I will try to answer some questions, but I am no expert, maybe some people might have better explanation. But I think our understanding level is the same. I have read too the gpg paper and it seems to be rather deep.

1) Right, to my understanding
2) I noticed that too, and I am as confused as you, as I can't find any satisfying response about the difference between default values and the documentation, and how to estimate the values for our own setup.
3) I think you're right about 'Number Periods for Interference', but Hy might tell us if it is correct.
4) You're right, Hy use a OAG with a long FL, and we use a guider for that (btw, we have the same guider FL and camera pixel is pretty the same too). For binning, as we have a fast guider (and low FL), I would stay in bin1. I have no issue of low snr with such setup. But I'm pretty sure the algorithm can handle big pixels (or bin2) as it can determine the centre of a star with fraction of pixel.
5) yes it's confusing me too, but I guess as Hy has a long FL for his oag, he needs to lower this number. But yes, I'd like to have a better explanation of that figure to determine it for my setup.
6) Right, GPG resets as you slew away and go back for a period of learning as it has no clue about the current position in RA.

And for your last question, my understanding is GPG adds some corrections to the guiding after several worm periods, but the guiding runs at the start, no matter GPG is learning.