I'm guessing this is more of a fun project than serious—the display on the nixie tubes won't refresh quickly enough to demonstrate the rubidium oscillator's ability to hold nanoseconds of time.
But in the case of potentially also using this board as a Stratum 1 clock (it has GPS/GNSS built in...), why would it not be as accurate as pretty much any other clock that serves up NTP time?
It would surely be a lot more accurate than any sync via NTP going over the public internet.
You could push nixies down from +200 us to roughly +/- 5 us with a current sense circuit feeding a tracking circuit (either a hardware PLL or digital DLL) with feedback from the PPS gen circuit.
Most of the inaccuracy of nixie displays (that are done properly and conventionally) is that the nixies require a few hundred us to build up enough charge to ionize the gas. If you send your HV trigger out before the actual start of second then you can account for the static part of this offset.
Reducing the random part would be... a trick. Low and constant temperature would be a decent starting point.
My next ideas are based around taking out the feedback loop and replacing it with static offsets based on an empirical model. Taking measurements of the nixie state transition delay (ie 1 to 2 takes a different amount of time than 2 to 3) would provide useful information. You could take temperature measurements and induce temperature over a range and build a matrix of the nixie state transition delay measurements. The temperature and transition matrix you can use as a priori that is weighted against actual feedback measurements.
But in the case of potentially also using this board as a Stratum 1 clock (it has GPS/GNSS built in...), why would it not be as accurate as pretty much any other clock that serves up NTP time?
It would surely be a lot more accurate than any sync via NTP going over the public internet.