n    assertIndependentEvaluationSplit(opts.claim, holdout, train)\n  }\n  if (opts.finalEvidence) {\n    if (holdoutDeferred) throw new Error('final evidence requires measured holdout')\n    if (opts.method) await reserveFinalEvidence(opts.finalEvidence, opts.claim, holdout, train)\n  }\n\n  if (opts.method) {\n    const partitions = splitMethodPartitions(\n      train,\n      opts.selectionScenarios,\n      budget.selectionFraction ?? 0.25,\n      splitUnitByScenario,\n    )\n    return runSelfImproveMethod({\n      opts: { ...opts, method: opts.method },\n      train: partitions.train,\n      selection: partitions.selection,\n      holdout,\n      costLedger,\n      storage,\n      runDir,\n      startedAt,\n    })\n  }\n  const generations = budget.generations ?? 3\n  const populationSize = budget.populationSize ?? 2\n  if (generations > 0 && !opts.proposer) {\n    throw new Error(\n      'selfImprove: method or proposer is required when budget.generations is greater than zero',\n    )\n  }\n  const proposer: SurfaceProposer<ProposalFinding> = opts.proposer ?? {\n    kind: 'baseline-only',\n    propose: async () => [],\n  }\n\n  const gate: Gate<TArtifact, TScenario> =\n    opts.gate ??\n    defaultProductionGate<TArtifact, TScenario>({\n      holdoutScenarios: holdout,\n      deltaThreshold: opts.claim?.minimumEffect ?? 0.05,\n      independentUnitByScenarioId: opts.claim ? evaluationUnitMap(opts.claim, holdout) : undefined,\n    })\n\n  if (opts.onProgress) {\n    opts.onProgress({ kind: 'baseline.started', scenarios: opts.scenarios.length })\n  }\n\n  const result = await runImprovementLoop<TScenario, TArtifact>({\n    scenarios: train,\n    baselineSurface: opts.baselineSurface,\n    premeasuredBaseline: opts.premeasuredBaseline,\n    dispatchWithSurface: opts.agent,\n    dispatchRef: opts.dispatchRef,\n    proposer,\n    judges: [opts.judge],\n    populationSize,\n    maxGenerations: generations,\n    candidateConcurrency: budget.candidateConcurrency,\n    reps: budget.reps,\n    maxImprovementShots: budget.maxImprovementShots,\n    holdoutScenarios: holdout,\n    claim: opts.claim,\n    finalEvidence: opts.finalEvidence,\n    holdout: holdoutMode,\n    gate,\n    neutralize: opts.neutralize,\n    autoOnPromote: opts.autoOnPromote ?? 'none',\n    ghOwner: opts.ghOwner,\n    ghRepo: opts.ghRepo,\n    storage,\n    runDir,\n    maxConcurrency,\n    cellPlacement: opts.cellPlacement,\n    dispatchTimeoutMs: opts.dispatchTimeoutMs,\n    cellRetry: opts.cellRetry,\n    costLedger,\n    expectUsage,\n    labeledStore: opts.labeledStore,\n    captureSource: opts.captureSource,\n    analyzeGeneration: opts.analyzeGeneration,\n    findings: opts.findings,\n    selectionRankKey: opts.selectionRankKey,\n    selectParent: opts.selectParent,\n    searchLedger: opts.searchLedger,\n  })\n\n  // Deferred holdout ran zero holdout cells, so the summary stats come from\n  // the improvement-set (search) campaigns — labeled as such on the result\n  // type — and `lift` is omitted rather than fabricated from empty campaigns.\n  const reportSplit: RunSplitTag = holdoutDeferred ? 'search' : 'holdout'\n  const reportBaselineCampaign = holdoutDeferred\n    ? result.baselineCampaign\n    : result.baselineOnHoldout\n  const reportWinnerCampaign = holdoutDeferred\n    ? winnerSearchCampaign(result)\n    : result.winnerOnHoldout\n  const reportUnitMap = opts.claim\n    ? evaluationUnitMap(opts.claim, holdoutDeferred ? train : holdout)\n    : undefined\n  const report = pairedCompositeSummary(reportBaselineCampaign, reportWinnerCampaign, reportUnitMap)\n  const baseline = report.baseline\n  const winnerStats = report.winner\n\n  // Repetitions refine a declared unit's mean without increasing the sample size.\n  let power: PowerPreflight | undefined\n  const baselineHoldoutComposites = holdoutDeferred ? [] : report.baselineComposites\n  if (baselineHoldoutComposites.length >= 3) {\n    // The shared judge's systematic bias remains outside this variance estimate.\n    power = powerPreflight({\n      baselineComposites: baselineHoldoutComposites,\n      deltaThreshold: opts.claim?.minimumEffect ?? 0.05,\n      sharedScorerChannel: true,\n    })\n    if (opts.onProgress) {\n      opts.onProgress({\n        kind: 'power.estimated',\n        n: power.n,\n        sd: power.sd,\n        mde: power.mde,\n        underpowered: power.underpowered,\n      })\n    }\n    if (power.underpowered && generations > 0) {\n      console.warn(`[selfImprove] ${power.recommendation}`)\n    }\n  }\n\n  if (opts.onProgress) {\n    opts.onProgress({\n      kind: 'baseline.completed',\n      compositeMean: baseline.compositeMean,\n      durationMs: Date.now() - startedAt,\n    })\n    opts.onProgress({\n      kind: 'gate.decided',\n      decision: result.gateResult.decision,\n      // Deferred holdout has no held-out measurement: in that mode the summary\n      // stats are search-split numbers, and emitting their delta as `lift`\n      // would misreport a train-split delta as a held-out one. Omit instead.\n      ...(holdoutDeferred ? {} : { lift: winnerStats.compositeMean - baseline.compositeMean }),\n    })\n  }\n\n  const cost = result.cost\n  const totalCost = cost.totalCostUsd\n\n  // Rigor packet: feed baseline + winner cells through analyzeRuns().\n  // The two candidates (`baseline` / `winner`) give the lift section a\n  // clean paired comparison; per-judge / per-dimension / cost-quality\n  // sections populate from the cells' judgeScores.\n  const insight = await analyzeRuns({\n    runs: [\n      ...cellsToRunRecords(\n        reportBaselineCampaign.cells,\n        'baseline',\n        runDir,\n        opts.baselineSurface,\n        reportSplit,\n        opts.model,\n      ),\n      ...(reportWinnerCampaign === reportBaselineCampaign\n        ? []\n        : cellsToRunRecords(\n            reportWinnerCampaign.cells,\n            'winner',\n            runDir,\n            result.winnerSurface,\n            reportSplit,\n            opts.model,\n          )),\n    ],\n    baselineCandidateId: 'baseline',\n    independentUnitByScenarioId: reportUnitMap,\n    decisionThreshold: opts.claim?.minimumEffect ?? 0.05,\n    ...(reportWinnerCampaign === reportBaselineCampaign ? {} : { candidateCandidateId: 'winner' }),\n  })\n\n  // ── Durable provenance: candidate→cell→gate→promote chain + rationale +\n  // diff + backend provenance. Always emitted; the +lift recomputes from it.\n  const durationMs = Date.now() - startedAt\n  const { record: provenance } = await emitLoopProvenance<TArtifact, TScenario>({\n    ...loopProvenanceArgsFromResult({\n      runId: `${runDir}#${startedAt}`,\n      runDir,\n      timestamp: new Date(startedAt).toISOString(),\n      baselineSurface: opts.baselineSurface,\n      result,\n      costReceipts: costLedger.list(),\n      totalCostUsd: totalCost,\n      totalDurationMs: durationMs,\n      independentUnitByScenarioId: holdoutDeferred ? undefined : reportUnitMap,\n    }),\n    storage,\n    hostedClient: opts.hostedTenant ? createHostedClient(opts.hostedTenant) : undefined,\n  })\n  if (opts.onProvenance) opts.onProvenance(provenance)\n\n  const summary: SelfImproveProposerResult<TScenario, TArtifact> = {\n    ...(opts.claim ? { claim: opts.claim } : {}),\n    ...(result.finalEvidence ? { finalEvidence: result.finalEvidence } : {}),\n    mode: 'proposer',\n    baseline,\n    winner: {\n      ...winnerStats,\n      surface: result.winnerSurface,\n      ...(result.winnerLabel ? { label: result.winnerLabel } : {}),\n      ...(result.winnerRationale ? { rationale: result.winnerRationale } : {}),\n    },\n    ...(holdoutDeferred ? {} : { lift: winnerStats.compositeMean - baseline.compositeMean }),\n    diff: result.promotedDiff,\n    provenance,\n    gateDecision: result.gateResult.decision,\n    generationsExplored: result.generations.length,\n    durationMs,\n    totalCostUsd: totalCost,\n    cost,\n    receipts: costLedger.list(),\n    ...