correspondence chess in the age of NNUE

[lkaufman]

I suppose the actual wording should read something like: If a move is played that repeats the position for the third time, either player may call it a repetition, with the repeating side getting 1/4 point, the other side 3/4 point. For engine vs. engine play I suppose it would be automatically scored as such. The main point isn't to avoid repetitions played to avoid a no draws by agreement rule, it is that testing has shown that prohibiiting (or penalising) repetitions reduces draws far more than the stalemate + bare king rules do. There are many drawn endings where if you aren't allowed to repeat, you will lose (many even material King and pawn endings would be in this category). I don't claim that this rule avoids all draws or even most draws; I do claim that it roughly doubles the percentage of decisive games when the level of play is high enough that that percentage is a fairly small number. If you want to eliminate all or even most draws we would need more rules.

Hi Larry.
I assume that you are aware, that this rule will change the character of the game massivly. Not only that KPvKP are effected, but also many other totally equally balanced endings. Even KRvKR and KQvKQ can only try to escape by the 50-move-rule. Chances tobe adopted by ICCF? In my oppinion close to zero. I seen to much reluctantness in accepting rule that affects the game characteristics, even ones, that didn't smell such "unfair" in many situations.
Cheers,

In one of my simulations I saw KRvKR ended in repetition. This is ugly separating them by 3/4-1/4, but perhaps another rule could be added like in this type of position just award 1/2-1/2.

Was this a simulation where the engine "knew" to avoid repeating lest it get just 1/4 point? Of course KRvKR will normally end in repetition if there is no penalty for the player who repeats, but if the players have incentive to avoid the rep it should be fairly rare in such endings.

[Rein Halbersma]

I suppose the actual wording should read something like: If a move is played that repeats the position for the third time, either player may call it a repetition, with the repeating side getting 1/4 point, the other side 3/4 point. For engine vs. engine play I suppose it would be automatically scored as such. The main point isn't to avoid repetitions played to avoid a no draws by agreement rule, it is that testing has shown that prohibiiting (or penalising) repetitions reduces draws far more than the stalemate + bare king rules do. There are many drawn endings where if you aren't allowed to repeat, you will lose (many even material King and pawn endings would be in this category). I don't claim that this rule avoids all draws or even most draws; I do claim that it roughly doubles the percentage of decisive games when the level of play is high enough that that percentage is a fairly small number. If you want to eliminate all or even most draws we would need more rules.

This is an interesting idea. I'd like to offer some experience from another game with similar repetition rules: Stratego. In Stratego equal pieces always trade during captures, so it takes a higher rank piece to capture another (lower ranked) piece. Stratego has two repetition provisions:

• No piece may move across a grid edge separating two squares on more than 3 consecutive moves. In chess terms: the sequence 1.Ra1-a2 black move 2. Ra2 black move 3. Ra1 would not be allowed. For long-ranged moves it works similarly.
• A piece cannot continuously directly threaten (i.e. move directly adjacent to) opponent pieces and play into an already repeated position that already occurred within the same chasing sequence. In chess terms: perpetual checks where the checking player repeats are not allowed. A non-threatening move resets the history for the purposes of this rule.

All other repetitive sequences are allowed and will typically lead to a draw agreement for lack of progress. There is 50 moves rule in Stratego, but tournament games are played with a Bronstein clock for each player and an overall game clock. Games not finished inside the total game time are declared a draw. Stratego has a very low draw rate, around 10% to 20% at the highest level. Most top player draws are due to the overall game clock. The repetition rules allow slightly advantageous positions to be won instead be held to a draw by the minority player.

I wonder how much the various chess endgames you mentioned in this thread rely on repeated threat sequences / perpetual checks.

[lkaufman]

I suppose the actual wording should read something like: If a move is played that repeats the position for the third time, either player may call it a repetition, with the repeating side getting 1/4 point, the other side 3/4 point. For engine vs. engine play I suppose it would be automatically scored as such. The main point isn't to avoid repetitions played to avoid a no draws by agreement rule, it is that testing has shown that prohibiiting (or penalising) repetitions reduces draws far more than the stalemate + bare king rules do. There are many drawn endings where if you aren't allowed to repeat, you will lose (many even material King and pawn endings would be in this category). I don't claim that this rule avoids all draws or even most draws; I do claim that it roughly doubles the percentage of decisive games when the level of play is high enough that that percentage is a fairly small number. If you want to eliminate all or even most draws we would need more rules.

