The 100 pct draw problem in ICCF (top) correspondence chess

[jefk]

In a recent thread about the latest ICCF (WC32) world championship
the high draw rate also was discussed (in fact 100 pct, when no input
mistakes are made and topclass -eg SFNnue- engines are used).

While i remembered an earlier discussion here about possible
rule modifications to reduce such a draw rate, i continued
with some points of that earlier discussion, but this went
a bit off-topic, so i now make a new thread/topic (this one)

The earlier discussion, almost two years ago, can be found here
https://talkchess.com/forum3/viewtopic. ... G9rMIn1wVc
with some suggestions by Larry kaufman and a reference
to an earlier proposal by Arno Nickel.

A summary list of proposals was made by Tom gunn CCE
on Facebook (in the ICCF group) and i repost here:
-Penalise for repetition
-Mandate bad openings / forced openings
-¾ point for stalemating opponent-
-¾ point for minor piece against bare king
-Abolish 50-move rule
-Remove a white pawn (c2, f2, a2 or h2)
-¼ point for perpetual check
-No black O-O, but black gets draw odds
-Change the starting positions
-Reduced time for white
-Use a countdown where white wins with draw & black wins at 0

Then I made some additional comments:
"the countdown proposal would only work if White has a significant (bigger) advantage (so forget it). Second, in addition to counting a piece advantage in the endgame (rare), i think B or W should also be rewarded when ending up two pawns more (less rare eg with opposite bishops) eg 0.6 for W and 0.7 for Black; we will need scores like 1/4 (0.25), 3/4 (0.75) so other scores as these (0.6 or 0.65) also should be possible; and no, i don't think it would change the nature of the game (with gambits) so much, the goal is to (significantly) reduce the draw margin."

Apparently 'forbidding' the 3 pos repetition draw rule can affect
the draw rate in the most significant way, without changing the
nature of the game of ches in an excessive way.
See here the earlier Kaufman posting
https://talkchess.com/forum3/viewtopic. ... 10#p880210

The usefulness of such a rule modifcation was demonstrated
by talkchess user 'Ferdy' , by adding a penalty (in the SF code
for the outcome of the game) for the side who (first) repeats
for the 3rd time
https://talkchess.com/forum3/viewtopic. ... 70#p880689
Don't know if Ferdt still active here otherwise such little SF modifcations
could easily also be made by others i guess); although it partly
also may be an GUI interface thing i suspect.

Pondering further about such a modification, i suspect the biggest
objection (at least for the short-medium term) is that modification
of tournament software would be required, to allow for scores as 1/4
3/4 and so on. However, as first investigation we could simulate
the result(s) if the side who repeats simply gets only zero points.
This looks like a drastic change, but with the adapted code,
the engine (or at first the operator) would avoid such a draw,
and thus in practice many of such zero's can be avoided i suspect
(but not all, thus making it an interesting possible modification
to reduce the draw margin in ICCF correspondence chess.

At first only as an ICCF alternative ofcourse, in order not to upset the
system too much, and to make the whole exercise voluntarily.

And before that, a simulation can be run eg. an engine tourn
with such modified rules (and/or possibly coding( to see
how the draw rate would be affected, and whether we get
realistic games (eg. not zero's for the White side in drawn position
like in that boring Catalan as mentioned earlier (Osipov game).

My 2 cnts.

(and constructive comments are welcome of course

[Nigel Robson]

Back in September LSS kindly agreed to host two or three trial CC tournaments with no black 0-0 and black draw odds. I am playing in the first. The chess is demanding and exciting. I had thought that black might have an advantage at CC time controls, but in my few games that has not been apparent. It also seems that relative player strength will be reflected in the results. It is early days, but this kind of event may prolong the life of CC.

[jefk]

thanks for your comment
any links ?
where you played this ?
and more clarification on the rules ?

cheerio

[lkaufman]

Back in September LSS kindly agreed to host two or three trial CC tournaments with no black 0-0 and black draw odds. I am playing in the first. The chess is demanding and exciting. I had thought that black might have an advantage at CC time controls, but in my few games that has not been apparent. It also seems that relative player strength will be reflected in the results. It is early days, but this kind of event may prolong the life of CC.

