fair drawless chess

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

jefk
Posts: 675
Joined: Sun Jul 25, 2010 10:07 pm
Location: the Netherlands
Full name: Jef Kaan

Re: fair drawless chess

Post by jefk »

There must be a (nearly) fair value for X.
ok then modified engine testing would be the way to go
(something for the Komodo team, maybe ?)

As for repetitions, in my correspondence games it didn't occur frequently so i wonder
how this could have such an (anti-draw) effect. But again with engine testing
(and this one is easy to program) this should become more apparent.
lkaufman
Posts: 5981
Joined: Sun Jan 10, 2010 6:15 am
Location: Maryland USA

Re: fair drawless chess

Post by lkaufman »

jefk wrote: Mon Jun 10, 2024 2:56 am
There must be a (nearly) fair value for X.
ok then modified engine testing would be the way to go
(something for the Komodo team, maybe ?)

As for repetitions, in my correspondence games it didn't occur frequently so i wonder
how this could have such an (anti-draw) effect. But again with engine testing
(and this one is easy to program) this should become more apparent.
Are you saying that most of your draws were fifty move rule draws (plus the occasional insufficient material and stalemate draws), or were they just agreed draws before it was apparent whether they would otherwise end in repetition or fifty move rule? I believe that fifty move rule is more common than repetition as a cause of draws when games are played to the end, but both are far more common than the other draw categories. Many drawn endgames become wins if repetition is forbidden, and even more do if this is combined with making stalemate a loss for the player who cannot move. Many perpetuals would be resignable if not allowed. It's not a cure-all, but would have quite a dramatic effect.
Komodo rules!
jefk
Posts: 675
Joined: Sun Jul 25, 2010 10:07 pm
Location: the Netherlands
Full name: Jef Kaan

Re: fair drawless chess

Post by jefk »

saying that most of your draws were fifty move rule draws (plus the occasional insufficient material ?
mostly a dead drawish position with 0.0 and no prospects in further analysis, ie. would
end in insufficient material or possible fifty move draw); i need to check this.

So if move repetition isn't allowed wouldnt this swing the balance towards White,
just wondering (you may have explained it earlier i guess). Anyway it would be
interesting to have an engine with this option, (with possible other options later).
User avatar
hgm
Posts: 27892
Joined: Fri Mar 10, 2006 10:06 am
Location: Amsterdam
Full name: H G Muller

Re: fair drawless chess

Post by hgm »

If under current rules 90+% of the games ends in a draw, then there would be far more to gain by dividing up that 90+% in a favorable way then by going for a normal win. Even when you manage 60-40 there you would do better than playing as usual, and using the tie-breaker as a sort of coin flip for acheiving a 50-50 split of the draws. This is why I think the focus would shift to getting the favorable tie breaker,

Note that for the 'nearly fatal opening' method there is no reason to play pairs of games. As long as a draw for the weak side is awarded the same as a win. There might be no intrinsic fairness there guaranteed by symmetry, so to make it fair enough requires careful tuning of the chosen line. I don't see that as a problem. As we kknow FIDE rules are pretty far from being fair, with a sizable white advantage. That doesn't stop us from having tournaments with one game per pairing. It should be easy to find an opening line that has a fairer win-vs-non-win distribution than FIDE.

Forbidding plain repetition would just turn most of these draws into 50-move draws. Except for very late end-games like K-vs-K it would be quite easy to avoid an exact repetition by aimless shuffling of pieces; even KR-KR offers thousands of constellations for a single player.
jefk
Posts: 675
Joined: Sun Jul 25, 2010 10:07 pm
Location: the Netherlands
Full name: Jef Kaan

Re: fair drawless chess

Post by jefk »

