/**
 * Capture Agent — Benchmark Harness
 *
 * Runs the same preset repeatedly, comparing:
 * - Total LLM cost
 * - Total wall-clock time
 * - Success rate
 * - Recovery rate (how often healing was needed)
 *
 * Useful for tracking determinism, recovery rate, and runtime cost over time.
 */
import type { ExecutionProgram, RunResult } from './execution-types.js';
export interface BenchmarkPreset {
    name: string;
    presetId: string;
    program: ExecutionProgram;
    /** Expected behavior description for manual verification */
    expectedBehavior: string;
    /** Difficulty tier */
    tier: 'simple' | 'multi_step' | 'auth_required' | 'clip' | 'video';
}
export interface BenchmarkRunResult {
    presetName: string;
    tier: string;
    success: boolean;
    totalDurationMs: number;
    llmCostEur: number;
    llmCallCount: number;
    totalOpcodes: number;
    recoveredOpcodes: number;
    failedOpcodes: number;
    healerInvocations: number;
    circuitBreakerTrips: number;
    variantCount: number;
    successfulVariants: number;
    error?: string;
}
export interface BenchmarkSummary {
    totalPresets: number;
    successCount: number;
    failCount: number;
    successRate: number;
    averageDurationMs: number;
    averageLlmCostEur: number;
    totalLlmCostEur: number;
    deterministicRate: number;
    recoveryRate: number;
    results: BenchmarkRunResult[];
    byTier: Record<string, {
        total: number;
        success: number;
        rate: number;
    }>;
}
export type RunFn = (program: ExecutionProgram) => Promise<RunResult>;
export declare function runBenchmark(presets: BenchmarkPreset[], runCapture: RunFn, options?: {
    runs?: number;
}): Promise<BenchmarkSummary>;
export declare function formatBenchmarkSummary(summary: BenchmarkSummary): string;