fair drawless chess

[lkaufman]

I have finally figured out how to change the draw rules in chess to eliminate draws without giving either White or Black a meaningful edge. Here is the proposed rule: "When the game would be drawn under current rules, the side who made the last capture (or promotion) after move number X wins the game. If no capture (or promotion) occurs after move number X, Black wins the game". To determine the proper value of X for this rule, engines would have to be taught this rule, and then top engine vs engine games played with various values to determine which value produces results closest to 50-50 for White and Black. A value close to zero should be winning for White, since White should be able to get a small edge and be in control of who should make the last capture, while a huge value like 300 is just normal Armageddon chess, which everyone agrees is a win for Black with best play. My guess is that some value for X of between 20 and 30 will result in nearly 50-50 results. Of course there is no guarantee that the best value for top engines will also be optimal for weak engines or humans; this could of course easily be tested with weak engines that play this variant or with top engines crippled by search depth. My feeling based on my experience is that the ideal value won't be drastically different for top engines and humans, some round number like perhaps 25 might work well enough to be suitable at all levels of play and time limits.

Some observations about this idea: 1. It is not necessary for the entire game point to be awarded by this tiebreak, it could also be used to give the tiebreak winner 0.75 or 0.6 or any other chosen fraction of the point. But then it won't be "fair" anymore unless the value of X is re-optimized for that scoring. 2. If you are more concerned about opening memorization/prep, this idea works just as well with Chess960, although the ideal value for X may be different. In fact for fully balanced play, a different value of X could be optimized for each of the 960 positions, although this is probably not needed for human play. 3. Note that all historical decisive games will still be valid, as will all opening lines that lead to a win, but drawing lines will now be theoretically winning for one side or the other, depending on the value of X. Most of the time, openings that produce a clear (but not winning) edge for White will be won for White, while openings that produce only a microscopic or cosmetic edge for White will be won for Black. There should be a large enough gray zone to keep things interesting, just as using one-sided positions as in TCEC does now with current rules. Not all normal openings will make the cut, but I expect that the only openings that would be considered good would be ones that are or were normal ones in the past.

[jefk]

wow, this is interesting (inspired Shogi or so (?), just wondering)

Besides engine testing, from (a) historical database(S) it could be figured out
via some pgnextract code (similar to maybe Pohls interesting moves algorithm)
for various values of X where the balance lies between Black and White.

While you suggest conventional opening hold, at least initially, i doubt whether
conventional playing strategies will hold, because of these capturing rule (or
avoiding captures); but i'll think about it, in the meantime becoming a more lazy person
nowadays with the AI's , i asked chatgpt40 (1) . Got an impressive answer(s)
but note that even this bot(4) can sometimes be bullshitting (as i noticed
when i explored some methods in game theory).

Certainly the endgame strategies would change, i think. It also reminds me of the old
adjournment rule for move 40, where players always should be alert when this would
happen in a game; similar some moves before move x (and estimating the endgame plan)
the players probably would need to adjust there strategy/evaluation;
as confirmed by my second ai question (2)

But when an engine would have such a rule builtin, then it will be figured out rather
quickly how the best playing strategies will look like i presume, which could mean
some changes over the years for the optimal x for humans (determining
x values for various rating classes might be a (wild?) idea). Opening strategies
will become different (which anyway isn't a bad thing), if the AI is correct
(3) more interesting games, with gambit and so on, something which i'm in
fact already am exploring (with the latest nnue evals some optimal gambit
lines have changed compared to many old book lines).

Talking about bs'ting, when asking about possible drawbacks of such a rule
it responded something about drawing tendencies, so i had to ask it again and
by then it had learned from itself ie adapted the earlier - wrong answer (4)

My prelimanry conclusion(S), very interesting proposal, certainly a topic for further
research/testing, and subsequently introducing it as extra variant may be the way
to go (at least for correspondence chess and engine chess at high level besides
the TCEC book cooks); more suggestion from the (imaginitive AI)

(1) https://poe.com/s/NzFdgFz7FBu5xTX8B5en
(2) https://poe.com/s/NzFdgFz7FBu5xTX8B5en
(3)https://poe.com/s/NzFdgFz7FBu5xTX8B5en
(4) https://poe.com/s/hHZaFkNTjVdj38KS77Fn
(5) https://poe.com/s/UF2wUiIh56fMgGCDL3kQ

[lkaufman]

wow, this is interesting (inspired Shogi or so (?), just wondering)

Besides engine testing, from (a) historical database(S) it could be figured out
via some pgnextract code (similar to maybe Pohls interesting moves algorithm)
for various values of X where the balance lies between Black and White.

While you suggest conventional opening hold, at least initially, i doubt whether
conventional playing strategies will hold, because of these capturing rule (or
avoiding captures); but i'll think about it, in the meantime becoming a more lazy person
nowadays with the AI's , i asked chatgpt40 (1) . Got an impressive answer(s)
but note that even this bot(4) can sometimes be bullshitting (as i noticed
when i explored some methods in game theory).

