Trial & Arbitration Simulation is generally available

Trial & Arbitration Simulation is generally available from today for all ZAAN accounts, including Open Bar.

For those who did not take part in the limited release: Simulation lets a team run a case before an opposing counsel, a bench or tribunal, and witnesses, all played by the system, using the matter’s own pleadings, bundle and transcripts. At the end it produces a scorecard: a liability confidence band, an award or damages range, the pleading the tribunal found weakest, and a transcript of every objection and ruling. It is a rehearsal room. It is not a prediction engine, and the first section of the scorecard says so in plain text.

What changed since limited release

Venues: 9 to 22. The limited release modelled nine venues, mostly English and Swiss arbitral seats and the English Commercial Court. General availability adds thirteen more across arbitral seats in Europe, Asia and North America and a small number of court venues, including two US federal districts. A venue model encodes procedural rules, typical hearing structure, evidentiary conventions, cost-allocation tendencies and the style in which reasoning is usually expressed. It does not model named judges or arbitrators. We will write separately about why.

Objections are on the record. Every objection raised by either side is ruled on with a one-paragraph reason, and the reason cites the venue’s procedural basis. In limited release, rulings were terse and users could not tell whether a sustained objection reflected the rules or the model’s mood. Now you can argue with it.

Scorecards are calibrated against prior runs. The liability confidence band now shows how it moved between runs of the same matter, so that a team can see whether a redrafted pleading changed the outcome. Ranges are deliberately wide. A band of 45–70% is a statement about uncertainty, not a hedge.

Witnesses are built from the record. Witness personas are generated only from documents in the matter vault: statements, deposition or examination transcripts, correspondence. Nothing is inferred from outside the matter. A witness with a thin record is a thin witness, and the scorecard flags where the record was too sparse to simulate credibly.

Rapid mode. A thirty-minute run that skips openings and runs cross-examination on the two witnesses the team selects. Built for the night before a hearing, when a full run is not an option.

Privilege. Every run executes inside the matter’s vault, in the tenant’s region, with the same per-tenant encryption and zero-retention guarantees as Review and Recall. Simulation transcripts are marked as work product on export. Nothing from a run is used to train anything.

What to watch for

Do not read the award range as a valuation. It is derived from the pleadings and evidence as simulated, under the venue’s conventions. It is useful for seeing which heads of claim the tribunal found persuasive. It is not a settlement number, and we have declined to add a feature that would present it as one.

Opposing counsel is competent, not omniscient. It argues from the bundle. If the other side has a document you have not seen, Simulation does not know about it either.

The weakest-pleading finding is the most useful and the most uncomfortable output. In the limited release, teams told us it was correct more often than they wanted it to be. Treat it as a question, not a verdict.

Venue selection matters more than you might expect. The same bundle run in two arbitral seats produces visibly different rulings on documentary evidence and on costs. Pick the real venue. If yours is not among the 22, pick the nearest in procedural tradition and note the gap in the scorecard header.

Known limitations

• Jury venues are not modelled. The two US federal districts assume a bench trial.
• Live voice in Simulation is available in 12 of the 38 languages Interview supports. Text mode supports all 38. We are testing wider voice coverage and will not promise dates.
• Very large bundles (above roughly 4,000 pages) are sampled rather than read in full for the opposing-counsel persona. The scorecard shows the sampling rate.
• Expert evidence is simulated as testimony, not re-computed. Simulation will not check an expert’s arithmetic.
• Multi-party proceedings with more than two sides are run as sequential two-party simulations, which loses some of the dynamics.
• Open Bar accounts have the full feature set with a monthly run allowance rather than an account limit; the allowance is shown in the workspace.

Upgrade notes

Limited-release accounts keep their run history and venue settings. The nine original venue models have been re-versioned; prior runs are tagged with the model version they used so that comparisons remain honest. Runs started before today cannot be re-opened in rapid mode.

A rehearsal does not tell you how the hearing will go. It tells you which parts of your case you have not yet had to defend out loud.