America's local law has been gathered into one place for the first time. LOCUS, the Local Ordinance Corpus for the United States, was published online in June by researchers at the University of California, Berkeley.
The corpus holds nearly 10,000 municipal and county law documents, amounting to about 7 million pages and millions of unique ordinances. It covers all 50 states, more than 3,000 counties and around 6,000 local jurisdictions, and it is free to use.
The material is the law that applies most often to most people: noise ordinances, construction and building codes, nuisance rules, dog leash laws, fishing licences and rules about roadside mango peddling.
Until now it was effectively unreadable in bulk. Local codes are published by thousands of separate bodies in thousands of separate formats, a great many of them as scanned documents.
The team built the corpus in layers. LightOnOCR, a vision-language optical character recognition system, extracted text and structure from the files; OpenAI models then tagged and categorised samples across the full corpus to produce standardised, machine-readable data.
The work was led by Diag Davenport, assistant professor of technology policy, governance and society, with Denis Peskoff, a postdoctoral scholar, Joe Barrow, an AI researcher, and Christopher Vu, an undergraduate.
"We've built a new kind of telescope—one pointed at governance itself," Davenport said.
The paper, "Freeing the Law with LOCUS: A Local Ordinance Corpus for the United States", has been accepted by the Conference on Neural Information Processing Systems and will be presented in December. A preprint is on arXiv.
The team expects the corpus to be used by researchers working on housing, zoning and the practical texture of American government, none of which has previously been measurable across the whole country.
The telescope is the right instrument, and this newspaper would urge that the comparison be taken literally rather than kindly.
A telescope does not bring its subject closer. It confirms that the subject is there, at the distance it always was, in quantities nobody can visit. Seven million pages spread across about 6,000 jurisdictions is roughly 1,200 pages each, which is now observable and remains, as before, unread.
There is a further property of telescopes that applies here. What the instrument shows is light that left some time ago. An ordinance scanned into the corpus is the law as it stood when the scan was taken, and a provision repealed since will go on shining in the data exactly as a dead star does.
The Department of Unnecessary Research notes the consequence for the oldest principle in the field, that ignorance of the law is no excuse. The law now has a page count. For the first time it is possible to state precisely how much of it a citizen has not read.
This is not a deterioration in anybody's position. Residents of about 6,000 jurisdictions were already bound by some 1,200 pages each, and complied, on the whole, by guessing.
Astronomy met the same difficulty and solved it by giving up on individual objects. LOCUS makes the equivalent move available to American law. The corpus can be surveyed, sampled and modelled, and the ordinances themselves need never be disturbed.