Release 4 min read

Simulation 1.4: Cross-examining the financial expert

We have added dedicated expert witness models to Trial & Arbitration Simulation, allowing counsel to test lines of attack against damages methodologies.

2026 · 10 · 23·admin

It has been just over a year since we made Trial & Arbitration Simulation generally available. In that time, the primary use case across our active accounts has been testing fact witnesses. Counsel upload a witness statement or a deposition transcript, and our model assumes the persona of the deponent. This is useful for testing a narrative, but fact witnesses generally only address liability. In commercial arbitration, the dispute frequently turns on quantum.

Today, we are releasing Simulation 1.4. This update introduces a separate architecture for expert witnesses, starting with financial and valuation experts. It allows legal teams to pressure-test damages calculations and valuation methodologies against a simulated expert before the hearing.

The difference between facts and methodology

Fact witnesses evade; expert witnesses defend. A fact witness might claim they do not recall a specific email or meeting. A financial expert will rarely claim they do not recall their own discounted cash flow (DCF) model. Instead, they will defend the assumptions beneath it: the weighted average cost of capital (WACC), the selection of comparable companies, or the terminal growth rate.

To model an expert, we had to adjust the underlying prompts and reasoning loops in ZAAN-7B Counsel. The model must no longer look for factual contradictions alone. It must hold a mathematical structure in memory and defend its components against standard cross-examination vectors.

If counsel challenges the illiquidity discount applied to a closely held subsidiary in a post-M&A dispute, the simulated expert will not simply agree. It will attempt to justify the percentage using standard valuation literature and the specific market conditions described in the uploaded report. It behaves like a professional whose reputation is tied to the methodology remaining intact.

Ingesting the expert report and exhibits

Setting up an expert simulation requires more than reading a PDF. Damages reports rely on spreadsheets. In Simulation 1.4, we now support the ingestion of tabular data and formulaic models alongside the narrative report.

When you upload a financial expert’s report, ZAAN extracts the core valuation mechanics. It isolates the dependent and independent variables. If the expert used a comparable transactions method, the system indexes the specific transactions chosen and the multiples applied (such as EV/EBITDA). If the dispute involves an earn-out calculation, the system maps the accounting definitions used in the report against the original share purchase agreement.

Counsel can then begin the moot. You might ask the simulated expert why a specific 2023 transaction was excluded from the comparable set. If the exclusion was addressed in the report, the model will cite the report’s reasoning—for example, distressed sale conditions. If the exclusion was not addressed, the model will generate a plausible defence based on standard accounting principles, while noting the vulnerability in the post-session scorecard.

Tribunal intervention during technical testimony

In our modelling of 22 distinct arbitration venues, we have observed that arbitrators behave differently during expert testimony than they do during fact testimony. A tribunal chair is far more likely to interrupt counsel during a highly technical cross-examination to ask for clarification, or to question the expert directly about the mechanics of a formula.

Simulation 1.4 replicates this dynamic. If counsel spends too long attacking an immaterial variable, or if the questioning becomes mathematically incoherent, the simulated tribunal will intervene. The chair may ask counsel to move on, or may ask the expert to explain the practical impact of the disputed variable on the final quantum. This forces the questioning advocate to keep their line of attack both technically accurate and accessible to the tribunal.

Calculating the impact on the award range

One of the core outputs of a ZAAN simulation is the scorecard, which estimates liability confidence and provides an award range. Previously, the award range was relatively static, derived entirely from the initial pleadings and the survival of fact-based claims.

With expert modelling, the award range is now dynamic. If counsel successfully traps the simulated expert—for instance, by exposing a double-counting error in the working capital adjustment, or by forcing the expert to concede that a different beta would drastically lower the discount rate—the simulation registers a structural hit to the quantum case.

The post-session scorecard will reflect this concession. The award range will narrow or shift downwards, and the “weakest pleading” section will explicitly flag the fragile assumption in the valuation model. This allows legal teams to see precisely which mathematical concessions yield the highest strategic value, rather than guessing which errors the tribunal will find most persuasive.

Limits on expert hallucination

Modelling financial expertise introduces a risk of hallucination that is arguably more dangerous than factual drift. An AI might invent a fictitious accounting standard to defend a weak model, which is a behaviour we must entirely suppress.

To prevent this, we have restricted the simulation’s knowledge base strictly to the uploaded report, its cited appendices, and a strict whitelist of standard valuation guidelines (such as IFRS standards and standard corporate finance texts). If counsel asks a question that requires an assumption outside this closed universe, the model is instructed to concede ignorance or state that the variable was outside the scope of its instruction.

Cross-examining an expert is an exercise in dismantling a machine, part by part. Simulation 1.4 gives counsel a way to test which parts are load-bearing, before the tribunal is seated.

See it on a contract you have already reviewed.

Send us a draft your team has already redlined and we will show you what ZAAN catches, and what it misses.