About

The most expensive guesses in law, retired.

A trial team preparing a nine-figure case gets one rehearsal: a mock trial, six to eight weeks out, at six figures, in front of a single hired panel. Whatever that one room happens to say becomes strategy. Change your mind about an exhibit afterward and there is no way to ask the same room again. The experiment cannot be rerun.

Project Scintilla started from a simple observation: the decisions lawyers agonize over (which expert, which opening, whether to fight for the contested exhibit) are exactly the kind of questions experiments answer, and litigation has never had an instrument that could run them. So we built it: hundreds of simulated juries, drawn from recorded human behavior, deliberating every version of your case, deterministically, in minutes.

What Scintilla is

A rehearsal instrument for trial teams
An experiment engine: same juries, every version
A diagnostic: where comprehension fails, who loses the room, what each decision is worth
Deterministic, auditable, and honest about uncertainty

What Scintilla is not

Not a prediction of any particular jury
Not legal advice, and not evidence
Not a survey of invented characters
Not a replacement for judgment, but an instrument for it

What we hold ourselves to

Six principles, none negotiable.

Rehearsal, not prophecy

We do not predict your jury. We show you how hundreds of plausible juries receive your case so you can fix it before a real one hears it. Anyone selling verdict prediction is selling weather control.

Behavior, not demographics

Jurors are composites of at least 100 real people's recorded behavior. Nobody here is generated from an age, an income, and a zip code. That method predicts at chance, and it is beneath the decision you are making.

The same jury, twice

You cannot un-tell a human juror something. Holding the panel constant while changing one element is the experiment litigation has never been able to run, and it is the entire point of simulation.

Uncertainty on every number

Every result ships with its confidence band, its jury count, and its resolution level. Hung juries are reported. A number without a band is a number you should not trust, ours or anyone's.

Deterministic and auditable

Identical runs reproduce bit-for-bit and any deliberation replays from its seed. If a number cannot be reproduced, attributed, and falsified, it is an anecdote with extra steps.

Ethics on the record

Composites are built from de-identified cohorts; no juror is ever conditioned on race, ethnicity, religion, or national origin; every export carries its disclaimer. The fence is part of the product.

The jury behind the product.

Every face below is a deterministic sketch of an opaque reference: illustration, never demography. The juror underneath is the recorded behavior of at least a hundred real people, and the same reference always deliberates the same way. That is the whole trick: people, held constant, so your case is the only variable.

Juries drawn from 304 behavioral composites, ≥100 real people behind each face

Read the method, then run it.

The architecture, the statistics, and the limitations, all on the record.

Research Open the console