import type { EvalResult } from '../core/types.ts';
import type { TokenUsage } from '../llm/types.ts';
import type { EvalJudge } from './EvalJudge.ts';
import type { EmbedFn } from './ReplayEvaluator.ts';
import type { EvalSuite } from './types.ts';
export declare const DEFAULT_METRICS: string[];
export interface MemberRunResult {
    output: string;
    durationMs: number;
    usage?: TokenUsage;
}
export type RunMemberFn = (task: string) => Promise<MemberRunResult>;
export type TaskStatus = 'completed' | 'error' | 'unevaluated';
export interface TaskScore {
    input: string;
    expectedOutput: string;
    output: string;
    durationMs: number;
    tokens: number;
    scores: Record<string, number>;
    status: TaskStatus;
    error?: string;
    assertions?: {
        score: number;
        passed: number;
        total: number;
    };
}
export interface EvalReport {
    suiteName: string;
    member: string;
    timestamp: string;
    tasks: TaskScore[];
    aggregate: Record<string, number>;
}
/**
 * Runs a member against every task in a suite, scores each run through an
 * EvalJudge, and aggregates per-metric means. `runner` is injected so tests
 * and the replay evaluator can drive it without an LLM.
 */
export declare class EvalRunner {
    private readonly runner;
    private readonly judge;
    private readonly metrics;
    private readonly embed?;
    constructor(runner: RunMemberFn, judge: EvalJudge, options?: {
        metrics?: string[];
        embed?: EmbedFn;
    });
    run(suite: EvalSuite, member: string): Promise<EvalReport>;
    private runTask;
}
/** Converts a report into EvalResult episodes consumable by the EpisodeLearner. */
export declare function buildEvalResults(report: EvalReport): EvalResult[];
//# sourceMappingURL=EvalRunner.d.ts.map