docketx/us-pro-se

US Pro Se — what the courts themselves tell people who have no lawyer

Code & tools: github.com/docketx — legal-scrambler pseudonymises a case file on your own hardware before a frontier model sees it; claude-for-legal is the Claude Code plugin (docketx-open-law) that loads these datasets and checks citations against them.

12,103 documents from 19 state court systems: 2,835 self-help guide pages, 727 instruction documents and 8,541 forms, 119,978,550 characters of text, each held verbatim with the exact URL it came from.

Most people in a US civil case have no lawyer. The material that actually serves them — how to ask for a fee waiver, how to serve papers, what a small-claims hearing looks like, which form number to file — is published by the courts, scattered across fifty websites, and collected by nobody. This is that layer.

This is not the law

Forms and instructions are not statutes and not case law. A court's self-help page explains a procedure; it does not create one, and it can be out of date or wrong about your situation. Nothing here should be quoted as authority, cited as law, or treated as legal advice. Where you need the law itself, use docketx/us-statutes, docketx/court-rules and docketx/us-caselaw — and the kind column here (self-help-guide / instructions / form) is there so a retrieval system never confuses the two.

What is in it

Every count is measured from the .jsonl.gz file in data/ on 2026-09-11 — not estimated, and not taken from any index page's own total. "still crawling" means the harvester's own queue is not yet empty for that state, so its count will grow; every other state's crawl ran until it had nothing left to fetch.

Fields

How it was collected, and what we refused to take

The first request to every host was https://<host>/robots.txt, sent as DocketRouterBot/1.0 (+https://docketrouter.ai/bot). That file is stored verbatim next to the harvest and its rules are enforced in code:

• —a Disallow matching a path we wanted means that path was not fetched;
• —a robots.txt we could not read — WAF 403, HTTP 401 — means the whole host was skipped, because unverifiable is not permission;
• —Crawl-delay was honoured, with a 1 request/second floor regardless;
• —we never changed user-agent to get around a block.

Per-state consequences of that policy:

• —Alabama (`al`) — Alabama's official self-help layer is genuinely this thin: the State Law Library publishes the small-claims guide and the appellate snapshot, and nothing else.
• —Iowa (`ia`) — `ai-train=no` — see the reservations section. Iowa also declares Crawl-delay: 30, which we honour, so this state fills in slowly: its crawl is still running.
• —Idaho (`id`) — Idaho Supreme Court tier only. The Court Assistance Office forms library (courtselfhelp.idaho.gov) sits behind a WAF that 403s its own robots.txt, so it is unverifiable and was not crawled.
• —Illinois (`il`) — Form PDFs are served from ilcourtsaudio.blob.core.windows.net, whose robots.txt returned HTTP 401 — unreadable, so that host was not crawled. HTML pages only.
• —Maryland (`md`) — mdcourts.gov/robots.txt carries Disallow: /courtforms — the Maryland forms library is excluded by robots and is not here.

Reservations that travel with the data

• —Iowa — iowacourts.gov/robots.txt carries, verbatim, Content-Signal: search=yes,ai-train=no,use=reference. That is a separate grant from Disallow, and we honour it: the Iowa records are for retrieval and reference only and must not be used as training data. The reservation is stamped into every Iowa record's license field.
• —Texas is absent because its host stopped answering. texaslawhelp.org/robots.txt had been readable (HTTP 200) when the source was verified on 2026-09-10, but on the harvest run it returned HTTP 202 with an empty body — a bot-management interstitial, not a robots file. Unreadable means unverifiable, so the crawl aborted before its first content request and Texas is not in this dataset. (Separately, when that host was readable its terms reserved commercial use: "Free. Not for sale ... you agree not to sell or make a profit in any way from any information or forms that you obtained through this website.")
• —Alaska is deliberately absent. Its source is verified and harvestable, but the Alaska Court System asserts copyright over its self-help publications — the Small Claims Handbook cover reads "reproduction of any part of this handbook for commercial purposes without the express written permission of the Alaska Court System is strictly prohibited." Rather than redistribute it, we hold Alaska pending a licensing decision. It is withheld, not missing.
• —Everything else is an official US state judiciary publication. State government works are not automatically public domain the way federal works are, so we do not claim a licence the source does not grant: each record's license field says exactly what the source asserted, which for most of these is nothing at all.

A state that is not in the table is UNVERIFIED, not withheld

That is the standing rule for every DocketX dataset. A missing state means we have not yet proved its source live — official host, robots.txt read for our honest bot, and one real passage extracted — not that the state refused. Twelve jurisdictions (ct dc fl ks ma mn nc nd nj ny ok sc) were checked on 2026-09-10 and are blocked today, mostly by WAFs that refuse our robots.txt request; they stay out until that changes. The rest are simply not done yet.

Companion datasets

docketx/us-caselaw · docketx/us-statutes · docketx/court-rules · docketx/us-judges