[pgn][Event "test"]
[Site "?"]
[Date "2024.08.27"]
[Round "2"]
[White "Scisys Kasparov Turbo 16k, level 7"]
[Black "Mephisto MMI, level 6"]
[Result "1/2-1/2"]
[ECO "C68"]
[PlyCount "167"]
1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 4. Bxc6 dxc6 5. O-O {last theo} Qd6 {last theo I
have also learned something from each of my machines, no matter how weak.
Specifically here - the opportunity to train this variant with white brought
me some valuable points later on.} 6. Nc3 {This is certainly not the way to
seek an advantage in Spanish, on the other hand, throwing your opponent out of
"his" variants also makes sense.} (6. Na3 $5) 6... Bg4 {0,25} 7. h3 Bh5 {h5 is
also possible here} 8. d3 (8. g4 Bg6 9. d4) 8... Nf6 (8... O-O-O) 9. Bg5 Be7 {
That's what I'm always afraid of with games of machines - that a position will
arise that neither understands and the whole game will degrade into quite
pointless stone-pulling.} 10. Qe2 O-O {-0,25} 11. Rad1 {Even in the next run,
the computer doesn't understand that it can't do it without g4. The malusy
valuation function for weakening the king's cover prevents it from doing so.} (
11. g4) 11... Qb4 (11... Qe6) 12. Rb1 (12. g4 Nxg4 13. a3 $1 Qd6 14. Bxe7 Qxe7
15. hxg4 Bxg4 16. Qe3 f5 17. exf5 Rxf5 18. Nh2 Bxd1 19. Rxd1 $16) 12... Qc5 13.
Be3 Qd6 14. Bg5 c5 {I was almost hoping for repeated moves before watching
this, but Black doesn't want to.} 15. Bh4 {He finally found a target, the e5
pawn.} Rad8 {-0,39} 16. Bg3 Nd7 17. Nd5 f6 18. Nxe7+ $6 {That's certainly not
a good trade, a centralized slider for a shooter who has nowhere to play.
Again, a valuation feature, shooters are usually slightly more valuable to
programs so they don't get put in bad positions (as often) against a pair of
shooters.} Qxe7 19. Rfd1 Qe6 20. Ra1 Qb6 {-0,33} 21. Rdb1 $6 Qd6 22. Qe3 Bf7 (
22... Bxf3 {That's what Stockfish says. I wouldn't want to. There's still a
chance the position will open up and then Bishop will be a shame.}) 23. Nd2 (
23. b3 {This belongs in the Spanish exchange, as soon as Black plays b5, the
move c4 takes away his chance to get a free pawn.}) 23... Qd4 24. Qxd4 {And
it's going to get worse, we'll most likely see an endgame, that was always a
disgusting experience among computers back then.} cxd4 25. Nb3 (25. f4 $14)
25... Bxb3 $2 {Strategically bad trade. The important black pawn e5 is on the
white bishop's color, if the position opens up, black will have big problems.}
(25... g5) 26. cxb3 $2 {Pawns are making their way downtown - with a few
exceptions... Hopefully he liked the free column, but the d3 pawn is now weak.}
f5 {Black doesn't understand this either, if he has a knight against a bishop,
he should time his opening position properly.} (26... Nc5 27. Rd1 g5 $17) 27.
exf5 Rde8 {Why?} (27... Rxf5) 28. Rc1 c5 (28... c6) 29. b4 $1 b6 30. Rc4 Rxf5 {
-0,55} ({It's not better, but I was curious how Stockfish imagines it.} 30...
b5 31. Rc2 cxb4 32. Re1) 31. bxc5 Nxc5 32. Rd1 Kf7 33. b4 Ne6 $2 {And that
already leads to some losses, too many weaknesses.} (33... Na4 34. Rd2 Nc3 $11)
34. Rdc1 $2 (34. Rc6 {and the pawn can no longer be defended} Rb8 $2 35. Re1)
34... a5 {additional weakness} ({more chances gave} 34... b5 35. Rc6 Nf4 36.
Bxf4 Rxf4) 35. bxa5 bxa5 {0,02} 36. Rb1 Kf6 37. Kf1 g6 38. Rb6 {wrong rook,
but should have been ready to go on b5} (38. Rc6) 38... Ra8 $2 {Black's
defense is tricky and a much better player would have trouble. Proper timing
is what matters here. Releasing the knight should have been a priority.} (38...
Kg5) 39. Rcc6 Re8 40. Ra6 e4 $2 {1,00} (40... Kf7 41. Rxa5 e4) 41. dxe4 Rb5 42.
e5+ $4 {He doesn't know what the plan is, but he knows that if he puts the
king in the center, the valuation function will add hundredths... The festival
of endgame mistakes begins... And specifically here, in the words of one Czech
grandmaster, the bishop is just pointing a gun up his own pawn's ass.} (42. Kg1
$18) 42... Kf5 $11 43. a4 (43. Bh2) 43... Rb1+ 44. Ke2 Nf4+ $4 {A new chance
for white. Black loses a pawn and should look to hold back the "a" advance.} (
44... h5 $11) 45. Bxf4 Kxf4 {1,77} 46. e6 Ke4 47. f3+ Kd5 48. Rd6+ Ke5 49. f4+
$2 (49. Rdb6 $18) 49... Kxf4 (49... Ke4 50. Kf2 Rb4 51. Kg3 Rxa4 52. Ra7 d3 53.
Rad7 Kf5 54. Rxd3 h5 55. e7 Kf6 $14) 50. Rxd4+ Ke5 51. Rd3 Ra1 $2 {1,64} (51...
