Scintilla

Scintilla puts every version of your trial in front of hundreds of simulated juries, built from recorded behavior rather than demographic guesswork, and shows you where the case is won or lost, element by element.

See a live matterHow it works

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

The whole loop in forty seconds: brief the case → hear 250 verdicts → find the weak point → change one thing → price it against the same juries.

250juries

deliberate every version of your case

<30min

from pasted case file to first verdict

≥100people

of recorded behavior behind every juror

100%reproducible

identical runs return identical verdicts, so every flip is your change

The instrument

Everything a verdict is made of, measured.

Strength, clarity, outrage, and whether this plaintiff's injury is actually tied to the product. One hundred documented verdicts sit behind those channels. The misses stay on the page.

The moment map

See where the case turns

Median juror leaning across the whole trial, with its uncertainty band. The exhibit that moves the room is visible as a cliff, not a hunch.

openingExhibit 14 ▲closing

Determinism

No noise, only signal

Run the same version twice and the verdicts match bit-for-bit, so any flip you see was caused by your change, not by chance.

100%reproducible

The cohort

People, not parameters

Each juror is a composite of ≥100 real people's recorded behavior. Faces are sketches; the data is the person.

+296

Comprehension

Watch understanding move

Per-issue comprehension and confusion debt across the trial. See exactly when the room stops following the science.

Paired testing

Same juries, both versions

Exact McNemar inference with bootstrap intervals, and an honest 'underpowered' label when the jury count can't resolve the effect.

76%→13%

one exhibit excluded · p < 0.001

How the channels are scored, and how the hundred-case suite is gated: the method. Every miss: benchmarks.

The record

Three things a live mock trial cannot hand you.

The verdict, with its uncertainty

250 juries, a confidence band, the hung rate, and the damages distribution. Never a bare number.

The room, replayed

Scrub any jury's deliberation. Watch who reframes the case, who holds out, and which argument moved the votes.

Every version, priced

Search the combinations of your case and learn what each decision is worth before you commit to one.

The method

Rehearsal is a loop, not a poll.

A survey tells you where the case stands. Preparation is what you do next, and every step below runs against the same juries, so each iteration is an experiment, not a fresh anecdote. The engine scores the record you actually try, including this plaintiff's causal chain, not a demographic guess about who is in the box.

  1. 01

    Brief the case

    Drop the case files in and a draft trial plan is written for you; you correct it like a junior associate's first pass.

  2. 02

    Hear the verdict

    250 juries deliberate the plan as it stands. You get a distribution with its uncertainty, never a bare number.

  3. 03

    Find the weak point

    The moment map shows where the case turns; the room shows who turned it; the comprehension map shows who stopped following.

  4. 04

    Change one thing, price it

    Swap the expert, cut the exhibit, plainen the direct. The same juries hear both versions, and the difference is exactly your change.

  5. 05

    Search every version

    Every combination of your open decisions, ranked, with the marginal worth of each call. Then back to the brief with a better plan.

The alternatives

What a mock trial costs, what a survey misses.

Live mock trialSurvey-style AI panelsScintilla
Juries that hear each versionOne, onceOne panel of individuals, polled250, deliberating
Time to a verdictSix to eight weeksHoursMinutes
Same jury hears both versionsImpossibleNo, a fresh panel each studyAlways. That is the design
What moved the numberAnecdote and instinctConfounded with panel changesExact: every flipped jury is attributable
DeliberationOne room, unrecorded dynamicsNone; individuals answer surveysEvery jury, replayable word for word
Uncertainty reportedNoneRarelyOn every number, always
Reruns reproduceNeverNo; model responses driftBit-identical

A live mock trial is still worth running: once, late, on your best version. Scintilla is how you decide which version that is.

Rehearsal, not prophecy.

Scintilla does not predict what your jury will do. It shows you how hundreds of plausible juries receive your case: where they stop following, which witness they stop believing, what the room argues about after the door closes. You fix it before a real jury hears it.

Questions

Asked, answered, on the record.

Does Scintilla predict what my actual jury will do?

No, and nothing honestly can. Scintilla shows you how hundreds of plausible juries receive your case (where comprehension fails, which witness loses the room, what the deliberation argues about) so you fix it before a real jury hears it. Rehearsal, not prophecy: every number ships with its uncertainty.

Where do the jurors come from?

Every simulated juror is a composite drawn from the recorded behavior of at least 100 real, de-identified people, with venue-realistic hardship attrition. No juror is ever generated from an age, an income, and a zip code. Demographic invention predicts at chance, and Scintilla never conditions a juror on a protected class.

How is testing two versions different from running two studies?

The same juries (the same seeds, the same people) hear both versions, and the only systematic difference is the element you changed. Identical runs reproduce bit-for-bit, so the noise floor is zero and every flipped verdict is attributable to your decision. A fresh panel per study can never give you that.

How do you know the engine is any good?

It is scored against one hundred documented real verdicts, rebuilt from their public records. Direction matches on 79 of 100 at the 250 juries the product runs, and on 56 of 69 cases where the engine commits to a side. That second number holds when we quadruple the juries. A new channel ships only if it improves a held-out split. Every miss is published on the benchmarks page.

Is the output legal advice? Can I use it in court?

No. Scintilla is a preparation tool, like a mock trial or a focus group. Its reports carry that disclaimer on every export, and nothing it produces is a prediction of any particular jury's behavior.

What happens to my case materials?

Case files are read in your browser, structured into a brief you edit, and used only to run your matter's simulations. Jurors react to what the brief says; the raw files are not what the jury sees, and your materials are never used to build anyone else's cohort.

How long until a first verdict?

Under half an hour from pasting the case file: the brief drafts itself in minutes, and 250 juries deliberate in seconds. Every further version you try is minutes, not weeks. That is the point: the value is in the iteration a calendar could never afford.

Rehearse your next case.

First verdict in under half an hour. Every version after that costs minutes.

Open the console