Dr. Socolar,
Thank you for the thorough and helpful response! This makes sense! And I’ve learned something today that has deepened my appreciation for using draws from the full joint posterior in analyses. After reading your response I am realizing that my main worry was that this correlation might be some sort of mild non-identifiability issue.. but I see now that my posterior has actually identified the parameters just fine, the samples are just reflecting uncertainty therein. This is a very neat thing!
I also am more clear now that true non-identifiable correlation is actually an infinite ridge, like that discussed here: Blog post: Identifying non-identifiability and that what I’m seeing here is not a pathology.
Thank you, @jsocolar, for your time!