Frequently asked questions
Straight answers to the questions litigators and their firms ask before adopting Supreme Mind. For the binding terms see our Terms of Service and Privacy Policy; for the security posture in depth see Security & trust.
Security & data
Where does our matter data live, and who can access it?
Firm accounts are stored in a US-region Postgres database and a US-region file store, isolated per firm at the row level. Only your firm's authenticated users can reach your matters, and no other firm can ever see them. Sign-in is passwordless and tied to a named individual.
Do you use our data to train super intelligence (SI) models?
No, and not by us either. Model calls go to Anthropic’s API under a zero data retention agreement, with inference pinned to the US: prompts and completions are not stored after the response is returned and are not used to train any model. Amazon Bedrock, in our own Amazon Web Services account, is held as a backup route on the same terms. Your matter content is not used to train any model, ours included.
How do you protect privilege and confidentiality?
You decide what to upload. Content is firm-isolated and US-stored, model calls run under zero data retention with US inference, so nothing is retained, and every call writes a metadata-only audit record (who, when, which task, which model and region). Privileged content is never written to logs.
Can we delete our data, and undo an accidental delete?
Yes. Deleting a matter removes it, its uploaded documents, and the deliverables generated from it after a 30-day recovery window. Within that window the matter can be restored, by the person who deleted it or by a firm administrator, so a mistaken delete is reversible.
What happens to our matters if Supreme Mind stops operating?
You get at least 60 days’ notice by email. In that time every brief stays downloadable and we send each firm an export of its matters and documents on request; afterwards all matter content is deleted within 30 days and we confirm it in writing. A sale of the company does not change these commitments. The detail is on Security & trust.
The experts
Are the experts real people?
No. Each one is a de-identified composite representing a class of expert witness, anchored to publicly documented methods and prior testimony. It is not any specific named individual and is not affiliated with, sponsored by, or endorsed by any real expert or firm. More on what a class is in How it works.
What exactly are you simulating?
A probabilistic synthesis of how a class of expert testifies, for example a defense event-study econometrician or a cartel-overcharge damages economist, bounded to the public record. Supreme Mind uses it to model the opposing expert's likely opinion, cross-examination weaknesses, and Daubert exposure on your matter. How it works sets out what a run reads and returns.
What is the settlement engine?
The next layer, and it is in development. Today a run simulates one expert on one fact pattern and returns a brief. The settlement engine argues the same matter out across a thousand scenarios, your experts against the other side’s, and returns one strategic recommendation: the spread of outcomes rather than a single read. The top five percent shows where your leverage is and what to build strategy from; the bottom quarter shows the paths that go wrong, so you can avoid them before settlement. If you want a say in how it is built, see pilots.
Which practice areas are covered?
Antitrust, securities litigation (Section 10(b) and Delaware Chancery), mass tort, personal injury and commercial litigation, each with its own library of expert classes, from the event-study econometrician to the causation toxicologist. Intellectual property, medical malpractice and other practices follow the matters firms bring. The expert class library describes the classes of expert by practice, and the quickest way to see how we would approach your matter is a demo.
Methodology & accuracy
Is this just a language model guessing?
No. The analysis is grounded in real data: event-study econometrics on market prices for securities matters, a cost pass-through analysis on public FRED input-cost series for antitrust, and Daubert and prior-testimony records from public court filings. Supporting quotations are verified against the public source, and any that cannot be verified are dropped. See How it works for the full method.
Can we rely on the output in court?
Treat it as decision-support, not legal advice. Counsel of record retains full professional responsibility, a credentialed human still testifies, and you should independently verify every fact, citation, quotation, and figure before use.
Could using this get our expert excluded?
The line courts are drawing is between judgment and substitution, not between using SI and not using it. In Kohls v. Ellison, No. 24-cv-3754 (D. Minn. Jan. 10, 2025), ECF 46, an expert’s declaration was struck because he had used a model to draft it and it carried fabricated citations. In Ferlito v. Harbor Freight Tools USA, Inc., No. 2:20-cv-05615 (E.D.N.Y. Apr. 23, 2025), ECF 48, the court denied a motion to exclude an expert who wrote his report from decades of experience and then used a model to confirm it, finding “no indication that Lehnert used ChatGPT to generate a report with false authority.” The same order that struck the declaration in Kohls was explicit that it did “not fault Professor Hancock for using AI for research purposes.”
Supreme Mind sits on the permitted side by construction, and deliberately. Nothing we produce is filed, disclosed under Rule 26, or offered as a testifying opinion. Your retained expert forms their own opinion, writes their own report and signs it. What we produce is internal preparation: what the other side’s expert is likely to argue, and where it is weak. Every supporting quote is checked verbatim against the public record and dropped when it cannot be verified, so there is no path by which a fabricated citation reaches a filing through us. How each court of appeals now applies the amended rule to experts generally is tracked on our Rule 702 tracker.
Discovery into how an expert used a model is an open question in at least one active case, so the safer position is a tool whose output traces back to a public source a reader can open, rather than one whose reasoning exists only in a prompt history. That is how ours is built.
How do you know the simulations are any good?
Two ways, and they are worth keeping apart. Every brief shows the data it is grounded in, and its quotations and citations trace back to a public source you can open; figures and the analysis itself you should still check. The quote and citation check is verifiable today, on your own matter, and it is the one we would ask you to judge us on now. The detail is in How it works.
Accuracy against outcomes is harder to prove honestly. A model trained on the public record may already have read how a decided case ended, so scoring it on decided cases measures recall as much as judgment. We run those comparisons internally and we do not present them as a track record. What we do instead is record a prediction on a pending public docket before it resolves, timestamped by the database rather than by us, and score it when the docket ends. Nothing there can be quietly revised or dropped after the fact. That record is young, so we are not quoting an accuracy figure from it yet, and we would rather say that than offer you one we cannot defend. What we do publish is a model of what better expert preparation could be worth in a close case, built on public research, with every input and how strong it is on the research page. It is an estimate, not a track record.
Access, pricing & control
Who at our firm can see our matters?
Matters are firm-scoped by design: your team shares one case file and collaborates on it, rather than each person keeping a private copy. Firm administrators manage users and firm-wide controls such as bulk deletion and recovery.
How is Supreme Mind priced?
Work is billed in credits, and one credit is $50. A full Exposure and Settlement Brief is a single run; lighter tasks such as cross-examination and record examination are priced below that in proportion to the computation they use. Example cases are free on a firm email address. Credits are bought in the app with a card, in any quantity you like: four is $200, a hundred is $5,000, and the rate is $50 either way. They sit with the firm rather than with a matter or a person, so they spend across as many matters as you want and anyone on your firm's email domain draws on the same balance. For a matter you expect to work hard, a matter plan gives unlimited runs on it for a year, from $25,000, invoiced. The full list is on the pricing page rather than provided on request.
Is our data ever shared with other firms?
No. Isolation is per firm, and the outputs generated for you are yours. We may generate similar outputs for other users from the same public record, but your content is never exposed to anyone outside your firm.
Getting started
How do we start?
Start on an example case, which is free and needs nothing but a firm email address: choose an expert and read a complete brief on a real public matter. When you want to run your own, you buy credits in the app with a card, one credit or a hundred, and upload the case file. For a practice group that wants several matters at once, a pilot sets out the shape: three months, named matters, and the criteria for success agreed before the first run.
What do you need from us to run a matter?
The public case materials you already have, such as the complaint, the opposing expert's report, and key filings, plus the basic facts of the case. Supreme Mind extracts the facts and grounds the analysis from there.
Have a question that is not answered here? Email richard@suprememind.ai.