One other piece of information, I increased the simulated data sampling rate by 10x thinking it might help if the issue is identifiability. So the data now looks like this:
However this model failed really spectacularly. From the trace plots it looks like the proposals almost never shift:
So this was surprising. I’m wondering if it indicates some kind of bug? The model specification is the same as above.
Just thought I’d add this to fill out the picture.