if under current rules 90+% of the games ends in a draw,
depends on the level, in correspondence chess at highest level its 100 pct (unless
someone dies). On playchess it's about 99 pct (unless a clear book mistake, or a disconnect;
the latter happens a few times /day depending on the location of the opponent)
Note that for the 'nearly fatal opening' method there is no reason to play pairs of games.
As long as a draw for the weak side is awarded the same as a win.
yep that's certainly an idea to consider (or even try) as the tourn software (ICCF and
playchess) then does not have to be modified.
There might be no intrinsic fairness there guaranteed by symmetry,
so to make it fair enough requires careful tuning of the chosen line.
this is a bit more difficult, we can pick some lines out of the TCEC which seem
most suitable, preferably several otherwise it becomes boring; the opening lines
(with full point for Black after draw) would be lines with a slight advantage
for White like d4 c5 (eg Benoni of Benko) or similar (i'll have a look at the TCEC lines;
taking an 2700+ database plus high level engines games with White lines scoring about 66 pct
may be suitable); and with a lottery/random system defining the White line for every game.
Forbidding plain repetition would just turn most of these draws into 50-move draws. it would be quite easy to avoid an exact repetition by aimless shuffling of pieces;
that's my feeling as well, but mr LK did some tests i think (with Komodo maybe), and
maybe Black can aim his play a bit more towards trying to get a move repetition (which
probably would slightly change the nature of the game, an almost unnoticeable effect)
In addition a stalemate rule may also be an idea (which gives a calibration problem
again maybe, and it will mess up the egtb 'theory' /tablebases).

Summarizing, at leasts the nr of ideas seems to converge, there now are three main options:
1) forbidding move repetition (and modification of an endgame rule eg. stalemate)
with the whole thing aimed at equal B/W chances
2) some special opening lines for White, tuned at approx equal scoring chances when Black
gets a full point with a draw it would become a bit more of a gamble but also more
interesting/exciting (certainly for the highest levels).
In theory it could be combined with above option 1 (and then a bit better white lines)
3) a new LK proposal of (trying to) make the last capture after ply X; this would avoid some
problems inherent in the above options, but will change the nature of middle/endgame play

PS as for human chess at highest level, i'm not really thinking about that, lateron -just
like with chess960- the Fide can pick out ideas as they consider suitable as nowadays
the Armageddon tie breaks (of which we also can wonder if it's really equal bw/chances) :|
jefk
Posts: 675
Joined: Sun Jul 25, 2010 10:07 pm
Location: the Netherlands
Full name: Jef Kaan

Re: fair drawless chess

Post by jefk »

here's a list:
http://chessopeningsforengines.wikidot. ... n-the-edge
Nb assuming that about 90 cp corresponds to (rougly) about 66 pct; although i'm surprised to see
the Sicilian Najdorf there (although not with a cp score); Najdorf is the strongest Sicilian defense,
maybe they are suggesting to pick out some special Najdorf lines with the 90 cp advantage.
More in detail lines can be chosen on the basis of w/d/l statistics found with Lc0.
The TCEC list is here:
http://chessopeningsforengines.wikidot. ... es-in-tcec
The openings for the cup events are the most useful, some user submitted
openings also look interesting. Such a system (the hgm proposal with random
White biased opening(s) can be implemented in Cutechess, as option, i suppose.
But how to sell such a system to ICCF or Playchess (initially as other option)
remains another matter.. :roll:
lkaufman
Posts: 5981
Joined: Sun Jan 10, 2010 6:15 am
Location: Maryland USA

Re: fair drawless chess

Post by lkaufman »

hgm wrote: Mon Jun 10, 2024 8:54 am If under current rules 90+% of the games ends in a draw, then there would be far more to gain by dividing up that 90+% in a favorable way then by going for a normal win. Even when you manage 60-40 there you would do better than playing as usual, and using the tie-breaker as a sort of coin flip for acheiving a 50-50 split of the draws. This is why I think the focus would shift to getting the favorable tie breaker,

Note that for the 'nearly fatal opening' method there is no reason to play pairs of games. As long as a draw for the weak side is awarded the same as a win. There might be no intrinsic fairness there guaranteed by symmetry, so to make it fair enough requires careful tuning of the chosen line. I don't see that as a problem. As we kknow FIDE rules are pretty far from being fair, with a sizable white advantage. That doesn't stop us from having tournaments with one game per pairing. It should be easy to find an opening line that has a fairer win-vs-non-win distribution than FIDE.

Forbidding plain repetition would just turn most of these draws into 50-move draws. Except for very late end-games like K-vs-K it would be quite easy to avoid an exact repetition by aimless shuffling of pieces; even KR-KR offers thousands of constellations for a single player.
All tie-breaker rules are not equal. The point of my proposed rule is that I believe it will correlate very highly with who is "better" according to normal principles and engine evals. The side with a slight edge in a drawish position will usually be able to control when action will start and therefore which side will get the last capture. For example, if you have an extra pawn, say two vs one in a rook endgame, at some point you push the extra pawn, opponent takes and you recapture, getting the last capture. Of course some modification of strategy is needed, I just think this minimizes how much change is needed compared to all other proposals.

