Use case that has come up is one where you wish to do inference on the complete set of samples where auto correlation is not (or less?) of an issue but then apply thinning to the samples used by rbmi to remove said autocorrelation.

At the moment we just uses Stans inbuilt thinning so this could be implemented simply by manually applying the thinning after fitting the full Stan model (as the retained Stan object would contain the full set of samples).