High number of divergences in DDM sampling

[morgbead]

I had initially posted this issue on a different thread, but the topic of that thread was distinct from the present issue. I'm hoping that by isolating this issue it can be better found by those with similar experiences - both to lend some insight, or to benefit from any solutions we can identify.

Briefly, the present issue I am having in modelling DDM's using HSSM is a terribly high number of divergences when sampling the model. The models appear to still be converging - examining trace plots and marginal posterior show no issues, and all Gelman-Rubin values are below the recommended value of 1.1. I elaborate a bit more on various attempts we've made to resolve this issue in the other thread (#844), and while these efforts seem to move the needle in the right direction, the divergences remain concerningly high.

We're unfortunately at a bit of stand-still at the moment in modelling our data - we don't trust the output of the models given these divergences, but we can't make the divergences go away. Our careful efforts to examine the data for potential issues reveal no red flags from what we can see. We have a backlog of different tasks (largely from novel pilot data) that should be amenable to modelling using this toolbox, but we aren't able to make any progress until this is resolved. We are hoping that someone might have suggestions / insight on how to mitigate these divergences to allow us to move forward- or, information on whether the data we have is simply incompatible with the current toolbox, and any recommendations for where to turn to if that's the case?