Certainly the endgame strategies would change, i think. It also reminds me of the old
adjournment rule for move 40, where players always should be alert when this would
happen in a game; similar some moves before move x (and estimating the endgame plan)
the players probably would need to adjust there strategy/evaluation;
as confirmed by my second ai question (2)

But when an engine would have such a rule builtin, then it will be figured out rather
quickly how the best playing strategies will look like i presume, which could mean
some changes over the years for the optimal x for humans (determining
x values for various rating classes might be a (wild?) idea). Opening strategies
will become different (which anyway isn't a bad thing), if the AI is correct
(3) more interesting games, with gambit and so on, something which i'm in
fact already am exploring (with the latest nnue evals some optimal gambit
lines have changed compared to many old book lines).

Talking about bs'ting, when asking about possible drawbacks of such a rule
it responded something about drawing tendencies, so i had to ask it again and
by then it had learned from itself ie adapted the earlier - wrong answer (4)

My prelimanry conclusion(S), very interesting proposal, certainly a topic for further
research/testing, and subsequently introducing it as extra variant may be the way
to go (at least for correspondence chess and engine chess at high level besides
the TCEC book cooks); more suggestion from the (imaginitive AI)

(1) https://poe.com/s/NzFdgFz7FBu5xTX8B5en
(2) https://poe.com/s/NzFdgFz7FBu5xTX8B5en
(3)https://poe.com/s/NzFdgFz7FBu5xTX8B5en
(4) https://poe.com/s/hHZaFkNTjVdj38KS77Fn
(5) https://poe.com/s/UF2wUiIh56fMgGCDL3kQ

I suppose the inspiration for the idea comes from the Komi rule in GO, since the Komi in Go has the same purpose as the X value in my proposal, to roughly equalize the chances in a game which would otherwise be a forced win for the first player. I looked at some comments by the chatgpt, and they are so full of obvious errors as to be useless. They don't seem to have understood the rule properly, and they often make comments that only a very weak chess player would make. It's like asking a 1200 rated player to comment on the relative merit of the Najdorf vs. the Sveshnikov Sicilian. Maybe in a couple years these bots will match top human experts in their commentary, but not now.
My hope (belief?) is that the proper value of X will not be very sensitive to the rating of the players or the time limit. Unfortunately, the proper time odds for conventional Armageddon to be fair is highly dependent on these factors; with human top players, something like 10 to 7 minutes is balanced, while with top engines, even an hour to a minute might not be enough. But I don't think X will be like that.

[jefk]

komi rule, yes i know it, and it's compensating the initial advantage for Black,
but because of the 0.5 addition value (eg. 6.5 initially) you can avoid draw.

Your proposal looks more far-reaching but i'll test it by playing some games against
the comp with my Dgt board, which i wanted to do anyway again and then
see how after move x play for my side can be adjusted against the comp -who
then will be unaware of this new rule and thus will play suboptimal.
For the rest agree ther AI responses were way too simple and often incorrect
but i use (part of) it as input for some brainstorming, input for further thoughts.

At the moment i'm already wondering how to win after move x, first think
for White then the goal is to get the last capture. But if White makes a capture in
an equal position, then a recapture by Black would be likely. Thus White will
avoid captures, but then he also will lose; but i presume you thought of this.
Now look at Black, similar, after a capture, a recapture is likely, thus Black usually
will avoid captures, and then -if no captures- Black will win. Ergo, it looks
like Black automatically gets a large advantage (but i'll test it, not always
positions are so balanced that such scenario's occur, and in endgames anything
can happen, well at least other things than normally in the middlegame.

[hgm]

I think this will cause a very drastic alteration of the game; the players would have to strive for a completely different goal. You might as well play Spartan Chess instead. This has a much lower draw rate, because end-games always involve unequal material. Eliminating draws completely is not very satisfactory; no matter what tie breaker you use, when both players performed equally it always seems arbitrary. You might as well flip a coin to determine the winner when FIDE rules would specify a draw.

To solve the draw problem with starting from an opening line that brings white on the verge of winning, and award the game to black when white doesn't win would probably have the least impact on strategy.

[jefk]

yep hgm, i did a testgame, and for humans it seems to 'work'' (is a viable variant), provided
the player(s) is (are) very alert and anticipative not to allow a recapture in the end. Other
objective indeed. So for engines it could work but would indeed be a (completely?) other
sort of chess, with the engine aiming for the expected capture sequences; with also
implications for the game before his nr x, i suspect. While -at least in my testgame- it seemed
to correlate with the original objectives (when one side achieves dominance much better
position, this side also can easier steer the game towards this (new/other) 'last capture'
goal (instead of the conventional checkmate). But still not so easy to fully comprehend
(initially) for people always having played the classical variant(*), especially after move x.
Nevertheless imo something to explore (as extra variant), possibly for further future;
and with some efforts, some new engines might even be able to bypass SF

As for selected special openings, well your (hgm) suggestion is possible but not easy to
calibrate wrt b/w chances (similar to Armageddon); but then i'm thinking about the
TCEC (or an improved set) with playing with both colors, and can work for
some suitable (**) applications (Iccf correspondence maybe); my 2 cnts.

