Issue with Variable Truncation Bounds Matching Distribution Bounds

[StaffanBetner]

While brms allows for variable truncation bounds, it struggles with untruncated observations, notably when bounds are set to extremes like -Inf and Inf in Gaussian distributions. This issue seems related to Stan's challenges with infinite gradients, as discussed in Stan issue #2881. Addressing this in brms could involve refining how such extreme truncation scenarios are handled, e.g. by implementing specific checks for these cases.

[paul-buerkner]

How would you suggest this to be translated to Stan code exactly?