(result.searchHistory ? { searchHistory: result.searchHistory } : {}),\n    insight,\n    ...(power ? { power } : {}),\n    raw: result,\n  }\n\n  // Opt-in hosted ingest. Failures are logged but never fail the loop: the\n  // local result is always returned.\n  if (opts.hostedTenant) {\n    try {\n      await shipEvalRunToHosted(opts.hostedTenant, opts, summary, result, runDir)\n    } catch (err) {\n      const msg = err instanceof Error ? err.message : String(err)\n      // eslint-disable-next-line no-console -- intentional: hosted-ingest is best-effort\n      console.warn(`[agent-eval] hosted ingest failed (continuing): ${msg}`)\n    }\n  }\n\n  return summary\n}\n\nasync function shipEvalRunToHosted<TScenario extends Scenario, TArtifact>(\n  tenant: HostedTenant,\n  opts: SelfImproveOptions<TScenario, TArtifact>,\n  summary: SelfImproveProposerResult<TScenario, TArtifact>,\n  raw: RunImprovementLoopResult<TArtifact, TScenario>,\n  runDir: string,\n): Promise<void> {\n  const client = createHostedClient(tenant)\n\n  function snapshotFromCampaign(\n    index: number,\n    surface: MutableSurface,\n    campaign: RunImprovementLoopResult<TArtifact, TScenario>['baselineCampaign'],\n    durationMs: number,\n  ): EvalRunGenerationSnapshot {\n    const cells: EvalRunCellScore[] = campaign.cells.map((cell) => {\n      const execution = campaignCellExecutionEvidence(cell)\n      return {\n        scenarioId: cell.scenarioId,\n        rep: cell.rep,\n        compositeMean: campaignCellTaskScore(cell) ?? null,\n        dimensions: campaignCellJudgeDimensions(cell),\n        terminalOutcome: execution.terminalOutcome,\n        executionErrorCount: execution.executionErrorCount ?? null,\n        errorMessage: cell.error ?? undefined,\n      }\n    })\n    const scoredCells = cells.flatMap((cell) =>\n      cell.compositeMean === null ? [] : [cell.compositeMean],\n    )\n    const compositeMean =\n      scoredCells.length === 0\n        ? null\n        : scoredCells.reduce((sum, score) => sum + score, 0) / scoredCells.length\n    return {\n      index,\n      surfaceHash: surfaceHash(surface),\n      surface,\n      cells,\n      compositeMean,\n      costUsd: campaign.aggregates.cost.totalCostUsd,\n      durationMs,\n    }\n  }\n\n  const generations: EvalRunGenerationSnapshot[] = []\n  // Baseline as generation 0.\n  generations.push(snapshotFromCampaign(0, opts.baselineSurface, raw.baselineCampaign, 0))\n  // Improvement generations as 1..N. Substrate stores per-surface campaigns\n  // per generation — we summarize the WINNING surface per generation here.\n  for (const gen of raw.generations) {\n    const winner = gen.surfaces.reduce(\n      (best, s) =>\n        s.campaign.aggregates.cellsExecuted > 0 &&\n        (best === undefined || averageComposite(s.campaign) > averageComposite(best.campaign))\n          ? s\n          : best,\n      gen.surfaces[0],\n    )\n    if (!winner) continue\n    generations.push(\n      snapshotFromCampaign(gen.record.generationIndex + 1, winner.surface, winner.campaign, 0),\n    )\n  }\n\n  const event: EvalRunEvent = {\n    runId: `${runDir}#${Date.now()}`,\n    runDir,\n    timestamp: new Date().toISOString(),\n    status: 'finished',\n    labels: opts.hostedLabels ?? {},\n    baseline: generations[0],\n    generations,\n    gateDecision: summary.gateDecision,\n    holdoutLift: summary.lift,\n    totalCostUsd: summary.totalCostUsd,\n    totalDurationMs: summary.durationMs,\n    insightReport: summary.insight,\n  }\n\n  await client.ingestEvalRun(event)\n}\n\nfunction averageComposite(\n  campaign: RunImprovementLoopResult<unknown, Scenario>['baselineCampaign'],\n): number {\n  const aggs = Object.values(campaign.aggregates.byScenario)\n  return aggs.length === 0 ? 0 : aggs.reduce((s, a) => s + a.meanComposite, 0) / aggs.length\n}\n","import type { RunEvalOptions } from '../campaign/presets/run-eval'\nimport { runEval } from '../campaign/presets/run-eval'\nimport { inMemoryCampaignStorage } from '../campaign/storage'\nimport { surfaceDispatchRef } from '../campaign/surface-identity'\nimport type {\n  CampaignResult,\n  DispatchContext,\n  JudgeConfig,\n  MutableSurface,\n  Scenario,\n} from '../campaign/types'\nimport type { HostedTenant } from '../hosted/client'\nimport {\n  type SelfImproveBudget,\n  type SelfImproveOptions,\n  type SelfImproveResult,\n  selfImprove,\n} from './self-improve'\n\nexport type AgentEvalAgent<TScenario extends Scenario, TArtifact> = (\n  surface: MutableSurface,\n  scenario: TScenario,\n  ctx: DispatchContext,\n) => Promise<TArtifact>\n\nexport type DefineAgentEvalOptions<TScenario extends Scenario, TArtifact> = SelfImproveOptions<\n  TScenario,\n  TArtifact\n>\n\nexport interface AgentEvalEvaluateOptions<TScenario extends Scenario, TArtifact>\n  extends Omit<\n    RunEvalOptions<TScenario, TArtifact>,\n    'dispatch' | 'judges' | 'runDir' | 'scenarios'\n  > {\n  /** Scenario set to evaluate. Defaults to the scenarios passed to `defineAgentEval`. */\n  scenarios?: TScenario[]\n  /** Surface to evaluate. Defaults to the baseline surface passed to `defineAgentEval`. */\n  surface?: MutableSurface\n  /** Agent to evaluate. Defaults to the agent passed to `defineAgentEval`. */\n  agent?: AgentEvalAgent<TScenario, TArtifact>\n  /** Single judge override. Ignored when `judges` is set. */\n  judge?: JudgeConfig<TArtifact, TScenario>\n  /** Full judge list override. Defaults to the single judge passed to `defineAgentEval`. */\n  judges?: JudgeConfig<TArtifact, TScenario>[]\n  /** Logical or filesystem run directory. Defaults to an in-memory run. */\n  runDir?: string\n}\n\nexport type AgentEvalImproveOptions<TScenario extends Scenario, TArtifact> = Omit<\n  Partial<SelfImproveOptions<TScenario, TArtifact>>,\n  'budget' | 'hostedTenant'\n> & {\n  budget?: Partial<SelfImproveBudget>\n  hostedTenant?: Partial<HostedTenant>\n}\n\nexport interface DefinedAgentEval<TScenario extends Scenario, TArtifact> {\n  /** The default scenarios used by `evaluate()` and `improve()`. */\n  readonly scenarios: readonly TScenario[]\n  /** The default baseline surface used by `evaluate()` and `improve()`. */\n  readonly baselineSurface: MutableSurface\n  /**\n   * Run one scored evaluation. Use this for a baseline score or to score one\n   * candidate surface without running an improvement loop.\n   */\n  evaluate(\n    opts?: AgentEvalEvaluateOptions<TScenario, TArtifact>,\n  ): Promise<CampaignResult<TArtifact, TScenario>>\n  /**\n   * Run the closed improvement loop. Per-call overrides replace the definition\n   * except for nested config objects (`budget`, `hostedTenant`), which\n   * are merged field-by-field so callers can override one knob without\n   * repeating secrets or budget defaults.\n   */\n  improve(\n    opts?: AgentEvalImproveOptions<TScenario, TArtifact>,\n  ): Promise<SelfImproveResult<TScenario, TArtifact>>\n}\n\n/**\n * Define an agent eval once, then either score a surface with `evaluate()` or\n * run the closed loop with `improve()`.\n *\n * This is a DX wrapper only: it delegates to `runEval()` and `selfImprove()` and\n * returns their native result shapes.\n */\nexport function defineAgentEval<TScenario extends Scenario, TArtifact>(\n  defaults: DefineAgentEvalOptions<TScenario, TArtifact>,\n): DefinedAgentEval<TScenario, TArtifact> {\n  const defaultEvaluateOptions = evaluateDefaults(defaults)\n\n  return {\n    scenarios: defaults.scenarios,\n    baselineSurface: defaults.baselineSurface,\n\n    async evaluate(opts = {}) {\n      const { agent, judge, judges, runDir, scenarios, surface, ...campaignOpts } = opts\n      const selectedAgent = agent ?? defaults.agent\n      const selectedSurface = surface ?? defaults.baselineSurface\n      const selectedRunDir = runDir ?? defaults.runDir ?? `mem://defineAgentEval-${Date.now()}`\n      const selectedStorage =\n        campaignOpts.storage ??\n        defaultEvaluateOptions.storage ??