Qu'est-ce que « CUPED » ?

CUPED (controlled experiments using pre-experiment data; Deng, Xu, Kohavi and Walker) compares the arms on the adjusted outcome $y - (x - \bar x)^\top\theta$, where $x$ are covariates the treatment cannot affect and $\theta$ is the coefficient of the regression of $y$ on $x$ pooled over both arms.