Is Karmageddon the future of chess?

[Ovyron]

Karmageddon is a variant of chess based on Armaggedon where black can't castle, but draws give black the win.

After several correspondence time control games against witchesbutt it was clear that I wasn't strong enough to beat him in a chess game. We ceased playing because it would just be a draw after another.

So we tried Karmageddon and it was an entirely different picture:

In this game I was black
[pgn]1. d4 d5 2. c4 c6 3. Nc3 Nf6 4. e3 Bf5 5. cxd5 cxd5 6. Qb3 Bc8 7. Nf3 e6 8. Bd3
h5 9. a3 Be7 10. O-O Kf8 11. Qc2 Bd7 12. b4 a6 13. Qb3 Nc6 14. Na4 Bd6 15. Nc5
b6 16. Nxd7+ Qxd7 17. Re1 Ne7 18. Bb2 Qb7 19. Ng5 Ra7 20. h4 {0-1/2}[/pgn]

Despite being unable to castle I managed to equalize and win the game.

In this game I was white
[pgn]1. c4 c5 2. e3 g6 3. d4 cxd4 4. exd4 Bg7 5. Nc3 d6 6. h3 Kf8 7. Nf3 Nc6 8. Be2 h5 9. d5 Ne5 10. Nd4 Nh6 11. O-O Nf5
12. Ncb5 Bf6 13. Nxf5 Bxf5 14. Be3 g5 15. f4 gxf4 16. Rxf4 Bg6 17. Qd2 a5 18. Raf1 Kg8 19. Rxf6 exf6 20. Bf2 Qf8 21. b3 h4 22. Bd4 Qd8 23. a4 Kg7 24. Bc3 Qb6+ 25. Kh2 Qd8 26. Kh1 b6 27. Qf4 {1/2-0}[/pgn]

Here I could take advantage of black's inability to castle and defeat them.

The only way to make this fair is to play a match of two games, but while on chess we were equal, here we were not, so Karmageddon has a much higher skill ceiling.

This variant could put an end to chess draw death, if it was unusable I'd have expected for black to hold the draw in both games, or for white to win both games, but it seems to have a healthy complexity.

Anybody that thinks black should always hold or white should always win is welcome to take on me, specially, I don't have the Komodo version with Armageddon scoring, it'd be interesting to see if it makes a difference.

[lkaufman]

Karmageddon is a variant of chess based on Armaggedon where black can't castle, but draws give black the win.

After several correspondence time control games against witchesbutt it was clear that I wasn't strong enough to beat him in a chess game. We ceased playing because it would just be a draw after another.

So we tried Karmageddon and it was an entirely different picture:

In this game I was black
[pgn]1. d4 d5 2. c4 c6 3. Nc3 Nf6 4. e3 Bf5 5. cxd5 cxd5 6. Qb3 Bc8 7. Nf3 e6 8. Bd3
h5 9. a3 Be7 10. O-O Kf8 11. Qc2 Bd7 12. b4 a6 13. Qb3 Nc6 14. Na4 Bd6 15. Nc5
b6 16. Nxd7+ Qxd7 17. Re1 Ne7 18. Bb2 Qb7 19. Ng5 Ra7 20. h4 {0-1/2}[/pgn]

Despite being unable to castle I managed to equalize and win the game.

In this game I was white
[pgn]1. c4 c5 2. e3 g6 3. d4 cxd4 4. exd4 Bg7 5. Nc3 d6 6. h3 Kf8 7. Nf3 Nc6 8. Be2 h5 9. d5 Ne5 10. Nd4 Nh6 11. O-O Nf5
12. Ncb5 Bf6 13. Nxf5 Bxf5 14. Be3 g5 15. f4 gxf4 16. Rxf4 Bg6 17. Qd2 a5 18. Raf1 Kg8 19. Rxf6 exf6 20. Bf2 Qf8 21. b3 h4 22. Bd4 Qd8 23. a4 Kg7 24. Bc3 Qb6+ 25. Kh2 Qd8 26. Kh1 b6 27. Qf4 {1/2-0}[/pgn]

Here I could take advantage of black's inability to castle and defeat them.

The only way to make this fair is to play a match of two games, but while on chess we were equal, here we were not, so Karmageddon has a much higher skill ceiling.

This variant could put an end to chess draw death, if it was unusable I'd have expected for black to hold the draw in both games, or for white to win both games, but it seems to have a healthy complexity.

Anybody that thinks black should always hold or white should always win is welcome to take on me, specially, I don't have the Komodo version with Armageddon scoring, it'd be interesting to see if it makes a difference.

