Stage 4 — Document Classification & Enrichment

Goal

Determine the legal type of each document and enrich it with document-specific structured data. This stage gives every document its legal identity before downstream title analysis.

Output: Enriched Legal Document.

Scope

This stage is responsible for:

• Document classification
• Category detection
• Subtype detection
• Type-specific field extraction
• Cross-document reference discovery
• Confidence scoring

This stage does not:

• Build chain of title
• Detect liens
• Perform legal reasoning

Input

• OCR Text
• Structured JSON
• Layout Information
• Document Summary
• Metadata

Processing Pipeline

Structured Document
        │
        ▼
Classification Engine
        │
        ▼
Category Detection
        │
        ▼
Subtype Detection
        │
        ▼
Schema Router
        │
        ▼
Document-specific Extraction
        │
        ▼
Reference Discovery
        │
        ▼
Enriched Document

Classification Engine

Use an LLM with structured output.

Responsibilities:

• Identify legal document category
• Identify subtype
• Produce confidence score
• Return deterministic JSON

Categories

• Deed
• Mortgage
• Assignment
• Release
• Easement
• Judgment
• HOA
• Tax
• Probate
• Bankruptcy
• UCC
• Survey
• Plat
• Other

Example Subtypes

Deeds

• Warranty Deed
• Quitclaim Deed
• Special Warranty Deed
• Trustee's Deed
• Sheriff's Deed
• Correction Deed

Mortgages

• Mortgage
• Deed of Trust
• HELOC
• Refinance
• Modification

Releases

• Mortgage Release
• Satisfaction
• Partial Release
• Lien Release

Schema Router

Each document type is routed into its own schema.

Examples:

• WarrantyDeedSchema
• MortgageSchema
• AssignmentSchema
• ReleaseSchema
• EasementSchema
• JudgmentSchema

This keeps downstream processing independent of document type.

Type-specific Extraction

After classification, extract fields unique to that document type.

Examples:

Warranty Deed

• Grantor
• Grantee
• Consideration
• Vesting Type
• Transfer Type

Mortgage

• Borrower
• Lender
• Original Loan Amount
• Interest Rate
• Maturity Date

Assignment

• Assignor
• Assignee
• Referenced Instrument

Release

• Released Instrument
• Release Date

Reference Discovery

Identify links between documents.

Examples:

• Release references Mortgage
• Assignment references Mortgage
• Modification references Mortgage
• Correction references Deed

Only establish references. Relationship reasoning happens in later stages.

Structured Output

Each document should contain:

• Category
• Subtype
• Generic Fields
• Type-specific Fields
• Cross-document References
• Confidence Score
• Classification Version

Edge Cases

• Multiple legal documents in one PDF
• Unknown document types
• Mixed document packets
• County-specific custom forms
• Illegible headers
• Low-confidence classification

Human Involvement

Only when classification confidence is very low or the document cannot be categorized.

Expected automation: 95–99%.

Definition of Done

• Category assigned.
• Subtype assigned.
• Type-specific schema applied.
• Type-specific fields extracted.
• Cross-document references identified.
• Confidence score generated.
• Ready for Stage 5 (Chain of Title).