My interest in roll punch advance is mainly along the lines of
practicality.  I am interested in the historical aspect, to the extent
that it provides useful information. I would like to scan rolls with a
degree of precision that captures the information as accurately as it
was punched. If it is possible to remove extraneous artifacts of the
punching process  and the effects of aging paper, and "recreate a master",
that would also be a desired goal. I was thinking that knowing the
original punch step would help in this process. Also it might help to
determine a step size for the punching of recuts that would provide
sufficient accuracy without slowing down the process.

I plan to step the roll .005" during the scanning process. Anything finer
would be very slow and produce too much raw data for the simple methods
I plan to use initially. Although I'm sure the data could be massaged
later to synchronize the note onset and duration with the original step
size (if that is ever determined) I would like to be convinced that I am
choosing a reasonable resolution from the start. Does anyone have any
ideas as to what timing accuracy is really necessary to not degrade the
final recut? In complex music with subtle rhythmic interactions, I'm sure
there is a greater chance of degradation, but what is a practical accuracy
for which to aim? The original punching is a compromise from the master,
but is it adequate?

About roll paper stability. From my experience with graphics and printing
I recall that paper, like wood, expands and contracts along the grain much
less (almost imperceptibly under normal variations of humidity) than
across the grain. (The old offset presses which used conventional plates
could only adjust the image stretch in one direction and this was adequate
for accurate registration of maps as long as the paper grain direction was
chosen correctly.) Of course seventy year old paper may have stretched as
the fibers are broken down and the paper is pulled through the spoolbox
many times, but there can be no assumed relationship between the change
in the two dimensions.

Wayne Stahnke's idea about matching note spacing with musical intervals
seems very reasonable as an explanation for choosing strange step sizes,
but was this just a bias on the part of the original engineers or is there
any validity in this choice. If a roll is hand played, to what extent does
the recording match musical intervals anyway? Then when you consider that,
at best, there are considerable differences in timing from one note to
another in the player. What is really necessary to make the music sound
good. To what extent does the human auditory apparatus correct for minor
timing variations? Did you ever notice how much better a roll can sound
after a glass or two of wine. Suddenly the music all falls into place and
you can hear how well the editors captured (or didn't) the original
performance. Is this better perception (less attention to distractions and
more to the music) or better fitting of the perception to a musical model
or ideal?

To sum up, I would like any input that would help me decide on a scan rate
that would not degrade the source data. I would hate to dedicate a lot of
time to building a scanner and scanning thousands of rolls, only to find
later that the data captured was less than adequate. With respect to roll
punching, I am less concerned. As long as the data is adequate, a method
can eventually be developed to produce rolls of sufficient accuracy.

If anyone is interested in details of my proposed scanner design, I would
be glad to provide such details. I would especially welcome advice as to
whether I am proceeding along a reasonable path.

Spencer Chase