About stare.pub

stare.pub resolves U.S. case citations to the full text of the decision and to the original scanned pages of the reporter it was printed in. Type 31 A.2d 647 and you get the opinion. It's a Codex Publicus project, built by lokkju.

Everything runs in your browser. The site, the citation index and the case data are all static files: there's no server doing work on your behalf, no database, and no API to go down. There's no account to create and nothing to sign up for. I count visits with Cloudflare Web Analytics, which sets no cookies and reports only aggregate numbers.

stare.pub aggregates published case law from the Caselaw Access Project and CourtListener, and produces it as static data you can host yourself: the citation index, the case records, the reporter map and the schemas, served as plain files anyone can mirror, with no account, key or per-request server. Agents and tools use those same files with plain GETs, or through tools that run on their own machine (the stare command line tool, in alpha, and a planned local MCP server) rather than a hosted API. llms.txt has the recipe, and the data page describes the files.

Published case law is the law. Reading it shouldn't depend on a subscription, or on any one institution's continued goodwill and funding.

The data

stare.pub holds 7.8 million decisions from 1658 to 2026, from the Caselaw Access Project and CourtListener.

Most of the cases come from the Caselaw Access Project (CAP). Between 2013 and 2018 the Harvard Law School Library scanned its collection of American case reporters: 40,622 volumes holding 6.9 million cases, decided between 1658 and 2020. The case data is public domain.

Ravel Law paid for the scanning in exchange for a period of exclusive commercial use, which passed to LexisNexis when it bought Ravel in 2017 and ran until March 2024. When it ended, Harvard published the whole corpus as bulk files at static.case.law, and on September 1, 2024 it retired the CAP website and its API. The bulk files are still up. Nobody has said for how long.

Codex Publicus holds a complete copy of that release, all 10.4 TB of it: the case text, the page scans, the OCR and the metadata. stare.pub reads its case text from that copy.

The rest come from CourtListener, the case law archive run by the Free Law Project. From CourtListener's bulk data stare.pub takes the opinions and metadata of cases CAP doesn't hold, including the decisions after CAP's last volumes, and the parallel citations CourtListener records, so a citation to one reporter finds the same decision printed in another. CourtListener marks its bulk data with the Public Domain Mark 1.0, "free of known copyright restrictions." The live index, v2026.10.26.2, is built from CourtListener's export of 2026-09-30, and data.stare.pub serves 900,342 case records from it. Each record names CourtListener and Free Law Project as its source, and the case view links the opinion on CourtListener. Those cases have text and metadata but no page scans.

How it works

When you look up a citation, your browser:

• fetches the volume's citation index, a gzipped text file usually under 2 KB, which says which case starts on which page;
• fetches the volume's case data, a zip of typically 1.5 MB, from the Codex Publicus copy at data.stare.pub, or for a case from CourtListener, that one case's record, a single gzipped JSON file;
• unpacks and renders the case itself, using a WebAssembly build of the same Rust code the stare command line tool runs;
• caches both files, so later lookups in that volume don't download them again.

A case has three views. Reader is its text. Archival PDF shows the case's pages from our archival PDF of the volume: the bitonal scans, with the OCR as a text layer you can select and copy. The PDFs are linearized, so the viewer reads the file's header and cross-reference table and then only the pages you look at, a few hundred KB of a file that runs to 84 MB for the largest volume. Raw Scans shows CAP's original color scans. Each page is a color JPEG of about 180 KB inside the volume's tar archive, which typically runs around 200 MB; the viewer reads the archive's table of contents and fetches only the bytes of the page you're looking at.

Where it stands

Done:

• The mirror. All 40,622 volumes of the CAP release are copied, counted and held read-only.
• The index. 40,622 CAP volumes are indexed and resolve today.
• Citation detection. citare, a Rust citation parser I wrote, finds the citations in every case you read and in anything you paste into the brief checker, including plain forms like 31 A.2d 647.
• Our own case text. Case data is served from the Codex Publicus copy, not from static.case.law.
• CourtListener. 900,342 records of cases CAP doesn't hold, from CourtListener's export of 2026-09-30, are published, and the index points their citations at them; CourtListener's parallel citations link a case's other printings to it.

In progress:

• Archival PDFs. One PDF/A per volume, built from the original bitonal scans with lossless compression, with the OCR as an invisible, searchable text layer and a bookmark for every case. Every page is checked against its source scan, pixel for pixel. They come out about a quarter smaller than CAP's own PDFs of the same volumes, which have no text layer at all.
• Structured case text. Every decision in Akoma Ntoso, the OASIS standard for legal documents, with a stable identifier for each opinion.

Planned:

• Full-text search, served as static files like everything else.
• Statutes and regulations: the U.S. Code, the Statutes at Large and the state codes, from the public sources that publish them.
• IPFS and torrents, for anyone to pin and seed.
• The site itself on IPFS, where the whole system can outlive any single host, this one included.

What's new

All updates · Atom feed

Help preserve case law

Harvard has kept the bulk files online so far. One host is still one point of failure, and every independent copy of the full release is one more place it survives if static.case.law goes away.

• Mirror it. The whole release is at static.case.law. If you have 10.4 TB to spare, take a copy, and keep it to a request or two a second.
• Pin it on IPFS, once we publish the CIDs.
• Seed it, once the per-reporter torrents exist.

Built on

• Caselaw Access Project: most of the cases on this site and every page scan, scanned and transcribed by the Harvard Law School Library and its Library Innovation Lab. The case law is public domain (CC0).
• CourtListener: the cases CAP doesn't hold and parallel citations, from the Free Law Project's bulk data, which it marks with the Public Domain Mark 1.0.
• citare: the citation parser behind the reader and the brief checker, written in Rust and compiled to WebAssembly. It's a Codex Publicus project, inspired by the Free Law Project's eyecite, and it uses their reporters-db for its reporter data.
• Akoma Ntoso: the OASIS standard for legal documents, which the structured case text will follow.
• PDF/A: the ISO archival PDF standard the per-volume PDFs are built to.
• Rust: the core lookup code, compiled to the stare command line tool and to WebAssembly for the browser. The index is one small gzipped text file per volume, a line for each case's first page, so the case on any page is one lookup away.
• Astro: the site, as plain HTML, JavaScript and CSS.
• Cloudflare: Workers serve the site and R2 holds the index and the case text. Any static host would do, whether S3, IPFS or a directory on your own machine.

stare.pub is one of the Codex Publicus projects. Write to info@codexpublicus.org with questions, corrections or a mirror you run.