Lawfare

Talking about RAGtime: the public-language guide

One line: the shared answers to "what is RAGtime?", at every length from one breath to two minutes, plus the habits that keep us accurate when we say them out loud.

Why this exists

We keep improvising the answer to the simplest question the project gets. The concept note runs nine pages and the skeptic brief runs seven, but neither fits in a hallway. The engineering habit is to answer "what is it?" with architecture, and the listener hears database words and stops listening. This doc ends the improvisation: same product, described the same way, by both of us, at whatever length the moment allows.

Two documents sit underneath it. why-not-point-at-apis is the full-depth conversation with a sharp skeptic. docs/reference/ragtime-concept-note-v3.md is the strategic narrative at full length. This doc is the spoken layer on top of both.

The ladder

Five lengths. Start at the top and go down only while the listener is still asking.

One sentence

One place to search the public record of American government.

One breath (the party answer)

It's a free-to-search research tool from Lawfare. The record of what our government does (court cases, laws, regulations, legal opinions) is public but scattered across a dozen clunky websites that don't talk to each other. RAGtime puts it all in one place, and you can ask it questions in plain English.

Thirty seconds (the elevator answer)

The record of how American government actually works is public: every federal court filing, every statute, every regulation, every formal legal opinion. But it lives on separate government sites, each with its own bad search box, and none of them connect. RAGtime, built by Lawfare, brings those records together. Searching is free. If you want more, you can ask questions in plain English and get answers where every claim links back to the underlying document, so you can check the work. Tools in this class normally cost thousands of dollars a year. This one is built for journalists, researchers, and the public.

The paragraph (for print: an about page, a description in someone else's piece)

RAGtime is a public research platform built and operated by the Lawfare Institute, live at ragtime.lawfaremedia.org. It brings the primary-source record of American government into a single searchable interface: federal court litigation, the U.S. Code, the Code of Federal Regulations, Office of Legal Counsel opinions, presidential documents, the Federal Register, congressional materials, the State Department's documentary history of U.S. foreign relations, and a growing list of other collections. Filtering and keyword search are free on every collection. AI features run on top: plain-English questions, analysis across collections, close reading of individual documents. Every assertion the AI makes is cited to a checkable primary source. RAGtime indexes and analyzes the record; it does not annotate it, editorialize about it, or offer legal advice.

Two minutes (the story version)

When you have the floor, tell it as a story, because it is one.

In early 2026, Lawfare set out to answer a specific question: in how many federal cases was the executive branch failing to comply with court orders? Doing it by hand, reading orders and following tips from lawyers, produced about 180 cases. So we built a computational method instead. We scanned all 94 federal district courts, collected metadata on more than 700,000 cases, trained a scoring system on more than a thousand human-labeled examples, and used AI to triage candidates for human review. The tracker grew to more than 450 confirmed cases across 45 districts, published on Lawfare.

Along the way we noticed what we had actually built: the most complete searchable archive of recent federal litigation outside the commercial legal databases — more than 1.5 million cases as of July 2026. The archive was worth more than the tracker. So we generalized it. We added the U.S. Code, the federal regulations, the Justice Department's formal legal opinions, the diplomatic record, and kept going; as of July 2026 there are eleven collections, and the list is growing.

The design has firm rules. Search is free on every collection; the paid part is the AI, not the data. Every claim the AI makes has to cite the document it came from, so you can always check it. And the platform never editorializes: it gives you the record, not our opinion of the record.

The point is to make a class of research cheap that used to be expensive or impossible. A reporter can move from one anecdote to a national pattern. A scholar can study judicial behavior at a scale that manual methods never allowed. A nonprofit can track enforcement without a commercial-database budget.

As long as it takes

Past two minutes, stop performing and open the source documents. The concept note has the full narrative and the use cases; why-not-point-at-apis has the honest version of every hard question below. No live conversation should go deeper than those docs. If one does, the doc needs updating, not the speech.

