Output type for legal AI hallucination cases

legal-ai-type.ts

const LegalHallucinationAnalysis = z.object({
  // Case identification
  case_id: z.string(),
  case_name: z.string(),
  date_occurred: z.string().regex(/^\d{4}-\d{2}-\d{2}$/),
  ai_model_used: z.string(),
  source_url: z.string().url().optional(),
  
  // Hallucination details
  hallucination_details: z.object({
    type: HallucinationType,
    specific_false_content: z.string(),
    correct_information: z.string(),
    user_query: z.string(),
    ai_response_excerpt: z.string(),
    fabricated_citations: z.array(z.string()).optional(),
    legal_domain: z.string() // e.g., "contract law", "criminal procedure"
  }),
  
  // Context gap analysis
  context_gap_analysis: z.object({
    primary_gap_type: GapType,
    missing_information: z.string(),
    information_availability: AvailabilityLevel,
    training_data_factor: z.boolean(),
    context_window_overflow: z.boolean(),
    jurisdictional_confusion: z.boolean(),
    temporal_information_needed: z.boolean(), // Recent legal changes
    citation_network_gap: z.boolean() // Missing legal precedent connections
  }),
  
  // Technical failure analysis
  technical_failure_mode: z.object({
    primary_failure: FailureMode,
    retrieval_failure: z.boolean(),
    reasoning_failure: z.boolean(),
    generation_failure: z.boolean(),
    verification_failure: z.boolean(),
    attention_mechanism_issue: z.boolean(),
    token_limit_exceeded: z.boolean(),
    failure_explanation: z.string()
  }),
  
  // Context preventability scoring (1-10 scale)
  context_preventability_scores: z.object({
    database_access_score: z.number().min(1).max(10),
    real_time_info_score: z.number().min(1).max(10),
    training_data_score: z.number().min(1).max(10),
    context_window_score: z.number().min(1).max(10),
    verification_score: z.number().min(1).max(10),
    overall_preventability: z.number().min(1).max(10),
    confidence_level: ConfidenceLevel,
    scoring_rationale: z.string()
  }),
  
  // Impact assessment
  impact_assessment: z.object({
    financial_cost: z.number().min(0), // USD
    court_sanctions: z.number().min(0), // USD
    reputational_damage: ImpactLevel,
    legal_consequences: z.array(z.string()),
    client_impact: ImpactLevel,
    systemic_risk: SeverityLevel,
    media_attention: z.boolean(),
    precedent_setting: z.boolean()
  }),
  
  // Solution requirements
  solution_requirements: z.object({
    required_data_sources: z.array(z.string()),
    technical_capabilities: z.array(z.string()),
    verification_mechanisms: z.array(z.string()),
    implementation_complexity: ComplexityLevel,
    estimated_cost: ComplexityLevel,
    development_timeline: TimelineEstimate,
    infrastructure_requirements: z.array(z.string())
  }),
  
  // Categorization
  categorization: z.object({
    primary_category: PreventabilityCategory,
    secondary_categories: z.array(z.string()),
    severity_level: SeverityLevel,
    precedent_value: SeverityLevel,
    industry_impact: SeverityLevel
  }),
  
  // Recommendations
  recommendations: z.object({
    immediate_actions: z.array(z.string()),
    technical_solutions: z.array(z.string()),
    policy_changes: z.array(z.string()),
    industry_implications: z.string(),
    prevention_strategies: z.array(z.string())
  }),
  
  // Quality control metadata
  analysis_metadata: z.object({
    analyst_confidence: ConfidenceLevel,
    analysis_date: z.string().regex(/^\d{4}-\d{2}-\d{2}$/),
    review_status: z.enum(['pending', 'reviewed', 'validated']),
    evidence_quality: SeverityLevel,
    methodology_version: z.string(),
    potential_biases: z.array(z.string()).optional()
  })
});