does it handle skewed distributions? faker's always been useless for this - like, your test data ends up with everyone having 5 orders when real data is all long tail