The simplest rule for fair Armageddon chess is that White starts with two moves (you can make this a subset of standard chess by requiring games to start 1.e3 Nc6 2.e4 Nb8 for example, if we assume that White would want to play e4 as one of his two moves). This isn't perfectly fair with Armageddon rule, probably White's edge is a bit less than enough to win by force, but it would perhaps be something like 45% White wins or so. Instead, "Black cannot castle short" (with or without a restriction on White castling Long) is much closer to 50-50 with the Armageddon rule, and can also be achieved by a move sequence.

Forbidding repetition will turn many (maybe "most") draws into 50 move draws, but there are many games where a draw can only be achieved by repetition, either perpetual check or just some endgame repetition pattern. There have been tests that surprised me by showing that such rules do have a huge effect. If the draw rate is 100%, it won't help much, but if it is 90 or 95%, it would help a lot.
Komodo rules!
User avatar
hgm
Posts: 27892
Joined: Fri Mar 10, 2006 10:06 am
Location: Amsterdam
Full name: H G Muller

Re: fair drawless chess

Post by hgm »

Yet another remark: guaranteed fairness can always be achieved by means of the pie rule: one player makes the first move, the other then decides which side he will play.
lkaufman
Posts: 5981
Joined: Sun Jan 10, 2010 6:15 am
Location: Maryland USA

Re: fair drawless chess

Post by lkaufman »

hgm wrote: Mon Jun 10, 2024 8:37 pm Yet another remark: guaranteed fairness can always be achieved by means of the pie rule: one player makes the first move, the other then decides which side he will play.
Yes, quite fair, but that variant is even more drawish than standard chess, unless you combine it with some rule that adjudicates most or all position that would otherwise be drawn, for example last capture wins (regardless of move number), or last check, or player who gave more checks, or player leading by some point count system, etc. Theoretically then it would be better to play second, since every first move is either winning for White or Black in that case, but of course in practice no one would know who is winning after some first moves with such tiebreak rules.
Komodo rules!
jefk
Posts: 675
Joined: Sun Jul 25, 2010 10:07 pm
Location: the Netherlands
Full name: Jef Kaan

Re: fair drawless chess

Post by jefk »

The simplest rule for fair Armageddon chess is that White starts with two moves (you can make this a subset of standard chess by requiring games to start 1.e3 Nc6 2.e4 Nb8 for example
yep and already a bigger set for such purpose was made by mr Pohl, the drawkiller
books, with advantage 0.90 -0.99 up to 150-159, so for balanced Armageddon
one of these sets can be used (and there are already prepared books for that)
https://www.sp-cc.de/index.htm

However it are very unconventional moves, and thus besides the biased opening list i specified
from a wiki (*), the UHO 090-0.99 already were made for such a purpose, and he even made
an Armaggadon rescoring tool. The latest version where books were made was the UCO21V3
set and i'll have a look there; polyglot for Cutechess or abk for Arena, and chosing random
lines from the book, and finally the Armageddon rescoring. So it all can be done (and was
done by mr P with his UHO tournament apparently with a full (UHO24) game pgn list as opening
input for Cutechess (not sure the Armageddon rescore method was used in the result(s)
https://www.sp-cc.de/files/uho_full_list.txt

The Armageddon rescoring is the most controversial i think for Iccf (and playchess).
So, also for fun i improved/finetuned my (big) gambit.bin a bit more on the basis
of some Playchess games, it has no edge for White so Armageddon isn't necessary;
whether it will reduce the draw rate as much as the UHO remains to be seen
(i'll probably do some testruns for that although not as systematically as mr P); only
drawback is that the lines usually are quite different, not as consistent +0.9 or so
as the UHO lines. But to average out the differences of various opening lines i could
to a round-robin for (a few) engines several times, and look at the total (draw rate).

Meanwhile, besides the chess324 variant (as improvement of chess960) i support the
new Kaufman anti-draw variant (last capture after move x), ie. worthwile for further
research (but i'm not an engine programmer and such draw-forbidding rules should be
included in the search algorithms for proper testing; not really easy i guess but feasible)

(*) http://chessopeningsforengines.wikidot. ... n-the-edge
but this list probaly is a lot more balanced than the Pohl UHCO set.