llmverify/dist/engines/classification/index.d.ts

AI Output Verification Toolkit — Local-first LLM safety, hallucination detection, PII redaction, prompt injection defense, and runtime monitoring. Zero telemetry. OWASP LLM Top 10 aligned.

/**
 * Classification Engine
 *
 * Comprehensive LLM output classification including intent detection,
 * JSON repair, instruction-following evaluation, hallucination risk,
 * and reasoning compression analysis.
 *
 * WHAT THIS DOES:
 * ✅ Detects output intent (summary, code, list, etc.)
 * ✅ Detects and repairs malformed JSON
 * ✅ Evaluates instruction-following compliance
 * ✅ Calculates hallucination risk indicators
 * ✅ Detects reasoning compression
 * ✅ Generates classification tags
 *
 * WHAT THIS DOES NOT DO:
 * ❌ Guarantee hallucination detection (heuristic-based)
 * ❌ Verify factual accuracy (no ground truth)
 * ❌ Replace human review
 *
 * @module engines/classification
 * @author Haiec
 * @license MIT
 */
export type { IntentTag, IntentCandidate, InstructionRuleType, FormatExpectation, InstructionRule, RuleResult, HallucinationSignals, HallucinationLabel, CompressionMetrics, ReasoningLabel, JsonRepairStep, BaseMetrics, ClassificationPolicy, ClassificationResult } from './types';
export { detectIntent, getPrimaryIntent } from './intent';
export { detectAndRepairJson } from './json-repair';
export { evaluateInstructionRules } from './instruction-eval';
export { calculateHallucinationSignals, calculateHallucinationRisk, getHallucinationLabel } from './hallucination';
export { calculateCompressionMetrics, calculateCompressionScore, getReasoningLabel } from './compression';
export { ClassificationEngine, classify } from './engine';