As Brenton commented on Twitter, posterior_predict() doesn’t seem to be the adequate function for my case. In fact, I want to generate the posterior predictive only accounting for the between-study variance-covariance matrix (“D”, as shown above), and further confirmed by Richard Riley here.
Thus, I want to “ignore” the data brought by brms::fcor(V) in the code above. Unfortunately, brms::posterior_epred() does not ignore it and yields the same error.
Not sure why though.