I have done a lot of testing of this along with Kai Laskos and S.Pohl over the past year, and our conclusion was that this precise variant is quite playable but somewhat favorable to White. With one rule change, namely that Black can still castle long but not short, the results became so close to 50-50 that it was impossible to say which side had the better chances, and this seems to be the best version. We call it NBSC Armageddon (No Black Short Castling). Check out https://www.sp-cc.de/ for more on this.

[Time]

This topic was mentioned in the TCEC chat a couple of days ago, but with the optional restriction that black can castle after a given number of moves, an idea which I remember that I also had myself after reading about the NB(S)C suggestions here around four years ago (where, though, I found the engines at the time suggested relatively uninspiring opening lines for either NBC or NBSC (I can't quite recall)).

My way to "operationalize" the idea is this: Assuming that chess is a draw, and that NBC (no black castling chess) is a win for white, an armageddon game can be arranged based on when each player thinks removing black short castling rights and long castling rights after move s and l, respectively, gives white a winning position. This is the most general approach, but one can also do a simplified one centered around NBSC (no black short castling chess) specificially.

Let me start with the latter: One way of determining the rules of the armageddon game would be to let both players write down an integer m which is either -2 or lower (-l) or 1 or higher (s). For m = -l, one believes that NBSC is no longer a draw if black gets the right to castle long only from move l and onwards. For m = s, one believes that NBSC is a win for white if black gets the right to castle long only from move s and onwards.

For m = -l, the higher the l, the more radically the player thinks modified NBSC is a draw. For m = s, the lower the s, the more radically the player thinks modified NBSC is a win.

Thus, the "assumption scale" becomes A = {-l, -l, -l + 1, ..., -2} U {s, s - 1, s - 2, ..., 1}, where l and s can be aribitrarily large. If both players select the same m, a pair of game is played. Otherwise, the choice of m corresponding to a modification of NBSC is selected by random within the subset {m1, ..., m2} \ {m1 or m2} (for the endpoint choice, see below).

Three cases: If both players think NBSC is a draw, and that they think it stops being a draw at moves -m1 and -m2 with regards to long castling rights, an armageddon game is created with regards to long castling rights in the move interval [-m2 + 1, -m1]. If both players think it is a win, and they think it starts being a win when black gets the right to castle from move m1 and m2, respectively, an armageddon game is created with regards to short castling rights in the move interval [m2, m1 - 1]. If they disagree on whether NBSC is a draw, the subset corresponding to the move +/-m to be chosen is {m1 + 1, ..., m2} (following the same order as A).

In the more general approach to modification of NBC, we do not discriminate NBLC, and each player has to determine his or her assumption about the separate move phases (0, 1, ..., l, ..., infinity) and (0, 1, ..., s, ..., infinity) where NBLC and NBLS is in force and when this leads to a white win, where l or s = 0 means it is never a win, and l or s = infinity means that it is always a win. Then, the players have to determine two-tuples, typically (0, s - 1), (1, s1 <= s - 1), (2, s2 <= s1), ..., (l - 1, s - 1) corresponding to when a generalized NBC with NBLC for moves 1 through l - 1 and through s - 1, respectively, is a draw. Corresponding to each players' set up two-tuples a "disagreement set" exists from which a two-tuple can be selected by random, determining an armageddon game.

This latter suggestion should only be regarded as an academic idea for the purpose that a consensus arrives around the former cut-off point, or, even worse, around its objective status as won or lost. I should of course have tried to explain the above in a simpler / less technical way, but I assume that I got the general idea across.

[lkaufman]

This topic was mentioned in the TCEC chat a couple of days ago, but with the optional restriction that black can castle after a given number of moves, an idea which I remember that I also had myself after reading about the NB(S)C suggestions here around four years ago (where, though, I found the engines at the time suggested relatively uninspiring opening lines for either NBC or NBSC (I can't quite recall)).

My way to "operationalize" the idea is this: Assuming that chess is a draw, and that NBC (no black castling chess) is a win for white, an armageddon game can be arranged based on when each player thinks removing black short castling rights and long castling rights after move s and l, respectively, gives white a winning position. This is the most general approach, but one can also do a simplified one centered around NBSC (no black short castling chess) specificially.

Let me start with the latter: One way of determining the rules of the armageddon game would be to let both players write down an integer m which is either -2 or lower (-l) or 1 or higher (s). For m = -l, one believes that NBSC is no longer a draw if black gets the right to castle long only from move l and onwards. For m = s, one believes that NBSC is a win for white if black gets the right to castle long only from move s and onwards.

