In earlier posts I quite deliberately decided not to venture into the
area regarding just what is being scanned -- the search for "best" roll.
I called it a "good" roll for the very simple reason that a good roll
well copied makes a good copy.  Why make things more complicated than
that?

Briefly consider three very different sources of differences: scanning
and image processing errors, production errors, and production
variations.

My feeling is that the scanning community were so pleased when scanners
were first up and running that many started plugging away with slightly
suspect scanner setups.  Not all scans are "archival", if you take that
to mean that the scan is an accurate representation of the roll and can
be used as a substitute for the roll in later years.  Things such as
paper skew, timing and measurement errors and focus errors all affect
scans, and can make the difference whether a scan can be processed with
reasonable confidence into a reconstructed master.

You can always get a master from a scan, but it may well not match
the original!  Scan processing software needs careful setup and the
reconstructed master must be checked in some detail.  Here, software
capabilities are the key: hand-aligning the grid to the image is
laborious but gives reasonable confidence of the result, while
automated processing is much faster but gives significantly less
certainty (think of the results from OCR processing, for instance).

Personally I would like to see a setup where the original scan image
is shown with an automatically-obtained master shown on top of it to
permit manual attention only where the automation fails.  One day,
perhaps.

Next, when you start reconstructing roll masters from scans you soon
become aware of some of the oddities observed in rolls.  An early point
noted was the number of manually-inserted punches in Ampico rolls.
Another was the huge variation in row step-sizes in British Duo-Art
rolls, presumably an artifact of perforator wear, which can trick
software into inserting false extra rows into these rolls, or require
huge amounts of manual synchronising.  These rolls seem particularly
susceptible to poor scans.

Then you have production variations.  QRS are the most famous
exemplar of the rejuvenated roll, and some titles have several only
partially-related variants.  I'm told that there are some examples
of alternative performances on reproducing systems, but more commonly
there are re-coded expression.  Occasionally small changes, such as
between British and American Duo-Arts, or major recoding such as seen
in many Duo-Art Audiographic issues.  To be thorough you'd need all
variants -- but there's no documentation of such variants.  If scans
are properly labelled their source ought to be known in later years to
allow comparisons; I wonder how many roll-scanners bother with this?

Another types of production variation was physical.  Duo-Art roll
perforators varied in the relative position of "snakebites" to notes,
for instance, and British pianos had the sustaining pedal positioned
slightly differently to American versions.  Different perforators used
different punch sizes, and the roll step-rates and marked speeds were
varied so that the size of the bridges and inter-note gaps were
preserved.  I have seen three copies, side-by-side, of the same Duo-Art
roll with different marked speeds and physical lengths.  The pattern of
the notes was the same, and all would have used the same master roll.
Undue importance is often given to the length of copied rolls, when
it's really only mimicking one example from a wide variety all of
equally historical validity.

All of this is much more complex than simply scanning three copies of
each roll!  Obviously there's a substantial element of "who cares"
about this sort of issue.  To many player owners all of the above will
seem like pointless intellectualising of a simple entertainment.
Others thrive on this sort minutiae but never did like the music much.

Julian Dyer