Testing depend largely on good Openings

[lkaufman]

You may argue that this is artificial, that only "balanced" (i.e. typical GM) openings should be used or else it's not real chess, and that is a valid argument, but unless we go with unbalanced openings computer chess is dying.

People have been saying that computer chess is dying for the past 15 years, yet these days there are more engines than there ever were in the past.

As a chessplayer, I prefer that engines play balanced lines, not given artificial wins.
Most draws have some interest involved anyway.

However, I'd say that unlike myself and Frank (and perhaps a few others), most testers aren't interested in watching decent games. They just want to spit out as many games as they can to construct a rating list.

I think you were the one who ran the recent gambit tournament, in which many of the gambits, especially those played by Black, were not close to equal or balanced. No one is proposing to use starting positions that are believed to be won for either color; the idea with unbalanced openings is to use positions that are bad for one side (normally Black), but still thought to be defensible with ideal play. This describes most gambits by Black against good White openings (for example, 1.e4 e5 2.Nf3 d5?! or 1.d4 d5 2c4 e5?! etc.) (Spanish Marshall is the big exception). The way I look at it, chess (and other such games) are interesting because they are a contest between two wills with opposing aims. In amateur human chess, each side aims to checkmate. But in top level chess (2800 humans or engines), White aims to checkmate, but Black aims to draw (though he'll switch to playing for a win if White makes serious errors). Unfortunately this contest is very unequal; Black's chances to draw are far greater than White's chances to win (at 3600 engine level with long time controls maybe 100 times greater or more). By starting with positions (such as the above gambits) where this ratio (at 3600 level) is more like 2 or 3 to 1 rather than 100 to 1, not only do we get much more information on which engine is stronger, we also tend to get more interesting games, as it is always clear that White and Black don't share the same (drawn) goal.

[Chessqueen]

You may argue that this is artificial, that only "balanced" (i.e. typical GM) openings should be used or else it's not real chess, and that is a valid argument, but unless we go with unbalanced openings computer chess is dying.

People have been saying that computer chess is dying for the past 15 years, yet these days there are more engines than there ever were in the past.

As a chessplayer, I prefer that engines play balanced lines, not given artificial wins.
Most draws have some interest involved anyway.

However, I'd say that unlike myself and Frank (and perhaps a few others), most testers aren't interested in watching decent games. They just want to spit out as many games as they can to construct a rating list.

I think you were the one who ran the recent gambit tournament, in which many of the gambits, especially those played by Black, were not close to equal or balanced. No one is proposing to use starting positions that are believed to be won for either color; the idea with unbalanced openings is to use positions that are bad for one side (normally Black), but still thought to be defensible with ideal play. This describes most gambits by Black against good White openings (for example, 1.e4 e5 2.Nf3 d5?! or 1.d4 d5 2c4 e5?! etc.) (Spanish Marshall is the big exception). The way I look at it, chess (and other such games) are interesting because they are a contest between two wills with opposing aims. In amateur human chess, each side aims to checkmate. But in top level chess (2800 humans or engines), White aims to checkmate, but Black aims to draw (though he'll switch to playing for a win if White makes serious errors). Unfortunately this contest is very unequal; Black's chances to draw are far greater than White's chances to win (at 3600 engine level with long time controls maybe 100 times greater or more). By starting with positions (such as the above gambits) where this ratio (at 3600 level) is more like 2 or 3 to 1 rather than 100 to 1, not only do we get much more information on which engine is stronger, we also tend to get more interesting games, as it is always clear that White and Black don't share the same (drawn) goal.

What is almost incredible to believe is that with the top 4 engines, all that it takes is one inferior move and the game turn in favor of the other side,, I believe it was move 7. Nbd2 from the Opening book that spoil the game for White, but if I let Komodo Dragon choose its own variation after move 7. the game ends in a Draw, from move 7. most top GMs believe that 7. Na3 is NOT a good move because it place the knight on the edge of the board having less control of squares, but in this case it is the best move. Na3, whereas, 7. Nbd2 which is the opening book turns out to be inferior and even Komodo -12.1 can hold a draw versus almighty latest version of Stockfish 15

