[Bug-gnubg] Inconsistent match analysis numbers

[Ian Shaw]

I find the signs of the error rates and luck rates (and the new ratings)
confusing.

I would expect all -ve numbers to indicate a reduced chance of winning the
match, and all +ve numbers to indicate an increased chance of winning.

This is the case for the luck rates. Bad luck is a -ve number and good luck
is +ve.

Error rates are reported with errors as +ve numbers in the match analysis
window. This is in contrast to the errors in the annotation window, where
they are shown as -ve relative to the best move. (The exception is an
incorrect take which correctly shows up as +ve for the doubler.)

Similarly, the player who is evaluated to be the favourite is shown as
having a -ve rating w.r.t. to the opponent. This seems counter-intuitive.
Similarly, the rating w.r.t. "perfect play" is a +ve number. This doesn't
seem right either.


I am also unsure what the "perfect opponent" referred to is. Three possible
scenarios come to mind.

1) The current analysis settings are assumed to be perfect.
2) An arbitrary GnuBg setting (e.g. 2-ply) is assumed to be perfect, and
GnuBg knows the relative strength of it's current setting. I notice that
currently gnu (2 ply?) is rated c. 2000 and gnu_zp is c. 1900 on FIBS.
3) There is a theoretical maximum rating (e.g. 2300) and GnuBg knows its own
FIBS ranking for the current analysis setting.

-------
Suggestion for improved usability:

It would be good to be able to launch rollouts of selected plays or cube
decisions directly from the annotation window without having to get a "hint"
first.
-------
Finally, the link <http://www.freeshell.org/~jth/html-images.zip> posted to
r.g.bg does not work and I can't get a decent export to html now. Can this
be sorted?
--
Ian Shaw