St. Louis chess960/FRC

[Modern Times]

[I do think that customized neural nets for each starting positions would be useful as the castling is completely different.

Ethereal has a net specifically for FRC.

[M ANSARI]

[I do think that customized neural nets for each starting positions would be useful as the castling is completely different.

Ethereal has a net specifically for FRC.

Would'nt it have to have 960 neural networks or one specific for each start position?

[lkaufman]

It is almost certain, both based on this statistic and on actual evals, that none of the 960 positions are winning for White, or even close to it.
[/quote]

OK now that I didn't know and is an interesting data point. Somehow some positions look completely lost ... that just shows how rich chess is! I would have expected at least 100 of the positions as forced wins for one side ... but that engines don't see that then for sure it means that this is not the case. I do think that customized neural nets for each starting positions would be useful as the castling is completely different.
[/quote]

I think that a few of the positions do give White roughly double his normal opening advantage, but White needs at least triple his normal advantage to have a theoretically won game, according to everything I've seen.

[M ANSARI]

Thanks for all this information ... certainly very interesting! I actually tried a few Chess 960 games and really it is so easy to immediately be lost! I think for sure a customized net for each start position would be very helpful ... at least for the first 10 or 15 moves. After 15 moves the game starts to look a lot like normal chess (if you survive that long). Of course the weird castling rules might have a major impact ... but for engines that is easy to add in.

[Chessqueen]

It is almost certain, both based on this statistic and on actual evals, that none of the 960 positions are winning for White, or even close to it.

OK now that I didn't know and is an interesting data point. Somehow some positions look completely lost ... that just shows how rich chess is! I would have expected at least 100 of the positions as forced wins for one side ... but that engines don't see that then for sure it means that this is not the case. I do think that customized neural nets for each starting positions would be useful as the castling is completely different.
[/quote]


In that case those 100 positions or whatever amount of positions are considered advantageous for one side which guarantee at least a draw should be removed to make it more fair for humans and call it chess860 or whatever number of positions.

[lkaufman]

It is almost certain, both based on this statistic and on actual evals, that none of the 960 positions are winning for White, or even close to it.

OK now that I didn't know and is an interesting data point. Somehow some positions look completely lost ... that just shows how rich chess is! I would have expected at least 100 of the positions as forced wins for one side ... but that engines don't see that then for sure it means that this is not the case. I do think that customized neural nets for each starting positions would be useful as the castling is completely different.

In that case those 100 positions or whatever amount of positions are considered advantageous for one side which guarantee at least a draw should be removed to make it more fair for humans and call it chess860 or whatever number of positions.
[/quote]

If single games are played from each position, without reversing colors, then there is an argument for excluding the more unequal (though still probably drawn) positions. If pairs of games are played (as they should be for fairness), then there is actually more of an argument to exclude the more equal positions, especially for engine play.

[lkaufman]

Correction: No black short castling (NBSC) was what I meant. (That white is probably winning under NBC is not new information.)

Personally, I would find it easier to play "restricted" classical chess with the black pieces, e.g. with black short castling not allowed for the first 15 moves, than a symmetrical Chess960 position, even if I would have an enormous defensive task in the former case.

A possible example line with the above restriction (based on what Chessify-SF suggests for pure NBSC for the initial moves, assuming for the sake of argument and example that the line is still relevant) could be [pgn]1. d4 d5 2. c4 e6 3. Nc3 a6 4. cxd5 exd5 5. g3 Nf6 6. Nf3 c6 7. Bg2 Be7 8. O-O
Nbd7 9. Re1 Ne4 10. Nd2 f5 11. Ndxe4 fxe4 12. f3 Nf6 13. fxe4 Nxe4 14. Nxe4
dxe4 15. Bxe4 Bf6 16. d5 O-O 17. dxc6 Qb6+ 18. Kg2 bxc6 19. Qd6 Rd8 20. Qxc6
Qxc6 21. Bxc6 Rb8 22. Bf4 Rb6 23. Bf3 Rxb2 24. Reb1 Bf5 25. Rxb2 Bxb2 26. Rf1
Ba3 27. Bg5 Rd4 28. Be3 Rd8 29. Bg5 Rd4 30. Be3 Rd8 31. Bb6 Rd6 32. Ba5 Rd4 33.
Bc3 Rd8 34. h4 *[/pgn] and it is still uncertain to me if white has a winning advantage in the endgame. Here, one could also imagine what Fischer described as a drawback but what I actually would find attractive as a scientific aspect of chess: the possibility of very late novelties. If these novelties actually matter, i.e. it might be the case that they improve over their predecessors in absolute terms rather than just practical ones, then I cannot see why this form of chess should be any less appealing than, say, FRC. I would rather think that people with a classical intuition should be able to enjoy it a lot more.

