Dear Peter,
It is up to you to decide if your model has measurement errors. If you decide to add measurement errors, then you will observe gaps between historical data and smoothed variables. The smaller are these differences the better is the fit. Obviously this comparison does not make sense if you do not have measurement errors (see also this related recent topic).
It may happen that, even without measurement errors, you observe differences between historical and smoothed variables (typically different means). This may reflect a problem in the specification of your measurement equations (missing constant)… Or a bug in Dynare itself.
Best,
Stéphane.