[pgn][Event "Computer chess game"]
[Site "DESKTOP-OFQ3C0P"]
[Date "2022.07.03"]
[Round "?"]
[White "Stockfish_15_x64_avx2"]
[Black "Komodo-12.1.1-64bit-bmi2"]
[Result "1/2-1/2"]
[BlackElo "3430"]
[Time "13:05:38"]
[WhiteElo "3555"]
[TimeControl "900+10"]
[Termination "adjudication"]
[PlyCount "131"]
[WhiteType "program"]
[BlackType "program"]


1. d4 Nf6 2. c4 g6 3. g3 Bg7 4. Bg2 O-O 5. Nf3 d5 6. O-O dxc4 7. Nbd2 b5 8.
Ne5 Nd5 9. b3 Nc3 10. Qe1 Qxd4 11. Nxf7 Nd5 12. Ng5 h6 13. bxc4 bxc4 14. e3
Qe5 15. Nxc4 Qxg5 16. e4 Qh5 17. exd5 Bxa1 18. d6 exd6 19. Bxa8 Bh3 20. Bg2
Bxg2 21. Qe6+ Kh7 22. Kxg2 Bg7 23. f3 Qb5 24. Ne3 Re8 25. Qf7 Qd7 26. Qc4
c5 27. Rd1 Qe6 28. Qxe6 Rxe6 29. f4 Nc6 30. Kf1 Nb4 31. a3 Nc6 32. Nd5 Re4
33. Kf2 Kg8 34. Nf6+ Bxf6 35. Rxd6 Bd4+ 36. Kf3 Re1 37. Bd2 Rf1+ 38. Ke2
Rf2+ 39. Ke1 Kf7 40. Rxd4 Rxh2 41. Rd6 Nd4 42. Bc3 Nc2+ 43. Kf1 Ne3+ 44.
Ke1 Ng4 45. Rc6 c4 46. Rxc4 Rh3 47. Rc5 Rxg3 48. f5 gxf5 49. Rxf5+ Ke6 50.
Rc5 h5 51. Kf1 h4 52. Be1 Rh3 53. Rg5 Nf6 54. Rg7 Rxa3 55. Bxh4 Nd5 56. Bd8
Ra2 57. Rh7 Ke5 58. Ke1 Kd4 59. Kd1 a5 60. Bxa5 Rxa5 61. Ke2 Ra2+ 62. Kf3
Ra3+ 63. Kf2 Nf6 64. Rh4+ Ne4+ 65. Ke2 Re3+ 66. Kf1 *[/pgn]

[Graham Banks]

I think you were the one who ran the recent gambit tournament, in which many of the gambits, especially those played by Black, were not close to equal or balanced................

The lines were checked beforehand, so I knew that they would not produce unfair games.

[lkaufman]

I think you were the one who ran the recent gambit tournament, in which many of the gambits, especially those played by Black, were not close to equal or balanced................

The lines were checked beforehand, so I knew that they would not produce unfair games.

I don't know what your criteria were, or what your definition of an "unfair game" is, but at least some of the positions were much closer to being lost than they were to being truly equal. For example, the Albin Counter Gambit gets a Stockfish eval of around +1 for White which is much closer to the win/draw line (maybe +1.3 or so for SF) than to 0.00, while Lc0 gives White about a 44% chance of winning outright, again not so far from the win/draw line. I am not criticizing you at all here, I fully approve of your decisions and appreciate that you ran the match. My point is that at least in some of the games, you were clearly embracing the spirit of unbalanced human openings, basically choosing openings that one side must play very well to draw. In fact the Albin Counter Gambit is surely an opening I would choose myself for a set of unbalanced human openings. I just don't like to refer to such openings as "balanced" or "equal" or "fair" just because they are (probably) drawn with perfect play; it is only because colors are reversed for every second game that such a match is in any sense fair or balanced. In fact, even standard openings like the Ruy Lopez for example shouldn't be called "balanced" or "equal", they should just be called "normal" or "standard". Chess is just not a fair or equal game, White starts off clearly better, and we shouldn't pretend otherwise. An equal opening only happens when White plays some inferior early move.

[Ovyron]

the path to a draw is too clear now.

No, the path to draw isn't clear at all. Try making a "balanced" book, one where no matter what the opponent plays, you'll reach a balanced opening that you will be able to draw.

