how complicated a "test fixture" could I employ before I degrade measurement
accuracy beyond any useful point
VNA works by measuring signals, then calculating based on differences from known references
namely open, short and 50 Ohms. So long as there are measurable differences among those references,
then measurements can be of interest.
To anticipate how interesting, watch the Smith plot while performing calibration with whatever "test fixture".
So long as point clusters for open, short and 50 Ohm are substantially separated,
then there is hope.
Oristo