/**
* LLM-based Evaluators
*
* This module provides intelligent evaluation capabilities that leverage
* LLM reasoning instead of rigid pattern matching.
*
* Key improvements over pattern-based evaluation:
* 1. Context-sensitive convergence assessment
* 2. Nuanced severity classification considering impact
* 3. Actual edge case analysis verification (not just keyword detection)
* 4. Evidence-based false positive detection
*/
// Types
export type {
SamplingFunction,
SamplingRequest,
SamplingResponse,
ConfidenceLevel,
EvaluationResult,
ConvergenceEvaluation,
SeverityEvaluation,
EdgeCaseEvaluation,
FalsePositiveEvaluation,
LLMEvaluatorConfig,
EvaluatorContext,
} from './types.js';
export { DEFAULT_EVALUATOR_CONFIG } from './types.js';
// Convergence Evaluator
export { evaluateConvergenceLLM } from './convergence-evaluator.js';
// Severity Evaluator
export {
evaluateSeverityLLM,
evaluateSeveritiesBatch,
type SeverityEvaluationInput,
} from './severity-evaluator.js';
// Edge Case Evaluator
export {
evaluateEdgeCasesLLM,
getRecommendedEdgeCases,
} from './edge-case-evaluator.js';
// False Positive Evaluator
export {
evaluateFalsePositiveLLM,
evaluateFalsePositivesBatch,
quickFalsePositiveCheck,
type FalsePositiveInput,
} from './false-positive-evaluator.js';
/**
* Unified evaluator that combines all evaluation types
*/
export interface UnifiedEvaluator {
evaluateConvergence: typeof import('./convergence-evaluator.js').evaluateConvergenceLLM;
evaluateSeverity: typeof import('./severity-evaluator.js').evaluateSeverityLLM;
evaluateEdgeCases: typeof import('./edge-case-evaluator.js').evaluateEdgeCasesLLM;
evaluateFalsePositive: typeof import('./false-positive-evaluator.js').evaluateFalsePositiveLLM;
}
/**
* Create a unified evaluator instance with a shared sampling function
*/
export function createUnifiedEvaluator(
samplingFn: SamplingFunction,
config?: Partial<LLMEvaluatorConfig>
): UnifiedEvaluator {
const { evaluateConvergenceLLM } = require('./convergence-evaluator.js');
const { evaluateSeverityLLM } = require('./severity-evaluator.js');
const { evaluateEdgeCasesLLM } = require('./edge-case-evaluator.js');
const { evaluateFalsePositiveLLM } = require('./false-positive-evaluator.js');
const { DEFAULT_EVALUATOR_CONFIG } = require('./types.js');
const mergedConfig = { ...DEFAULT_EVALUATOR_CONFIG, ...config };
return {
evaluateConvergence: (context: EvaluatorContext) =>
evaluateConvergenceLLM(context, samplingFn, mergedConfig),
evaluateSeverity: (input: SeverityEvaluationInput) =>
evaluateSeverityLLM(input, samplingFn, mergedConfig),
evaluateEdgeCases: (context: EvaluatorContext) =>
evaluateEdgeCasesLLM(context, samplingFn, mergedConfig),
evaluateFalsePositive: (input: FalsePositiveInput) =>
evaluateFalsePositiveLLM(input, samplingFn, mergedConfig),
};
}
// Import types for the unified evaluator
import type { SamplingFunction, LLMEvaluatorConfig, EvaluatorContext } from './types.js';
import type { SeverityEvaluationInput } from './severity-evaluator.js';
import type { FalsePositiveInput } from './false-positive-evaluator.js';
