Thanks to Craig Brougher for his thorough description of the hysteresis
effect in vacuum valves (970730 MMDigest).
It has become obvious that an optical roll reader needs this hysteresis,
whether built into the sensing hardware or simulated by software as in
Jody's design. And those parameters should be adjustable.
In addition, one should probably simulate the delays in the air system and
the piano action. Finally, if reading reproducing piano rolls, one needs to
implement an accurate transfer function (simulation) for that system, which
is far from trivial, as Robbie Rhodes pointed out.
The above is my understanding of some of the minimum requirements (on the
electrical/software side) for a roll-to-MIDI machine.