Does the Stan documentation have any description of what is sometimes referred to as “potentials”?

Since they are basically probability distributions on parameters or combinations of parameters, in an extreme (and possibly less problematic case) they can be not unlike constrained parameters or combinations of parameters (where the potential is one or minus infinity).
Alternatively, they can be seen as just priors on parameters or combination thereoff, in which case maybe there is some discussion on the consequences of having “two priors”, so to speak.
Thanks for the help.

The third basic class, Potential , represents ‘factor potentials’ ([Lauritzen1990],[Jordan2004]_), which are not variables but simply log-likelihood terms and/or constraints that are multiplied into joint distributions to modify them.

You can add whatever you want to the target log density. It doesn’t have to correspond to a known density or anything.

A sampling statement like y ~ normal(mu, sigma) gets translated to something like target += normal_lpdf(y | mu, sigma) internally, but you can just add whatever.

Thanks. I looked at the sections below and completely missed the references in the beginning. I’ll look them up.
The Stan implementation makes sense, and I don’t suppose it should interfere with the sample anymore than any arbitrary priors would. Thanks again.