\n        (selectedRunDir.startsWith('mem://') ? inMemoryCampaignStorage() : undefined)\n      const evalOptions: RunEvalOptions<TScenario, TArtifact> = {\n        ...defaultEvaluateOptions,\n        ...campaignOpts,\n        ...(selectedStorage ? { storage: selectedStorage } : {}),\n        runDir: selectedRunDir,\n        scenarios: scenarios ?? defaults.scenarios,\n        dispatch: (scenario, ctx) => selectedAgent(selectedSurface, scenario, ctx),\n        dispatchRef: surfaceDispatchRef(\n          selectedSurface,\n          campaignOpts.dispatchRef ?? defaults.dispatchRef,\n        ),\n        judges: evaluateJudges(judges, judge ?? defaults.judge),\n      }\n      if (evalOptions.reps !== undefined)\n        evalOptions.reps = requirePositiveInteger(evalOptions.reps, 'reps')\n      return runEval<TScenario, TArtifact>(evalOptions)\n    },\n\n    async improve(opts = {}) {\n      const {\n        budget: budgetOverride,\n        hostedTenant: hostedTenantOverride,\n        ...topLevelOverrides\n      } = opts\n      const merged = mergeDefined(defaults, topLevelOverrides)\n      const budget = mergeBudget(defaults.budget, budgetOverride)\n      const hostedTenant = mergeHostedTenant(defaults.hostedTenant, hostedTenantOverride)\n      return selfImprove<TScenario, TArtifact>({\n        ...merged,\n        ...(budget ? { budget } : {}),\n        ...(hostedTenant ? { hostedTenant } : {}),\n      })\n    },\n  }\n}\n\ntype SharedEvaluateDefaults<TScenario extends Scenario, TArtifact> = Omit<\n  RunEvalOptions<TScenario, TArtifact>,\n  'dispatch' | 'judges' | 'runDir' | 'scenarios'\n>\n\nfunction evaluateDefaults<TScenario extends Scenario, TArtifact>(\n  defaults: DefineAgentEvalOptions<TScenario, TArtifact>,\n): SharedEvaluateDefaults<TScenario, TArtifact> {\n  const out: SharedEvaluateDefaults<TScenario, TArtifact> = {}\n  if (defaults.storage) out.storage = defaults.storage\n  if (defaults.labeledStore) out.labeledStore = defaults.labeledStore\n  if (defaults.captureSource) out.captureSource = defaults.captureSource\n  if (defaults.cellPlacement) out.cellPlacement = defaults.cellPlacement\n  if (defaults.expectUsage) out.expectUsage = defaults.expectUsage\n  if (defaults.budget?.dollars !== undefined) out.costCeiling = defaults.budget.dollars\n  if (defaults.budget?.maxConcurrency !== undefined)\n    out.maxConcurrency = defaults.budget.maxConcurrency\n  if (defaults.budget?.reps !== undefined)\n    out.reps = requirePositiveInteger(defaults.budget.reps, 'budget.reps')\n  return out\n}\n\nfunction mergeBudget(\n  defaults: SelfImproveBudget | undefined,\n  overrides: Partial<SelfImproveBudget> | undefined,\n): SelfImproveBudget | undefined {\n  const merged = mergeOptionalObject(defaults, overrides)\n  if (merged?.reps !== undefined) merged.reps = requirePositiveInteger(merged.reps, 'budget.reps')\n  return merged\n}\n\nfunction mergeHostedTenant(\n  defaults: HostedTenant | undefined,\n  overrides: Partial<HostedTenant> | undefined,\n): HostedTenant | undefined {\n  const merged = mergeOptionalObject(defaults, overrides)\n  if (!merged) return undefined\n  if (!merged.endpoint?.trim() || !merged.apiKey?.trim() || !merged.tenantId?.trim()) {\n    throw new Error(\n      'defineAgentEval.improve: hostedTenant requires endpoint, apiKey, and tenantId after merging defaults and overrides',\n    )\n  }\n  return merged\n}\n\nfunction mergeDefined<T extends object>(defaults: T, overrides: Partial<T> | undefined): T {\n  if (!overrides) return defaults\n  const merged = { ...defaults } as Record<string, unknown>\n  for (const [key, value] of Object.entries(overrides)) {\n    if (value !== undefined) merged[key] = value\n  }\n  return merged as T\n}\n\nfunction mergeOptionalObject<T extends object>(\n  defaults: T | undefined,\n  overrides: Partial<T> | undefined,\n): T | undefined {\n  if (!defaults && !overrides) return undefined\n  return mergeDefined(defaults ?? ({} as T), overrides)\n}\n\nfunction evaluateJudges<TArtifact, TScenario extends Scenario>(\n  judges: JudgeConfig<TArtifact, TScenario>[] | undefined,\n  defaultJudge: JudgeConfig<TArtifact, TScenario>,\n): JudgeConfig<TArtifact, TScenario>[] {\n  if (judges !== undefined) {\n    if (judges.length === 0) {\n      throw new Error('defineAgentEval.evaluate: judges must not be empty')\n    }\n    return judges\n  }\n  return [defaultJudge]\n}\n\nfunction requirePositiveInteger(value: number, field: string): number {\n  if (!Number.isInteger(value) || value < 1) {\n    throw new Error(`defineAgentEval: ${field} must be a positive integer`)\n  }\n  return value\n}\n"],"mappings":";;;;;;;;;AAYA,SAAgB,uBACd,UACA,QACA,6BACA;CACA,MAAM,SAAS,yBAAyB,UAAU,QAAQ,2BAA2B;CACrF,MAAM,eAAe,aACnB,OAAO,YACL,OAAO,QAAQ,SAAS,WAAW,UAAU,CAAC,CAAC,KAAK,CAAC,IAAI,eAAe,CACtE,IACA,UAAU,aACZ,CAAC,CACH;CACF,OAAO;EACL,UAAU;GAAE,eAAe,OAAO;GAAY,aAAa,YAAY,QAAQ;EAAE;EACjF,QAAQ;GAAE,eAAe,OAAO;GAAW,aAAa,YAAY,MAAM;EAAE;EAC5E,oBAAoB,OAAO;EAC3B,cAAc,OAAO;CACvB;AACF;;;;;AAMA,SAAS,WAAW,GAAmB;CACrC,IAAI,IAAI;CACR,KAAK,IAAI,IAAI,GAAG,IAAI,EAAE,QAAQ,KAAK;EACjC,KAAK,EAAE,WAAW,CAAC;EACnB,IAAI,KAAK,KAAK,GAAG,QAAQ,MAAM;CACjC;CACA,OAAO,EAAE,SAAS,EAAE,CAAC,CAAC,SAAS,GAAG,GAAG;AACvC;;;;;;;;AASA,SAAgB,kBACd,OACA,aACA,OACA,SACA,UACA,eACa;CACb,MAAM,aAAa,mBAAmB,OAAO;CAC7C,MAAM,aAAa,mBAAmB,WAAW;CACjD,OAAO,MAAM,KAAK,SAAS;EACzB,MAAM,gBAAgB,KAAK,mBAAmB,KAAK,gBAAgB,CAAC,KAAK,aAAa,IAAI,CAAC;EAC3F,IAAI,cAAc,SAAS,GACzB,MAAM,IAAI,gBACR,oBAAoB,KAAK,OAAO,+BAA+B,cAAc,KAAK,IAAI,GACxF;EAEF,MAAM,QAAQ,cAAc,MAAM;EAClC,IAAI,CAAC,OACH,MAAM,IAAI,gBACR,2CAA2C,KAAK,OAAO,gCACzD;EAEF,IAAI,CAAC,iBAAiB,KAAK,GACzB,MAAM,IAAI,gBACR,sBAAsB,MAAM,sCAAsC,KAAK,QACzE;EAEF,OAAO,wBAAwB,MAAM;GACnC,OAAO,GAAG,MAAM,IAAI,YAAY,IAAI,KAAK;GACzC,cAAc;GACd;GAEA,MACE,KAAK,MAAM,MACX,WAAW,KAAK,UAAU,CAAC,CACxB,MAAM,GAAG,CAAC,CAAC,CACX,MAAM,EAAE,CAAC,CACT,QAAQ,GAAG,MAAO,IAAI,KAAK,EAAE,WAAW,CAAC,MAAO,GAAG,CAAC;GACzD;GACA;GACA;GACA,WAAW;GACX;EACF,CAAC;CACH,CAAC;AACH;;;;ACSA,eAAsB,qBAA4D,MAUvB;CACzD,MAAM,EAAE,MAAM,OAAO,WAAW,SAAS,YAAY,SAAS,QAAQ,cAAc;CACpF,MAAM,SAAS,KAAK,UAAU,CAAC;CAC/B,IAAI,KAAK,kBAAkB,SAAS,CAAC,KAAK,WAAW,CAAC,KAAK,SACzD,MAAM,IAAI,MAAM,2DAA2D;CAE7E,MAAM,kBAAkB,KAAK,aAAa,KAAA,IAAY,KAAA,IAAY,gBAAgB,KAAK,QAAQ;CAC/F,IAAI,mBAAmB,OAAO,YAAY,YACxC,MAAM,IAAI,MAAM,yDAAyD;CAC3E,MAAM,kBAAkB,gBAAgB,KAAK,eAAe;CAC5D,MAAM,QAAQ,KAAK;CACnB,MAAM,EACJ,UACA,MAAM,YACN,YACE,MAAM,0BAA0B;EAClC,QAAQ,KAAK;EACb;EACA,qBAAqB,KAAK;EAC1B,2BAA2B,KAAK;EAChC,OAAO;GACL;GACA,gBAAgB;GAChB,oBAAoB;GACpB,qBAAqB,KAAK;GAC1B,QAAQ,CAAC,KAAK;GACd,QAAQ,GAAG,OAAO,gBAAgB,KAAK,OAAO,KAAK,QAAQ,oBAAoB,GAAG;GAClF,MAAM;GACN,YAAY;IACV;IACA,gBAAgB,OAAO,kBAAkB;IACzC,MAAM,OAAO;IACb,aAAa,KAAK;IAClB,mBAAmB,KAAK;IACxB,WAAW,KAAK;IAChB,eAAe,KAAK;IACpB,cAAc,KAAK;IACnB,eAAe,KAAK;IACpB,aAAa,KAAK,eAAe;GACnC;GACA;EACF;CACF,CAAC;CACD,MAAM,aAAa;CACnB,MAAM,kBACJ,uBACE;EAAC;EAAoB;EAAkB;EAAuB;CAAgB,CAAC,CAAC,KAC7E,WAAW;EACV,OAAO;EACP,MAAM,sBAAsB,WAAW,QAAQ,EAAE,OAAO,GAAG,WAAW,GAAG,QAAQ,CAAC,CAAC;CACrF,EACF,CACF;CACF,MAAM,kBACJ,uBAAuB,CACrB;EAAE,OAAO,KAAK,OAAO;EAAM,MAAM;CAAW,GAC5C;EAAE,OAAO;EAAoB,MAAM,UAAU;CAAE,CACjD,CAAC;CACH,MAAM,OACJ,KAAK,QACL,sBAA4C;EAC1C,kBAAkB;EAClB,gBAAgB,KAAK,OAAO,iBAAiB;EAC7C,6BAA6B,KAAK,QAAQ,kBAAkB,KAAK,OAAO,OAAO,IAAI,KAAA;CACrF,CAAC;CACH,MAAM,aAAa,MAAM,mBAAmB;EAC1C,OAAO,KAAK;EACZ,eAAe,KAAK;EACpB;EACA,eAAe,SAAS;EACxB,WAAW;EACX,qBAAqB,KAAK;EAC1B,aAAa,KAAK;EAClB,QAAQ,CAAC,KAAK;EACd,SAAS,OAAO;EAChB,YAAY,KAAK;EACjB,MAAM;GACJ,GAAG;GACH,MAAM,OAAO,SAAS;IACpB,MAAM,OAAO,UAAU;IACvB,IACE,OAAO,YAAY,KAAA,MAClB,CAAC,KAAK,sBAAsB,KAAK,eAAe,OAAO,UAExD,OAAO;KACL,UAAU;KACV,SAAS,CAAC,8DAA8D;KACxE,mBAAmB,CACjB;MACE,MAAM;MACN,QAAQ;MACR,QAAQ;OAAE;OAAM,WAAW,OAAO;MAAQ;KAC5C,CACF;IACF;IAEF,OAAO,KAAK,OAAO,OAAO;GAC5B;EACF;EACA;EACA;EACA;EACA,WAAW;EACX,MAAM,OAAO;EACb,gBAAgB,OAAO,kBAAkB;EACzC,mBAAmB,KAAK;EACxB,WAAW,KAAK;EAChB,eAAe,KAAK;EACpB,aAAa,KAAK,eAAe;EACjC,OAAO;CACT,CAAC;CACD,MAAM,WAAW,kBACb;EACE,UAAU,8BAA8B;GACtC,UAAU,WAAW;GACrB,SAAS;GACT,SAAS;EACX,CAAC;EACD,QAAQ,8BAA8B;GACpC,UAAU,WAAW;GACrB,SAAS,SAAS;GAClB,SAAS;EACX,CAAC;CACH,IACA,KAAA;CACJ,MAAM,WAAW,WAAW,YAAY;CACxC,MAAM,SAAS,WACX,OACA,uBACE,WAAW,mBACX,WAAW,iBACX,KAAK,QAAQ,kBAAkB,KAAK,OAAO,OAAO,IAAI,KAAA,CACxD;CACJ,MAAM,WAAW,QAAQ,YAAY;CACrC,MAAM,SAAS,QAAQ,UAAU;CACjC,MAAM,OAAO,YAAY,SAAS,OAAO,gBAAgB,SAAS,gBAAgB,KAAA;CAClF,MAAM,UAAU,WACZ,KAAA,IACA,MAAM,YAAY;EAChB,MAAM,CACJ,GAAG,kBACD,WAAW,kBAAkB,OAC7B,YACA,QACA,iBACA,WACA,KAAK,KACP,GACA,GAAI,WAAW,oBAAoB,WAAW,oBAC1C,CAAC,IACD,kBACE,WAAW,gBAAgB,OAC3B,UACA,QACA,SAAS,eACT,WACA,KAAK,KACP,CACN;EACA,qBAAqB;EACrB,6BAA6B,KAAK,QAC9B,kBAAkB,KAAK,OAAO,OAAO,IACrC,KAAA;EACJ,mBAAmB,KAAK,OAAO,iBAAiB;EAChD,GAAI,WAAW,oBAAoB,WAAW,oBAC1C,CAAC,IACD,EAAE,sBAAsB,SAAS;CACvC,CAAC;CACL,MAAM,OAAO,UAAU;CACvB,MAAM,eAAe;EACnB,MAAM,KAAK,OAAO;EAClB,MAAM;EACN,GAAI,SAAS,eAAe,KAAA,IAAY,CAAC,IAAI,EAAE,YAAY,SAAS,WAAW;EAC/E,GAAI,SAAS,eAAe,KAAA,IAAY,CAAC,IAAI,EAAE,YAAY,SAAS,WAAW;CACjF;CACA,MAAM,aAAa,KAAK,IAAI,IAAI;CAChC,MAAM,WAAW,WAAW,KAAK;CACjC,MAAM,SAAS;EACb,QAAQ;EACR,OAAO,GAAG,OAAO,GAAG;EACpB;EACA,WAAW,IAAI,KAAK,SAAS,CAAC,CAAC,YAAY;EAC3C,qBAAqB,mBAAmB,eAAe;EACvD,mBAAmB,mBAAmB,SAAS,aAAa;EAC5D,MAAM,WAAW;EACjB,oBAAoB;EACpB,GAAI,KAAK,QAAQ,EAAE,OAAO,KAAK,MAAM,IAAI,CAAC;EAC1C,UAAU;GACR,kBAAkB,oBAAoB,OAAO,OAAO,QAAQ,CAAC;GAC7D,sBAAsB,oBAAoB,WAAW,OAAO,QAAQ,CAAC;GACrE,oBAAoB,oBAAoB,SAAS,OAAO,QAAQ,CAAC;GACjE,wBAAwB,0BAA0B,WAAW,iBAAiB;GAC9E,sBAAsB,0BAA0B,WAAW,eAAe;GAC1E,oBAAoB,gBAAgB,QAAQ;GAC5C,GAAI,SAAS,EAAE,qBAAqB,OAAO,aAAa,IAAI,CAAC;GAC7D,GAAI,WAAW,uBACX,EAAE,2BAA2B,0BAA0B,WAAW,oBAAoB,EAAE,IACxF,CAAC;EACP;EACA,MAAM,WAAW;EACjB,GAAI,WAAW,EAAE,SAAS,WAAoB,IAAI,CAAC;EACnD,GAAI,YAAY,SACZ;GACE,0BAA0B,SAAS;GACnC,wBAAwB,OAAO;GAC/B,aAAa;EACf,IACA,CAAC;EACL;EACA,cAAc,KAAK;EACnB,iBAAiB;CACnB;CAEA,MAAM,aAAgE,KAAK,MACzE,KAAK,UAAU,MAAM,CACvB;CACA,MAAM,aAA0C;EAC9C,GAAG;EACH,cAAc,gBAAgB,UAAU;CAC1C;CACA,QAAQ,UAAU,MAAM;CACxB,QAAQ,MAAM,GAAG,OAAO,0BAA0B,KAAK,UAAU,YAAY,MAAM,CAAC,CAAC;CACrF,KAAK,eAAe,UAAU;CAC9B,KAAK,aAAa;EAChB,MAAM;EACN,UAAU,WAAW,WAAW;EAChC,GAAI,SAAS,KAAA,IAAY,CAAC,IAAI,EAAE,KAAK;CACvC,CAAC;CACD,MAAM,WACJ,KAAK,kBAAkB,QAAQ,WAAW,WAAW,aAAa,SAC9D,WAAW;EACT,QAAQ,WAAW;EACnB,MAAM,WAAW;EACjB,cAAc,WAAW;EACzB,SAAS,KAAK;EACd,QAAQ,KAAK;CACf,CAAC,IACD,KAAA;CACN,MAAM,SAAwD;EAC5D,GAAI,WAAW,EAAE,SAAS,IAAI,CAAC;EAC/B,MAAM;EACN,GAAI,KAAK,QAAQ,EAAE,OAAO,KAAK,MAAM,IAAI,CAAC;EAC1C,GAAI,WAAW,gBAAgB,EAAE,eAAe,WAAW,cAAc,IAAI,CAAC;EAC9E,uBAAuB;EACvB;EACA,QAAQ;GACN,SAAS,SAAS;GAClB,eAAe,QAAQ,iBAAiB;GACxC,aAAa,QAAQ,eAAe,CAAC;GACrC,OAAO,KAAK,OAAO;EACrB;EACA,GAAI,SAAS,KAAA,IAAY,CAAC,IAAI,EAAE,KAAK;EACrC,MAAM,WAAW;EACjB,cAAc,WAAW,WAAW;EACpC;EACA;EACA;EACA,YAAY,WAAW,QAAQ;EAC/B,cAAc,KAAK;EACnB;EACA;EACA,GAAI,UAAU,EAAE,QAAQ,IAAI,CAAC;EAC7B,GAAI,SAAS,gBAAgB,EAAE,eAAe,SAAS,cAAc,IAAI,CAAC;EAC1E,KAAK;GACH,GAAG;GACH,MAAM;GACN;GACA,eAAe,SAAS;GACxB,mBAAmB,YAAY,SAAS,aAAa;GACrD,QAAQ;GACR;GACA,GAAI,WAAW,EAAE,SAAS,IAAI,CAAC;EACjC;CACF;CACA,IAAI,KAAK,cAAc;EACrB,MAAM,YACJ,OACA,SACA,UACA,mBAC+B;GAC/B;GACA,aAAa,YAAY,OAAO;GAChC;GACA,OAAO,SAAS,MAAM,KAAK,SAAS;IAClC,MAAM,YAAY,8BAA8B,IAAI;IACpD,OAAO;KACL,YAAY,KAAK;KACjB,KAAK,KAAK;KACV,eAAe,sBAAsB,IAAI,KAAK;KAC9C,YAAY,4BAA4B,IAAI;KAC5C,iBAAiB,UAAU;KAC3B,qBAAqB,UAAU,uBAAuB;KACtD,GAAI,KAAK,QAAQ,EAAE,cAAc,KAAK,MAAM,IAAI,CAAC;IACnD;GACF,CAAC;GACD;GACA,SAAS,SAAS,WAAW,KAAK;GAClC,YAAY,SAAS;EACvB;EACA,IAAI;GACF,MAAM,mBAAmB,KAAK,YAAY,CAAC,CAAC,cAAc;IACxD,OAAO,WAAW;IAClB;IACA,WAAW,WAAW;IACtB,QAAQ;IACR,QAAQ;KAAE,GAAG,KAAK;KAAc,MAAM;IAAS;IAC/C,UAAU,SACR,GACA,iBACA,WAAW,mBACX,UAAU,iBAAiB,IAC7B;IACA,aAAa,CACX,SACE,GACA,SAAS,eACT,WAAW,iBACX,QAAQ,iBAAiB,IAC3B,CACF;IACA,cAAc,OAAO;IACrB,GAAI,SAAS,KAAA,IAAY,CAAC,IAAI,EAAE,aAAa,KAAK;IAClD,cAAc,OAAO;IACrB,iBAAiB;IACjB,GAAI,UAAU,EAAE,eAAe,QAAQ,IAAI,CAAC;GAC9C,CAAC;EACH,SAAS,OAAO;GACd,QAAQ,KACN,mDAAmD,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK,GAC1G;EACF;CACF;CACA,OAAO;AACT;;;;ACJA,IAAa,sBAAb,cAAyC,MAAM;CAC7C;CACA;CAEA,YAAY,OAAgB,QAA0B;EACpD,MAAM,WAAW,iBAAiB,QAAQ,QAAQ,IAAI,MAAM,OAAO,KAAK,CAAC;EACzE,MAAM,SAAS,SAAS,EAAE,OAAO,SAAS,CAAC;EAC3C,KAAK,OAAO;EACZ,KAAK,OAAO,OAAO,QAAQ;EAC3B,KAAK,WAAW,OAAO,KAAK;CAC9B;AACF;AAEA,SAAS,4BACP,MACM;CACN,oCAAoC,IAAI;CACxC,IACE,CAAC,KAAK,WACL,KAAK,wBAAwB,KAAA,KAC5B,KAAK,8BAA8B,KAAA,KACnC,KAAK,aAAa,KAAA,IAEpB,MAAM,IAAI,MAAM,uDAAuD;CACzE,IAAI,KAAK,UAAU,KAAK,UACtB,MAAM,IAAI,MAAM,yDAAyD;CAE3E,IAAI,CAAC,KAAK,QAAQ;EAChB,IAAI,KAAK,uBAAuB,KAAA,GAC9B,MAAM,IAAI,MAAM,iDAAiD;EAEnE;CACF;CACA,IACE,OAAO,KAAK,OAAO,SAAS,YAC5B,CAAC,KAAK,OAAO,KAAK,KAAK,KACvB,KAAK,OAAO,KAAK,KAAK,MAAM,KAAK,OAAO,QACxC,OAAO,KAAK,OAAO,aAAa,YAEhC,MAAM,IAAI,MAAM,kEAAkE;CAEpF,MAAM,SAAS,KAAK;CACpB,IAAI,QAAQ,gBAAgB,KAAA,GAC1B,MAAM,IAAI,MACR,uFACF;CAEF,IAAI,QAAQ,mBAAmB,KAAA,GAC7B,MAAM,IAAI,MACR,8FACF;CAEF,IACE,QAAQ,yBAAyB,KAAA,KACjC,QAAQ,wBAAwB,KAAA,KAChC,KAAK,sBAAsB,KAAA,KAC3B,KAAK,aAAa,KAAA,KAClB,KAAK,iBAAiB,KAAA,KACtB,KAAK,wBAAwB,KAAA,KAC7B,KAAK,qBAAqB,KAAA,KAC1B,KAAK,iBAAiB,KAAA,GAEtB,MAAM,IAAI,MACR,wLACF;AAEJ;AAEA,SAAS,sBACP,iBACA,mBACA,UACA,gBACgD;CAChD,IAAI,CAAC,OAAO,SAAS,QAAQ,KAAK,YAAY,KAAK,YAAY,GAC7D,MAAM,IAAI,MAAM,yDAAyD;CAE3E,MAAM,uBAAO,IAAI,IAAuB;CACxC,KAAK,MAAM,YAAY,iBAAiB;EACtC,IAAI,KAAK,IAAI,SAAS,EAAE,GACtB,MAAM,IAAI,MAAM,uCAAuC,SAAS,GAAG,EAAE;EAEvE,KAAK,IAAI,SAAS,IAAI,QAAQ;CAChC;CACA,IAAI,mBAAmB;EACrB,IAAI,kBAAkB,WAAW,GAC/B,MAAM,IAAI,MAAM,mDAAmD;EAErE,MAAM,+BAAe,IAAI,IAAY;EACrC,KAAK,MAAM,YAAY,mBAAmB;GACxC,IAAI,CAAC,KAAK,IAAI,SAAS,EAAE,GACvB,MAAM,IAAI,MACR,oCAAoC,SAAS,GAAG,qCAClD;GAEF,IAAI,aAAa,IAAI,SAAS,EAAE,GAC9B,MAAM,IAAI,MAAM,iDAAiD,SAAS,GAAG,EAAE;GAEjF,aAAa,IAAI,SAAS,EAAE;EAC9B;EACA,MAAM,QAAQ,gBAAgB,QAAQ,aAAa,CAAC,aAAa,IAAI,SAAS,EAAE,CAAC;EACjF,IAAI,MAAM,WAAW,GACnB,MAAM,IAAI,MAAM,0CAA0C;EAE5D,IAAI,gBAAgB;GAClB,MAAM,gBAAgB,IAAI,IAAI,CAAC,GAAG,YAAY,CAAC,CAAC,KAAK,OAAO,eAAe,IAAI,EAAE,CAAC,CAAC;GACnF,IAAI,MAAM,MAAM,aAAa,cAAc,IAAI,eAAe,IAAI,SAAS,EAAE,CAAC,CAAC,GAC7E,MAAM,IAAI,MAAM,6DAA6D;EAEjF;EACA,OAAO;GACL;GACA,WAAW,kBAAkB,KAAK,aAAa,KAAK,IAAI,SAAS,EAAE,CAAE;EACvE;CACF;CACA,IAAI,gBAAgB,SAAS,GAC3B,MAAM,IAAI,MAAM,+DAA+D;CAEjF,IAAI,gBAAgB;EAClB,MAAM,QAAQ,kBAAkB,iBAAiB,UAAU,cAAc;EACzE,OAAO;GAAE,OAAO,MAAM;GAAO,WAAW,MAAM;EAAQ;CACxD;CACA,MAAM,SAAS,CAAC,GAAG,eAAe,CAAC,CAAC,MACjC,GAAG,MAAM,mBAAmB,EAAE,EAAE,IAAI,mBAAmB,EAAE,EAAE,CAC9D;CACA,MAAM,QAAQ,KAAK,IAAI,GAAG,KAAK,IAAI,OAAO,SAAS,GAAG,KAAK,MAAM,OAAO,SAAS,QAAQ,CAAC,CAAC;CAC3F,OAAO;EACL,WAAW,OAAO,MAAM,GAAG,KAAK;EAChC,OAAO,OAAO,MAAM,KAAK;CAC3B;AACF;;;;;;;;AASA,SAAS,mBAAmB,OAAuB;CACjD,IAAI,OAAO;CACX,KAAK,IAAI,QAAQ,GAAG,QAAQ,MAAM,QAAQ,SAAS;EACjD,QAAQ,MAAM,WAAW,KAAK;EAC9B,OAAO,KAAK,KAAK,MAAM,QAAQ,MAAM;CACvC;CACA,OAAO;AACT;;AAGA,SAAS,kBACP,WACA,UACA,gBAC8C;CAC9C,IAAI,gBAAgB;EAClB,MAAM,QAAQ,CAAC,GAAG,IAAI,IAAI,UAAU,KAAK,aAAa,eAAe,IAAI,SAAS,EAAE,CAAE,CAAC,CAAC,CAAC,CAAC,MACvF,GAAG,MAAM,mBAAmB,CAAC,IAAI,mBAAmB,CAAC,MAAM,IAAI,IAAI,KAAK,IAAI,IAAI,IAAI,EACvF;EACA,IAAI,MAAM,SAAS,GACjB,MAAM,IAAI,MAAM,6DAA6D;EAC/E,MAAM,QAAQ,KAAK,IAAI,GAAG,KAAK,IAAI,MAAM,SAAS,GAAG,KAAK,MAAM,MAAM,SAAS,QAAQ,CAAC,CAAC;EACzF,MAAM,aAAa,IAAI,IAAI,MAAM,MAAM,GAAG,KAAK,CAAC;EAChD,OAAO;GACL,SAAS,UAAU,QAAQ,aAAa,WAAW,IAAI,eAAe,IAAI,SAAS,EAAE,CAAE,CAAC;GACxF,OAAO,UAAU,QAAQ,aAAa,CAAC,WAAW,IAAI,eAAe,IAAI,SAAS,EAAE,CAAE,CAAC;EACzF;CACF;CACA,MAAM,SAAS,CAAC,GAAG,SAAS,CAAC,CAAC,MAAM,GAAG,MAAM,mBAAmB,EAAE,EAAE,IAAI,mBAAmB,EAAE,EAAE,CAAC;CAChG,MAAM,WAAW,KAAK,IAAI,GAAG,KAAK,IAAI,OAAO,SAAS,GAAG,KAAK,MAAM,OAAO,SAAS,QAAQ,CAAC,CAAC;CAC9F,OAAO;EACL,SAAS,OAAO,MAAM,GAAG,QAAQ;EACjC,OAAO,OAAO,MAAM,QAAQ;CAC9B;AACF;;;;;;AAOA,SAAS,qBACP,QACoE;CACpE,KAAK,IAAI,IAAI,OAAO,YAAY,SAAS,GAAG,KAAK,GAAG,KAAK;EACvD,MAAM,WAAW,OAAO,YAAY,EAAE,EAAE,SAAS,MAC9C,MAAM,EAAE,gBAAgB,OAAO,iBAClC;EACA,IAAI,UAAU,OAAO,SAAS;CAChC;CACA,OAAO,OAAO;AAChB;AAqCA,eAAsB,YACpB,MAIkD;CAClD,OAAO;EACL,GAAG;EACH,OAAO,aAAa,KAAK,KAAK;EAC9B,GAAI,KAAK,SAAS,KAAK,gBACnB;GACE,OAAO,KAAK,SAAS,sBAAsB,KAAK,KAAK;GACrD,eAAe,KAAK,iBAAiB,2BAA2B,KAAK,aAAa;GAClF,WAAW,gBAAgB,KAAK,SAAS;GACzC,oBAAoB,KAAK,sBAAsB,gBAAgB,KAAK,kBAAkB;GACtF,QAAQ,KAAK,UAAU,gBAAgB,KAAK,MAAM;EACpD,IACA,CAAC;CACP;CACA,MAAM,YAAY,KAAK,IAAI;CAI3B,MAAM,SAAS,cAFb,KAAK,WACJ,KAAK,SAAS,iCAAiC,cAAc,qBAAqB,YACzC;CAC5C,MAAM,UACJ,KAAK,YAAY,OAAO,WAAW,QAAQ,IAAI,wBAAwB,IAAI,kBAAkB;CAC/F,MAAM,aAAa,oBAAoB;EACrC;EACA;EACA,gBAAgB,KAAK,QAAQ;CAC/B,CAAC;CACD,IAAI;EACF,OAAO