This is an interesting idea. I'd like to offer some experience from another game with similar repetition rules: Stratego. In Stratego equal pieces always trade during captures, so it takes a higher rank piece to capture another (lower ranked) piece. Stratego has two repetition provisions:

• No piece may move across a grid edge separating two squares on more than 3 consecutive moves. In chess terms: the sequence 1.Ra1-a2 black move 2. Ra2 black move 3. Ra1 would not be allowed. For long-ranged moves it works similarly.
• A piece cannot continuously directly threaten (i.e. move directly adjacent to) opponent pieces and play into an already repeated position that already occurred within the same chasing sequence. In chess terms: perpetual checks where the checking player repeats are not allowed. A non-threatening move resets the history for the purposes of this rule.

All other repetitive sequences are allowed and will typically lead to a draw agreement for lack of progress. There is 50 moves rule in Stratego, but tournament games are played with a Bronstein clock for each player and an overall game clock. Games not finished inside the total game time are declared a draw. Stratego has a very low draw rate, around 10% to 20% at the highest level. Most top player draws are due to the overall game clock. The repetition rules allow slightly advantageous positions to be won instead be held to a draw by the minority player.

I wonder how much the various chess endgames you mentioned in this thread rely on repeated threat sequences / perpetual checks.

It's pretty clear that in the vast majority of QvQ or RvR endings neither player should have to repeat to hold a draw; normally there won't be a winning plan for either side. But there will be plenty of other endings where the only way to save a draw is to repeat. Just banning perpetual check (as is done in Japanese, Chinese, and Korean chess) will reduce draw significantly, but not nearly as much as banning repetitions in general.

[Ferdy]

I suppose the actual wording should read something like: If a move is played that repeats the position for the third time, either player may call it a repetition, with the repeating side getting 1/4 point, the other side 3/4 point. For engine vs. engine play I suppose it would be automatically scored as such. The main point isn't to avoid repetitions played to avoid a no draws by agreement rule, it is that testing has shown that prohibiiting (or penalising) repetitions reduces draws far more than the stalemate + bare king rules do. There are many drawn endings where if you aren't allowed to repeat, you will lose (many even material King and pawn endings would be in this category). I don't claim that this rule avoids all draws or even most draws; I do claim that it roughly doubles the percentage of decisive games when the level of play is high enough that that percentage is a fairly small number. If you want to eliminate all or even most draws we would need more rules.

Hi Larry.
I assume that you are aware, that this rule will change the character of the game massivly. Not only that KPvKP are effected, but also many other totally equally balanced endings. Even KRvKR and KQvKQ can only try to escape by the 50-move-rule. Chances tobe adopted by ICCF? In my oppinion close to zero. I seen to much reluctantness in accepting rule that affects the game characteristics, even ones, that didn't smell such "unfair" in many situations.
Cheers,

In one of my simulations I saw KRvKR ended in repetition. This is ugly separating them by 3/4-1/4, but perhaps another rule could be added like in this type of position just award 1/2-1/2.

Was this a simulation where the engine "knew" to avoid repeating lest it get just 1/4 point? Of course KRvKR will normally end in repetition if there is no penalty for the player who repeats, but if the players have incentive to avoid the rep it should be fairly rare in such endings.

I added a penalty for the player that repeats.

[Thomas A. Anderson]

I can not imagine of any 4-piece endgame position (KRvKR, KQvKQ, etc) where one side has an "advantage" that "deserves" 3/4 of the point. The effect of this rule (3-fold-position-repetition isn't 0.5 points) would be that any of today's dead drawn endgame must be reevaluated, if one side can escape with a 50-move-0.5 before it is forced to step in the 3-fold-position-repetition-trap and only get 0.25.

Hi Neo,

The goal of the rule is to transform some draws into something other, by design. To deserve half a point you have to do better and see further than your opponent, so that you don't fall into those draw looking endgames (or positions) were you're actually getting just 1/4 of a point, and maybe anticipate so much than you actually get 3/4 of a point, and not just half. This has always how chess has been played, but we're focusing on something other than wins now, because those are becoming so scarce than the difference in playing strength no loger gets reflected on the scoreboard.

