For the vast majority of tasks, it really isn't necessary to have 'absolute' accuracy of the 10MHz reference. I still think that most users would be better off (most of the time) if they simply used a decent OCXO as a master reference if they need to lock two or more instruments together. I've done this in the past with vector sig gens and also when grabbing data from vector sig gens using an RTSA. In this case I use the reference from the RTSA as the master reference and connect it to the sig gen(s). This makes the captured IQ data look more 'conventional' as it won't have frequency offset errors baked into it.
It is generally a good idea to use a master OCXO that has low spurious and low phase noise. So something from a decent spectrum analyser or sig gen should be fine. Using a cheapo GPSDO is going to be risky. The Bodnar GPSDO doesn't appear to have good phase noise performance.? The Stanford Rb standard is better but still not anything special. It definitely doesn't qualify as 'excellent' in terms of spectral purity.
Also, sometimes it is useful (as in intuitive) to see cw signals with subtle frequency offsets beating together on an analyser. This can give insights into what contributes to a cw signal on an analyser. Part of it may come from a sig gen or from the DUT or maybe it is within the analyser itself.?
