[Bug-gnubg] Gap in guestimated ratings?

[James F. Kibler]

Thanks to Maaereyes Solomon (author of 3DFiBs) I downloaded gnubg for
Windows a couple of weeks ago and find it an excellent learning tool.  

The attached .gif image has a simple plot of the gnubg guestimated
ratings from matches I've played on FIBS vs. the date the match was
played.  The blue diamonds are the ratings for each match, the pink
squares are the daily averages, and the black line is a polynomial fit
to the data.  You'll quickly notice that there are no match ratings
between 1300 and 1500.  One would expect a Rayleigh distribution of
match ratings with a mean approximately equal to one's FIBS rating.  My
current FIBS rating is about 1500.  The average of these ratings is
1615.  Well, I could understand if the gnubg rating was biased relative
to the FIBS rating, but how could it have a 200 point gap in the
middle?  Gnubg says I'm either pretty good or very bad, not the
middle-of-the-road player that I really am!

Can you tell me what factors are used in gnubg to estimate the rating? 
I was thinking to collect some more data and do a least squares fit to
the estimator in an attempt to improve the coefficients.

Thanks, Jim Kibler