Roland uses a note off with velocity zero (0) when the key is retriggered without activating the top sensor.
So if the highest contact point is s0, and middle contact point is s1, and the lowest contact point is s2, then somebody pushes the note down all the way to the bottom, then all three sensors are triggered, and stay triggered/active.
And the note velocity can be measured from the time difference t12 (s1 triggered first, followed by s2 triggered second).
And when the note gets released, then maybe a release velocity can be measured v21 (where s2 becomes untriggered, followed by s1 becoming untriggered).
Here - it looks like -- if the released key gets adequately released to the point where s0 becomes untriggered again, then the measured release velocity gets replaced with a ZERO velocity.
But if the released key doesn't get to the point of untriggering s0 - meaning that the key doesn't get back to near the horizontal position - (and assuming s1 and s2 are both back to untriggered state) and if the person pushes the key down again, then s1 and s2 both become triggered again, resulting in another note-on velocity measurement. But just before broadcasting the new note-on velocity, the system first broadcasts a note-off with a release (ie. note off) velocity -- except the actual release velocity is replaced with a value of '1'.
So actual release note-off velocities aren't broadcasted. They're replaced with either a 0 or a 1, depending on the situation.
