Usual rig is:
signal generator -> attenuator pad -> series fixture -> attenuator pad -> power detector
The only bit missing here ^ however why pads when you have adjustable stimulus and sensitivity ?
Attempting this with the new SNA5004A rather that SVA1032X as I need to get used to the quite different UI in the thing to know how to drive it properly.
The pads are there to provide a resistive termination to the crystal. If you have any capacitance in the crystal fixture it can "pull" the crystal parameters. In fact that's why crystals have a load specification in pF usually.
Good example here with the Marconi 2022E replacement oscillator.
If you remove the resistors and model C62/D13 as a single variable capacitor (around 22-40pF) in series with C63, C64 (120pF each) that'll give a total loop capacitance with the parasitic capacitance of the transistor of around 30pF . This is by no coincidence the loading of the crystal. D13 is voltage variable to compensate for the frequency. The reason the frequency needs compensation is that it's a property of the parasitic loading of the crystal so when you tune an oscillator you're actually just trying to get rid of the manufacturing variations of parasitics. Fun fun fun.
Same goes for a VNA fixture. Transmission line cables are parasitic as are bits of fixture. Attenuation isolates some of those properties from the DUT. In this case 3dB is probably a good starting point but 10dB would give it better termination.
Lots of pads also complicate some VNA measurements due to the total use of dynamic range plus the S21 loss of the DUT so can be interesting test case and exploratory exercise in VNA fun.
NanoVNA pukes doing this as well as there's only 70dB and you have to inject a low signal of -30dBm into the DUT anyway. Difficult to get good peaks and dips to get any reasonable resolution. Grr. In that case you're actually better with a signal generator, counter and RF voltmeter!