The Play — Issue #11: The Next Bad Citation in Your Case Won't Be Yours

A federal judge in Michigan struck the plaintiff’s brief last week. Then she struck the defendant’s, for the same reason. Both sides had signed a certification swearing they had personally verified every citation. Both certifications were worthless. This issue is about the bad citations you didn’t write.

On July 21, Judge Judith Levy of the Eastern District of Michigan struck objections filed by both parties in Evans v. Robertson. The court’s term for the problem is “factitious citations,” and this was not the first round. The parties were sanctioned in November 2025 for the same conduct. Magistrate Judge Elizabeth Stafford had ordered back in September 2025 that every future filing carry a written representation that the filer had personally checked each citation and confirmed it stands for the proposition attached to it. Both parties filed that representation. Both filings were defective anyway.

Here is the part that should change how you work. The cases were real. Plaintiff cited Mitan v. Campbell, 474 Mich. 21 (2005), for the proposition that a republication can restart the clock on a defamation claim. Mitan holds close to the opposite: one year from first accrual, with no extension for republication. Defendant cited Delay v. Rosenthal Collins Group, 585 F.3d 1003 (6th Cir. 2009), for the plaintiff bearing the burden on diversity jurisdiction, and Everett v. Verizon Wireless, 460 F.3d 818 (6th Cir. 2006), for the plaintiff bearing the amount-in-controversy burden. Everett puts that burden on the removing defendant. Run all three through Westlaw and they come back real, published, and good law. The verification most firms actually perform is an existence check. This failure mode passes it, cleanly, every time.

Now the volume question, because one case is an anecdote. Damien Charlotin’s tracker logged 1,809 decisions as of July 25, 1,250 of them in the US. Nine US entries are dated July 21 through 24 alone. And the composition has quietly inverted: pro se litigants appear in 1,060 entries, lawyers in 703. Misrepresentation of real authority shows up in 751 entries and false quotations in 490, so the invented case is no longer the only story, or even the most interesting one. If you defend companies against self-represented plaintiffs, or run any volume docket, the fabricated authority in your matters is increasingly arriving from the other side of the table.

Courts have started noticing who stayed quiet, and the consequence is already financial. Last September, in Noland v. Land of the Free, L.P., 114 Cal.App.5th 426, California’s Second District sanctioned an appellant’s counsel $10,000, payable into the court’s general fund, over briefs in which nearly all the quotations were fabricated, and referred him to the State Bar. The panel published the opinion as a warning, and its rule was flat: no filing should carry a citation the responsible attorney has not “personally read and verified.” The order to show cause had warned that sanctions might include fees and costs awarded to the defense. The court then declined to award the defense its attorney fees, because respondents had not detected the fabricated citations either. They took their appellate costs and ate the rest. That is the first published decision to put a price on failing to catch your opponent’s fake authority, and the price landed on the side that did nothing wrong except miss it.

Ten issues of this newsletter have told you to check your own work. The other side’s brief is the part nobody has been assigned.

Legora announced this morning that it is acquiring Wexler, a London startup whose engine runs case documents through a gated pipeline built to isolate discrete, verifiable events: who said what, to whom, when, and why it matters to the case. Not summaries. Not loosely sourced excerpts. It operates at over a million documents per case, and its users include Clifford Chance, Goodwin, and Herbert Smith Freehills Kramer. Eighteen people, revenue up eightfold year over year. This is Legora’s fifth acquisition of 2026, following a $600 million round at a $5.6 billion valuation. Wexler becomes the fact layer underneath Legora’s agentic workflows, and its engineers anchor a new London engineering hub.

Notice what got bought. Not another research tool, not another drafting assistant. Legora’s chief executive called fact-finding the most manual and expensive part of legal work, and expects fact queries to multiply by orders of magnitude as agents absorb more of it. The architecture is the tell: a pipeline deliberately built to refuse to summarize, because a summary is where provenance goes to die. That is a well-funded vendor concluding that the valuable property is not fluency but traceability. Which is the same lesson the Lead Play delivers from the opposite direction. The failure in Evans was not an invented case. It was a real source severed from what it actually said.

You don’t have to buy anything to act on this. Better still, a traceability claim is one of the few AI marketing claims you can falsify in an afternoon.

Legal IT Insider’s Orange Rag reported on July 28 that the legal team at Prosus, the technology group and investor, is running close to 300 agents. Each one is scoped to a single workflow. EU AI Act compliance scanning. Deal summary generation. Some connect directly into outside firms’ systems. Coverage runs across corporate, IP, compliance, and M&A. The account is explicit that these are not pilots bolted onto existing tasks. The team rebuilt the work around the agents instead of automating what its lawyers were already doing.

The lesson is not “build 300 agents.” Nobody reading this has that budget or that appetite, and you don’t need either. The lesson is the unit of work. Nearly 300 agents inside one legal department means the average agent does one narrow, boring thing. That is the exact inverse of how most mid-market AI projects get scoped, where the ambition is a system that handles contract review, and eighteen months later there is a pilot, a steering committee, and a license nobody uses. A narrow agent has a definable input, a definable output, and a test you can run on Friday afternoon. It also fails cheaply, which is the underrated feature.

Caveats, honestly: this is one company, reported secondhand, and the legal team of a technology investor is not a 90-lawyer litigation shop. Treat the number as directional. The scoping lesson holds whether the real figure is 300 or 30.

For ten issues the advice here has pointed inward. Check your output, preserve your prompts, fix your policy. This week it points across the table. Before you re-read your own brief, read their three best citations. See you in the next one.