REX
What people actually want.
Measured, not asked.
Human preference and social calibration, captured continuously from three hundred thousand people a month who are not paid, not recruited, and not performing. Priced by the judgment. Weighted by the judge.
The best readers of other people are two and a half times as accurate as the worst. Preference data prices them the same.
Models are aligned to "human preference" scored by small pools of paid raters, treated as interchangeable. They are not. On a 300,000-prediction sample, scored on whether one person could predict another's choice, accuracy ran from a quarter to nearly two thirds. The spread held at three sample sizes as the sample doubled.
Preference, with the loser recorded
Not just what a person chose. What it beat, from the same hand, in the same moment. Most of the information in a choice lives in the options that lost, and we keep them.
A quality score for every annotator
How well a person reads other people, measured on a separate task with a verifiable answer. Weight every judgment by the judge. Nobody else has a second instrument to check the first against.
Where the line sits
Tolerance, tone and timing. How far a person will go, how far a room will let them, and the gap between what someone enjoys and what they will admit to. Two numbers, both held.
Four conditions. No dataset on the market satisfies all of them.
Nobody is compensated for an opinion. There is no incentive to produce one, so what comes out is the one they had.
No panel, no application, no vetting by other vetted people. Participants arrived on their own and stayed on their own.
Every judgment is made in front of people the participant knows, for stakes they care about. Nobody performs a laugh for a researcher.
Consent is asked at the moment of capture, scoped and revocable. Records carry it with them. Nothing here was scraped.
Rates, not totals. The corpus refills every night whether or not anyone is buying it. The instrument is play: two titles, one of them with a verifiable answer key, which is what makes a calibration score possible at all.
We work with a small number of partners.
Model labs, frontier labs, and the research teams inside them. We publish neither our methods nor our map, and that is deliberate.
Request access