It's the whole self-similar thing:
The first 3 notes are A-B-A. A-B-A repeats a lot of places so let's replace it with α.
So now play the instrument longer (imagine zooming out one level). A-B-A-C-A-B-A = α-C-α. Hey, that looks similar. Let's call it β for short.
Zoom out the timescale again and now the instrument plays β-D-β, which also repeats in a bunch of places so…