/**
 * Evaluation Engine — avalia resultados de agentes e workflows.
 *
 * Deriva um verdict (PASS / PASS_WITH_WARNINGS / FAIL / BLOCKED / UNKNOWN)
 * a partir de métricas em [0,1] ponderadas, resultados de testes e regressões.
 *
 * Uso:
 *   const engine = new EvaluationEngine();
 *   const result = engine.evaluate({ metrics: {...}, tests: {...} });
 */
import type { EvaluationReport, EvaluationResult, EvaluationThresholds, MetricName, MetricWeightings, Metrics, TestSummary, Verdict } from '../types.js';
export declare const DEFAULT_WEIGHTINGS: MetricWeightings;
export declare const DEFAULT_THRESHOLDS: EvaluationThresholds;
export interface EvaluateInput {
    metrics?: Partial<Record<MetricName, number>>;
    tests?: TestSummary;
    regressions?: string[];
    recommendations?: string[];
    weightings?: Partial<MetricWeightings>;
    thresholds?: Partial<EvaluationThresholds>;
    /** Se true, retorna BLOCKED em vez de FAIL quando há regressões críticas. */
    blockOnCriticalRegression?: boolean;
}
export declare class EvaluationEngine {
    private readonly defaults;
    constructor(defaults?: {
        weightings?: MetricWeightings;
        thresholds?: EvaluationThresholds;
    });
    /**
     * Computa o score ponderado das métricas informadas.
     * Métricas ausentes não penalizam o score (são ignoradas).
     */
    weightedScore(metrics: Partial<Record<MetricName, number>>, weightings: MetricWeightings): number;
    /**
     * Deriva o verdict a partir de score, testes e regressões.
     */
    computeVerdict(score: number, opts: {
        tests?: TestSummary;
        regressions?: string[];
        thresholds: EvaluationThresholds;
        metrics: Metrics;
    }): Verdict;
    /**
     * Métricas a partir do resultado de testes.
     */
    testMetrics(tests: TestSummary): {
        testResults: number;
        confidence: number;
    };
    /**
     * Avalia um resultado de agente/workflow.
     */
    evaluate(input: EvaluateInput): EvaluationResult;
    /**
     * Confiança derivada de quantas métricas foram medidas de fato.
     */
    computeConfidence(metrics: Partial<Record<MetricName, number>>, tests?: TestSummary): number;
    /**
     * Constrói um relatório completo com metadados do run.
     */
    buildReport(input: EvaluateInput & {
        taskId: string;
        task: string;
        agentId?: string;
    }): EvaluationReport;
}
/** Helper: clamp de métrica para [0,1]. */
export declare function clamp01(v: number): number;
//# sourceMappingURL=engine.d.ts.map