Re: [Bug-gnubg] Player Records Overall ratings

[Joern Thyssen]

On Sat, Mar 22, 2003 at 11:17:13AM -0300, Albert Silver wrote
> Hmmm... Well, first of all, yes it was done using an older build, but
> the question wasn't whether I *could* add the results as I was always
> able to. The problem was with the result shown for the Overall result.

Yes, but my point is that this could be garbage if you added them with
an pre-March 13 version of gnubg. 

Note that the total error rate is kept as separate field in the player's
record -- it's  *not* some trivial function of the chequer play and cube
play error rates.

You can try the following with a post-March 13 version:

(i) find and make a backup of gnubgpr
(ii) delete gnubgpr
(iii) load the matches you sent in the previous mail, and add the
statistics to the player records.  If you still have "wrong" player
records, then we have a problem!
(iv) restore you old gnubgpr

> I'm attaching here a zipped screenshot so you can see exactly what I
> get. BTW, since you mentioned discrepancies between evaluation results
> with the Linux and Windows version, did you test this on a Windows
> version?

It has nothing to do with discrepancies in evaluations; no matter how
wrong an evaluation is, the match statistics would still obey the
inequality I set up in a previous mail.

I did the four steps I decribe above on win98 (March 17 build) and linux
(March 23). Here are links to the resulting player's records:

<URL:http://jth.freeshell.org/win.jpg>
<URL:http://jth.freeshell.org/linux.jpg>

As you can see the results are complete identical, and all of them obey
"overall <= chequer + cube".

Jørn