Most firms can't answer that. Ratia can. We build the tools and services that measure legal AI against the court-adjudicated record — is the case it cites real, did it read the holding correctly, is it still good law — so you get AI's enormous upside without betting your name on an unverified answer.
On the list. You'll see the Minnesota results first.
Something went wrong — please try again.
Early access is opening to a small group of design partners.
Every legal-AI tool reports impressive accuracy. Almost none of it is measured against what a court actually held. Tools cite cases that don't stand for what they're quoted for, miss that a holding was reversed, and say all of it convincingly.
This isn't hypothetical. Courts are sanctioning lawyers for it, right now:
The headline says "AI." The facts say software: one lawyer's fake cases came from a legal drafting platform, the other's from a research tool that wasn't designed for her jurisdiction's law. All four lawyers disqualified; the trial cancelled.
A New York appeals court fined a lawyer and his firm for a brief citing nonexistent cases and fictitious quotations — AI-assisted research he failed to verify.
One of the most prestigious firms in the country — with AI safeguards on paper — caught by opposing counsel, not by its own review.
Judges have warned. Courts have sanctioned. The filings keep coming — because warnings don't verify anything. Measurement does.
Read those stories closely and "AI" is never the whole story. A specific tool did it — a product, a version, a configuration. In the Mississippi case, one of the tools wasn't even built for that state's law. Some tools are grounded in the jurisdiction you practice in. Some aren't. The headline can't tell you which is which. An audit can.
Nobody independent is checking the brakes. The builders grade their own work, and the buyers can't tell grounded from fluent — because grounding is a lawyer's knowledge, not a software feature.
The upside of AI in legal work is enormous. The sanctions don't say "don't use it." They say "don't use it blind."
The lawyers getting sanctioned aren't fools — they're using a powerful tool without a way to know what it actually does. That's not an AI problem; it's a measurement problem. Understand your tool and AI is the biggest leverage the profession has seen. That understanding is what we build.
The hand-wringing about AI isn't really about AI. It's about not knowing what it does. We know — because we measure it, case by case, against the court's own words. That's the difference between fear and diligence.
Is the cited case real, and does the opinion actually say what the tool claims — checked at the pincite, not a headnote, not a summary.
Did the tool read the holding correctly on the issue that decides the matter? The court is the ground truth — never the tool, never an editor.
Was it reversed, overruled, or superseded? A case can exist, say exactly what's claimed, and still be dead law. Existence is not validity.
Evaluating a legal-AI tool before you commit? We test it against court-adjudicated ground truth so you choose on measured grounding — not on the demo.
Your firm already relies on AI — in research, drafting, review. We audit what it's actually doing on your jurisdiction's law, before a court finds out for you.
Every engagement delivers the Ratia Report:
How the tool performed on each check — existence, holding, validity — reported by court level, never blended into one flattering number.
How the tool fails when it fails: invented cases, misread holdings, dead law cited as good. The section that tells you what you're actually risking.
Blind scoring, independent attorney review, ground truth anchored to the court's own words at the pincite. Built to survive an expert's cross-examination.
A Minnesota pilot is underway, with ground truth built from the state's court-adjudicated record. We're starting in one jurisdiction, proving the measurement, and opening it to a small group of design partners before it goes wider.
Built on the conviction that grounding is measurable — and that someone independent has to measure it.
On the list. You'll see the Minnesota results first.
Something went wrong — please try again.
No spam. One note when early access opens.