About the project
Why one analysis is not enough
In 2018, 29 teams of scientists got the same football data and the same question: do referees give more red cards to players with darker skin? The teams chose different methods and got different answers. Our project asks: can AI show this problem for any dataset, in a few minutes?
The problem
Every analysis has many small choices: which columns to control for, what to do with outliers, how to handle missing values. Each choice is reasonable, but together they can change the result. A single study only shows one of these paths.
Our solution
The AI only decides how to analyze. It never calculates the numbers itself, so every result can be checked and repeated. We also check the AI answers for mistakes, like controlling for a column that is part of the outcome.
What we found
With data from 173 countries, all AI analysts agreed that richer countries live longer. But the size of the effect changed by 3.6 times, only because some analysts also controlled for infant mortality. The best formula was logarithmic: every time income doubles, life expectancy grows by about 3.5 years.
Related research
Two research groups tested a similar idea in 2026: Carnegie Mellon University ("Many AI Analysts, One Dataset") and Stanford University ("The Agentic Garden of Forking Paths"). Our results agree with theirs. Our goal is different: a free tool that anyone can use without writing code.