Freeing the Law with LOCUS: A Local Ordinance Corpus for the United States

2026-05-07 · Source: cs.CL updates on arXiv.org · Field: Legal & Regulatory — Legal Technology (LegalTech), Regulatory Affairs & Government Relations · Depth: Expert, extended

Summary

The Local Ordinance Corpus for the United States (LOCUS) is a new, comprehensive dataset addressing the critical absence of machine-readable local laws. This corpus includes codes from 9,239 U.S. cities and counties, with a county-harmonized access layer covering 2,309 of 3,144 counties, representing most of the population. LOCUS employs LightOnOCR-2-1B, a 1B parameter vision-language model, to process diverse PDF formats into Markdown, and uses GPT-5.4-nano for initial classification. ModernBERT-based classifiers and regressors then annotate laws across dimensions like function, topic, opacity, paternalism, enforcement discretion, and salience. Available on Hugging Face, LOCUS-v1 facilitates advanced legal AI research, enabling national-scale retrieval, regulatory extraction, and comparative policy analysis of local regulations.

Key takeaway

For legal AI scientists and research teams developing systems for U.S. regulatory analysis, LOCUS-v1 offers an unprecedented, machine-readable corpus of local ordinances. You should integrate this dataset to build more sophisticated models capable of navigating layered legal authority and performing national-scale comparative policy analysis. Be aware that while LOCUS provides a crucial geographic substrate, your systems must still reason about specific jurisdictional control and state-local overlaps.

Key insights

LOCUS provides a national, machine-readable corpus of U.S. local ordinances, enabling large-scale legal AI research.

Principles

• Local law is a layered system requiring reasoning about jurisdictional overlap.
• Local codes exhibit a recurring documentary form and functional division.
• Harmonization of legal data must be explicit about its abstractions.

Method

Collect PDFs via browser automation, OCR to Markdown using LightOnOCR-2-1B, segment, then classify (GPT-5.4-nano) and score (ModernBERT regressors trained on LLM-as-a-Judge pairwise comparisons).

In practice

• Perform search and Q&A over local rules with varying terminology.
• Extract regulated activities, permits, fees, and penalties.
• Develop benchmarks for multi-layered legal reasoning systems.

Topics

• Local Ordinances
• Legal AI
• Corpus Development
• Optical Character Recognition
• ModernBERT
• Regulatory Analysis

Best for: NLP Engineer, AI Scientist, Research Scientist, Legal Professional

Related on AIssential

• Freeing the Law with LOCUS: A Local Ordinance Corpus for the United States — Local legal ordinances, critical for daily life, are now machine-readable at scale through LOCUS.
• Grounded in Law: A Multi-Stage Anti-Hallucination Pipeline for Legal RAG Systems in Brazilian Portuguese — A multi-stage RAG pipeline significantly reduces legal citation hallucinations in Brazilian Portuguese LLMs.
• Concepts and measures of bureaucratic constraints in European Union laws from hand-coding to machine-learning — Models learn from manually annotated text to analyze EU law constraints.
• When accurate AI is still dangerously incomplete — In high-stakes domains, AI requires "completeness" and authority beyond mere accuracy and relevancy.
• Discovery of Legal Patterns in Civil Petitions via LLM-Based Fact Extraction and Density Clustering — LLM-based fact extraction significantly improves legal document clustering by reducing noise and enhancing semantic coherence.
• Retrieval-Based Multi-Label Legal Annotation: Extensible, Data-Efficient and Hallucination-Free — Retrieval-based legal annotation offers a data-efficient, hallucination-free alternative to generative LLMs for high-cardinality label spaces.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CL updates on arXiv.org.