import type { EvalSuite, RunRecord, TaskFrequencies, Usage } from './types.js';
import { type UsageSummary } from './usage.js';
export interface ConfigSummary {
    configId: string;
    total: number;
    passed: number;
    failed: number;
    errored: number;
    difficult: number;
    retries: number;
    /** null when any record is missing latency. */
    totalLatencyMs: number | null;
    /** null when any record is missing usage attribution; never silently zero. */
    totalUsage: Usage | null;
    usageAttributionComplete: boolean;
    usageBreakdown: UsageSummary;
    /** Operator effort stays separate so model savings cannot hide human work. */
    totalOperatorMinutes: number | null;
    operatorAttributionComplete: boolean;
}
export declare function summarize(records: RunRecord[], configId: string): ConfigSummary;
export interface WeightedResult {
    configId: string;
    weightedPassRate: number;
}
/**
 * Workload-weighted results are only produced when every evaluated fixture has
 * a documented frequency. Otherwise callers must report per-task results and
 * make no representative-savings claim.
 */
export declare function weightedPassRates(suite: EvalSuite, records: RunRecord[], frequencies: TaskFrequencies | null): WeightedResult[] | null;
export declare function renderReport(suite: EvalSuite, records: RunRecord[], frequencies?: TaskFrequencies | null, evidence?: 'synthetic' | 'observed'): string;
//# sourceMappingURL=report.d.ts.map