The assumption that time-averaged statistics of a single long simulation equal ensemble averages across many samples.