Well, I know people that would be able to bust one of those lines, and prove the "draw that was clear" wasn't so at all. It's just that testers of engines never try to find a hole on those "balanced" openings, so if there's a move that gives a side some big advantage, it'll not be found, because nobody is looking for it.

This is specially true on the age of NNUE, where you can build a big book on an off-beat line and win just because your side has a lot more time on the clock than the opponent missing your lines. Computer chess on the competitive side is still live and kicking, if anybody tried to use those balanced openings on playchess they'd be utterly destroyed by better moves by people that dedicate their lives to finding holes in lines.

I've seen positions where Stockfish NNUE has the score backwards, both engines show some +0.20 but after the game continues both engines agree it's really -0.40, so a black player wants to get into this line, while engine testers will be happy to settle to play some +0.15 position that is actually 0.00 missing black's improvement.

I've seen completely lost positions that are scored at +0.30, but there's already a sequence of moves that force some +2.00 advantage, so white looks great aiming for this line in case the opponent allows it and loses. An engine tester already knows what moves black will play so has no reason to try this.

What engine testing is missing is competition, trying to do the best that you can for your side causes you to find and play better openings. Using balanced openings with reversed colors leads to draw death, because there's no incentive to find better lines for either side.

[Lazy_Frank]

At the moment i am very interested to spot out where is chess draw limit (pick the bad line in range ~ 1.3-1.5 evaluation and sort out it is drawable or not), clearly opposite approach.

[Chessqueen]

At the moment i am very interested to spot out where is chess draw limit (pick the bad line in range ~ 1.3-1.5 evaluation and sort out it is drawable or not), clearly opposite approach.

If you take this game after the opening move 7. Nd2 which is clearly an inferior move the score dropped to 1.1 and after that Komodo older version 12.1 was able to hold a draw against the very best engine Stockfish 15 with the Back pieces, that move 7.Nd2 was not made by Stockfish 15, it was part of an opening book which went up to move 16. If it was up to Stockfish 15 the correct move was 7.Na3! Which none of the top 8 engines will ever miss.

[pgn][Event "Computer chess game"]
[Site "DESKTOP-OFQ3C0P"]
[Date "2022.07.03"]
[Round "?"]
[White "Stockfish_15_x64_avx2"]
[Black "Komodo-12.1.1-64bit-bmi2"]
[Result "1/2-1/2"]
[BlackElo "3430"]
[Time "13:05:38"]
[WhiteElo "3555"]
[TimeControl "900+10"]
[Termination "adjudication"]
[PlyCount "131"]
[WhiteType "program"]
[BlackType "program"]


1. d4 Nf6 2. c4 g6 3. g3 Bg7 4. Bg2 O-O 5. Nf3 d5 6. O-O dxc4 7. Nbd2 b5 8.
Ne5 Nd5 9. b3 Nc3 10. Qe1 Qxd4 11. Nxf7 Nd5 12. Ng5 h6 13. bxc4 bxc4 14. e3
Qe5 15. Nxc4 Qxg5 16. e4 Qh5 17. exd5 Bxa1 18. d6 exd6 19. Bxa8 Bh3 20. Bg2
Bxg2 21. Qe6+ Kh7 22. Kxg2 Bg7 23. f3 Qb5 24. Ne3 Re8 25. Qf7 Qd7 26. Qc4
c5 27. Rd1 Qe6 28. Qxe6 Rxe6 29. f4 Nc6 30. Kf1 Nb4 31. a3 Nc6 32. Nd5 Re4
33. Kf2 Kg8 34. Nf6+ Bxf6 35. Rxd6 Bd4+ 36. Kf3 Re1 37. Bd2 Rf1+ 38. Ke2
Rf2+ 39. Ke1 Kf7 40. Rxd4 Rxh2 41. Rd6 Nd4 42. Bc3 Nc2+ 43. Kf1 Ne3+ 44.
Ke1 Ng4 45. Rc6 c4 46. Rxc4 Rh3 47. Rc5 Rxg3 48. f5 gxf5 49. Rxf5+ Ke6 50.
Rc5 h5 51. Kf1 h4 52. Be1 Rh3 53. Rg5 Nf6 54. Rg7 Rxa3 55. Bxh4 Nd5 56. Bd8
Ra2 57. Rh7 Ke5 58. Ke1 Kd4 59. Kd1 a5 60. Bxa5 Rxa5 61. Ke2 Ra2+ 62. Kf3
Ra3+ 63. Kf2 Nf6 64. Rh4+ Ne4+ 65. Ke2 Re3+ 66. Kf1 *
[/quote][/pgn]

[Graham Banks]

I think you were the one who ran the recent gambit tournament, in which many of the gambits, especially those played by Black, were not close to equal or balanced................

