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

A team of researchers has released LOCUS, a corpus that compiles municipal and county ordinance codes from 9,239 U.S. cities and counties, addressing a long-standing gap in machine-readable local law [1][2]. The raw corpus, described in a paper submitted to arXiv on June 17, 2026, aggregates nearly all publicly available local codes, which govern areas such as zoning, housing, business licensing, and public health [1][2]. The researchers note that these texts have historically been “fragmented across vendor platforms designed for human browsing rather than bulk research access” [2]. To convert the documents into a usable format, the team applied optical character recognition to handle the diverse file types in which the ordinances were stored [1][2]. Alongside the full collection, the group produced a county-harmonized access layer that covers 2,309 of the 3,144 U.S. counties, a subset that accounts for a majority of the national population [1][2]. This harmonized layer is intended to make it easier to compare regulations across jurisdictions. The corpus is being released with detailed coverage metadata to support reproducibility and to aid downstream legal AI research [1][2]. The absence of local law from machine-readable corpora has been a recognized obstacle for computational legal studies. While statutes and case law from federal and state levels are widely available, local ordinances represent a foundational tier of the legal system that directly shapes daily life [2][3]. The LOCUS project aims to enable the incremental expansion of machine-readable access to this layer of American law [2]. To demonstrate the corpus’s utility, the researchers trained a collection of classifiers and scorers based on the ModernBERT architecture [1][2]. These models are designed to analyze local law along dimensions such as opacity and paternalism, which the authors state have not previously been studied at this scale [1][2]. The dataset, named LOCUS-v1, and its derivative models have been made available through the Hugging Face platform [2].