if-legal-corpus

Self-hosted legal document corpus for contract analysis and freelancer rights research. Contains international legislation, case law, and industry standards indexed in ChromaDB for semantic search.

Quick Start

Setup

# Clone repository (already done)
cd /home/setup/if-legal-corpus

# Create virtual environment
python3 -m venv .venv
source .venv/bin/activate

# Install dependencies
pip install -r requirements.txt

Running Downloads

To download and process documents from the inventory:

# Download all documents from LEGAL_CORPUS_IMPORT_LIST.md
python scripts/download_all.py

# This will:
# 1. Read LEGAL_CORPUS_IMPORT_LIST.md
# 2. Download each document to raw/ subdirectories
# 3. Log downloads to logs/download_log.csv
# 4. Record manifest entries in manifests/download_manifest.csv

Running Chroma Ingestion

To index downloaded documents into ChromaDB for semantic search:

# Ingest documents into Chroma vector database
python scripts/ingest_chromadb.py \
  --manifest manifests/download_manifest.csv \
  --db-dir indexes/chromadb

# This will:
# 1. Read all documents with status=success from manifest
# 2. Extract text (supports .md, .pdf, .html, .json, .xml)
# 3. Chunk text (1500 chars per chunk, 200 char overlap)
# 4. Generate embeddings (using Chroma's default embedding)
# 5. Store vectors with metadata in indexes/chromadb/

Corpus Structure

Raw Documents (raw/)

raw/
├── uk/                    # United Kingdom legislation (P0: COMPLETE)
│   ├── employment/        # Employment law
│   │   └── Employment_Rights_Act_1996.md
│   ├── ip/                # Intellectual property
│   │   ├── Patents_Act_1977.md
│   │   ├── Trade_Secrets_Enforcement_Regulations_2018.md
│   │   └── Copyright_Rights_Databases_Regulations_1997.md
│   └── tax/               # Tax & IR35 regulations
│       └── Social_Security_Intermediaries_Regulations_2000.md
├── us_federal/            # US federal statutes and regulations
├── us_state/              # State-level law (CA, NY, TX, etc.)
├── eu/                    # EU directives and regulations
├── germany/               # German civil code (BGB)
├── france/                # French labor and IP law
├── canada/                # Canadian federal and provincial law
├── australia/             # Australian legislation
├── datasets/              # Pre-labeled contract datasets (CUAD, etc.)
├── caselaw/               # Landmark cases (CourtListener, etc.)
└── industry/              # Industry standards (AIGA, GAG, IGDA, etc.)

Manifest & Logs (manifests/ and logs/)

• manifests/download_manifest.csv - Complete inventory with download status, file sizes, and SHA-256 hashes
• logs/download_log.csv - Timestamped download operations

Vector Database (indexes/chromadb/)

• Chroma persistent storage - 5,290 vectors from all ingested documents
• Collection name: if_legal_corpus
• Metadata per vector: inventory_path, document_name, local_path, sha256

UK P0 Documents - Integration Complete

All 5 critical UK documents for freelancer contract analysis have been integrated (2025-11-28):

Document	Category	Size	Vectors	Status
Employment Rights Act 1996	Employment law	1,031 KB	6	✓ Complete
Patents Act 1977	IP law	455 KB	3	✓ Complete
Trade Secrets Enforcement Regulations 2018	IP law	18 KB	1	✓ Complete
Social Security (Intermediaries) Regulations 2000	IR35 Tax	19 KB	1	✓ Complete
Database Rights Regulations 1997	IP law	18 KB	1	✓ Complete

Total UK collection: 12 documents, 12+ vectors, 16 MB

Corpus Coverage Summary

• Total Inventory Items: 153
• Successfully Downloaded: 91 (59%)
• Download Errors: 45 (29%)
• No Direct Link: 17 (11%)
• Chroma Vectors: 5,290
• Raw Size: ~16 MB

Important Notes

Chroma Database Size

The ChromaDB indexes are large and should not be committed to git:

• Add indexes/chromadb/ to .gitignore
• Regenerate vectors locally using scripts/ingest_chromadb.py

Virtual Environment

The .venv/ directory is excluded from git. Install dependencies locally after cloning.

Document Formats Supported

• Markdown (.md)
• PDF (.pdf)
• HTML (.html)
• XML (.xml)
• JSON (.json)
• Plain text

Chunking Strategy

• Default chunk size: 1,500 characters
• Overlap: 200 characters
• Configurable in scripts/ingest_chromadb.py via chunk_text() function

License & Attribution

This corpus includes:

• Public domain legislation from UK, US, EU, Canada, Australia
• Creative commons licensed industry standards (AIGA, GAG, IGDA)
• Dataset acknowledgments: CUAD, ContractNLI, LEDGAR

See individual source documents for their specific licensing terms.