The lines were checked beforehand, so I knew that they would not produce unfair games.

I don't know what your criteria were, or what your definition of an "unfair game" is, but at least some of the positions were much closer to being lost than they were to being truly equal. For example, the Albin Counter Gambit gets a Stockfish eval of around +1 for White which is much closer to the win/draw line (maybe +1.3 or so for SF) than to 0.00, while Lc0 gives White about a 44% chance of winning outright, again not so far from the win/draw line. I am not criticizing you at all here, I fully approve of your decisions and appreciate that you ran the match. My point is that at least in some of the games, you were clearly embracing the spirit of unbalanced human openings, basically choosing openings that one side must play very well to draw. In fact the Albin Counter Gambit is surely an opening I would choose myself for a set of unbalanced human openings. I just don't like to refer to such openings as "balanced" or "equal" or "fair" just because they are (probably) drawn with perfect play; it is only because colors are reversed for every second game that such a match is in any sense fair or balanced. In fact, even standard openings like the Ruy Lopez for example shouldn't be called "balanced" or "equal", they should just be called "normal" or "standard". Chess is just not a fair or equal game, White starts off clearly better, and we shouldn't pretend otherwise. An equal opening only happens when White plays some inferior early move.

My main goal in playing the 16 cores gambit match at 2 hours + 30 sec increments, was to show players that most gambits are safe to play if you understand the nuances of the positions that arise.
I was also hoping that Stockfish and Dragon by Komodo would show new moves or 'plans' for such players.
Pretty sure that I achieved both of these.

[Ovyron]

At the moment i am very interested to spot out where is chess draw limit (pick the bad line in range ~ 1.3-1.5 evaluation and sort out it is drawable or not), clearly opposite approach.

It's probably drawable. A more interesting limit is the 1.90 eval, I think the ones I've seen have some 50% chance of being drawable as the score falls or already winning if the score increases. At 2.00 it's probably winning.

These scores hold at about early middle game, before I've seen 2.00 positions that are drawable still near the opening, later, say, by move 50 a score of 1.40 can be decisive. Probably someone smart could map these out, but Stockfish NNUE overestimates its chances on the opening and underestimates them near the endgame.

[lkaufman]

the path to a draw is too clear now.

No, the path to draw isn't clear at all. Try making a "balanced" book, one where no matter what the opponent plays, you'll reach a balanced opening that you will be able to draw.

Well, I know people that would be able to bust one of those lines, and prove the "draw that was clear" wasn't so at all. It's just that testers of engines never try to find a hole on those "balanced" openings, so if there's a move that gives a side some big advantage, it'll not be found, because nobody is looking for it.

This is specially true on the age of NNUE, where you can build a big book on an off-beat line and win just because your side has a lot more time on the clock than the opponent missing your lines. Computer chess on the competitive side is still live and kicking, if anybody tried to use those balanced openings on playchess they'd be utterly destroyed by better moves by people that dedicate their lives to finding holes in lines.

I've seen positions where Stockfish NNUE has the score backwards, both engines show some +0.20 but after the game continues both engines agree it's really -0.40, so a black player wants to get into this line, while engine testers will be happy to settle to play some +0.15 position that is actually 0.00 missing black's improvement.

I've seen completely lost positions that are scored at +0.30, but there's already a sequence of moves that force some +2.00 advantage, so white looks great aiming for this line in case the opponent allows it and loses. An engine tester already knows what moves black will play so has no reason to try this.

What engine testing is missing is competition, trying to do the best that you can for your side causes you to find and play better openings. Using balanced openings with reversed colors leads to draw death, because there's no incentive to find better lines for either side.

Have people tried entering latest Stockfish on good hardware in Rapid (not blitz) online tournaments with NO BOOK? If so, does it lose any games, and if so what percent? Or perhaps same with just a very short generic book from GM games, like 5 or 6 moves only?