import { KnowledgeGraph } from '../../contracts/graph.js';
import type { ContextSessionDiagnostics, ContextSessionState } from '../../contracts/context-session.js';
import { type RetrieveResult } from '../../runtime/retrieve.js';
import { type PromptRunnerUsage } from '../prompt-runner.js';
export type BenchmarkPromptTokenSource = 'estimated_cl100k_base' | 'claude_reported_input' | 'gemini_reported_input';
export interface BenchmarkPromptArtifacts {
    prompt: string;
    answer: string;
    report: string;
    share_safe_report: string;
}
export interface BenchmarkPromptExecution {
    mode: 'madar';
    question: string;
    promptFile: string;
    outputFile: string;
    command: string;
}
export interface BenchmarkPromptRunnerResult {
    exitCode: number;
    stdout: string;
    stderr: string;
    elapsedMs: number;
}
export interface RunBenchmarkPromptOptions {
    graphPath: string;
    graph: KnowledgeGraph;
    question: string;
    execTemplate: string;
    outputDir?: string;
    now?: Date;
    retrievalBudget?: number;
    retrieval?: RetrieveResult;
    session?: ContextSessionState;
    runner?: (execution: BenchmarkPromptExecution) => Promise<BenchmarkPromptRunnerResult>;
}
export interface BenchmarkPromptRun {
    prompt_tokens_estimated: number;
    query_tokens: number;
    effective_query_tokens: number;
    reused_context_tokens: number;
    session_diagnostics: ContextSessionDiagnostics;
    total_tokens: number | null;
    prompt_token_source: BenchmarkPromptTokenSource;
    usage: PromptRunnerUsage | null;
    answer_text: string | null;
    elapsed_ms: number;
    artifacts: BenchmarkPromptArtifacts;
    session_state: ContextSessionState;
}
export declare function retrieveBenchmarkContext(graph: KnowledgeGraph, graphPath: string, question: string, budget: number): RetrieveResult;
export declare function runBenchmarkPrompt(options: RunBenchmarkPromptOptions): Promise<BenchmarkPromptRun>;