The statement that White is probably winning under NBC prompted me to investigate further. It does seem to be true, based on deep evals of the top three engines (SF, Dragon, Lc0). The evals after very deep searches suggest that White has about a 60% probability to win outright, and in view of the general agreement of the three and the tendency of the evals to climb with depth does seem to confirm the claim. So I thought about the simplest, cleanest modification that doesn't require special code to play the game properly (a move limit on castling is a rule change and requires new engines to play it properly). So I came up with this: White can only castle short, Black can only castle long. Obviously this is less favorable for White than NBC, since White has lost an option but Black has not. It is not a lot less favorable, since White normally castles short, especially if he wants to win and knows that Black can only castle long. But it is enough of a change to reduce White's average win prob. to about 46%, so most likely this variant is not quite a forced win, but quite close to it. Anyway it is much closer to the win/draw line than NBC appears to be, and it is better to be just short of the line than over it, since normal chess is presumed to be a theoretical draw. This variant also has the advantage of "symmetry" in that each side can castle, just on opposite sides, and of course it is not a rule change, just a mod of castling rights to the initial position. Maybe this is the ideal startposition for Armageddon chess; the most like normal chess of any that I know of that are roughly balanced with Armageddon scoring. Now it just needs a cool name!

[Chessqueen]

Correction: No black short castling (NBSC) was what I meant. (That white is probably winning under NBC is not new information.)

Personally, I would find it easier to play "restricted" classical chess with the black pieces, e.g. with black short castling not allowed for the first 15 moves, than a symmetrical Chess960 position, even if I would have an enormous defensive task in the former case.

A possible example line with the above restriction (based on what Chessify-SF suggests for pure NBSC for the initial moves, assuming for the sake of argument and example that the line is still relevant) could be [pgn]1. d4 d5 2. c4 e6 3. Nc3 a6 4. cxd5 exd5 5. g3 Nf6 6. Nf3 c6 7. Bg2 Be7 8. O-O
Nbd7 9. Re1 Ne4 10. Nd2 f5 11. Ndxe4 fxe4 12. f3 Nf6 13. fxe4 Nxe4 14. Nxe4
dxe4 15. Bxe4 Bf6 16. d5 O-O 17. dxc6 Qb6+ 18. Kg2 bxc6 19. Qd6 Rd8 20. Qxc6
Qxc6 21. Bxc6 Rb8 22. Bf4 Rb6 23. Bf3 Rxb2 24. Reb1 Bf5 25. Rxb2 Bxb2 26. Rf1
Ba3 27. Bg5 Rd4 28. Be3 Rd8 29. Bg5 Rd4 30. Be3 Rd8 31. Bb6 Rd6 32. Ba5 Rd4 33.
Bc3 Rd8 34. h4 *[/pgn] and it is still uncertain to me if white has a winning advantage in the endgame. Here, one could also imagine what Fischer described as a drawback but what I actually would find attractive as a scientific aspect of chess: the possibility of very late novelties. If these novelties actually matter, i.e. it might be the case that they improve over their predecessors in absolute terms rather than just practical ones, then I cannot see why this form of chess should be any less appealing than, say, FRC. I would rather think that people with a classical intuition should be able to enjoy it a lot more.

The statement that White is probably winning under NBC prompted me to investigate further. It does seem to be true, based on deep evals of the top three engines (SF, Dragon, Lc0). The evals after very deep searches suggest that White has about a 60% probability to win outright, and in view of the general agreement of the three and the tendency of the evals to climb with depth does seem to confirm the claim. So I thought about the simplest, cleanest modification that doesn't require special code to play the game properly (a move limit on castling is a rule change and requires new engines to play it properly). So I came up with this: White can only castle short, Black can only castle long. Obviously this is less favorable for White than NBC, since White has lost an option but Black has not. It is not a lot less favorable, since White normally castles short, especially if he wants to win and knows that Black can only castle long. But it is enough of a change to reduce White's average win prob. to about 46%, so most likely this variant is not quite a forced win, but quite close to it. Anyway it is much closer to the win/draw line than NBC appears to be, and it is better to be just short of the line than over it, since normal chess is presumed to be a theoretical draw. This variant also has the advantage of "symmetry" in that each side can castle, just on opposite sides, and of course it is not a rule change, just a mod of castling rights to the initial position. Maybe this is the ideal startposition for Armageddon chess; the most like normal chess of any that I know of that are roughly balanced with Armageddon scoring. Now it just needs a cool name!

