Ernie is on the right track, but this one is different to a toggle.  These IR signals decode as OrtekMCE, and it employs 3 states:
For reference, in IRP notation OrtekMCE is: {38.6k,480}<1,-1|-1,1>([P=0][P=1][P=2],4,-1,D:5,P:2,F:6,C:4,-48m)+{C=3+#D+#P+#F}
The issue is that the repeat frames are not all identical. P is a position code: 0 for the start frame, 2 for the end frame and 1 for all frames in between.   So a frame is sent with P = 0 when the button is first pressed.  While the button is being held down, P has the value 1. When the button is released, the last frame is sent with P=2.

This is an excellent system which allows the receiver to know exactly what the user is doing (any computer, phone, tablet keyboard uses a similar system.)  However the formats used to describe IR signals by Pronto Hex, Global Cache, etc aren't capable of representing this sequence.

UEI remotes use executors (subroutines) to send signals, so it is easy for them to generate the 3 states.  I don't know if other manufacturers can do that, but it is clear that a system which depends on the learning approach will be  sub-optimal.