How do we ensure random effects have correct distribution?

Using HMC will permit propagation of uncertainty, but that would only be noticeable in quantities/visualizations computed across iterations. See here for discussion of approaches to checking hierarchical distribution structure.