Practise the part that doesn’t come naturally.
Most experts know their field. Almost none get the rehearsal hours they need to defend that knowledge under cross-examination — out loud, in real time, against someone whose job is to make them look uncertain.
CrossCoach is a private flight simulator for that experience.
An exchange from a Level 3 latent-prints session.
Counsel is the model. You respond in your own words. The questioning adapts to what you say.
Annotated by the evaluator, run after the session ends.
Can you tell the court your conclusion?
Yes. In my opinion the latent print was made by the same finger as the exemplar.
And on what basis is that judgment defensible, given that no agreed ground truth exists for latent print examination?
I rely on the comparison protocol I applied — Level 2 minutiae and selective Level 3 detail, evaluated against the AAAS 2017 guidance. The judgment is a conclusion of correspondence, not a probability statement.
Are you familiar with the Ulery et al. 2011 false-positive rate of 0.1%?
Yes.
So you accept that a one-in-a-thousand error is possible in your own work?
Four steps, in private. Repeatable until the articulation is routine.
Set the context
Choose your discipline and a difficulty level. Optionally describe the specific case — your conclusion, your methodology, where the weaknesses are.
Sit through the cross
A simulated opposing counsel asks questions in real time. The questioning adapts: evasions get followed up, unsupported claims are challenged with the literature.
Get evaluated
A separate evaluator reviews the transcript and scores you on five dimensions, with turn-by-turn feedback identifying strong and weak moments.
Read up
Study the underlying source material — PCAST, NAS, AAAS, key case law — with an AI tutor before your next session.
Five dimensions of credible expert testimony.
The dimensions are deliberately generic to expert testimony — not specific to any single field. They map almost directly onto what courts everywhere actually scrutinise about expert evidence.
Read the research behind it →Foundation of method
Can you defend the validity of the approach you used, with appropriate reference to the evidence base?
Expression of uncertainty
Do you quantify or qualify your conclusions accurately, or overclaim?
Independence & management of bias
Are you transparent about what context you had, and aware of how it could have influenced you?
Observation vs inference
Can you separate what you saw from what you concluded, and explain the inferential step?
Transparency & communication
Are you clear, willing to acknowledge limits, and free of jargon-shielding?
The reports they’ll quote at you, read before they do.
Each one distils a landmark report or study into a short, plain-English walk-through, with the practice built in. The same source material grounds every cross-examination session.
Built for forensic science. Architected for any field whose conclusions can be contested.
Latent fingerprints
Print quality, feature levels, complications, conclusion type. AAAS 2017, PCAST, Ulery, Hicklin.
Crime scene reconstruction
Scene type, contributing disciplines, conclusion type, context exposure. Chisum & Turvey, NAS.
Firearms & toolmarks
Cartridge case identification, bullet comparison, AFTE conclusions, score-based likelihood ratios.
DNA mixture interpretation
Allelic dropout, contributor estimation, probabilistic genotyping, LR thresholds.
Medical experts
Radiology, pathology, psychiatry. Malpractice, coronial, criminal matters.
Engineering & accounting
Failure analysis, structural assessment, valuation methodology, audit judgment.
Onboarding a new vertical is a structured content exercise, not a re-platforming: a discipline prompt module, a case-context schema, a knowledge base, evaluator calibrations, UI labels.
Propose a disciplinePerformance on the stand is articulation, not knowledge.
An expert’s knowledge rarely fails on the stand. What fails is the articulation of it: explaining a method in plain words, owning its limits, holding a line of reasoning steady under questioning designed to bend it.
Articulation is a skill, and it builds the way skills build: repetition, with feedback. But a real cross-examination might come once or twice a year, and the witness box is no place to learn from mistakes.
CrossCoach supplies the repetitions. Low-stakes sessions, a structured evaluation after each one, in private, as often as it takes for the articulation to become routine.
The questions are grounded in real source material.
Counsel doesn’t confabulate. Its questioning is retrieved from a curated knowledge base of standards, peer-reviewed research, and leading case law for each discipline.
Competence gets you to court. Confidence gets you through it.
Currently in pilot with select Australian forensic laboratories and police agencies. Access is by invitation while the rubric stabilises.
hello@crosscoach.ai · CrossCoach™ · Brisbane