Ok, I see what you mean. I actually went back to my original implicit function which was simplified so a solution would be trivial to find. What I was doing wrong before was passing in some parameters that were no longer needed so that probably drove MCMC nuts (bad crazy nuts, not good sampler nuts).
I reran without the unused parameters and I’m down to one divergent transition, verifying what you said.
My mistake is that I was thinking of the algebra_solver() solver like any math operation which should have no side-effect on the sampler. I neglected to consider the problem of failing to converge to a solution.
Thanks Ben for helping out with this. I’ll continue taking baby steps towards the nonlinear form of the implicit function I want solved.