I've been watching the discussion about scanning and MIDI, etc.
While accuracy has been mentioned, one aspect that hasn't been
discussed is storage longevity.

I was reminded of the seriousness of this issue by two things; one is
an article in the MIT Technology Journal that discusses the problem of
long term storage of digital media, and the other was an article in one
of the pop news magazines about the long term storage of data from
digital cameras.

The MIT article discussed the problem of data coding and decoding in
a technology where standards are largely driven by market forces.  For
instance, users of Encore 4.1 found, much to their distress, that an
upgrade to Windows XP rendered the program inoperable, and thus their
data and music unreadable.  A required upgrade to Encore 4.2 (and its
new data format) then rendered the data unreadable to anyone running
Encore 4.1.

A similar situation happens to any user of Microsoft Office.  Software
upgrades render documents unreadable to previous versions.  The only
solution is to upgrade.  These are format changes that occur at least
every three years. What MIT was worried about was whether or not the
historian 100 or 200 years from now would be able to decode important
and historic documents, pictures, music etc. when it was created under
such circumstances.

Aside from format problems, there is the question of the storage media.
(The oft-repeated horror story is that there is data from the Voyager
probe that cannot be read.  Either the oxide falls off the tape, or the
data has 'faded', or there are no machines available that can read the
tape; they were all sold off during hardware upgrades or can't be fixed
for lack of parts.)  One of the media mentioned as being the best hope
for long term storage was paper (albeit a very specialized archival
kind of paper).

The pop magazine article was on a similar issue.  Users of digital
cameras are discovering (too late) that their images are very fragile.
A simple hard disk or floppy crash destroys them.  More interestingly
they are finding that common transfers of those images using ink jet
printers as equally as fragile.  Ordinary jet paper and ink don't hold
color quality.  Mixtures of paper types and ink types have the same
result.

The only thing that appears (for the short term) to have any hope of
lasting is paper and ink obtained directly from the printer manufacturer;
because that paper and ink is specially engineered for that printer.
For the long term, the only thing that appears to have any hope of
surviving decades or longer is to have the digital images transferred
to ... high quality photo paper using traditional methods!

The reason to store the data for mechanical musical instruments on
paper is that it has been empirically proven to last longer.  The
reason to store the data as holes punched in the paper is because the
data will not fade or distort or otherwise lose meaning and thus be
un-retrievable.  We do not have to worry whether or not the data can be
read 100 or 200 years from now.  Even if badly or naively stored, the
data will still be readable.  This media has an excellent long-term
track record.  Modern digital media has an awful track record.  In only
30 years, valuable unique historical data from Voyager is irretrievably
lost.

Imagine the horror that awaits the roll manufacturer who has their
masters stored on 5 1/2 inch floppies playable only on 20 year old
hardware.  Run forward another 20 years to the archivist faced with
cracked CD-W media that can only be read on 20 year old CD readers
mounted on 20 year old hardware, or 20 year old backup tapes that
can't be read at all because the hardware can't be found to read them.
The list goes on and on.

When the current members of the roll scanners group go to their
great reward, who will take over the maintenance of the storage media
on which their work has been preserved?  It's a roll of the dice to
know if the media will fall into the hands of interested and caring
archivists, or dolts who just pour all the stuff in cardboard boxes and
shelve it.  Digital media requires much more intensive care and ongoing
attention if the information is to survive.

We know what happened to punched paper media when the archivists
just piled the stuff in cardboard boxes and shelved it.  It is still
readable 100+ years later, having been given no better care than to make
sure it didn't catch fire or the rats and bugs didn't eat it.

Aside from longevity, there is the issue of accuracy.  If you are with
me so far, and agree that paper is the media that has the best chance
of being around 100+ years from now, then you'll now see the reason for
extremely accurate scanning and perforation.

I now turn to a technology that is producing the closest thing
available today to perfect recreations of the original data -- the
combination of Wayne Stahnke's data sourcing technology, and Dave
Saul's perforating technology.  I say closest thing because there is
still the question of punch diameter and paper advance mechanics (is
it capstan advanced or take-up-spool advanced?).

