Benchmarks

One hundred real verdicts.

Each case was rebuilt from the public record and run through 250 simulated juries. Open a row for the real verdict, what we called, and why it is a match, a miss, or too close to call.

direction
79/100
when it commits
56/69
too close to call
18
calibration
11/13
Real
P plaintiff, D defense, H hung.
Called
The side 250 simulated juries favored.
% plaintiff
Share of those juries, not a score.
Result
Match same side. Miss other side. Even is 45-55%.

100/100

CaseCalledResult

Calibration

11/13 matched

3M earplugs

12/12 matched

Patent and commercial

7/8 matched

Roundup

6/8 matched

Talc

6/7 matched

Vioxx

4/6 matched

Pelvic mesh

4/5 matched

AndroGel

3/4 matched

Pinnacle hip

3/4 matched

Xarelto

3/4 matched

Auto

2/3 matched

Hernia mesh

3/3 matched

Tobacco

2/3 matched

Other

13/20 matched

How to read the score, including the misses

Raw direction is 79 of 100 at the 250 juries the product runs, and 74 of 100 when we use 1,000. Every flip at the larger run sits inside a sixteenth of a point of even. On cases where the engine commits, the score is 56 of 69, identical at both run sizes. That is the number we would want a skeptic to hold us to.

The frozen calibration set is 11 of 13. Ernst was reversed on appeal for insufficient causation; the simulation sides with the appellate court. Uno took 4.5 days on a plaintiff-picked bellwether; we call it the coin flip it nearly was.

18 stress cases land within five points of even. Real juries also split inside those programs: the same Vioxx record won in Texas and lost in New Jersey, and 3M won some earplug bellwethers and lost others. If even real juries split on near-identical records, the reliable move is a paired test: the same juries hear both versions of your case.

Compensatory damages on a single plaintiff are in scope. Punitive and multi-plaintiff awards are not. Records, seeds, and the engine version are pinned; the suite fails the build if these numbers degrade.

Now run your case.

The same engine, on the record you are actually trying.