Research

Field report · our own run

We tried to break our own premise

We built Loop's method, then ran the one experiment designed to disprove it. Our agent got stuck where a real person would, even with the scripting removed. Here's the honest result, limitations included.

The say–do gap: why asking people isn't measuring them

Stated intentions explain about a quarter of what people actually do. That number is why Loop watches behavior instead of collecting opinions.

Why we won't quote you an accuracy number

Every tool in this category leads with a percentage. The peer-reviewed literature says those numbers are the least trustworthy thing about them.

Can you trust a synthetic user? An honest reading of the evidence

The research says simulated people reproduce human patterns well in some conditions and badly in others. Here is where the line actually falls, including the papers that argue against us.

Agents on real screens: the research under Loop

Simulating interviews is a solved-looking problem. Simulating behavior on a live interface is the harder one, and it is where the field has moved.