How it works

One run simulates one class of expert on one fact pattern and returns a brief in four sections. This page sets out what goes in, what comes out, how it is checked, and where the limits are. To see the output itself, read the sample brief on In re Luckin Coffee; it is complete and needs no account.

What a run reads

Your 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 from them: parties, class period, disclosure dates, the issues in dispute. You decide what to upload, and you are never required to upload anything beyond a caption and a docket number.

Real data, not only text. The analysis is grounded in data the expert would use: an event study on actual market prices around the disclosure dates 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. In the sample brief, that is an OLS market-model event study on Luckin’s prices against the S&P 500, SEC EDGAR filings, and the judicial record on the methodology from CourtListener.

What a run returns

• Likely opinion: what the expert will conclude, in their own voice, with the methodological basis.
• Cross-examination weaknesses: ranked, with the question to ask, the deflection to expect, and citations to the case file.
• Methodology challenges: the Daubert and Rule 702 attack surface, with the authorities looked up (see how each circuit applies the amended rule in the Rule 702 tracker).
• Settlement range: how the testimony moves it, directional, with the uncertainty stated.

The same matter can then be put to work: rehearse the cross-examination live against the simulated expert, stress-test a report before it is served, or ask the case file a question and get an answer cited to the page.

How every quote is checked

Every cited authority is located by name in public case-law databases at generation time. Every supporting quotation is checked verbatim against the public source, and a quote that cannot be verified is dropped, not guessed. The passage a run relied on can be read in place, so a quotation or citation traces back to a source a reader can open. Figures, calculations and the analysis itself are not checked this way; counsel should verify them before relying on them. The quote and citation check is the part of the product you can verify today, on your own matter, and it is the one we ask to be judged on.

What a class of expert is

Each expert is a de-identified composite representing a class of expert witness, for example a defense event-study econometrician or a cartel-overcharge damages economist. It is anchored to publicly documented methods and prior testimony and bounded to the public record. It is not any specific named individual, it is not affiliated with, sponsored by or endorsed by any real expert or firm, and it never claims to predict what a named individual will say in a specific deposition.

What we do not claim

• No accuracy percentage yet. 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 do not present those comparisons as a track record. Instead, predictions are recorded on pending public dockets before they resolve, timestamped by the database, and scored when the docket ends. That record is young, and we would rather say so than quote a figure we cannot defend.
• Not testimony, and not filed. Nothing Supreme Mind produces is filed, disclosed under Rule 26, or offered as a testifying opinion. Your retained expert forms their own opinion, writes and signs their own report, and takes the stand.
• Decision support, not legal advice. Counsel of record keeps full professional responsibility, and every fact, citation, quotation and figure should be independently verified before use.

Where your data goes

Model calls run in our own Amazon Web Services account through Amazon Bedrock, in US regions, where prompts and completions are not stored and are not used to train any model. Matter content is isolated per firm in US-region storage and is never used to train any model, ours included. The detail is on Security & trust; the questions firms ask most are in the FAQ.

Questions about method? Email richard@suprememind.ai.