Since #1513 we have some machinery for checking this, but only when used inside the embedded laplace functions. There are some modes of e.g. https://github.com/roualdes/bridgestan or cmdstanr that would like to be able to check this compatibility for the entire likelihood, so it would be nice to let them opt into it.