There's surprisingly little information about it. All I can find is that it was originally build in 1950 and was renovated in 2008, and it used to use relays, which were broken for some time.
So, looking at Wikipedia and comparing with his circuit PDF and what he said, it sounds like there wasn't thought given to current stability. Apparently these bulbs are very sensitive to too much current, so it sounds as if he was literally breaking the bulbs to get it to work.
I wonder then if there might have been some way to control for this properly?
Neon bulbs require a certain voltage to strike (for small ones, several 10's of volts). Until they are struck they are very high resistance, once alight they are low resistance until the voltage drops well below the striking voltage.
This leads to an easy way to make a neon oscillator; put a capacitor across the bulb, and a high value resistor in series to a 90v supply. The capacitor charges to the strike voltage, then discharges until the neon extinguishes; the oscillation rate and ratio characterises the particular bulb.
Don't know if it's still true, but they used to include radioactive tracers in the neon to reduce the striking voltage (also I believe the striking voltage varies with incident illumination).
Thanks for that! Great demo, and using strong UV shows the effect in spades (I'd guess in this case you wouldn't want the radioactivity, so the illumination effect from the photons is more pronounced).
You would need one for each bulb as the problem is that the bulbs characteristics are not tightly controlled like they are for vacuum tubes and vary widely even in bulbs from the same lot.
Circuit for multiple output current mirror [0] is very simple but then you start using transistors to run the neons and I think not using transistors was point of the project
Digital counting took too many parts before transistors. The Dekatron was one attempt to get the parts count down.
[1] https://www.youtube.com/watch?v=o3HLZOJKEQU