(simulation file in this project's files section, or click here and save-as... Then go to the falstad circuit simulator and select File-Open...)

The bottom two decade-counters merely simulate a rotating drum with magnets and reed switches spinning at 10 rotations/sec. Seems realizable. The left one just creates a slight delay between certain states, essentially a larger magnet. Similarly, the relays connecting the bulbs are also part of the drum-simulator.

For each bulb there are three states: Read/refresh-start, refresh, and stop-refreshing.

The third (upper) decade counter is a delay and loop for periodic write-testing. It could be little more than an RC circuit, but here it tests the bit being written periodically until the read-back value matches the value we're writing. This circuit's been optimized for writing that bit as fast as possible, and still allowing refresh of all bulbs as-usual.

I've also added write-hysteresis, by plopping a resistor in series with the bulb or the read-back resistor, depending if writing 0 or 1. Thus the bulb will be written until it's either 10ohms below or above the threshold.

Note that after reaching steady-state Ron is roughly 116ohms, and Roff roughly 44. The readback-threshold seems to be roughly 60ohms. I haven't checked the write-hysteresis thresholds very carefully. That was a late-addition upon finding an unusual case. But, I think it should be quite handy to assure reliability e.g. when the ambient temperature varies dramatically.

It's a lot of circuitry for three bits, but the numbers suggest it could be extended a bit. And... without write-automation the circuit's actually pretty small. Another possibility, maybe, is to use one write-circuit for multiple rows. And, obviously, multiple rows can use the same rotating-drum refresh-sequencer.

Actually, I should clean this up a bit... it looks way more complicated than it is.