[x] flocker should have working posterior predictions for the visit-level data.
[x] flocker should have working integration with loo.
[x] flocker should have it's own methods to simulate from the posterior Z matrix, both conditional on the observed data and conditional just on the covariates.
Note that this cannot use brms infrastructure because the likelihood doesn't factor row-wise.
To close,
loo
.Note that this cannot use brms infrastructure because the likelihood doesn't factor row-wise.