,MAAM,eACX,MACA,YACA,WACA,QACA,OACF;CACF,SAAS,OAAO;EACd,MAAM,IAAI,oBAAoB,OAAO,UAAU;CACjD;AACF;AAEA,eAAe,eACb,MACA,YACA,WACA,QACA,SACkD;CAClD,MAAM,SAAS,KAAK,UAAU,CAAC;CAC/B,4BAA4B,IAAI;CAChC,MAAM,iBAAiB,OAAO,kBAAkB;CAChD,MAAM,kBAAkB,OAAO,mBAAmB;CAClD,MAAM,cAAc,OAAO,WAAW;CACtC,MAAM,kBAAkB,gBAAgB;CACxC,MAAM,cAAc,KAAK,eAAe;CACxC,MAAM,iBAAiB,KAAK,QAAQ,kBAAkB,KAAK,OAAO,KAAK,SAAS,IAAI,KAAA;CACpF,MAAM,sBACJ,KAAK,OAAO,mBAAmB,cAAc,iBAAiB,KAAA;CAMhE,MAAM,kBAAkB,OAAO;CAC/B,MAAM,EAAE,OAAO,YAAY,kBACvB;EACE,OAAO,KAAK,UAAU,QAAQ,MAAM,CAAC,gBAAgB,MAAM,MAAM,EAAE,OAAO,EAAE,EAAE,CAAC;EAC/E,SAAS;CACX,IACA,kBACE;EAAE,OAAO,KAAK;EAAW,SAAS,CAAC;CAAiB,IACpD,kBAAkB,KAAK,WAAW,iBAAiB,mBAAmB;CAE5E,IAAI,MAAM,WAAW,GACnB,MAAM,IAAI,MACR,mFACF;CAEF,IAAI,QAAQ,WAAW,KAAK,CAAC,iBAC3B,MAAM,IAAI,MAAM,2DAA2D;CAG7E,IAAI,KAAK,OAAO,mBAAmB,aACjC,iCAAiC,KAAK,OAAO,SAAS,KAAK;CAE7D,IAAI,KAAK,eAAe;EACtB,IAAI,iBAAiB,MAAM,IAAI,MAAM,0CAA0C;EAC/E,IAAI,KAAK,QAAQ,MAAM,qBAAqB,KAAK,eAAe,KAAK,OAAO,SAAS,KAAK;CAC5F;CAEA,IAAI,KAAK,QAAQ;EACf,MAAM,aAAa,sBACjB,OACA,KAAK,oBACL,OAAO,qBAAqB,KAC5B,mBACF;EACA,OAAO,qBAAqB;GAC1B,MAAM;IAAE,GAAG;IAAM,QAAQ,KAAK;GAAO;GACrC,OAAO,WAAW;GAClB,WAAW,WAAW;GACtB;GACA;GACA;GACA;GACA;EACF,CAAC;CACH;CACA,MAAM,cAAc,OAAO,eAAe;CAC1C,MAAM,iBAAiB,OAAO,kBAAkB;CAChD,IAAI,cAAc,KAAK,CAAC,KAAK,UAC3B,MAAM,IAAI,MACR,0FACF;CAEF,MAAM,WAA6C,KAAK,YAAY;EAClE,MAAM;EACN,SAAS,YAAY,CAAC;CACxB;CAEA,MAAM,OACJ,KAAK,QACL,sBAA4C;EAC1C,kBAAkB;EAClB,gBAAgB,KAAK,OAAO,iBAAiB;EAC7C,6BAA6B,KAAK,QAAQ,kBAAkB,KAAK,OAAO,OAAO,IAAI,KAAA;CACrF,CAAC;CAEH,IAAI,KAAK,YACP,KAAK,WAAW;EAAE,MAAM;EAAoB,WAAW,KAAK,UAAU;CAAO,CAAC;CAGhF,MAAM,SAAS,MAAM,mBAAyC;EAC5D,WAAW;EACX,iBAAiB,KAAK;EACtB,qBAAqB,KAAK;EAC1B,qBAAqB,KAAK;EAC1B,aAAa,KAAK;EAClB;EACA,QAAQ,CAAC,KAAK,KAAK;EACnB;EACA,gBAAgB;EAChB,sBAAsB,OAAO;EAC7B,MAAM,OAAO;EACb,qBAAqB,OAAO;EAC5B,kBAAkB;EAClB,OAAO,KAAK;EACZ,eAAe,KAAK;EACpB,SAAS;EACT;EACA,YAAY,KAAK;EACjB,eAAe,KAAK,iBAAiB;EACrC,SAAS,KAAK;EACd,QAAQ,KAAK;EACb;EACA;EACA;EACA,eAAe,KAAK;EACpB,mBAAmB,KAAK;EACxB,WAAW,KAAK;EAChB;EACA;EACA,cAAc,KAAK;EACnB,eAAe,KAAK;EACpB,mBAAmB,KAAK;EACxB,UAAU,KAAK;EACf,kBAAkB,KAAK;EACvB,cAAc,KAAK;EACnB,cAAc,KAAK;CACrB,CAAC;CAKD,MAAM,cAA2B,kBAAkB,WAAW;CAC9D,MAAM,yBAAyB,kBAC3B,OAAO,mBACP,OAAO;CACX,MAAM,uBAAuB,kBACzB,qBAAqB,MAAM,IAC3B,OAAO;CACX,MAAM,gBAAgB,KAAK,QACvB,kBAAkB,KAAK,OAAO,kBAAkB,QAAQ,OAAO,IAC/D,KAAA;CACJ,MAAM,SAAS,uBAAuB,wBAAwB,sBAAsB,aAAa;CACjG,MAAM,WAAW,OAAO;CACxB,MAAM,cAAc,OAAO;CAG3B,IAAI;CACJ,MAAM,4BAA4B,kBAAkB,CAAC,IAAI,OAAO;CAChE,IAAI,0BAA0B,UAAU,GAAG;EAEzC,QAAQ,eAAe;GACrB,oBAAoB;GACpB,gBAAgB,KAAK,OAAO,iBAAiB;GAC7C,qBAAqB;EACvB,CAAC;EACD,IAAI,KAAK,YACP,KAAK,WAAW;GACd,MAAM;GACN,GAAG,MAAM;GACT,IAAI,MAAM;GACV,KAAK,MAAM;GACX,cAAc,MAAM;EACtB,CAAC;EAEH,IAAI,MAAM,gBAAgB,cAAc,GACtC,QAAQ,KAAK,iBAAiB,MAAM,gBAAgB;CAExD;CAEA,IAAI,KAAK,YAAY;EACnB,KAAK,WAAW;GACd,MAAM;GACN,eAAe,SAAS;GACxB,YAAY,KAAK,IAAI,IAAI;EAC3B,CAAC;EACD,KAAK,WAAW;GACd,MAAM;GACN,UAAU,OAAO,WAAW;GAI5B,GAAI,kBAAkB,CAAC,IAAI,EAAE,MAAM,YAAY,gBAAgB,SAAS,cAAc;EACxF,CAAC;CACH;CAEA,MAAM,OAAO,OAAO;CACpB,MAAM,YAAY,KAAK;CAMvB,MAAM,UAAU,MAAM,YAAY;EAChC,MAAM,CACJ,GAAG,kBACD,uBAAuB,OACvB,YACA,QACA,KAAK,iBACL,aACA,KAAK,KACP,GACA,GAAI,yBAAyB,yBACzB,CAAC,IACD,kBACE,qBAAqB,OACrB,UACA,QACA,OAAO,eACP,aACA,KAAK,KACP,CACN;EACA,qBAAqB;EACrB,6BAA6B;EAC7B,mBAAmB,KAAK,OAAO,iBAAiB;EAChD,GAAI,yBAAyB,yBAAyB,CAAC,IAAI,EAAE,sBAAsB,SAAS;CAC9F,CAAC;CAID,MAAM,aAAa,KAAK,IAAI,IAAI;CAChC,MAAM,EAAE,QAAQ,eAAe,MAAM,mBAAyC;EAC5E,GAAG,6BAA6B;GAC9B,OAAO,GAAG,OAAO,GAAG;GACpB;GACA,WAAW,IAAI,KAAK,SAAS,CAAC,CAAC,YAAY;GAC3C,iBAAiB,KAAK;GACtB;GACA,cAAc,WAAW,KAAK;GAC9B,cAAc;GACd,iBAAiB;GACjB,6BAA6B,kBAAkB,KAAA,IAAY;EAC7D,CAAC;EACD;EACA,cAAc,KAAK,eAAe,mBAAmB,KAAK,YAAY,IAAI,KAAA;CAC5E,CAAC;CACD,IAAI,KAAK,cAAc,KAAK,aAAa,UAAU;CAEnD,MAAM,UAA2D;EAC/D,GAAI,KAAK,QAAQ,EAAE,OAAO,KAAK,MAAM,IAAI,CAAC;EAC1C,GAAI,OAAO,gBAAgB,EAAE,eAAe,OAAO,cAAc,IAAI,CAAC;EACtE,MAAM;EACN;EACA,QAAQ;GACN,GAAG;GACH,SAAS,OAAO;GAChB,GAAI,OAAO,cAAc,EAAE,OAAO,OAAO,YAAY,IAAI,CAAC;GAC1D,GAAI,OAAO,kBAAkB,EAAE,WAAW,OAAO,gBAAgB,IAAI,CAAC;EACxE;EACA,GAAI,kBAAkB,CAAC,IAAI,EAAE,MAAM,YAAY,gBAAgB,SAAS,cAAc;EACtF,MAAM,OAAO;EACb;EACA,cAAc,OAAO,WAAW;EAChC,qBAAqB,OAAO,YAAY;EACxC;EACA,cAAc;EACd;EACA,UAAU,WAAW,KAAK;EAC1B,GAAI,OAAO,gBAAgB,EAAE,eAAe,OAAO,cAAc,IAAI,CAAC;EACtE;EACA,GAAI,QAAQ,EAAE,MAAM,IAAI,CAAC;EACzB,KAAK;CACP;CAIA,IAAI,KAAK,cACP,IAAI;EACF,MAAM,oBAAoB,KAAK,cAAc,MAAM,SAAS,QAAQ,MAAM;CAC5E,SAAS,KAAK;EACZ,MAAM,MAAM,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;EAE3D,QAAQ,KAAK,mDAAmD,KAAK;CACvE;CAGF,OAAO;AACT;AAEA,eAAe,oBACb,QACA,MACA,SACA,KACA,QACe;CACf,MAAM,SAAS,mBAAmB,MAAM;CAExC,SAAS,qBACP,OACA,SACA,UACA,YAC2B;EAC3B,MAAM,QAA4B,SAAS,MAAM,KAAK,SAAS;GAC7D,MAAM,YAAY,8BAA8B,IAAI;GACpD,OAAO;IACL,YAAY,KAAK;IACjB,KAAK,KAAK;IACV,eAAe,sBAAsB,IAAI,KAAK;IAC9C,YAAY,4BAA4B,IAAI;IAC5C,iBAAiB,UAAU;IAC3B,qBAAqB,UAAU,uBAAuB;IACtD,cAAc,KAAK,SAAS,KAAA;GAC9B;EACF,CAAC;EACD,MAAM,cAAc,MAAM,SAAS,SACjC,KAAK,kBAAkB,OAAO,CAAC,IAAI,CAAC,KAAK,aAAa,CACxD;EACA,MAAM,gBACJ,YAAY,WAAW,IACnB,OACA,YAAY,QAAQ,KAAK,UAAU,MAAM,OAAO,CAAC,IAAI,YAAY;EACvE,OAAO;GACL;GACA,aAAa,YAAY,OAAO;GAChC;GACA;GACA;GACA,SAAS,SAAS,WAAW,KAAK;GAClC;EACF;CACF;CAEA,MAAM,cAA2C,CAAC;CAElD,YAAY,KAAK,qBAAqB,GAAG,KAAK,iBAAiB,IAAI,kBAAkB,CAAC,CAAC;CAGvF,KAAK,MAAM,OAAO,IAAI,aAAa;EACjC,MAAM,SAAS,IAAI,SAAS,QACzB,MAAM,MACL,EAAE,SAAS,WAAW,gBAAgB,MACrC,SAAS,KAAA,KAAa,iBAAiB,EAAE,QAAQ,IAAI,iBAAiB,KAAK,QAAQ,KAChF,IACA,MACN,IAAI,SAAS,EACf;EACA,IAAI,CAAC,QAAQ;EACb,YAAY,KACV,qBAAqB,IAAI,OAAO,kBAAkB,GAAG,OAAO,SAAS,OAAO,UAAU,CAAC,CACzF;CACF;CAEA,MAAM,QAAsB;EAC1B,OAAO,GAAG,OAAO,GAAG,KAAK,IAAI;EAC7B;EACA,4BAAW,IAAI,KAAK,EAAA,CAAE,YAAY;EAClC,QAAQ;EACR,QAAQ,KAAK,gBAAgB,CAAC;EAC9B,UAAU,YAAY;EACtB;EACA,cAAc,QAAQ;EACtB,aAAa,QAAQ;EACrB,cAAc,QAAQ;EACtB,iBAAiB,QAAQ;EACzB,eAAe,QAAQ;CACzB;CAEA,MAAM,OAAO,cAAc,KAAK;AAClC;AAEA,SAAS,iBACP,UACQ;CACR,MAAM,OAAO,OAAO,OAAO,SAAS,WAAW,UAAU;CACzD,OAAO,KAAK,WAAW,IAAI,IAAI,KAAK,QAAQ,GAAG,MAAM,IAAI,EAAE,eAAe,CAAC,IAAI,KAAK;AACtF;;;;;;;;;;AC1+BA,SAAgB,gBACd,UACwC;CACxC,MAAM,yBAAyB,iBAAiB,QAAQ;CAExD,OAAO;EACL,WAAW,SAAS;EACpB,iBAAiB,SAAS;EAE1B,MAAM,SAAS,OAAO,CAAC,GAAG;GACxB,MAAM,EAAE,OAAO,OAAO,QAAQ,QAAQ,WAAW,SAAS,GAAG,iBAAiB;GAC9E,MAAM,gBAAgB,SAAS,SAAS;GACxC,MAAM,kBAAkB,WAAW,SAAS;GAC5C,MAAM,iBAAiB,UAAU,SAAS,UAAU,yBAAyB,KAAK,IAAI;GACtF,MAAM,kBACJ,aAAa,WACb,uBAAuB,YACtB,eAAe,WAAW,QAAQ,IAAI,wBAAwB,IAAI,KAAA;GACrE,MAAM,cAAoD;IACxD,GAAG;IACH,GAAG;IACH,GAAI,kBAAkB,EAAE,SAAS,gBAAgB,IAAI,CAAC;IACtD,QAAQ;IACR,WAAW,aAAa,SAAS;IACjC,WAAW,UAAU,QAAQ,cAAc,iBAAiB,UAAU,GAAG;IACzE,aAAa,mBACX,iBACA,aAAa,eAAe,SAAS,WACvC;IACA,QAAQ,eAAe,QAAQ,SAAS,SAAS,KAAK;GACxD;GACA,IAAI,YAAY,SAAS,KAAA,GACvB,YAAY,OAAO,uBAAuB,YAAY,MAAM,MAAM;GACpE,OAAO,QAA8B,WAAW;EAClD;EAEA,MAAM,QAAQ,OAAO,CAAC,GAAG;GACvB,MAAM,EACJ,QAAQ,gBACR,cAAc,sBACd,GAAG,sBACD;GACJ,MAAM,SAAS,aAAa,UAAU,iBAAiB;GACvD,MAAM,SAAS,YAAY,SAAS,QAAQ,cAAc;GAC1D,MAAM,eAAe,kBAAkB,SAAS,cAAc,oBAAoB;GAClF,OAAO,YAAkC;IACvC,GAAG;IACH,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC;IAC3B,GAAI,eAAe,EAAE,aAAa,IAAI,CAAC;GACzC,CAAC;EACH;CACF;AACF;AAOA,SAAS,iBACP,UAC8C;CAC9C,MAAM,MAAoD,CAAC;CAC3D,IAAI,SAAS,SAAS,IAAI,UAAU,SAAS;CAC7C,IAAI,SAAS,cAAc,IAAI,eAAe,SAAS;CACvD,IAAI,SAAS,eAAe,IAAI,gBAAgB,SAAS;CACzD,IAAI,SAAS,eAAe,IAAI,gBAAgB,SAAS;CACzD,IAAI,SAAS,aAAa,IAAI,cAAc,SAAS;CACrD,IAAI,SAAS,QAAQ,YAAY,KAAA,GAAW,IAAI,cAAc,SAAS,OAAO;CAC9E,IAAI,SAAS,QAAQ,mBAAmB,KAAA,GACtC,IAAI,iBAAiB,SAAS,OAAO;CACvC,IAAI,SAAS,QAAQ,SAAS,KAAA,GAC5B,IAAI,OAAO,uBAAuB,SAAS,OAAO,MAAM,aAAa;CACvE,OAAO;AACT;AAEA,SAAS,YACP,UACA,WAC+B;CAC/B,MAAM,SAAS,oBAAoB,UAAU,SAAS;CACtD,IAAI,QAAQ,SAAS,KAAA,GAAW,OAAO,OAAO,uBAAuB,OAAO,MAAM,aAAa;CAC/F,OAAO;AACT;AAEA,SAAS,kBACP,UACA,WAC0B;CAC1B,MAAM,SAAS,oBAAoB,UAAU,SAAS;CACtD,IAAI,CAAC,QAAQ,OAAO,KAAA;CACpB,IAAI,CAAC,OAAO,UAAU,KAAK,KAAK,CAAC,OAAO,QAAQ,KAAK,KAAK,CAAC,OAAO,UAAU,KAAK,GAC/E,MAAM,IAAI,MACR,oHACF;CAEF,OAAO;AACT;AAEA,SAAS,aAA+B,UAAa,WAAsC;CACzF,IAAI,CAAC,WAAW,OAAO;CACvB,MAAM,SAAS,EAAE,GAAG,SAAS;CAC7B,KAAK,MAAM,CAAC,KAAK,UAAU,OAAO,QAAQ,SAAS,GACjD,IAAI,UAAU,KAAA,GAAW,OAAO,OAAO;CAEzC,OAAO;AACT;AAEA,SAAS,oBACP,UACA,WACe;CACf,IAAI,CAAC,YAAY,CAAC,WAAW,OAAO,KAAA;CACpC,OAAO,aAAa,YAAa,CAAC,GAAS,SAAS;AACtD;AAEA,SAAS,eACP,QACA,cACqC;CACrC,IAAI,WAAW,KAAA,GAAW;EACxB,IAAI,OAAO,WAAW,GACpB,MAAM,IAAI,MAAM,oDAAoD;EAEtE,OAAO;CACT;CACA,OAAO,CAAC,YAAY;AACtB;AAEA,SAAS,uBAAuB,OAAe,OAAuB;CACpE,IAAI,CAAC,OAAO,UAAU,KAAK,KAAK,QAAQ,GACtC,MAAM,IAAI,MAAM,oBAAoB,MAAM,4BAA4B;CAExE,OAAO;AACT"}

