import { analyzeAgentTranscript } from './transcript-metrics.js';
export type BenchmarkRepoTier = 'small-familiar' | 'large-unfamiliar';
export interface PresetBenchmarkTask {
    id: string;
    class: 'shared' | 'governance';
    prompt: string;
    expectedTools: string[];
    distractors: string[];
}
export interface PresetBenchmarkCorpus {
    schemaVersion: 1;
    id: string;
    environment: {
        image: string;
        digest: string;
    };
    agentConfig: {
        models: string[];
        runs: number;
    };
    repoTiers: BenchmarkRepoTier[];
    tasks: PresetBenchmarkTask[];
}
export interface CorpusValidationIssue {
    code: 'invalid-corpus' | 'stale-tool-id' | 'missing-distractor';
    message: string;
    taskId?: string;
    preset?: string;
    toolId?: string;
}
export interface TrajectoryScore {
    selectedTool: string | null;
    selectionCorrect: boolean;
    steps: number;
    tokenCost: number;
    costUsd: number;
    reread: ReturnType<typeof analyzeAgentTranscript>;
}
export interface SelectionScore {
    selectedTool: string | null;
    selectionCorrect: boolean;
    tokenCost: number;
}
export declare function validatePresetBenchmarkCorpus(corpus: PresetBenchmarkCorpus, knownToolIds: ReadonlySet<string>, surfaces: Readonly<Record<string, ReadonlySet<string>>>): CorpusValidationIssue[];
/** Deterministically score one logged first-tool selection response. */
export declare function scoreSelectionResponse(task: PresetBenchmarkTask, rawResponse: string): SelectionScore;
/** Deterministically score one logged agent trajectory. No model is called here. */
export declare function scorePresetTrajectory(task: PresetBenchmarkTask, rawTranscript: string): TrajectoryScore;
//# sourceMappingURL=preset-protocol.d.ts.map