If one doesn't consider these for the moment, then this combined
methodology clearly produces perfect copies of the original data.
I say this because in the transfer from original to copy, there is no
impedance mismatch.  One data point in a row in a matrix represents one
punch in a discrete position on the paper.  The next row in the matrix
represents one advance of the paper.  More to the point, the software
that scans, manipulates and encodes the data and the equipment to punch
the roll are designed precisely to do only the above.  One data point
represents one punched hole.

MIDI does not describe a discrete position in two dimensional space.
It describes an event in time. To achieve a punch point description
in strict MIDI requires a MIDI On at a point in time, and a MIDI OFF
at another point in time.  One can jigger this to achieve a mock
singularity (the on time is the same as the off time, or one just
chooses to use only ON times), but the act of jiggering the standard
at all implies that the standard is being used in a non-standard way --
hence impedance mismatch.

A similar mistake is to assume that a finer granularity in the data
representation will achieve the same level of perfection in the
finished product.  It cannot do this for the same reason that one
cannot blindly convert between floating point representation of numbers
and integer representation of numbers and be dead certain of the
outcome.  1 divided by 3 stored in integer arithmetic does not yield
1/3 when subsequently retrieved (it comes back as 1).  1 divided by
3 stored in floating point comes back as a close approximation of 1/3.

A value stored in integer arithmetic can be compared with another
integer and the result (is there a match or not) is a completely
reliable answer.  A value stored as floating point, compared in the
same manner is NOT reliable.  A lot of work must go on inside computer
programs to correctly determine the outcome of a test for value match
between floating point numbers.  So you can see that naively converting
between integer and floating point during calculations can produce
significant data skew.  For this reason, most financial packages do
all the calculations in integer arithmetic and specify where the decimal
point should appear in the printout.

Similarly, scanning and perforating methodologies that convert between
different data representations are by definition inaccurate.  They
cannot be accurate because there are constant compromises and
recalculations that occur when determining the discrete punch position
that the data represents.  In the jargon, these are called "sampling
errors."  At its worst, this produces rolls like the early days of
recutting where onset and offset of ports can vary by as much as a
1/16-1/8 of an inch.  Chords come out as arpeggios.  Dancers lose their
balance.  Every subsequent generation using such methods only gets
worse.

The Stahnke/Saul combination introduces no impedance mismatch and no
sampling errors and hence is reliably accurate; subsequent generations
using the same methods will not introduce further errors.  One single
data point represents one punch in a discrete positional matrix. There
is no intermediate interpretation or recalculation performed during the
transfer between data point and punch operation.

One does not have to interpret something like "the original slot is AA
inches long at tempo BB, which sort of equates to CC midi ticks which
sort of equates to DD micro punches."  Instead, the transfer between
a data point and a punch operation is "The machine should now punch
a single hole at row AA, port BB."  Thus there is no necessity to adjust
for things like the effect of errors introduced by tracker bar port
size during pneumatic scanning.  A two-punch slot is two punches in
length, not NN ticks long being translated to YY micropunches long.

(As a side benefit for the archivist 100 or 200 years hence, since the
scalloping produced by the perforator will be preserved during the
recreation, the original master matrix will also be available in the
same way that it is available to us today.  Thus even if the digital
version is lost, the accurate matrix data will still remain.)

To conclude, I strongly urge those who are scanning rolls to consider
using perforated media as the long-term storage of choice.  High
quality (even low quality) paper is proven to last 100+ years even
under hostile storage conditions.  I further urge those who are
perforating rolls to consider very carefully the methods used to read
the original data and perforate the final output.  100 or 200+ years
from now, long after the primary sources have crumbled into dust,
a copy that "sounds good enough" or is a "reasonable facsimile" will
not cut it.  Years from now do you want to be remembered for "Clark"
quality or "Capitol" quality?

Unless you adopt methods that produce accurate punch for punch copies
from expertly restored master images, like the ones described above,
you will be producing distortions (high-quality representations but
distortions none the less) and thus doing a disservice to yourself
and posterity.

George Bogatko