Also, calibration always has a residue and the residue
is what you live with in production. Residue can depend
on attributes not tested at WAT, like the skew between
the overlap capacitances and thresholds of NMOS, PMOS
devices in all of the switches (affecting charge injection
balance, which can only be designed to center, not span).
Span is what you're after with Monte Carlo analysis.
Look to how simulator paths and such may change between
regular and MC modes - maybe you are losing some veriloga
include-path, or losing the compile activity that loads the
C-code version?
What's wrong with substituting a switch from analogLib?