There have been a few threads recently about game commentaries that
seemingly depend too heavily on engine annotation without the
application of some common sense. I recently played a quick sparring
game against Shredder 10 where the engine annotation by blundercheck
was found wanting.

Strangely even though all the engines in infinite analysis or deep
positional analysis could easily see that the move I played was
inferior none of them said so in blundercheck mode. (Tried Fritz8,
Shredder10, Crafty19.19, Rybka2.3.2, Comet B68). The annotation time
is 20s with an 8 cp window. But they did in full analysis mode even
with a 10s shorter time..

The odd thing is that the position where the fault shows up is already
massively won for black, but the move that I played was not the
strongest way to finish the game by far. I was spooked by 25. Qh4
believing incorrectly that Shredder had found a quick way to mate me
before I could finish him off. Instead of going for the kill with b3
(my original plan) or Bb7 (Rybka thinks the latter better) I played
the considerably weaker 25. .... Qf8.

Only Rybka commented on this move once(not reproducibly and it did not
leave a "Rybka 2.3.2 32 bit" annotation tag), whereas in all the other
positions where it found an improvement it did. Several engines show
the right move(s) and evaluations in the active window but the correct
result and line of play does *not* get added to the game annotation!!!

Here is the full game from start to end.

[Event "Blitz:10'+2""]
[Date "2007.08.31"]
[White "Shredder 10"]
[Black "Brown, Martin"]
[Result "0-1"]
[ECO "B63"]
[WhiteElo "2145"]
[BlackElo "2078"]
[PlyCount "70"]

1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 Nc6 6. Bg5 e6 7. Qd2
Be7 8.
O-O-O O-O 9. f4 a6 10. Nb3 b5 11. Bxf6 gxf6 12. f5 Kh8 13. g3 b4 14.
Ne2 a5 15.
Nf4 a4 16. Nd4 Nxd4 17. Qxd4 e5 18. Qd5 Ra5 19. Bb5 Qb6 20. Kb1 Rxb5
21. Qd3
exf4 22. gxf4 Ba6 23. Rhg1 Qb8 24. Qg3 Rg8 25. Qh4 Qf8 26. Rg3 Rxg3
27. hxg3
Qg7 28. g4 Rc5 29. Qh5 b3 30. cxb3 axb3 31. axb3 Be2 32. b4 Bxd1 33.
bxc5 dxc5
34. Qh4 Qxg4 35. Qh2 Qe2 {-12.81/12} 0-1

35. ... Qf3 would have been even stronger, but the engine still
resigned after Qe2 (me under time pressure).

Here is a short snippet to isolate move 25 where blundercheck is less
than helpful.

[Event "Blitz:10'+2""]
[Date "2007.08.31"]
[White "Shredder 10"]
[Black "Brown, Martin"]
[Result "0-1"]
[ECO "B63"]
[WhiteElo "2145"]
[BlackElo "2078"]
[Annotator "Crafty 19.19"]
[SetUp "1"]
[FEN "1q3r1k/4bp1p/b2p1p2/1r3P2/pp2PP2/6Q1/PPP4P/1K1R2R1 b - - 0 24"]
[PlyCount "4"]

24... Rg8 {-4.47/13} 25. Qh4 {-5.39/10} ({Shredder 10:} 25. Qd3 Rxg1
26. Rxg1
a3 27. Qg3 Qf8 28. Qd3 axb2 29. h4 Ra5 30. Qd4 Qh6 31. Qxb4 Rc5 32.
Qb6 Bb5 {
-6.41/13}) 25... Qf8 {-4.37/12} 26. Rg3 {-4.66/12} 0-1

Actually the last engine annotator was Rybka2.3.2 not Crafty 19.19.
This is the shortest game segment that the blundercheck and annotate
functions will accept. I would be interested in suggestions why when
the engine clearly does find the right line and the conditions for
including it in a blundercheck are met nothing useful happens.
Note that there is a 102 cp jump in the evaluation moving from 25. Qh4
to 25. .... Qf8 and no remarks at all!

It seems that full analysis sometimes finds important things that
blundercheck misses. This is a pity as annotate tends to be
excessively chatty. I prefer blundercheck, but what else is it
missing?

Thanks for any enlightenment.

Regards,
Martin Brown