Dave Saul's excellent response to this question explained how he -- and
others using either original masters or a computer-aided system like
Wayne Stahnke's WEB program -- achieves 100% accurate recutting, or
nearly so.

What I would like to talk about is why any other one-to-one system of
recutting is almost bound to introduce a certain amount of inaccuracy.
Such a system reads a constantly moving original roll as it passes over
a tracker bar in the same way that the roll would pass over the tracker
bar of the instrument it was meant to play on.  The tracker can either
read the roll pneumatically or electrically by brush contact through
the holes of the paper.

The perforator which this system is meant to drive necessarily does
not move the paper under the punch pins in the same constantly moving
fashion, but in a series of steps: advance/stop/punch/advance/stop/
punch ...   At each punch cycle the perforator "reads" what holes in
the original happen to be passing over the tracker bar at the moment
and punches those holes into the paper.  At some moments a hole in the
original is open enough to trigger a punch; at others, it is only
partially open, and does not.  Sometimes the perforator takes two reads
of the same single hole and punches it twice; in other cases it may
skip a read, all depending randomly on what is happening at the tracker
bar at the moment of punch.

The accuracy of the reading can be improved by various techniques,
including configuration of the tracker bar.  For example a tracker
hole that is not very high (a hole shaped like this: -- , rather than
like this: o ) may help; Mike Grant uses this configuration, I believe.
But nothing can completely eliminate the introduction of slight
inaccuracies into the copies.  The question is, are these inaccuracies
enough to disturb the musicality of the roll and enough for the
listener to notice?  For band organ rolls, probably not.  For
reproducing piano rolls, probably so -- to a trained pianist at least.
The other question is, as Robbie Rhodes said, are buyers willing to pay
for improved accuracy?

Matthew Caulfield