For m = -l, the higher the l, the more radically the player thinks modified NBSC is a draw. For m = s, the lower the s, the more radically the player thinks modified NBSC is a win.

Thus, the "assumption scale" becomes A = {-l, -l, -l + 1, ..., -2} U {s, s - 1, s - 2, ..., 1}, where l and s can be aribitrarily large. If both players select the same m, a pair of game is played. Otherwise, the choice of m corresponding to a modification of NBSC is selected by random within the subset {m1, ..., m2} \ {m1 or m2} (for the endpoint choice, see below).

Three cases: If both players think NBSC is a draw, and that they think it stops being a draw at moves -m1 and -m2 with regards to long castling rights, an armageddon game is created with regards to long castling rights in the move interval [-m2 + 1, -m1]. If both players think it is a win, and they think it starts being a win when black gets the right to castle from move m1 and m2, respectively, an armageddon game is created with regards to short castling rights in the move interval [m2, m1 - 1]. If they disagree on whether NBSC is a draw, the subset corresponding to the move +/-m to be chosen is {m1 + 1, ..., m2} (following the same order as A).

In the more general approach to modification of NBC, we do not discriminate NBLC, and each player has to determine his or her assumption about the separate move phases (0, 1, ..., l, ..., infinity) and (0, 1, ..., s, ..., infinity) where NBLC and NBLS is in force and when this leads to a white win, where l or s = 0 means it is never a win, and l or s = infinity means that it is always a win. Then, the players have to determine two-tuples, typically (0, s - 1), (1, s1 <= s - 1), (2, s2 <= s1), ..., (l - 1, s - 1) corresponding to when a generalized NBC with NBLC for moves 1 through l - 1 and through s - 1, respectively, is a draw. Corresponding to each players' set up two-tuples a "disagreement set" exists from which a two-tuple can be selected by random, determining an armageddon game.

This latter suggestion should only be regarded as an academic idea for the purpose that a consensus arrives around the former cut-off point, or, even worse, around its objective status as won or lost. I should of course have tried to explain the above in a simpler / less technical way, but I assume that I got the general idea across.

If you are talking about human vs human games, then of course it makes no difference what the theoretical result of the game should be, it just has to be fair practically. The problem with bidding in Armageddon is that it rewards a skill unrelated to chess strength, namely knowing and interpreting statistics on how to bid. In practice many Armageddon bids are just insane, in both directions. For engine vs engine, the issue is that you can't just set castling rights at the start, you would need to modify the engines to know about the move number rule. For NBSC, there's no need, it is close enough to even to just toss for color. Another more elegant solution would be to just say that neither side can castle on the side where the opponent has already castled; in practice this would be almost the same as NBSC, since White would castle short first almost every game. But again the problem is that current engines and GUIs don't support this variant. Another variant that is close to fair is to start the game with 1.e4 already played and still White to move (i.e. two move odds), with Black winning draws. Probably Black can barely hold with perfect play, but for humans or even 3000 level engines it might be pretty fair in practice, and requires no new engine mods or GUI.

[pohl4711]

https://www.sp-cc.de/anti-draw-openings.htm

https://www.sp-cc.de/files/pohl_antidra ... gs_v1.6.7z


In this download you find my opening sets for this, so engine tests and tournaments can be done with this:
NBC = No Black castle allowed
NBSC = No Black short-castle allowed

And in the Armageddon-folder, there are tools for rescoring a pgn-file with Armageddon-scoring.

[Time]

I agree that at present, NBSC and/or what I learn is already called "white one more move" (WOMM), e.g. in the above suggestion rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1 of GM Kaufman (can also be reached by entering 1. e3 Nf6 2. e4 Ng8), might be "interesting enough". (It also seems quite likely that if given any choice for move 1 as well, white's optimal choice of moves is 1/2. e4 and 2/1. d4.)

However, if given sufficient attention, I would assume that a consensus about the likely objective status of both variants would arise. (For NBSC, one could e.g. start with Mr. Pohl's .ctg book and build an IDeA tree from it. I did not find a corresponding book folder for WOMM.) If so, it would be easier for NBSC to make modifications. As for the complication about Armageddon bidding processes, at least in the scenario where the players actually disagree about whether NBSC is a win, one can of course just play pure NBSC-Armageddon for simplicity, while in the other two cases, there might be more fair bidding systems than randomizing the move number from the "disagreement interval". E.g., one can play a pair of Armageddon games from each move endpoint. (This would make the situation the same as in the scenario where the players agree on the move number.)