Hope that helps

Hi Juan. Nice to hear from you. I know the goals of the project and I am totally convinced of the need for change. "Unfortunately" after a lot of correspondence on this topic I am already quite disillusioned with the feasibility in the ICCF. As with CoVid-19, the right rules are only one part of a solution, the more important part is the acceptance of this solution by people. Experience has shown that it is essential for acceptance that the character of the game is not significantly influenced.
The rule proposed here that repetition of positions is valued at 0.75-0.25 would mean that (only one example) almost no pawn ending would result in 0.5-0.5. That would be less of a problem for engines, but very unnatural for humans. Likewise, many other endgames that are perfectly balanced for us humans would suddenly be avoided by the engines. As a result, the engine game would become less and less understandable and we would either have to leave the game entirely in the hands of the engines or learn a new kind of game. Both are not desirable or at least not enforceable in the ICCF. Til today Chess960 does not play a role in the ICCF, instead duplicates of whole games are piling up.

[Ozymandias]

As with CoVid-19, the right rules are only one part of a solution, the more important part is the acceptance of this solution by people.
[...]
The rule proposed here that repetition of positions is valued at 0.75-0.25 would mean that (only one example) almost no pawn ending would result in 0.5-0.5.

True, but these are the smallest changes with some tangible effect we have think of so far. If anything, Larry is entertaining more drastic measures. In essence, if we change even the smallest thing, people might leave because of it, but if we do nothing, everyone eventually will.

Just to visualize it, can you give me an example of a pawn ending position you consider to be totally equal, but were one side would half lose with these rules?

[Thomas A. Anderson]

What about something simple as:
W: Ke1 Pe2
B: Ke8 Pe7
[d]4k3/4p3/8/8/8/8/4P3/4K3 w - - 0 1

[Ozymandias]

What about something simple as:
W: Ke1 Pe2
B: Ke8 Pe7
[d]4k3/4p3/8/8/8/8/4P3/4K3 w - - 0 1

It illustrates what I was talking about earlier, material might be equal (even symmetrical), but white moves first, which means that black can be forced to give up space if it can't repeat the position, and space would ultimately mean losing the pawn. Such an advantage has always been deemed too small to deserve a win, and I agree, but that's not the same as splitting the point, we can give white something here, to distinguish this one from a truly equal position.

[Ferdy]

One improvement to 3/4-1/4 repetition scoring is to only apply it if total material (excluding kings and pawns) remaining on the board is equal or more than 2*queen value + 2*minor value where material is defined as q=10, r=5, b=n=3. That is about 26. If material is below that, normal 1/2-1/2 will be applied. This would solve lots of problems on familiar position with somewhat unreasonable results based from what I observed on game simulations with this rule. That 26 is my initial guess based on the fact that a Q vs Q should just be rewarded as 1/2-1/2 there must be a lot struggle and positions to learn from when this material balance is reached.

[lkaufman]

One improvement to 3/4-1/4 repetition scoring is to only apply it if total material (excluding kings and pawns) remaining on the board is equal or more than 2*queen value + 2*minor value where material is defined as q=10, r=5, b=n=3. That is about 26. If material is below that, normal 1/2-1/2 will be applied. This would solve lots of problems on familiar position with somewhat unreasonable results based from what I observed on game simulations with this rule. That 26 is my initial guess based on the fact that a Q vs Q should just be rewarded as 1/2-1/2 there must be a lot struggle and positions to learn from when this material balance is reached.

Based on the tests we've run with Komodo, even banning all repetitions (or scoring them 1/4-3/4) will still leave the draw percentage between top engines on big hardware way too high, although I predict that the frequency of nondraws (including 1/4-3/4 results) will roughly double. Any limitation like the above will leave the draw percentage so high as to still be almost unplayable. I would expect in any case that most even material simple endgames (such as Q vs Q) will still be drawn even with these rules, with the major exception of K and pawn endgames where the rules will produce many 3/4 - 1/4 results. In most other simple equal material endgames one side can make fifty somewhat random moves without repeating and then claim the draw; in cases where one side can't do this he will usually be clearly worse by normal chess criteria. Basically, if we're going to make a real effort to change the rules to reduce draws, the changes need to accomplish a big enough reduction to justify the effort. I think that FIDE has already rated at least one tournament that had unequal scoring for stalemate, so we're already 1/3 of the way there!