LexGrit

The guide

  1. Getting started

    What LexGrit is, what you need to use it, and the shape of the app once you are in.

  2. Courts and their rules

    Create the forum you appear before, choose where its procedural rules come from, and teach it how your court actually applies them.

  3. Preparing a case

    The case profile, the pleading that grounds everything, the must-hit checklist, and the pre-mortem that attacks your theory before opposing counsel does.

  4. Judicial affidavits

    The witness testimony an affidavit cross-examination is confined to — enter your real one, or draft one from the record and accept it after reading it.

  5. Who you face

    Opposing counsel, the bench, the witness — configured before the session, so the room you rehearse in is the room you are walking into.

  6. Running a session

    Quick start or the studio, what the proceeding decides, how objections and rulings work live, and how a session ends.

  7. After a session

    The scorecard and its breakdown, the replay, re-arguing an exchange, the practice record you can export, and teaching the court what it got wrong.

Questions

Why can't I sign up?

Because LexGrit is in private development, and it publishes no accuracy figures yet. The order of operations is deliberate: measurement first, then invitations. A rehearsal tool that has not proven its own instruments would be asking you to practise against something nobody has checked — see what it refuses to claim. The status page is where availability will change first.

Is my voice recorded?

No. Your voice is processed live to run the session — speech to text, and the room’s replies back to speech — and no session audio is stored. What is kept is the written transcript and derived numbers: pace, timings, objection counts. The replay is built from those, which is why it has no play button.

Is my case data used to train AI models?

No. Your pleadings, transcripts, and corrections are yours, visible only to your account, and LexGrit trains no models on them. When you “teach the court”, what you build is corpus and retrieval tuning — your corrections are stored, ranked, and quoted back with provenance — not model fine-tuning, and the product says so in exactly those words because that is what the code does. During a session, your material is sent to the AI provider that powers that session’s responses, and nowhere else.

Where does the AI get the law it cites?

From documents, never from the model’s imagination. A forum’s rules come through one of three doors — official PDFs you uploaded, a curated corpus LexGrit ships for that catalogued forum, or that corpus plus your own corrections. Rulings cite passages retrieved from those documents, each citation is checked against the text it claims, and a citation that cannot be grounded is flagged rather than trusted. What the corpus does not contain, the bench does not cite.

Why does my court say it is not grounded in its curated rules?

Because the curated corpus it points at has no published edition it can retrieve from — most commonly, the edition is still awaiting review. Curated rule text reaches a courtroom only after it has been through a review gate; until then your forum retrieves none of it, and the product tells you rather than running a hearing that silently rules on nothing. The notice on the court page states the specific reason and the way out: upload the forum’s rules yourself, or wait for the edition to be published.

Can I delete a case?

Yes, in two tiers, and the difference matters. Archive hides the case; sessions and scorecards are kept, and nothing is destroyed. Purge permanently deletes the case and everything under it — sessions, transcripts, scorecards, documents — and requires typing the case title back before it will run. Courts work the same way, with one addition: a forum cannot be purged while any case still references it.

Does a good score mean I'll win?

No, and LexGrit will not imply it. A rehearsal is not a prediction of how a court will rule, and the scores are not calibrated against human judges yet — they exist to compare your own rehearsals against each other, so you can see whether Thursday’s argument beat Monday’s. That is also why the exportable practice record deliberately carries no score at all.

Something looks wrong. What should I do?

During private development the people using LexGrit are in direct contact with the people building it — raise it through the channel you already have, with the case and rough time of the session if it happened in a room. There is no public support channel yet; one arrives with public availability, and pretending otherwise with an unstaffed inbox would be worse than saying this.