package/dist/define-agent-eval-DblGedQk.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"define-agent-eval-DblGedQk.d.ts","names":[],"sources":["../src/campaign/presets/run-final-comparison.ts","../src/contract/self-improve-method.ts","../src/contract/self-improve.ts","../src/contract/define-agent-eval.ts"],"mappings":";;;;;;;;;UAaiB,uBAAuB,kBAAkB,UAAU,mBAC1D,KAAK,mBAAmB,WAAW;EAC3C,iBAAiB;EACjB,eAAe;EACf,sBACE,SAAS,gBACT,UAAU,WACV,KAAK,WAAW,mBAAmB,WAAW,+BAC3C,QAAQ;EACb,MAAM,KAAK,WAAW;EACtB,QAAQ;EACR,gBAAgB;EAChB;EACA;EACA,cAAc,QAAQ,gBAAgB,UAAU,mBAAmB;;;iBAI/C,mBAAmB,kBAAkB,UAAU,WACnE,MAAM,uBAAuB,WAAW,aAAU;;;;;;;;;;;;kBAA5C;;;;;;;;;;;UCKS;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,oBAAoB,YAAY,0BAA0B;EAC1D,QAAQ;EACR;IACE;IACA;IACA;IACA;IACA;IACA;IACA;IACA,sBAAsB;;EAExB,MAAM,QAAQ,kBAAkB;EAChC;EACA;EACA;EACA;EACA,MAAM;EACN;EACA;;UAGe,wBAAwB,kBAAkB,UAAU,mBAC3D,KACN,0BAA0B,WAAW;EAWvC;IAAa,UAAU;IAAiB,QAAQ;;EAChD,uBAAuB;EACvB;;EAEA,UAAU,0BAA0B,WAAW;EAC/C,QAAQ,KAAK,0BAA0B,WAAW;IAChD;;EAEF,YAAY;EACZ,cAAc,YAAY,0BAA0B,WAAW;;EAE/D,MAAM;;EAEN,YAAY;EACZ,UAAU,0BAA0B,WAAW;EAC/C,KAAK,QAAQ,kBAAkB,mBAAmB,WAAW;IAC3D;IACA,iBAAiB,0BAA0B,WAAW;IACtD,eAAe,0BAA0B,WAAW;IACpD;IACA,QAAQ;IACR,MAAM;IACN,WAAW,kBAAkB;;;;;UCnBhB;;;EAGf;;;;EAIA;;EAEA;;EAEA;;;EAGA;;;EAGA;;;;EAIA;;EAEA,mBAAmB;;;;;;;;;EASnB;;EAEA;;;;;EAKA;;KAGU;EACN;EAA0B;;EAC1B;EAA4B;EAAuB;;EACnD;EAA4B;EAAe;;EAC3C;EAA8B;EAAe;EAAuB;;EAGpE;EAAsB;EAAkB;;EACxC;EAAyB;EAAW;EAAY;EAAa;;UAElD,mBAAmB,kBAAkB,UAAU,mBACtD;;;;;;;;;;;;;;EAcR,QAAQ,SAAS,gBAAgB,UAAU,WAAW,KAAK,oBAAoB,QAAQ;;;;;;;EAQvF;;EAEA;;;EAIA,WAAW;;EAGX,OAAO,YAAY,WAAW;;;EAI9B,iBAAiB;;EAGjB,SAAS;;;;;;;;;;;;;EAcT,sBAAsB,gCAAgC,WAAW;;;;;EAMjE,WAAW,gBAAgB;;;;;;EAO3B,SAAS,mBAAmB,WAAW;;;EAIvC,qBAAqB;;;EAIrB,OAAO,KAAK,WAAW;;;;;;;;EASvB,cAAc,eAAe,gBAAgB,iBAAiB,mBAAmB;;;;EAKjF,UAAU;;;;EAKV;;;EAIA,gBAAgB,QAAQ,uBAAuB;;;;;EAM/C,iBAAiB;IACf,UAAU;IACV;IACA;;;;EAKF;;;;;;;EAQA,YAAY;;;EAIZ,cAAc,OAAO;;;;EAKrB;EACA;EACA;;;;;;;;;;;;;;EAeA,eAAe;;;EAIf,eAAe;;;;EAKf,eAAe;;EAGf;;;;;;;;EASA;;;;;;;;;EAUA,oBAAoB,uBAAuB,WAAW;;;EAItD,WAAW;;;;;;;EAQX,mBAAmB,uBAAuB,WAAW;;;;;;EAOrD,eAAe,uBAAuB,WAAW;;;;EAKjD,eAAe,uBAAuB,WAAW;;EAEjD,WAAW;;EAEX,QAAQ;;EAER,gBAAgB;;UAGD,0BAA0B,kBAAkB,UAAU;EACrE;EACA,QAAQ;EACR,gBAAgB;;;;EAIhB;IACE;IACA,aAAa;;;;;EAKf;IACE;IACA,aAAa;IACb,SAAS;;;IAGT;;;;IAIA;;;;;;EAMF;;;EAGA;;;;EAIA,YAAY;;EAEZ;;;EAGA;;EAEA;;EAEA;;EAEA,MAAM;;;EAGN,UAAU;;;EAGV,gBAAgB;;EAEhB;IACE;IACA,MAAM;IACN;IACA,aAAa;;;;;;;;EAQf,SAAS;;;;EAIT,QAAQ;;;;;;EAMR,KAAK,yBAAyB,WAAW;;KAG/B,kBAAkB,kBAAkB,UAAU,aACtD,0BAA0B,WAAW,aACrC,wBAAwB,WAAW;KAE3B,yBAAyB,kBAAkB,UAAU,aAAa,KAC5E,mBAAmB,WAAW;EAG9B,QAAQ,mBAAmB,WAAW;EACtC;EACA,gBAAgB,QAAQ;;KAGd,2BAA2B,kBAAkB,UAAU,aAAa,KAC9E,mBAAmB,WAAW;EAG9B;EACA,gBAAgB,QAAQ;;;cAIb,4BAA4B;WAC9B,MAAM;WACN,UAAU;EAEnB,YAAY,gBAAgB,QAAQ;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAsNtB,YAAY,kBAAkB,UAAU,WACtD,MAAM,yBAAyB,WAAW,aACzC,QAAQ,wBAAwB,WAAW;iBAC9B,YAAY,kBAAkB,UAAU,WACtD,MAAM,2BAA2B,WAAW,aAC3C,QAAQ,0BAA0B,WAAW;iBAChC,YAAY,kBAAkB,UAAU,WACtD,MAAM,mBAAmB,WAAW,aACnC,QAAQ,kBAAkB,WAAW;;;KC/oB5B,eAAe,kBAAkB,UAAU,cACrD,SAAS,gBACT,UAAU,WACV,KAAK,oBACF,QAAQ;KAED,uBAAuB,kBAAkB,UAAU,aAAa,mBAC1E,WACA;UAGe,yBAAyB,kBAAkB,UAAU,mBAC5D,KACN,eAAe,WAAW;;EAI5B,YAAY;;EAEZ,UAAU;;EAEV,QAAQ,eAAe,WAAW;;EAElC,QAAQ,YAAY,WAAW;;EAE/B,SAAS,YAAY,WAAW;;EAEhC;;KAGU,wBAAwB,kBAAkB,UAAU,aAAa,KAC3E,QAAQ,mBAAmB,WAAW;EAGtC,SAAS,QAAQ;EACjB,eAAe,QAAQ;;UAGR,iBAAiB,kBAAkB,UAAU;;WAEnD,oBAAoB;;WAEpB,iBAAiB;;;;;EAK1B,SACE,OAAO,yBAAyB,WAAW,aAC1C,QAAQ,eAAe,WAAW;;;;;;;EAOrC,QACE,OAAO,wBAAwB,WAAW,aACzC,QAAQ,kBAAkB,WAAW;;;;;;;;;iBAU1B,gBAAgB,kBAAkB,UAAU,WAC1D,UAAU,uBAAuB,WAAW,aAC3C,iBAAiB,WAAW"}

package/dist/external-optimizer-process-BnSmrkgL.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"external-optimizer-process-BnSmrkgL.js","names":[],"sources":["../src/campaign/single-run-lock.ts","../src/campaign/external-optimizer-callback.ts"],"sourcesContent":["/**\n * Single-run lock for evaluations that share one mutable environment.\n *\n * Two concurrent runs against a shared stateful gym silently corrupt each\n * other: each resets/mutates environment state mid-cell of the other, and\n * every score from both becomes garbage that LOOKS like worker variance\n * (agent-lab R357 burned hours on flip-flopping scores before tracing them\n * to exactly this). The fix is a pid lockfile: refuse to start while a live\n * holder exists, reclaim stale locks whose pid is gone, release only if the\n * lock is still ours.\n *\n * `alsoCheck` exists because independent runners can guard the same shared\n * resource with differently named lockfiles; a runner must respect all of\n * them even though it writes only its own.\n */\n\nimport {\n  type AtomicFileLockUnavailable,\n  probeAtomicFileLock,\n  tryAcquireAtomicFileLock,\n} from '../ledger-core/atomic-file-lock'\n\nexport interface SingleRunLockOptions {\n  /** Lockfile this runner writes (and checks). */\n  readonly lockPath: string\n  /** Other runners' lockfiles guarding the same resource; checked, never written. */\n  readonly alsoCheck?: readonly string[]\n  /** Install a process 'exit' hook that releases the lock. Default true. */\n  readonly releaseOnExit?: boolean\n  /** Owner pid recorded in the lockfile metadata. Default process.pid. */\n  readonly pid?: number\n}\n\nexport interface SingleRunLock {\n  /** Remove the lockfile if this process still owns it. Idempotent. */\n  release(): void\n}\n\nfunction assertAvailable(path: string): void {\n  const unavailable = probeAtomicFileLock({ lockPath: path })\n  if (unavailable) throw unavailableError(path, unavailable)\n}\n\nfunction unavailableError(path: string, unavailable: AtomicFileLockUnavailable): Error {\n  if (unavailable.reason === 'recovery') {\n    return new Error(\n      `single-run lock recovery is already in progress (${path}); refusing a concurrent run`,\n    )\n  }\n  return new Error(\n    `single-run lock held by live pid ${unavailable.holder.pid} (${path}); refusing a concurrent run on the shared resource`,\n  )\n}\n\n/**\n * Acquire the lock or throw naming the live holder. A stale lock (holder pid\n * no longer running) is reclaimed by one contender. An interrupted reclaim\n * leaves a marker that fails closed instead of admitting overlapping runs.\n */\nexport function acquireSingleRunLock(opts: SingleRunLockOptions): SingleRunLock {\n  const pid = opts.pid ?? process.pid\n  if (!Number.isSafeInteger(pid) || pid <= 0)\n    throw new Error('single-run lock pid must be a positive integer')\n  for (const path of opts.alsoCheck ?? []) assertAvailable(path)\n  const acquisition = tryAcquireAtomicFileLock({\n    lockPath: opts.lockPath,\n    pid,\n  })\n  if (!acquisition.acquired) throw unavailableError(opts.lockPath, acquisition)\n  const releaseOnExit = opts.releaseOnExit ?? true\n  let released = false\n  const release = (): void => {\n    if (released) return\n    released = true\n    if (releaseOnExit) process.off('exit', release)\n    try {\n      acquisition.lock.release()\n    } catch {\n      // release is best-effort; a leftover stale lock is reclaimed on next acquire\n    }\n  }\n  try {\n    for (const path of opts.alsoCheck ?? []) assertAvailable(path)\n  } catch (error) {\n    release()\n    throw error\n  }\n  if (releaseOnExit) process.on('exit', release)\n  return { release }\n}\n","import { createServer, type IncomingMessage, type ServerResponse } from 'node:http'\nimport { contentHash } from '../verdict-cache'\nimport {\n  type ExternalOptimizerCallback,\n  type ExternalOptimizerCallbackLimits,\n  type ExternalOptimizerEvaluationObservation,\n  type ExternalTextCandidate,\n  type ExternalTextEvaluationRequest,\n  isExternalTextCandidate,\n  isRecord,\n  resolveExternalOptimizerCallbackLimits,\n} from './external-optimizer-contracts'\nimport {\n  closeServer,\n  listenLocal,\n  sendJsonIfOpen,\n  waitForActiveHandlers,\n} from './external-optimizer-http'\n\ntype UnsequencedObservation = ExternalOptimizerEvaluationObservation extends infer T\n  ? T extends ExternalOptimizerEvaluationObservation\n    ? Omit<T, 'sequence'>\n    : never\n  : never\n\nexport async function startExternalOptimizerCallback<TResponse>(args: {\n  token: string\n  maxEvaluations: number\n  acceptEvaluation?: () => number | undefined\n  evaluate: (request: ExternalTextEvaluationRequest, signal: AbortSignal) => Promise<TResponse>\n  observe?: (observation: ExternalOptimizerEvaluationObservation) => void\n  /** Loopback JSON byte limits. Omitted fields use finite defaults. */\n  limits?: Partial<ExternalOptimizerCallbackLimits>\n  signal?: AbortSignal\n}): Promise<ExternalOptimizerCallback> {\n  assertCallbackConfig(args)\n  const limits = resolveExternalOptimizerCallbackLimits(args.limits)\n  args.signal?.throwIfAborted()\n  let evaluations = 0\n  let accepting = true\n  let closePromise: Promise<void> | undefined\n  const activeControllers = new Set<AbortController>()\n  const activeHandlers = new Set<Promise<void>>()\n  const proposedCandidateHashes = new Set<string>()\n  let observationSequence = 0\n  const server = createServer((request, response) => {\n    if (!accepting) {\n      sendJsonIfOpen(response, 503, { error: 'external optimizer callback is closing' })\n      return\n    }\n\n    const controller = new AbortController()\n    const abortRequest = (): void => {\n      request.destroy()\n      response.destroy()\n    }\n    activeControllers.add(controller)\n    controller.signal.addEventListener('abort', abortRequest, { once: true })\n\n    let handler!: Promise<void>\n    handler = handleCallback(\n      request,\n      response,\n      controller.signal,\n      args,\n      limits,\n      () => {\n        const accepted = args.acceptEvaluation ? args.acceptEvaluation() : evaluations + 1\n        if (accepted === undefined) return undefined\n        if (!Number.isSafeInteger(accepted) || accepted <= 0) {\n          throw new Error('external optimizer callback: invalid accepted evaluation count')\n        }\n        if (accepted > args.maxEvaluations) return undefined\n        evaluations += 1\n        return accepted\n      },\n      (observation) => {\n        if (!args.observe) return\n        args.observe({ ...observation, sequence: ++observationSequence })\n      },\n      proposedCandidateHashes,\n    ).finally(() => {\n      controller.signal.removeEventListener('abort', abortRequest)\n      activeControllers.delete(controller)\n      activeHandlers.delete(handler)\n    })\n    activeHandlers.add(handler)\n    void handler.catch(() => undefined)\n  })\n  const port = await listenLocal(server)\n  const close = (): Promise<void> => {\n    closePromise ??= closeCallbackServer()\n    return closePromise\n  }\n  const onAbort = (): void => {\n    void close().catch(() => undefined)\n  }\n  args.signal?.addEventListener('abort', onAbort, { once: true })\n  if (args.signal?.aborted) onAbort()\n  return {\n    url: `http://127.0.0.1:${port}/evaluate`,\n    token: args.token,\n    evaluations: () => evaluations,\n    close,\n  }\n\n  async function closeCallbackServer(): Promise<void> {\n    args.signal?.removeEventListener('abort', onAbort)\n    accepting = false\n    const closingServer = closeServer(server)\n    server.closeIdleConnections?.()\n    for (const controller of activeControllers) controller.abort()\n    const [serverResult] = await Promise.allSettled([\n      closingServer,\n      waitForActiveHandlers(activeHandlers),\n    ])\n    if (activeControllers.size !== 0 || activeHandlers.size !== 0) {\n      throw new Error('external optimizer callback closed with active request work')\n    }\n    if (serverResult?.status === 'rejected') throw serverResult.reason\n  }\n}\n\nasync function handleCallback<TResponse>(\n  request: IncomingMessage,\n  response: ServerResponse,\n  signal: AbortSignal,\n  args: {\n    token: string\n    maxEvaluations: number\n    acceptEvaluation?: () => number | undefined\n    evaluate: (request: ExternalTextEvaluationRequest, signal: AbortSignal) => Promise<TResponse>\n    observe?: (observation: ExternalOptimizerEvaluationObservation) => void\n  },\n  limits: ExternalOptimizerCallbackLimits,\n  nextEvaluation: () => number | undefined,\n  observe: (observation: UnsequencedObservation) => void,\n  proposedCandidateHashes: Set<string>,\n): Promise<void> {\n  try {\n    if (request.method !== 'POST' || request.url !== '/evaluate') {\n      sendJsonIfOpen(response, 404, { error: 'not found' })\n      return\n    }\n    if (request.headers.authorization !== `Bearer ${args.token}`) {\n      sendJsonIfOpen(response, 401, { error: 'unauthorized' })\n      return\n    }\n    let body: unknown\n    try {\n      body = await readJson(request, limits.maxRequestBytes)\n    } catch {\n      observe({ kind: 'refusal', reason: 'invalid-request' })\n      sendJsonIfOpen(response, 400, { error: 'request body must be bounded valid JSON' })\n      return\n    }\n    if (\n      !isRecord(body) ||\n      !isExternalTextCandidate(body.candidate) ||\n      typeof body.exampleId !== 'string'\n    ) {\n      observe({ kind: 'refusal', reason: 'invalid-request' })\n      sendJsonIfOpen(response, 400, { error: 'candidate and exampleId are required strings' })\n      return\n    }\n    const candidate = cloneCandidate(body.candidate)\n    const candidateHash = contentHash({ kind: 'external-text-candidate', candidate })\n    if (!proposedCandidateHashes.has(candidateHash)) {\n      observe({ kind: 'proposal', candidate, candidateHash })\n      proposedCandidateHashes.add(candidateHash)\n    }\n    const count = nextEvaluation()\n    if (count === undefined) {\n      observe({\n        kind: 'refusal',\n        reason: 'evaluation-limit',\n        candidate,\n        candidateHash,\n        exampleId: body.exampleId,\n      })\n      sendJsonIfOpen(response, 429, { error: 'evaluation limit reached' })\n      return\n    }\n    let result: TResponse\n    try {\n      result = await args.evaluate({ candidate, exampleId: body.exampleId }, signal)\n    } catch (error) {\n      observe({\n        kind: 'refusal',\n        reason: 'evaluation-failed',\n        candidate,\n        candidateHash,\n        exampleId: body.exampleId,\n      })\n      // The thrown detail is the only diagnostic for a failed evaluation —\n      // an opaque 500 costs the caller a blind debugging round.\n      sendJsonIfOpen(response, 500, {\n        error: `evaluation failed: ${String(error).slice(0, 400)}`,\n      })\n      return\n    }\n    let encoded: string\n    try {\n      encoded = encodeJsonResponse(result)\n    } catch {\n      observe({\n        kind: 'refusal',\n        reason: 'evaluation-failed',\n        candidate,\n        candidateHash,\n        exampleId: body.exampleId,\n      })\n      sendJsonIfOpen(response, 500, { error: 'evaluation response must be valid JSON' })\n      return\n    }\n    if (Buffer.byteLength(encoded) > limits.maxResponseBytes) {\n      observe({\n        kind: 'refusal',\n        reason: 'evaluation-failed',\n        candidate,\n        candidateHash,\n        exampleId: body.exampleId,\n      })\n      sendJsonIfOpen(response, 413, { error: 'evaluation response too large' })\n      return\n    }\n    observe({\n      kind: 'evaluation',\n      candidate,\n      candidateHash,\n      exampleId: body.exampleId,\n      evaluationNumber: count,\n      response: structuredClone(result),\n    })\n    sendEncodedJsonIfOpen(response, 200, encoded)\n  } catch {\n    sendJsonIfOpen(response, 500, { error: 'evaluation failed' })\n  }\n}\n\nfunction readJson(request: IncomingMessage, maxRequestBytes: number): Promise<unknown> {\n  return new Promise((resolvePromise, reject) => {\n    let size = 