I was intrigued by Jeffrey Wood's comments about the two versions of
Chopin's Fantaisie in F minor, Op. 49, played by Harold Bauer, and had
a look at the London-made shorter version which I have.  I ran this
through the roll-scanner and used Warren Trachtman's software to
re-create the master file to have a look at it in more detail.

Only certain perforator advance rates make sense for roll cutting,
because of the need to have the correct width of strengthening bridge.
Aeolian used two rates for Duo-Art rolls, notionally 21 rows per inch
(rpi) or 31.5 rpi.  This was a perforator setting, not really a case
of "two to one" or "three to one" masters. The same form of master
would have been used for all rolls, which were all cut on the same
perforators with different cogs fitted to the paper-drive mechanism as
needed.  Some surviving perforators still have these alternative drive
cogs.

The effect on the master roll is that at 21 rpi a bridge is created by
omitting one punch row, and at 31.5 rpi a bridge is created by omitting
two punch rows.  Clearly for the same paper speed, adopting 31.5 rpi
gives substantially better time resolution than 21 rpi, and many
classical rolls are cut at the 31.5 setting -- but by no means all of
them, with even quite late rolls being cut at 21 rpi.  The choice of
step rate would be made when the "original" roll from the recording
perforator was converted into a master.

The longest Duo-Art rolls that I have seen are uniformly no more than
90 feet of music, plus leader and runout.  The copy of 6266 which I
have is 66 feet long at tempo 65, cut at 31.5 rpi.  The longer version
would presumably have been cut at 21 rpi, and so would have been 99
feet long (and tempo 95 or 100), about 10% more than the normal
maximum.  You can see why it was adapted into a shorter version.

What's interesting is how it was shortened.  Clearly very little was
done to the original master roll other than adapting the strengthenin
bridges.  Throughout the roll there are many very short inter-note
gaps, which means that the gap between notes barely covers the tracker
bar and so gives the piano mechanism almost no time to reset.

Looking at the reconstructed master, there are many cases when it would
have been possible to trim a punch off the end of the preceding note to
make it slightly easier on pianos.  It simply wasn't done.  With the
wonders of modern computer-controlled recutting this is easy enough to
do, if anybody wants such a roll.

What's odd with this roll is why it was issued at tempo 100 and 21 rpi
in the first place, when tempo 90 at 31.5 rpi would have given a more
accurate roll of the normal maximum length.  Unfortunately, no amount
of over-detailed technical observation such as the above can give any
true insight into what went through the minds of the folks at the time.
So we'll never know!

Julian Dyer