[Bug-gnubg] GNUBG strange game stats (?) + random questions

[Massimiliano . Maini]

Hi all,
I've just downloaded GNUBG (Win) build-030820 to check the FIBS rating thing.
First match : 7pt, I won in 3 games (1-1, then in 3rd game the cube went to 8).
But there's something strange in the match statistics (here's the cut&paste).
For info, GNU was playing at World Class, Analysis and Evaluation at Supremo
(GNU not analyzed), Tutor on (same as Analysis), all running under NT.

                                      gnubg                MaX
Checkerplay statistics:
Total moves                           40                   40
Unforced moves                        0                    39
Moves unmarked                        36                   0
Moves marked good                     4                    35
Moves marked doubtful                 0                    4
Moves marked bad                      0                    1
Moves marked very bad                 0                    0
Error rate (total)                    0                    -0,386 (-4,477%)
Error rate (pr. move)                 n/a                  -0,010 (-0,115%)
Checker play rating                   Supernatural         Advanced
Rolls marked very lucky               0                    1
Rolls marked lucky                    0                    2
Rolls unmarked                        0                    36
Rolls marked unlucky                  0                    1
Rolls marked very unlucky             0                    0
Luck rate (total)                     0                    +1,229 (+37,344%)
Luck rate (pr. move)                  0                    +0,031 (+0,934%)
Luck rating                           None                 Good dice, man!
Cube decision statistics:
Total cube decisions                  5                    25
Actual or close cube decisions        5                    9
Doubles                               2                    3
Takes                                 2                    1
Pass                                  1                    1
Missed doubles around DP              0                    0
Missed doubles around TG              0                    0
Wrong doubles around DP               0                    0
Wrong doubles around TG               0                    0
Wrong takes                           0                    1 (-0,003, -0,042%)
Wrong passes                          0                    0
Error rate (total)                    0                    -0,003 (-0,042%)
Error rate (per cube decision)        0                    -0,000 (-0,005%)
Cube decision rating                  Supernatural         Supernatural
Overall
Overall error rate (total)            0                    -0,388 (-4,519%)
Overall error rate (per decision)     0                    -0,008 (-0,094%)
Equivalent Snowie error rate          0                    -4,9
Overall rating                        Supernatural         Advanced
Actual result                         -50,00%              +50,00%
Luck adjusted result                  -12,66%              +12,66%
FIBS rating difference                -169,89
(based on luck adj. result)
Estimated abs. rating                 2050,0               1925,8
(based on error rate per decision)

Remarks :

1) I got 1 bad, 4 doubtful and 35 moves marked good over 40.

I've noticed something strange that you should be able to reproduce.
Start a match where you move first, play and wait for GNU to play, then
look at the game statistics : you'll have one move marked good and if
you open the game record and annotation window, clicking on your move
you'll see that it is not marked in the game record window,but in the
annotation window the listbox for checker play is on 'good' (and also
the one for cube decision). Now save the match (.sgf) and reload : in
the annotation window cube decision and checker play are unmarked
(empty listbox). This  agrees with the game record window, but the game
stats still report a good move.
The same happens when GNU starts : after having played your first move
(this time you have a cube decision here) wait for GNU to play, then
open the game record window and the annotation window, and look
at that move : your cube decision (usually a no double at this stage) is
marked good and your cheker play too (and game stats reports one good
move for you). Again, analyzing the game at this opint doesn't change
anything, while saving the match and reloading 'fixes' the problem in
the annotation window, but not the stats.

I also have an example with a 1pt match analyzed giving 40 moves, 35
unforced, 40 marked as good.

2) I got 25 total cube decisions, 9 actual or close cube deciosions.

Just to be sure, what's the 'Total cube decisions' ?
Does it count all the possible double/no-double plus all the take/drop ?
And what's the 'Actual or close cube decision' ?
Does it contain all the take/drop plus the "close" double/no-double ?

3) GNU got 4 good moves.

How can GNU Supremo detect 4 good moves from GNU World Class ?
And in general, how are good moves detected ?
Also, I wasn't analyzing GNU ...

I have saved the 7pt match (analyzed,.sgf), if needed ...

MaX.

P.S.
COSMETIC : The FIBS rating difference (-169.89) is shown only in GNU's
column while the luck adj result is shown in both (-12.66%, +12.66%).
I think that showing it in both columns with opposite signs removes any
possible ambiguity.
Also, now that both fibs rating estimations are available (based on luck
adj result and based on error rate) it would be nice to store them in
player records to check if they seem to converge nithe "long run".