The May/June 1996 AMICA Bulletin has several articles on how Ampico
recorded the velocity (and therefore the note's dynamics) of the hammer.
Ampico's recording device contained two recorders: one was used to
record the notes played by the pianists on paper, similar to a standard
marking piano of the day. Integrated into and right next to the note
recorder was the dynamics recorder.
According to the diagrams, the dynamics recorder had 3 wires going to
each note on the piano. Each note has two sets of contacts and a
common ground (totaling 3 wires) that are 'made' just as the hammer
impacts the string: one contact calibrated to connect _just_ before the
second. The dynamics recorder's spark chronograph then graphically
showed on a second roll of moving paper the difference in time between
the two contact connections, allowing the arrangers to determine the
*hammer velocity* at that point (faster moving hammer = shorter time
difference between the 2 connections, slower hammer = longer time).
As a side-note, when the recording was later arranged, the Ampico
keyboard was equally split into 8 segments encompassing 9 and 12 notes
grouped in 3 or 4 3-note clusters evenly distributed across the compass
of the keyboard so that each 12-note section could have their dynamics
controlled independently of the other.
It's worth noting that this is oftentimes hardly what the pianist
actually played. For instance: what happens when 2 or more notes (as
in a chord) are struck with differing forces within a 12-note cluster?
I would think the editors approximate the volume and, with no choice,
make those notes the same volume. If true, realize that in this
situation, the original dynamics are now lost! But this was their
way of approximating a real recording by arranging the expression in
12-note segregated 'clumps'.
To be sure, there is only so much room at the margins of a piano
roll to put the dynamic information, but this hardly represents a
"reproduced" performance in my mind, though this was the mantra of
the 'reproducing' sales pitch, then and now.