The main challenge I think is, as mentioned by above by Mr. Kaufman, that one probably needs to modify engines for this variant. I don't know what would be simpler: modify Stockfish or train a Leela version for the specified version.

By the way, though, I know that in Aquarium, one can select a set of moves which the engine ought not to analyse. One might imagine a more general choice several ply down, i.e. the principal variation given is based on a given number of nodes being impermissible. I would assume it extending this function beyond the position being analysed might be impossible, and at least assigning a filter rule for the impermissible nodes, for which a modification to the engine appears to be needed if nothing else for the sake of efficiency.

[Ozymandias]

Black can still castle long but not short

If that's the desired goal, why no use the old FICS wild Style 0 variant?

White has the typical set-up at the start. Black's pieces are the same, except that the King and Queen are reversed, so they are not on the same files as White's King and Queen. Castling is done similarly to normal chess: o-o-o indicates long castling and o-o short castling.

It's the variant that gets the less attention, even though it's probably the most (fortuitously) played. Every beginner will place the pieces this way, and we have to correct them all time until they get it "right", but maybe it would be better to let nature take its course and let the pieces in the same disposition for both colors.

[CornfedForever]

Please say "The future of engine vs engine chess" as 'human vs human chess' is just fine, thank you.

[Rebel]

On top of my head, these are the differences between human vs comp play as we practice it at the moment.

1. Humans know the name of the opponent.
2. Humans know the elo rating of the opponent.
3. Humans know the playing style of the opponent.
4. Humans know the weak and strong points of the opponent.
5. Humans know the opening lines the opponent favors.

All this can be applied to computer chess, based on the elo rating one can calculate the contempt factor. Based on the opponent name one can switch playing style, switch opening book. Lots of new challenges for programmers.

[chrisw]

Karmageddon is a variant of chess based on Armaggedon where black can't castle, but draws give black the win.

After several correspondence time control games against witchesbutt it was clear that I wasn't strong enough to beat him in a chess game. We ceased playing because it would just be a draw after another.

So we tried Karmageddon and it was an entirely different picture:

In this game I was black
[pgn]1. d4 d5 2. c4 c6 3. Nc3 Nf6 4. e3 Bf5 5. cxd5 cxd5 6. Qb3 Bc8 7. Nf3 e6 8. Bd3
h5 9. a3 Be7 10. O-O Kf8 11. Qc2 Bd7 12. b4 a6 13. Qb3 Nc6 14. Na4 Bd6 15. Nc5
b6 16. Nxd7+ Qxd7 17. Re1 Ne7 18. Bb2 Qb7 19. Ng5 Ra7 20. h4 {0-1/2}[/pgn]

Despite being unable to castle I managed to equalize and win the game.

In this game I was white
[pgn]1. c4 c5 2. e3 g6 3. d4 cxd4 4. exd4 Bg7 5. Nc3 d6 6. h3 Kf8 7. Nf3 Nc6 8. Be2 h5 9. d5 Ne5 10. Nd4 Nh6 11. O-O Nf5
12. Ncb5 Bf6 13. Nxf5 Bxf5 14. Be3 g5 15. f4 gxf4 16. Rxf4 Bg6 17. Qd2 a5 18. Raf1 Kg8 19. Rxf6 exf6 20. Bf2 Qf8 21. b3 h4 22. Bd4 Qd8 23. a4 Kg7 24. Bc3 Qb6+ 25. Kh2 Qd8 26. Kh1 b6 27. Qf4 {1/2-0}[/pgn]

Here I could take advantage of black's inability to castle and defeat them.

The only way to make this fair is to play a match of two games, but while on chess we were equal, here we were not, so Karmageddon has a much higher skill ceiling.

This variant could put an end to chess draw death, if it was unusable I'd have expected for black to hold the draw in both games, or for white to win both games, but it seems to have a healthy complexity.

Anybody that thinks black should always hold or white should always win is welcome to take on me, specially, I don't have the Komodo version with Armageddon scoring, it'd be interesting to see if it makes a difference.

I have done a lot of testing of this along with Kai Laskos and S.Pohl over the past year, and our conclusion was that this precise variant is quite playable but somewhat favorable to White. With one rule change, namely that Black can still castle long but not short, the results became so close to 50-50 that it was impossible to say which side had the better chances, and this seems to be the best version. We call it NBSC Armageddon (No Black Short Castling). Check out https://www.sp-cc.de/ for more on this.

Have you spoken with Kai Laskos over the past year?