I believe that the contrary is more even with NWSC ( No White Short Castling), whereas Black can castle either way makes the standard chess more even, since it already has the initial first move.

[lkaufman]

Correction: No black short castling (NBSC) was what I meant. (That white is probably winning under NBC is not new information.)

Personally, I would find it easier to play "restricted" classical chess with the black pieces, e.g. with black short castling not allowed for the first 15 moves, than a symmetrical Chess960 position, even if I would have an enormous defensive task in the former case.

A possible example line with the above restriction (based on what Chessify-SF suggests for pure NBSC for the initial moves, assuming for the sake of argument and example that the line is still relevant) could be [pgn]1. d4 d5 2. c4 e6 3. Nc3 a6 4. cxd5 exd5 5. g3 Nf6 6. Nf3 c6 7. Bg2 Be7 8. O-O
Nbd7 9. Re1 Ne4 10. Nd2 f5 11. Ndxe4 fxe4 12. f3 Nf6 13. fxe4 Nxe4 14. Nxe4
dxe4 15. Bxe4 Bf6 16. d5 O-O 17. dxc6 Qb6+ 18. Kg2 bxc6 19. Qd6 Rd8 20. Qxc6
Qxc6 21. Bxc6 Rb8 22. Bf4 Rb6 23. Bf3 Rxb2 24. Reb1 Bf5 25. Rxb2 Bxb2 26. Rf1
Ba3 27. Bg5 Rd4 28. Be3 Rd8 29. Bg5 Rd4 30. Be3 Rd8 31. Bb6 Rd6 32. Ba5 Rd4 33.
Bc3 Rd8 34. h4 *[/pgn] and it is still uncertain to me if white has a winning advantage in the endgame. Here, one could also imagine what Fischer described as a drawback but what I actually would find attractive as a scientific aspect of chess: the possibility of very late novelties. If these novelties actually matter, i.e. it might be the case that they improve over their predecessors in absolute terms rather than just practical ones, then I cannot see why this form of chess should be any less appealing than, say, FRC. I would rather think that people with a classical intuition should be able to enjoy it a lot more.

The statement that White is probably winning under NBC prompted me to investigate further. It does seem to be true, based on deep evals of the top three engines (SF, Dragon, Lc0). The evals after very deep searches suggest that White has about a 60% probability to win outright, and in view of the general agreement of the three and the tendency of the evals to climb with depth does seem to confirm the claim. So I thought about the simplest, cleanest modification that doesn't require special code to play the game properly (a move limit on castling is a rule change and requires new engines to play it properly). So I came up with this: White can only castle short, Black can only castle long. Obviously this is less favorable for White than NBC, since White has lost an option but Black has not. It is not a lot less favorable, since White normally castles short, especially if he wants to win and knows that Black can only castle long. But it is enough of a change to reduce White's average win prob. to about 46%, so most likely this variant is not quite a forced win, but quite close to it. Anyway it is much closer to the win/draw line than NBC appears to be, and it is better to be just short of the line than over it, since normal chess is presumed to be a theoretical draw. This variant also has the advantage of "symmetry" in that each side can castle, just on opposite sides, and of course it is not a rule change, just a mod of castling rights to the initial position. Maybe this is the ideal startposition for Armageddon chess; the most like normal chess of any that I know of that are roughly balanced with Armageddon scoring. Now it just needs a cool name!

I believe that the contrary is more even with NWSC ( No White Short Castling), whereas Black can castle either way makes the standard chess more even, since it already has the initial first move.

You are missing the point; I'm not trying to make the normal game more even, I'm trying to make Armageddon chess even, where Black wins draws. White needs to have a much larger advantage than in normal chess for this to be fair (with equal clock times), and this proposal seems to give about the proper advantage to offset draw odds.

[Time]

Correction: No black short castling (NBSC) was what I meant. (That white is probably winning under NBC is not new information.)

Personally, I would find it easier to play "restricted" classical chess with the black pieces, e.g. with black short castling not allowed for the first 15 moves, than a symmetrical Chess960 position, even if I would have an enormous defensive task in the former case.

