Hi David.
I understand and it is often the case when measuring with Network analyzers. In the first setup you find all the flaws and the second is much better. I have experienced it so many times, but on the same time it is where the experience really builds up. Sounds good about your new thoughts.
About plot and S2P file. You can easily import the S2P file in fx LTspice and do your plots there. I fully agree with you that this is a good way to move forward, because when you can also add additional components and immediately see the impact. In the end you can then simulate the complete circuit around the filter before building anything. And of cause if perfect, you also extract a S1P of the passives you intend to use and use that in the simulation also.
My experience is for the ceramic filters, that if Group delay or attenuation in stop-band is bad, it is often a filter impedance matching problem. Matching problems does often not affect pass-band.
