@ricardoV94 , thank you for having a look at my problem. I forgot to mention (or include in my code sample) the fact that printing out the observed and unobserved RVs, or the graph visualization, or the point_logps all show that only the first set of RVs corresponding to the first time-series are in the model. So, I take your point about the print statements, but there is definitely some deeper issue at hand here. 
I will see what moving RVs out of the scan function does.