I've enjoyed the excellent correspondence about
the effects of solder mask, manufacturing tolerances,
etc., on Zo.
I would only add two things to the discussion:
1. Understand what your possible measurement inaccuracies
2. Most closed form algorithims are based on work from the
1950s to nearly the present day that relied heavily on
curve fitting of measured results. Computer driven field
solution methods didn't become practical until today's
afforable computing power became available.
I have found Zo calculated by closed form algorithims to
be from a percent or two lower to ten percent or so higher
(depending on which algorithims I used) than Zo predicted by
computer driven field solver methods. The more your particular
design diverges from the measured, approximated test setup -
the more out on a limb you are.
Verification measurements have been done on the field solver
predictions that show them to be accurate within a couple of
In summary, I am not surprised that measured results vs. simulated
results for Zo for a 90 ohm system might be 5 to 9 ohms lower using
the curve fitted algorithim approach. Unless the software supplier
has taken special, usually proprietary, pains to improve what is
available in the literature, I wouldn't expect much better.