/**
 * Evaluation Harness -- orchestrates running the corpus through the ATR engine
 * and produces a structured EvalReport.
 *
 * Supports:
 *   - Regex-only evaluation (Tier 2)
 *   - Regex + Embedding evaluation (Tier 2 + 2.5)
 *   - Full pipeline evaluation (all tiers)
 *   - Per-sample latency measurement
 *   - Regression check against baseline thresholds
 *
 * @module agent-threat-rules/eval/eval-harness
 */
import type { CorpusSample } from './corpus.js';
import { getCorpusStats } from './corpus.js';
import type { EvalReport, BaselineThresholds, RegressionCheck } from './metrics.js';
import type { RuleQualityReport } from './rule-metrics.js';
export interface EvalConfig {
    /** Directory containing ATR YAML rules */
    readonly rulesDir: string;
    /** Path to attack embeddings JSON (optional, for Tier 2.5) */
    readonly embeddingsPath?: string;
    /** Custom corpus (defaults to built-in EVAL_CORPUS) */
    readonly corpus?: readonly CorpusSample[];
    /** Baseline thresholds for regression check */
    readonly thresholds?: BaselineThresholds;
    /** Path to save report JSON */
    readonly outputPath?: string;
    /** Enable Tier 2.5 embedding evaluation (default: auto-detect) */
    readonly enableEmbedding?: boolean;
}
/**
 * Run the full evaluation harness.
 * Returns the EvalReport and RegressionCheck.
 */
export declare function runEval(config: EvalConfig): Promise<{
    report: EvalReport;
    regression: RegressionCheck;
    corpusStats: ReturnType<typeof getCorpusStats>;
    tiersUsed: readonly string[];
    ruleQuality: RuleQualityReport;
}>;
export declare function runEvalCLI(): Promise<void>;
//# sourceMappingURL=eval-harness.d.ts.map