To expand on yesterday's note and Robbie's comment about the archival
qualities of MIDI files: MIDI is an ideal medium for archiving roll
data, but has to be used intelligently to do so.

MIDI files with dynamics simulated from roll expression coding are
not archives of the roll data because you cannot get back to the
original expression coding from the MIDI velocities.  (You can estimate
it, but to get anywhere near you would need to know the original
conversion algorithm, and even then you'd be guessing -- which is not
good archival practice!)

However, the MIDI file format handles up to 127 notes, each with 'on'
and 'off' codes (as well as loads of stuff).  This is quite enough for
directly storing all the 'punch' or 'no punch' information of most roll
formats, mapping every perforation column (both notes and ancillary
coding) on the roll directly to a MIDI note.  Indeed, Richard Stibbons
has defined use of MIDI format for all of the various Rollscanners
derivative files.

The original scan is a graphical file, but as soon as it is interpreted
to work out where the perforation columns are across the rolls and when
actual perforations occur along the roll, this can be stored in MIDI.

The initial analogue interpretation simply allocates the right 'note
on' and 'note off' numbers and records the time to the nearest MIDI
time point.  The 'holy grail' recreated roll master quantises the
timing by identifying the actual position of each punch, and this can
be stored by redefining the MIDI timing so a single count represents
one punch row.  You get loads of short notes, with every bridge being
an 'off'-'on' pair.  Stripping the bridges out uses exactly the same
file format but ends up with longer notes.  Converting the file to
'real' MIDI removes the notes representing ancillary coding and imposes
that information as controllers and MIDI velocities on the notes
themselves (as well as reinstating the timing to ordinary clock
values).

The beauty of using MIDI for all these various representations of the
roll is that the same file-reading software can be used throughout --
which makes a great deal more sense than having bespoke file formats at
each stage!  What's more, MIDI is a widely-understood standard so using
it makes the archived roll data more accessible.  A small drawback is
that MIDI files are a little bigger because they store additional data
that is not required when representing rolls.  Given disk sizes these
days this isn't a problem.

Julian Dyer