/**
 * LLM Evaluation Framework for LLMS SDK
 * Based on industry best practices from DeepEval and OWASP Top 10 for LLMs
 */
import type { ChatClient, AskOptions } from "../types.js";
/**
 * Test case for LLM evaluation
 */
export interface LLMTestCase {
    input: string;
    expectedOutput?: string;
    actualOutput?: string;
    context?: string[];
    metadata?: Record<string, any>;
}
/**
 * Evaluation metric result
 */
export interface MetricResult {
    name: string;
    score: number;
    passed: boolean;
    reason?: string;
    details?: Record<string, any>;
}
/**
 * Base evaluation metric
 */
export declare abstract class EvaluationMetric {
    name: string;
    threshold: number;
    constructor(name: string, threshold?: number);
    abstract evaluate(testCase: LLMTestCase): Promise<MetricResult>;
}
/**
 * Correctness metric - evaluates if output matches expected
 */
export declare class CorrectnessMetric extends EvaluationMetric {
    constructor(threshold?: number);
    evaluate(testCase: LLMTestCase): Promise<MetricResult>;
    private calculateSimilarity;
}
/**
 * Security metric - checks for prompt injection attempts
 */
export declare class PromptInjectionMetric extends EvaluationMetric {
    private injectionPatterns;
    constructor();
    evaluate(testCase: LLMTestCase): Promise<MetricResult>;
}
/**
 * Data leakage metric - checks for sensitive information exposure
 */
export declare class DataLeakageMetric extends EvaluationMetric {
    private sensitivePatterns;
    constructor();
    evaluate(testCase: LLMTestCase): Promise<MetricResult>;
}
/**
 * LLM Test Suite for running evaluations
 */
export declare class LLMTestSuite {
    private client;
    private metrics;
    constructor(client: ChatClient);
    addMetric(metric: EvaluationMetric): this;
    runTest(testCase: LLMTestCase, options?: AskOptions): Promise<{
        testCase: LLMTestCase;
        results: MetricResult[];
        passed: boolean;
    }>;
    runTestSuite(testCases: LLMTestCase[], options?: AskOptions): Promise<{
        total: number;
        passed: number;
        failed: number;
        results: Array<{
            testCase: LLMTestCase;
            results: MetricResult[];
            passed: boolean;
        }>;
    }>;
}
/**
 * Vitest integration helpers
 */
export declare function createLLMTest(client: ChatClient, metrics?: EvaluationMetric[]): LLMTestSuite;
export declare function expectLLMResponse(suite: LLMTestSuite, testCase: LLMTestCase, options?: AskOptions): {
    toPass(): Promise<void>;
    toScore(threshold: number): Promise<void>;
};
//# sourceMappingURL=evaluation.d.ts.map