Saying it without the jargon

The internal vocabulary does not survive contact with a general audience. Translations:

We say internally Say instead
corpus, corpora collection, body of records
semantic search, embeddings, vectors search by meaning, not just exact words
RAG, retrieval-augmented generation the AI answers from the documents and cites them
hub and spoke search everything at once, or go deep on one collection
hallucination the AI making things up
ingestion, pipeline adding new collections
docket entries every filing in a case
MCP connector plugging RAGtime into your own AI tools
free metadata floor search is free

The test: if the sentence would need a footnote, say the footnote instead.

The exception runs the other way with a technical audience. For developers and AI researchers the internal vocabulary is the correct register, and translating down reads as evasive.

The name

People ask, and the answer is short:

It's a pun. The RAG part is a technical term for AI that has to quote its sources. And ragtime is American music.

Written form: RAGtime, capital R-A-G. If the asker is technical, expand the acronym (retrieval-augmented generation) and move on; the pun does not improve with explanation.

What we always say

Four commitments recur in every version of the pitch. They are commitments, not marketing:

What we don't say

Hard questions, short honest answers

Longer versions of all of these live in why-not-point-at-apis. The short forms:

"Isn't all of this already free online?" Yes, and search on RAGtime is free too. What you pay for, if you pay at all, is AI computing time, not the records. The honest pitch is what the free sites can't do: some of our collections exist nowhere else in usable form, the records are connected to each other, and we keep a durable copy of a record that sometimes gets taken down. Where a free site serves a single collection well, we say so.

"Won't the AI just make things up?" The free tier has no AI in it at all, just filters and keyword search over the records. When you do use the AI, every claim links to the document it came from, so you never have to take the machine's word for anything. That discipline is the design, not a disclaimer.

"Why is a media organization building software?" Because we needed it. We built the first version to count cases where the government was defying court orders, and no existing tool could do it. The archive that came out of that reporting turned out to be worth more than the story, so we kept building. Then tell the origin story if there's time.

"Is this legal advice?" No. It's a research tool over public records, for journalists, researchers, and anyone who needs the record itself. What you do with the record is between you and, if you have one, your lawyer.

"Are you competing with CourtListener and the free-access projects?" No. We build on their work and credit it, and for a single collection the canonical free site is often exactly where you should go. Our value starts when a question crosses collections.

"What does it cost?" Give the structure, not the prices, because prices go stale: searching is free; AI features cost money because they cost us money, paid either through your own AI account or prepaid credit with Lawfare. For current specifics, open the site rather than quoting from memory. While the supporter beta lasts, note the gate honestly: full access is for Lawfare supporters and volunteer beta-testers today, and the mass public release is the next milestone.

Reading the audience

Audience Lead with Careful about
General public The party answer, then the name pun Any word from the jargon table's left column
Journalists The tracker story; from anecdote to pattern Sounding like a vendor. Offer a query, not a speech
Academics Research at a scale manual methods never allowed Overclaiming completeness; name the windows and gaps
Lawyers, legal orgs Coverage they can't otherwise afford; citations on everything Anything that smells like legal advice or Westlaw-killing
Technologists The real architecture, jargon welcome Unannounced product plans
Lawfare readers Lawfare's product, same editorial values Implying commentary and the record are blended

Rules of the road

  1. Problem first, technology last. Nobody outside the field cares how it works until they care what it's for. The scattered-record problem is the hook; the architecture is an appendix.
  2. Concede what's true. The strongest credibility move we have is saying "for that, the free site is fine" out loud. Overclaiming is what a sharp listener waits for. This is why-not-point-at-apis compressed to a habit.
  3. Date every number. Stale numbers stated as current are small lies with long lives.
  4. The story beats the feature list. The compliance tracker is the best two minutes we have. Use it.
  5. Stop early. The ladder exists so you can stop at any rung. Ending on the party answer with the listener still curious beats reaching the architecture with their eyes glazed.

Related

Feedback on these talking points?

Something reads wrong, or people keep asking a question this page doesn't answer? Tell us here — every note lands directly with the team.