Rocking motion is sensitive to the boundary and initial conditions of a rocking structure, making experiments nonrepeatable. Thus, the claims that numerical rocking motion models are not only inaccurate but that all rocking structures behave unpredictably. Hence, rocking is not used as a seismic design approach. This paper revisits the issue of rocking motion unpredictability. Seismic behavior of structures is inherently stochastic, because the loading is stochastic. Therefore, the question of interest is not whether models can predict the seismic response to a single ground motion, but if the statistical characteristics of the ensemble of responses to a set of ground motions that define the seismic hazard can be predicted. For this purpose, a rocking podium, which is a three‐dimensional structure comprising an aluminum slab supported by four tubular steel columns, was tested on a shake table excited by two sets of 100 consistently generated ground motions. It was found that the cumulative distribution function (CDF) of the experimentally obtained displacements is statistically stable. Next, a blind prediction contest was organized. The contestants were invited to predict the CDFs of the slab lateral displacement. They were able to predict the slab displacement CDF relatively well. Both finite element and discrete element modeling approaches were used, but no clear pattern emerged as it was found that the performance of either approach depends on the input parameters used and the assumptions made. It was also observed that the contestants who did not use Rayleigh damping in their models produced better predictions.