(*) but then 'normal' chess also isn't easy and takes many years to learn
(and you need to start young to get at GM level); so it might be something
for a new generation maybe
(**) setting it up in an engine room like 'playchess would be a hassle
because of the two-game '(b/w and w/b) requirement.

[towforce]

You cannot beat simple!

Code: Select all

+-------------------------------+--------+
| Outcome                       | Points |
+-------------------------------+--------+
| Win                           | 2      |
+-------------------------------+--------+
| Draw As Black Using Less Time | 0.7    |
+-------------------------------+--------+
| Draw As White Using Less Time | 0.6    |
+-------------------------------+--------+
| Draw As Black Using More Time | 0.4    |
+-------------------------------+--------+
| Draw As White Using More Time | 0.3    |
+-------------------------------+--------+
| Lose                          | 0      |
+-------------------------------+--------+

[lkaufman]

You cannot beat simple!

Code: Select all

+-------------------------------+--------+
| Outcome                       | Points |
+-------------------------------+--------+
| Win                           | 2      |
+-------------------------------+--------+
| Draw As Black Using Less Time | 0.7    |
+-------------------------------+--------+
| Draw As White Using Less Time | 0.6    |
+-------------------------------+--------+
| Draw As Black Using More Time | 0.4    |
+-------------------------------+--------+
| Draw As White Using More Time | 0.3    |
+-------------------------------+--------+
| Lose                          | 0      |
+-------------------------------+--------+

I assume you meant 1 point for a win, since the way you have it, it is just a more extreme version of the football scoring where draws effectively count as 1/3 of a win (in your chart it would be just 1/4). This of course just turns chess into a game of "chicken" (rather like the mutually assured distruction of nuclear weapons) where each player hopes the other will do something inferior to avoid a draw. If you did mean 1, then the issue is that awarding more points for the player using less time encourages whoever gets ahead on time to cling to a tiny time advantage, even just a few seconds out of hours, to "win" in case of a draw. This is not desirable either. Simply giving more to Black than to White for a draw is an old idea with little downside, but it won't reduce draws, just tied scores.

[lkaufman]

I think this will cause a very drastic alteration of the game; the players would have to strive for a completely different goal. You might as well play Spartan Chess instead. This has a much lower draw rate, because end-games always involve unequal material. Eliminating draws completely is not very satisfactory; no matter what tie breaker you use, when both players performed equally it always seems arbitrary. You might as well flip a coin to determine the winner when FIDE rules would specify a draw.

To solve the draw problem with starting from an opening line that brings white on the verge of winning, and award the game to black when white doesn't win would probably have the least impact on strategy.

Changes to draw rules don't affect current strategy aiming to win games, they only provide alternate avenues to win if the normal ones fail. As long as the tiebreak rules usually favor the same side who would have the better winning chances given flawed play, they shouldn't change strategy too much. Using TCEC-like openings or no Black castle short or White starts with two moves or the like together with Armageddon rule is fine, but it obviously changes the opening totally and I think changes the game more than a good draw tiebreak like mine. At the very least, forbidding repetitions (either repeated positions or moves of a piece back where it just left without an intervening capture or pawn move) (you can debate which rule is superior) would dramatically reduce draws without changing too much other than play in positions where a repetition can be foreseen.

[lkaufman]

komi rule, yes i know it, and it's compensating the initial advantage for Black,
but because of the 0.5 addition value (eg. 6.5 initially) you can avoid draw.

Your proposal looks more far-reaching but i'll test it by playing some games against
the comp with my Dgt board, which i wanted to do anyway again and then
see how after move x play for my side can be adjusted against the comp -who
then will be unaware of this new rule and thus will play suboptimal.
For the rest agree ther AI responses were way too simple and often incorrect
but i use (part of) it as input for some brainstorming, input for further thoughts.

At the moment i'm already wondering how to win after move x, first think
for White then the goal is to get the last capture. But if White makes a capture in
an equal position, then a recapture by Black would be likely. Thus White will
avoid captures, but then he also will lose; but i presume you thought of this.
Now look at Black, similar, after a capture, a recapture is likely, thus Black usually
will avoid captures, and then -if no captures- Black will win. Ergo, it looks
like Black automatically gets a large advantage (but i'll test it, not always
positions are so balanced that such scenario's occur, and in endgames anything
can happen, well at least other things than normally in the middlegame.

Captures before move X don't matter, so if X is large Black has a large advantage due to the Armageddon rule, but if X is small then the Armageddon rule has little meaning, since almost all games have captures. In that case the player with the better position, which should normally be White, should be able to provoke the first capture by the opponent (which he can recapture), to avoid something worse happening. There must be a (nearly) fair value for X.