A possible example line with the above restriction (based on what Chessify-SF suggests for pure NBSC for the initial moves, assuming for the sake of argument and example that the line is still relevant) could be [pgn]1. d4 d5 2. c4 e6 3. Nc3 a6 4. cxd5 exd5 5. g3 Nf6 6. Nf3 c6 7. Bg2 Be7 8. O-O
Nbd7 9. Re1 Ne4 10. Nd2 f5 11. Ndxe4 fxe4 12. f3 Nf6 13. fxe4 Nxe4 14. Nxe4
dxe4 15. Bxe4 Bf6 16. d5 O-O 17. dxc6 Qb6+ 18. Kg2 bxc6 19. Qd6 Rd8 20. Qxc6
Qxc6 21. Bxc6 Rb8 22. Bf4 Rb6 23. Bf3 Rxb2 24. Reb1 Bf5 25. Rxb2 Bxb2 26. Rf1
Ba3 27. Bg5 Rd4 28. Be3 Rd8 29. Bg5 Rd4 30. Be3 Rd8 31. Bb6 Rd6 32. Ba5 Rd4 33.
Bc3 Rd8 34. h4 *[/pgn] and it is still uncertain to me if white has a winning advantage in the endgame. Here, one could also imagine what Fischer described as a drawback but what I actually would find attractive as a scientific aspect of chess: the possibility of very late novelties. If these novelties actually matter, i.e. it might be the case that they improve over their predecessors in absolute terms rather than just practical ones, then I cannot see why this form of chess should be any less appealing than, say, FRC. I would rather think that people with a classical intuition should be able to enjoy it a lot more.

The statement that White is probably winning under NBC prompted me to investigate further. It does seem to be true, based on deep evals of the top three engines (SF, Dragon, Lc0). The evals after very deep searches suggest that White has about a 60% probability to win outright, and in view of the general agreement of the three and the tendency of the evals to climb with depth does seem to confirm the claim. So I thought about the simplest, cleanest modification that doesn't require special code to play the game properly (a move limit on castling is a rule change and requires new engines to play it properly). So I came up with this: White can only castle short, Black can only castle long. Obviously this is less favorable for White than NBC, since White has lost an option but Black has not. It is not a lot less favorable, since White normally castles short, especially if he wants to win and knows that Black can only castle long. But it is enough of a change to reduce White's average win prob. to about 46%, so most likely this variant is not quite a forced win, but quite close to it. Anyway it is much closer to the win/draw line than NBC appears to be, and it is better to be just short of the line than over it, since normal chess is presumed to be a theoretical draw. This variant also has the advantage of "symmetry" in that each side can castle, just on opposite sides, and of course it is not a rule change, just a mod of castling rights to the initial position. Maybe this is the ideal startposition for Armageddon chess; the most like normal chess of any that I know of that are roughly balanced with Armageddon scoring. Now it just needs a cool name!

Interesting that your analysis confirms what mine indicated. I also independently made the same consideration about the possibility of modifying white's castling opportunities. Of the 16 possible configurations, in addition to NBSC (KQ-q) and WSC&BLC (K--q), there also is "only white short castling" (OWSC (K---)) remaining as an interesting variant, which appears to be somewhere in the middle between the former two eval-wise. This variation might yield a little lesser resemblance with standard chess as there will be no lines where both sides can castle.

So, e.g. assuming that NBSC is a white win and WSC&BLC a draw, OWSC will be either a harder win (for white) or a harder draw (for black) by some measure assumed to be correlated with both differences, e.g. the number of moves black and white, respectively, can force the game to last (assuming the fifty moves rule). (As a variant to resolve the draw problem, I considered this by itself instead of e.g. mobility chess, but without any refinement it is too far from what is considered practically optimal chess. But it is probably good enough to distinguish between two sufficiently different positions that are still objectively equal under classical rules.)

There is the philosophical question about how much we actually want to know, since one could imagine a state of knowledge where all these attempts are more or less irrelevant. It seems to me that these variants are "uninteresting" as soon as we know (to a reasonably high degree of certainty) their objective status, while this is at the same time what one is trying to resolve when analysing. A bit paradoxical. Perhaps what we really want is a game where all knowledge for the forseeable future will always be far from perfect, somewhat like go, i.e. somewhat far from the precision with which one can analyse the vast majority of at least the non-opening chess positions today. But again, the possibility of this precision is also somewhat attractive in that it forms a cumulative body of knowledge about e.g. endgame theory. It seems that one should try to refine chess in a way that retains as much existing theory while making solving it harder. So e.g. if WSC&BLC or OWSC for a short while is unsolved, i.e. Stockfish and Dragon will not achieve the same results over and over playing each other, when they eventually are solved, one could perhaps try to make a further refinement by specifying move intervals. The exact borderline between wins and draws under these refinements would assumably be so sensitive to the status of a few select number of lines that it could take a very long time to resolve their probable status. Intuitively, this should to a lesser extent be the case comparing NBSC and WSC&BLC, but sorting this out eventually seems to be quite manable for today's engines.