I think that dielectric constant (in dielectrics with no mobile
charge) is relatively frequency-independent, and error terms
in capacitor value vs frequency can be ascribed to series R,
L and shunt C (to where?) elements lumped into the
measurement.
In a RF IC PDK you would see capacitor elements with
R, L, C parasitics embedded, at least covering what lies
within the physical extents of, and respond to geometry
of, the capacitor cell.
Losses vs frequency is a different matter but properly
measured these ought not to affect C itself.