The laws closest to your life are the hardest for machines to read

The rules that decide whether you can build a fence, run a food truck, or keep three dogs aren't written in Congress. They're written by your city and county — local ordinances covering zoning, housing, noise, business licenses, public health. And these rules, the ones that touch daily life most, are almost completely missing from the digital text collections that train legal AI.

Why? Because local codes live on vendor websites built for a human to click through one page at a time, not for a researcher to download in bulk. Many exist only as scanned images, not searchable text.

LOCUS fixes this. The authors gathered nearly every publicly available local code into one collection — ordinances from 9,239 cities and counties. To turn scanned pages into usable text, they ran OCR (software that reads characters off an image, the way your phone copies text from a photo). On top of the raw pile, they built a cleaned, organized layer covering the 2,309 largest of America's 3,144 counties — together home to most of the U.S. population.

They also trained small AI models to rate ordinances on things like opacity (how hard the law is to understand) and paternalism (how much it restricts personal choice) — measurements never done at this scale. These scores are early tools, not settled verdicts on what any law actually does.

Why you should care: The regulations that most directly shape your home and business have been effectively invisible to research. LOCUS makes them searchable — so someone can finally measure, city by city, which local laws are needlessly confusing or controlling.