Rb2+) 52. Rxa5+ Kf6 53. Rf3+ Kxe6 54. Re3+ Kf7 55. Ra7+ Kf8 {1,91} 56. Rxe8+
Kxe8 57. Kd3 $4 {A big problem at the time. The evaluation function could not
be complex because it would hamper the calculation. But simple sometimes led
to such stupidity, it's not about anything other than some hundredth for the
king in the center.} (57. Kf3 {cca 5,35 Stockfish}) 57... Ra3+ $4 {A silly
loss of tempo puts White back into the win.} (57... Rg1 $11 {A clear draw
according to Stockfish, but Mephisto would probably struggle to defend the "a"
pawn.}) 58. Kc4 h5 59. Kd5 $2 (59. a5 Kd8 60. a6 Kc8 61. Ra8+ Kc7 62. a7 Kb7
63. Rg8 Kxa7 64. Rxg6 {And since White is closer to the pawn, it's done. Which
is too many moves for the machine to count.}) 59... h4 60. Ra8+ $4 {And now
it's a tie if Black could defend it.} (60. a5 {It still went, black doesn't
have time to take on g2.}) 60... Kf7 $11 {0,94} 61. Ra5 {If a junior played
something like that, his coach would freak out. The cause is probably the king
in the center again, white won't let black in.} Ra2 62. Ra7+ Kf6 63. Kc5 {Only
when g2 falls does the need to cover the pawn outweigh the bonus for the king
in the centre.} Rxg2 64. Rh7 Kg5 65. Rd7 Rg3 {-0,84} 66. Rd5+ Kh6 67. a5 Rxh3
68. Kb4 $4 {It doesn't add up to how dangerous the black pawns already are.
And so he prevents black from getting to the "a" column.} (68. a6 g5 69. Kb6
Ra3 70. Ra5 Rb3+ 71. Kc5 Rc3+ 72. Kd4 Rc8 73. a7 Ra8 74. Ke3 Kh5 75. Kf3 $11)
68... Rh2 {Just a waste of time} (68... Rf3) 69. a6 Ra2 $4 (69... Rb2+ 70. Ka5
g5 $19) 70. Ra5 $11 Rb2+ {-0,56} 71. Kc3 Rb8 72. a7 Ra8 73. Kd2 h3 {There's
been a mistake somewhere. Turbo, when he plays white, automatically flips the
board. And I must have entered something wrong here, I have 74.Ra6 h2 75.Ke3
h1Q and here Turbo protested that there is no pawn...} 74. Ra4 $4 {I
interrupted the game, it was 3:00 a.m. CET. In the morning the match started
from this position. Is it my mistake in the assignment, or is there some
self-learning going on? In any case, this move also loses - or at least it
should.} (74. Ra3 h2 75. Rh3+ Kg5 76. Rxh2 Rxa7 $11) 74... h2 $4 {But it still
results in a rook ending with the black "g" pawn, which is a draw, but with
the performance of the opponents the result is uncertain.} (74... Kg5 $19) 75.
Rh4+ $11 Kg7 {-1,20} 76. Rxh2 Rxa7 77. Ke3 Ra1 $6 {Behind the pawn, not in
front of it. And cut off the king when there's a chance.} 78. Ke4 Re1+ {So
clearly the best chance to win is to cut off the king Rf1, though that's not
enough here, White would work his ass off - and probably lose the position
with his performance.} 79. Kd5 $4 {-#35 tablebase What to say. He can't see
until the pawn transformation, so the bonus that his valuation function gives
for the king in the center overrides his decision making, oh yeah.} Kf6 $4 {
Complete five-stone tablebases have a "huge" 32MB, more than these machines
could afford.} ({The only one winning} 79... g5 $1 $19) 80. Rf2+ {Surprisingly,
the only move that holds a tie. With the king cut off, the usual pattern of
rooks from behind is no longer enough.} Kg5 {It can take a long time, a break
for breakfast. They both have pondering, so Turbo has a definite advantage
here, if he hits a move in the game he can make it count for about an hour.
Still, I think we'll see plenty more mistakes, so whatever.} 81. Rg2+ Kf5 82.
Rf2+ Kg4 83. Rg2+ Kf4 {Surprisingly conciliatory, I expected him to push the
"g" pawn forward as far as it would go and then White would either be able to
draw or not.} 84. Rxg6 {Estimate - draw. I'm not going to spend time on
another 50 moves and the machine won't make a mistake here.} 1/2-1/2
[/pgn]
I've lost a day of my life and I may lose some ELO.... First a position that neither program understands and then an endgame... I was once in charge of entering games into a database at a tournament in Prague, including the weakest ones, which I commented were of strategic value at the level of fighting children in kindergarten. I was usually so exhausted after a week of the tournament that I lost about 250 ELOs on the server, and it took another month to get them back.
I influenced the game late last night when I entered the wrong move. The turbo automatically rotates the board when playing white, making it more difficult for me to transfer moves. I think I'll turn Mephisto next time... So at about 3:00 CET, aborted, found the error in the morning and finished.
I don't have power to the "East German" yet, so the next opponent will be
DGT Centaur, level friendly. It could be fast, I can almost always beat him, but it doesn't get worse than 1600. So we'll see how the promised adaptivity works out. With such
SparkChess (a program for android, not a uci engine) it works by playing the game (sometimes) like a pig, but then in a losing position it starts playing full out. Good for practicing technique, but it doesn't feel like a human game.
