Fraudulent Document Detection Failures

class FraudDetector:
    """Detect fraudulent mortgage documents"""
    
    INCOME_VARIANCE_THRESHOLD = 0.05  # 5% variance allowed
    
    def __init__(self):
        self.image_forensics = ImageForensicsEngine()
        self.font_analyzer = FontConsistencyAnalyzer()
        self.format_validator = DocumentFormatValidator()
    
    def analyze_document(self, document: dict) -> dict:
        """Comprehensive fraud analysis"""
        flags = []
        
        # Image-level analysis
        if document.get("image"):
            image_flags = self.image_forensics.analyze(document["image"])
            flags.extend(image_flags)
        
        # Font consistency
        font_flags = self.font_analyzer.check_consistency(document)
        flags.extend(font_flags)
        
        # Format validation
        format_flags = self.format_validator.validate(
            document["type"], 
            document["content"]
        )
        flags.extend(format_flags)
        
        # Cross-reference if available
        if document.get("verification_source"):
            xref_flags = self.cross_reference(
                document["content"],
                document["verification_source"]
            )
            flags.extend(xref_flags)
        
        return {
            "fraud_flags": flags,
            "risk_score": self.calculate_risk(flags),
            "recommendation": self.get_recommendation(flags)
        }
    
    def cross_reference_income(self, 
                               w2_income: float,
                               tax_transcript_income: float) -> dict:
        """Cross-reference W-2 with IRS transcript"""
        variance = abs(w2_income - tax_transcript_income) / tax_transcript_income
        
        if variance > self.INCOME_VARIANCE_THRESHOLD:
            return {
                "flag": "income_mismatch",
                "severity": "high",
                "w2_income": w2_income,
                "transcript_income": tax_transcript_income,
                "variance": variance,
                "action": "manual_review_required"
            }
        
        return {"flag": None, "verified": True}
    
    def detect_altered_digits(self, document_image) -> list:
        """Detect digitally altered numbers"""
        flags = []
        
        # Analyze numeric regions for:
        # - Font inconsistency
        # - JPEG artifacts
        # - Color depth variations
        # - Alignment anomalies
        
        numeric_regions = self.extract_numeric_regions(document_image)
        
        for region in numeric_regions:
            analysis = self.image_forensics.analyze_region(region)
            
            if analysis["manipulation_probability"] > 0.7:
                flags.append({
                    "flag": "potential_alteration",
                    "region": region["location"],
                    "confidence": analysis["manipulation_probability"],
                    "indicators": analysis["indicators"]
                })
        
        return flags


class CrossDocumentValidator:
    """Validate consistency across mortgage documents"""
    
    def validate_income_consistency(self, documents: dict) -> dict:
        """Check income matches across all sources"""
        income_sources = {
            "w2": documents.get("w2", {}).get("wages"),
            "paystub": documents.get("paystub", {}).get("ytd_gross"),
            "tax_return": documents.get("tax_return", {}).get("agi"),
            "employment_letter": documents.get("voe", {}).get("annual_salary"),
            "transcript": documents.get("4506t", {}).get("wages")
        }
        
        # Remove None values
        available = {k: v for k, v in income_sources.items() if v}
        
        if len(available) < 2:
            return {"status": "insufficient_sources"}
        
        # Check consistency
        values = list(available.values())
        max_val = max(values)
        min_val = min(values)
        variance = (max_val - min_val) / max_val
        
        if variance > 0.10:  # More than 10% variance
            return {
                "status": "mismatch",
                "severity": "high" if variance > 0.25 else "medium",
                "sources": available,
                "variance": variance,
                "action": "investigate_discrepancy"
            }
        
        return {"status": "consistent", "sources": available}

instructions: |
  ## FRAUD DETECTION FOR MORTGAGE DOCUMENTS
  
  When processing mortgage documents, flag these red flags:
  
  DOCUMENT AUTHENTICITY:
  - Font inconsistencies within same document
  - Poor image quality on specific sections
  - Misaligned text or numbers
  - Unusual formatting for document type
  - Metadata showing recent creation/editing
  
  INCOME VERIFICATION:
  - W-2 income doesn't match 4506-T transcript
  - Paystub YTD doesn't align with W-2
  - Employment letter salary differs from W-2
  - Self-employment income lacks Schedule C support
  
  ASSET VERIFICATION:
  - Large deposits without documentation
  - Account numbers in wrong format for institution
  - Statement layout doesn't match bank's format
  - Balance inconsistent across statement pages
  
  IDENTITY:
  - SSN format invalid or recycled
  - Address is known mail drop
  - Employment unverifiable
  - Multiple applications with similar info
  
  FLAG for human review, don't auto-reject:
  High confidence fraud indicators require escalation.