I had noted this issue before when checking the KC901V insertion loss measured in field strength mode using my HP reference stepped attenuator set pictured below. Note all connections here are made SHF rated semi rigid hard line so cable and connector quality should not be an issue.
The KC901V is in VNA S21 mode, sweeping the full 7GHz range with 1dB/div range.
The first result you can see side on in the first picture is after a basic user calibration with the both stepped attenuators set at 0dB. Note there is a small step up from -0.4 dB to 0dB just at the start. The blue hold trace is the calibration result and the yellow is after a 1dB step. I assume the ripple at the higher end is a result of reflections because as stated in the specifications Ports 1 and 2 are not a perfect 50 ohm match.
This result is for a 5dB step.
The next result is a user calibration with the attenuator set at 3dB. The second sweep is with the attenuation set to 5dB to give a 2dB change with respect to the user calibrated reference point. Note that frequency at which the 0.4dB step is has changed. The 2db result is somewhat smoother at higher frequencies now that that of the first display picture.
The last picture compares the hold user cal result (for 3db setting) with a new 0dB user calibration.
Many tests later with this and other fixed reference attenuators I have found that the user calibration step disappears above 7GHz with more than 7dB insertion loss.
I am interested to know what causes this and if it can be changed or compensated for.