0\n    const chunks: Buffer[] = []\n    request.on('data', (chunk: Buffer) => {\n      size += chunk.length\n      if (size > maxRequestBytes) {\n        reject(new Error('callback body too large'))\n        request.destroy()\n        return\n      }\n      chunks.push(chunk)\n    })\n    request.on('error', reject)\n    request.on('end', () => {\n      try {\n        resolvePromise(JSON.parse(Buffer.concat(chunks).toString('utf8')))\n      } catch (error) {\n        reject(error)\n      }\n    })\n  })\n}\n\nfunction encodeJsonResponse(value: unknown): string {\n  const encoded = JSON.stringify(value)\n  if (encoded === undefined) throw new Error('evaluation response must be JSON-serializable')\n  return encoded\n}\n\nfunction sendEncodedJsonIfOpen(response: ServerResponse, status: number, encoded: string): void {\n  if (response.destroyed || response.writableEnded) return\n  response.writeHead(status, {\n    'content-type': 'application/json; charset=utf-8',\n    'content-length': String(Buffer.byteLength(encoded)),\n  })\n  response.end(encoded)\n}\n\nfunction assertCallbackConfig(args: {\n  token: string\n  maxEvaluations: number\n  acceptEvaluation?: () => number | undefined\n  evaluate: (request: ExternalTextEvaluationRequest, signal: AbortSignal) => Promise<unknown>\n  observe?: (observation: ExternalOptimizerEvaluationObservation) => void\n  limits?: Partial<ExternalOptimizerCallbackLimits>\n  signal?: AbortSignal\n}): void {\n  if (typeof args.token !== 'string' || !args.token.trim()) {\n    throw new Error('external optimizer callback: token must be non-empty')\n  }\n  if (!Number.isSafeInteger(args.maxEvaluations) || args.maxEvaluations <= 0) {\n    throw new Error('external optimizer callback: maxEvaluations must be a positive safe integer')\n  }\n  if (args.acceptEvaluation !== undefined && typeof args.acceptEvaluation !== 'function') {\n    throw new Error('external optimizer callback: acceptEvaluation must be a function')\n  }\n  if (typeof args.evaluate !== 'function') {\n    throw new Error('external optimizer callback: evaluate must be a function')\n  }\n  if (args.observe !== undefined && typeof args.observe !== 'function') {\n    throw new Error('external optimizer callback: observe must be a function')\n  }\n}\n\nfunction cloneCandidate(candidate: ExternalTextCandidate): ExternalTextCandidate {\n  return typeof candidate === 'string' ? candidate : { ...candidate }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;AAsCA,SAAS,gBAAgB,MAAoB;CAC3C,MAAM,cAAc,oBAAoB,EAAE,UAAU,KAAK,CAAC;CAC1D,IAAI,aAAa,MAAM,iBAAiB,MAAM,WAAW;AAC3D;AAEA,SAAS,iBAAiB,MAAc,aAA+C;CACrF,IAAI,YAAY,WAAW,YACzB,uBAAO,IAAI,MACT,oDAAoD,KAAK,6BAC3D;CAEF,uBAAO,IAAI,MACT,oCAAoC,YAAY,OAAO,IAAI,IAAI,KAAK,oDACtE;AACF;;;;;;AAOA,SAAgB,qBAAqB,MAA2C;CAC9E,MAAM,MAAM,KAAK,OAAO,QAAQ;CAChC,IAAI,CAAC,OAAO,cAAc,GAAG,KAAK,OAAO,GACvC,MAAM,IAAI,MAAM,gDAAgD;CAClE,KAAK,MAAM,QAAQ,KAAK,aAAa,CAAC,GAAG,gBAAgB,IAAI;CAC7D,MAAM,cAAc,yBAAyB;EAC3C,UAAU,KAAK;EACf;CACF,CAAC;CACD,IAAI,CAAC,YAAY,UAAU,MAAM,iBAAiB,KAAK,UAAU,WAAW;CAC5E,MAAM,gBAAgB,KAAK,iBAAiB;CAC5C,IAAI,WAAW;CACf,MAAM,gBAAsB;EAC1B,IAAI,UAAU;EACd,WAAW;EACX,IAAI,eAAe,QAAQ,IAAI,QAAQ,OAAO;EAC9C,IAAI;GACF,YAAY,KAAK,QAAQ;EAC3B,QAAQ,CAER;CACF;CACA,IAAI;EACF,KAAK,MAAM,QAAQ,KAAK,aAAa,CAAC,GAAG,gBAAgB,IAAI;CAC/D,SAAS,OAAO;EACd,QAAQ;EACR,MAAM;CACR;CACA,IAAI,eAAe,QAAQ,GAAG,QAAQ,OAAO;CAC7C,OAAO,EAAE,QAAQ;AACnB;;;AChEA,eAAsB,+BAA0C,MASzB;CACrC,qBAAqB,IAAI;CACzB,MAAM,SAAS,uCAAuC,KAAK,MAAM;CACjE,KAAK,QAAQ,eAAe;CAC5B,IAAI,cAAc;CAClB,IAAI,YAAY;CAChB,IAAI;CACJ,MAAM,oCAAoB,IAAI,IAAqB;CACnD,MAAM,iCAAiB,IAAI,IAAmB;CAC9C,MAAM,0CAA0B,IAAI,IAAY;CAChD,IAAI,sBAAsB;CAC1B,MAAM,SAAS,cAAc,SAAS,aAAa;EACjD,IAAI,CAAC,WAAW;GACd,eAAe,UAAU,KAAK,EAAE,OAAO,yCAAyC,CAAC;GACjF;EACF;EAEA,MAAM,aAAa,IAAI,gBAAgB;EACvC,MAAM,qBAA2B;GAC/B,QAAQ,QAAQ;GAChB,SAAS,QAAQ;EACnB;EACA,kBAAkB,IAAI,UAAU;EAChC,WAAW,OAAO,iBAAiB,SAAS,cAAc,EAAE,MAAM,KAAK,CAAC;EAExE,IAAI;EACJ,UAAU,eACR,SACA,UACA,WAAW,QACX,MACA,cACM;GACJ,MAAM,WAAW,KAAK,mBAAmB,KAAK,iBAAiB,IAAI,cAAc;GACjF,IAAI,aAAa,KAAA,GAAW,OAAO,KAAA;GACnC,IAAI,CAAC,OAAO,cAAc,QAAQ,KAAK,YAAY,GACjD,MAAM,IAAI,MAAM,gEAAgE;GAElF,IAAI,WAAW,KAAK,gBAAgB,OAAO,KAAA;GAC3C,eAAe;GACf,OAAO;EACT,IACC,gBAAgB;GACf,IAAI,CAAC,KAAK,SAAS;GACnB,KAAK,QAAQ;IAAE,GAAG;IAAa,UAAU,EAAE;GAAoB,CAAC;EAClE,GACA,uBACF,CAAC,CAAC,cAAc;GACd,WAAW,OAAO,oBAAoB,SAAS,YAAY;GAC3D,kBAAkB,OAAO,UAAU;GACnC,eAAe,OAAO,OAAO;EAC/B,CAAC;EACD,eAAe,IAAI,OAAO;EAC1B,QAAa,YAAY,KAAA,CAAS;CACpC,CAAC;CACD,MAAM,OAAO,MAAM,YAAY,MAAM;CACrC,MAAM,cAA6B;EACjC,iBAAiB,oBAAoB;EACrC,OAAO;CACT;CACA,MAAM,gBAAsB;EAC1B,MAAW,CAAC,CAAC,YAAY,KAAA,CAAS;CACpC;CACA,KAAK,QAAQ,iBAAiB,SAAS,SAAS,EAAE,MAAM,KAAK,CAAC;CAC9D,IAAI,KAAK,QAAQ,SAAS,QAAQ;CAClC,OAAO;EACL,KAAK,oBAAoB,KAAK;EAC9B,OAAO,KAAK;EACZ,mBAAmB;EACnB;CACF;CAEA,eAAe,sBAAqC;EAClD,KAAK,QAAQ,oBAAoB,SAAS,OAAO;EACjD,YAAY;EACZ,MAAM,gBAAgB,YAAY,MAAM;EACxC,OAAO,uBAAuB;EAC9B,KAAK,MAAM,cAAc,mBAAmB,WAAW,MAAM;EAC7D,MAAM,CAAC,gBAAgB,MAAM,QAAQ,WAAW,CAC9C,eACA,sBAAsB,cAAc,CACtC,CAAC;EACD,IAAI,kBAAkB,SAAS,KAAK,eAAe,SAAS,GAC1D,MAAM,IAAI,MAAM,6DAA6D;EAE/E,IAAI,cAAc,WAAW,YAAY,MAAM,aAAa;CAC9D;AACF;AAEA,eAAe,eACb,SACA,UACA,QACA,MAOA,QACA,gBACA,SACA,yBACe;CACf,IAAI;EACF,IAAI,QAAQ,WAAW,UAAU,QAAQ,QAAQ,aAAa;GAC5D,eAAe,UAAU,KAAK,EAAE,OAAO,YAAY,CAAC;GACpD;EACF;EACA,IAAI,QAAQ,QAAQ,kBAAkB,UAAU,KAAK,SAAS;GAC5D,eAAe,UAAU,KAAK,EAAE,OAAO,eAAe,CAAC;GACvD;EACF;EACA,IAAI;EACJ,IAAI;GACF,OAAO,MAAM,SAAS,SAAS,OAAO,eAAe;EACvD,QAAQ;GACN,QAAQ;IAAE,MAAM;IAAW,QAAQ;GAAkB,CAAC;GACtD,eAAe,UAAU,KAAK,EAAE,OAAO,0CAA0C,CAAC;GAClF;EACF;EACA,IACE,CAAC,SAAS,IAAI,KACd,CAAC,wBAAwB,KAAK,SAAS,KACvC,OAAO,KAAK,cAAc,UAC1B;GACA,QAAQ;IAAE,MAAM;IAAW,QAAQ;GAAkB,CAAC;GACtD,eAAe,UAAU,KAAK,EAAE,OAAO,+CAA+C,CAAC;GACvF;EACF;EACA,MAAM,YAAY,eAAe,KAAK,SAAS;EAC/C,MAAM,gBAAgB,YAAY;GAAE,MAAM;GAA2B;EAAU,CAAC;EAChF,IAAI,CAAC,wBAAwB,IAAI,aAAa,GAAG;GAC/C,QAAQ;IAAE,MAAM;IAAY;IAAW;GAAc,CAAC;GACtD,wBAAwB,IAAI,aAAa;EAC3C;EACA,MAAM,QAAQ,eAAe;EAC7B,IAAI,UAAU,KAAA,GAAW;GACvB,QAAQ;IACN,MAAM;IACN,QAAQ;IACR;IACA;IACA,WAAW,KAAK;GAClB,CAAC;GACD,eAAe,UAAU,KAAK,EAAE,OAAO,2BAA2B,CAAC;GACnE;EACF;EACA,IAAI;EACJ,IAAI;GACF,SAAS,MAAM,KAAK,SAAS;IAAE;IAAW,WAAW,KAAK;GAAU,GAAG,MAAM;EAC/E,SAAS,OAAO;GACd,QAAQ;IACN,MAAM;IACN,QAAQ;IACR;IACA;IACA,WAAW,KAAK;GAClB,CAAC;GAGD,eAAe,UAAU,KAAK,EAC5B,OAAO,sBAAsB,OAAO,KAAK,CAAC,CAAC,MAAM,GAAG,GAAG,IACzD,CAAC;GACD;EACF;EACA,IAAI;EACJ,IAAI;GACF,UAAU,mBAAmB,MAAM;EACrC,QAAQ;GACN,QAAQ;IACN,MAAM;IACN,QAAQ;IACR;IACA;IACA,WAAW,KAAK;GAClB,CAAC;GACD,eAAe,UAAU,KAAK,EAAE,OAAO,yCAAyC,CAAC;GACjF;EACF;EACA,IAAI,OAAO,WAAW,OAAO,IAAI,OAAO,kBAAkB;GACxD,QAAQ;IACN,MAAM;IACN,QAAQ;IACR;IACA;IACA,WAAW,KAAK;GAClB,CAAC;GACD,eAAe,UAAU,KAAK,EAAE,OAAO,gCAAgC,CAAC;GACxE;EACF;EACA,QAAQ;GACN,MAAM;GACN;GACA;GACA,WAAW,KAAK;GAChB,kBAAkB;GAClB,UAAU,gBAAgB,MAAM;EAClC,CAAC;EACD,sBAAsB,UAAU,KAAK,OAAO;CAC9C,QAAQ;EACN,eAAe,UAAU,KAAK,EAAE,OAAO,oBAAoB,CAAC;CAC9D;AACF;AAEA,SAAS,SAAS,SAA0B,iBAA2C;CACrF,OAAO,IAAI,SAAS,gBAAgB,WAAW;EAC7C,IAAI,OAAO;EACX,MAAM,SAAmB,CAAC;EAC1B,QAAQ,GAAG,SAAS,UAAkB;GACpC,QAAQ,MAAM;GACd,IAAI,OAAO,iBAAiB;IAC1B,uBAAO,IAAI,MAAM,yBAAyB,CAAC;IAC3C,QAAQ,QAAQ;IAChB;GACF;GACA,OAAO,KAAK,KAAK;EACnB,CAAC;EACD,QAAQ,GAAG,SAAS,MAAM;EAC1B,QAAQ,GAAG,aAAa;GACtB,IAAI;IACF,eAAe,KAAK,MAAM,OAAO,OAAO,MAAM,CAAC,CAAC,SAAS,MAAM,CAAC,CAAC;GACnE,SAAS,OAAO;IACd,OAAO,KAAK;GACd;EACF,CAAC;CACH,CAAC;AACH;AAEA,SAAS,mBAAmB,OAAwB;CAClD,MAAM,UAAU,KAAK,UAAU,KAAK;CACpC,IAAI,YAAY,KAAA,GAAW,MAAM,IAAI,MAAM,+CAA+C;CAC1F,OAAO;AACT;AAEA,SAAS,sBAAsB,UAA0B,QAAgB,SAAuB;CAC9F,IAAI,SAAS,aAAa,SAAS,eAAe;CAClD,SAAS,UAAU,QAAQ;EACzB,gBAAgB;EAChB,kBAAkB,OAAO,OAAO,WAAW,OAAO,CAAC;CACrD,CAAC;CACD,SAAS,IAAI,OAAO;AACtB;AAEA,SAAS,qBAAqB,MAQrB;CACP,IAAI,OAAO,KAAK,UAAU,YAAY,CAAC,KAAK,MAAM,KAAK,GACrD,MAAM,IAAI,MAAM,sDAAsD;CAExE,IAAI,CAAC,OAAO,cAAc,KAAK,cAAc,KAAK,KAAK,kBAAkB,GACvE,MAAM,IAAI,MAAM,6EAA6E;CAE/F,IAAI,KAAK,qBAAqB,KAAA,KAAa,OAAO,KAAK,qBAAqB,YAC1E,MAAM,IAAI,MAAM,kEAAkE;CAEpF,IAAI,OAAO,KAAK,aAAa,YAC3B,MAAM,IAAI,MAAM,0DAA0D;CAE5E,IAAI,KAAK,YAAY,KAAA,KAAa,OAAO,KAAK,YAAY,YACxD,MAAM,IAAI,MAAM,yDAAyD;AAE7E;AAEA,SAAS,eAAe,WAAyD;CAC/E,OAAO,OAAO,cAAc,WAAW,YAAY,EAAE,GAAG,UAAU;AACpE"}