Only a few readers—other than my parents—seem to have noticed that my column, The Situation, sometime in early April suddenly stopped continuing tomorrow.
The reason was not that The Situation itself had been substantially ameliorated. It has kept right on situationing along over the past two months—almost as though it doesn’t need my column.
A variety of factors led to the hiatus. I needed a break. Certain Lawfare editing and managerial matters required my sustained focus. Also, I was building this thingy which we are rolling out this week in a limited beta release to Lawfare’s material supporters.
I know what you’re thinking. You’re thinking, “But wait! I’m not a Lawfare material supporter? How do I get access to the limited beta release of RAGtime?”
Well, first off, shame on you! You should be a material supporter of Lawfare. And the good news for you is that you can become one right here and right now. And as soon as you do so, you will suddenly be eligible for the limited beta release of RAGtime.
There. Doesn’t that feel better?
I also know what you’re thinking now. You’re thinking, “But hang on! What is RAGtime?”
I’m glad you asked, if only in your thoughts.
RAGtime is a set of tools I have been developing over the last several months in the course of helping some of my colleagues develop and analyze datasets for use in doing rule-of-law journalism. It all began when Lawfare Associate Editor Katherine Pompilio approached me with the problem of how to identify all of the instances in which the government had violated court orders in immigration habeas cases—by failing to release illegally-detained individuals, refusing to provide information about immigrants’ whereabouts, or declining to return personal property like driver’s licenses. It’s hard to find these instances. Pompilio had identified roughly 180 such cases—significantly more than had been identified in published news stories—but she had run out of obvious leads for identifying them manually. She knew there were more, but how do you find them in courts all over the country?
It’s actually a tricky problem. Databases don’t code for violated court orders. They can be expensive to query. The cases do not all cite each other. And the violations of court orders do not generally get noted in published opinions.
Pompilio’s problem led me to explore an interesting hypothesis: What if the best way to identify these cases lay in the texts of federal dockets? The theory was that one doesn’t need to read the whole case or its actual orders to figure out which cases are at high risk of containing violated court orders. One need, rather, read only the list of documents filed in the cases, which often contains key language signaling judicial fury at government government conduct.
So I downloaded every federal docket in every habeas case since the beginning of the second Trump administration. And I trained Anthropic’s Claude on the dockets Pompilio had identified as containing actual violations of court orders. I then had it read the tens of thousands of other habeas dockets to flag for Pompilio the ones likely to yield fruit. This gave her a dataset of many hundreds of cases, most of which were duds. But more than 200 were not duds. And she emerged with a list of 415 cases in which we could document actual violations of actual court orders.
I did what any red-blooded American would do in response: I downloaded the dockets of all federal litigation filed in the United States since Jan. 20, 2025, and I built an AI-enabled search engine on top of it. This was the core of what became RAGtime. Other journalists—some at Lawfare and some elsewhere—began asking me to create datasets to examine specific questions.
Could we identify all criminal cases against protesters? Why yes!
What about cases brought under a particular statute? No problem.
Could we produce lists of cases that had a high probability of reflecting politically-motivated abuse of law enforcement? Funny you should ask.
What about civil cases challenging DOGE actions that had substantial administrative records? Done.
The database, it occurred to me, would be more powerful still if I started ingesting large numbers of the underlying documents within the cases: the complaints, the indictments, the opinions, the arguments over dispositive motions.
The result was RAGtime’s federal court litigation corpus, which currently contains 1.2 million case dockets, nearly 6.9 million docket entries, and more than 380,000 key documents within those cases.
The idea was not to make this material public. The material is already public—thanks to the magnificent work of the Free Law Project and its Courtlistener database. The idea is to make it searchable across jurisdictions, across case types, and across levels of the judiciary using both traditional, local keyword searches and also using AI-enabled analytical tools in a fashion tuned for investigations, journalism, and other specialized research applications.
The goal, to be clear, was not to have an AI answer research questions for humans. It was, rather, to enable the use of an AI to go through impossibly large, impossibly messy piles of datasets to tee up useful, manageable questions for human review. As I write these words, the litigation database continues to populate. It is adding new documents. It is pushing its start date back before the advent of the second Trump administration. It can answer an astonishing array of questions, and it gets more powerful by the day.
I initially started building RAGtime for Lawfare’s own internal use. But then it occurred to me to turn it outwards and make it available to other journalists. And for that matter, why not make it available to everyone else too?
Litigators can examine how the judge presiding over their case thinks about a particular legal issue that never goes up on appeal, or how an opposing party tends to conduct itself during discovery, or whether a particular witness has been problematic in other proceedings, or whether an attorney has been misbehaving but so far evading actual sanctions.
Activists planning protests or community organizers strategizing campaigns might want to get a sense for which demonstrations of dissent are being targeted and how to best protect their exercise of First Amendment rights. And if they’re already facing charges, they can dig up evidence to support motions to dismiss for things like vindictive prosecution or due process violations (“see, judge, it’s not just our case—there’s a pattern of government conduct here!”).
Researchers might be interested in the ability to do macro studies of what types of cases are being brought in what jurisdictions—and then to be able to zoom in on the particular cases that make up their macro data.
And so what started as a vibe-coding project to support my reporters turned into a software development project for the public at large.
As the ambition for RAGtime grew, it occurred to me that the tool would be even more powerful if I gobbled up non-litigation documents. Why not, say, ingest all public Office of Legal Counsel (OLC) opinions—so that people might trace the development, say, of OLC’s view of war powers over time?
And why not also the United States Code and the Code of Federal Regulations, so that people could do detailed AI-enabled searches of the actual law?
My colleague Scott R. Anderson was doing research on the history of war powers and was interested in doing searches in the Foreign Relations of the United States series, which contains hundreds of volumes published by the State Department that record “the official documentary historical record of major U.S. foreign policy decisions and significant diplomatic activity.” So I ingested that too.
And then there was Lawfare itself, which is perhaps the largest database of thought in the world on the subject of national security law and policy. Why not make it possible for a user, as I did the other day, to ask: “Do Benjamin Wittes and Jack Goldsmith agree or disagree about presidential immunity?” and get a useful and nuanced summary report of 20 Lawfare articles?
Over the next several months, we will be ingesting large numbers of new datasets into RAGtime. We will be integrating them with one another and developing tools to make them ever-more useful. And as soon as it is ready, we will be making it available to the general public.
We are not charging for access to RAGtime. We are limiting beta access to it to material supporters of the site, because the system is not yet robust enough to support huge numbers of users and we have to limit the size of the user population somehow. The goal, however, is to make it available publicly. Unfortunately, API calls to AI companies are not free—not free to us, and thus not free to the user. Users who have their own API keys can use those on RAGtime and pay us nothing at all. Users who need to use Lawfare’s API access are welcome to do so, and we have arranged a means for them to reimburse the Lawfare Institute the cost of the API calls.
So please check out RAGtime, which will be growing substantially over the next few months.
And I will resume my examinations of The Situation soon—though not quite tomorrow.
Note: A video demonstration of RAGtime is available here and here.