When we initially did some tests on this idea a few years ago, it seemed to be almost perfectly balanced, impossible to tell whether White or Black was favored. But now with much stronger engines thanks to NNUE, it looks like White has the advantage (is White leading in your events?). If so, a small modification moves the needle a bit in Black's favor, making it too close to call; namely that White cannot 0-0-0. Given that Black can only castle long and that White must win, he would only rarely want to castle long anyway, but losing this option does occasionally matter, enough to balance the chances according to current engine analysis. Current SF dev, which now defines +1.00 as 50% win prob. (so fair for Armageddon), gives this option exactly +1.00 on my computer after 34 plies, although of course it jumps around a bit and depends on number of threads.

For the pawn odds version of Armageddon, it seems that removing d2 or e2 with White winning draws is also perfectly fair (as best I can tell), with Stockfish dev. also giving a -1.00 eval (fluctuating up or down a centipawn or two) for both around depth 37. Removing c2 is a bit worse for White, since it doesn't aid bishop development, but still just a bit over the magic 1.00 eval in Black's favor.

The other solution, changing the draw rules (or draw scoring), rather than the start position, is also viable but a bit trickier. If we forbid (or penalize) repetitions, and make stalemate and bare king losses (or 1/4 point or so), and extend 50 move rule to say 75 moves, draws go way down but White's advantage becomes much larger. It would be nice if we could then award draws to Black (or give Black 3/4 point or so), but probably this would favor Black too much. So for this to work, some subset of 50 move rule (or 75 move rule or whatever) would have to become wins, with remaining draws scored in Black's favor. Just what that subset would be, would be a matter of debate.

[Nigel Robson]

LSS is the CC server Lechenicher SchachServer. It has a large membership, probably second only to the ICCF, and many strong ICCF players play there too.
The rules are as described. Black may not castle kingside, but black takes the point from a draw. Otherwise normal chess rules apply.

[Nigel Robson]

When we initially did some tests on this idea a few years ago, it seemed to be almost perfectly balanced, impossible to tell whether White or Black was favored. But now with much stronger engines thanks to NNUE, it looks like White has the advantage (is White leading in your events?).

Too few results so far to be able to take a view, but in my games I have found playing with white a little easier than with black. Very interesting to note your other comments.

[jefk]

about the LSS chess server: thx for the info, found them.
https://www.chess-server.net
So they have some special tournaments (even 'test
tournaments'); as for the Black draw odds (without castling),
looks like an interesting variant, but with strong play it should
have implications for the opening choice. as for engine
analysis, i'm not aware of strong (eg SF) engine which can
play such a variant; there exists a SF derivative, 'Fairychess'
https://fairy-stockfish.github.io/variants/
but i don't see the draw odds variant in the list of variants...

Thus i still advocate a variant without the 3 pos repetition draw rule,
as example, i had a look again at the Moll vs Osipov game in the fianl
(the Bogo-Indian defense against the White Catalan setup).
After some more inspection, to prevent move repetition tih
a move as Nb6d5, he has to play another move, eg. Ba7b8
But then White is slightly better. And in such a way a lot
of draws can be prevented, i suspect.
With such a modified 3 pos rep (loss) rule, the draw
rate probably can be reduced significantly; whereas it
may look a bit drastic to penalize such a 3 x position-repetition
with a loss, the engines can often avoid this (except in
some endgames maybe). Maybe it would lead to a winning rate
of approx. 30 pct for White and 10 pct Black,
and at the highest levels maybe only 10 pct for W and
5 pct for B, don't know, but it would make correspondence
chess competitive again. NB wondering what mr Kaufman
woudl think of this (zero points instead of 1/4 after such
repetition) ? To me it looks worth-wile investigate. I'll send
a PM to Ferdy to ask about his SF version which avoids draws.
Worth a try, i would suggest, first maybe only at this LSS
server, but after more experience with such games, and/or
possibly other modifications could (maybe should) be made.
But then, when we got a solid system, it could be advised
for the ICCF to use as well, i.e. become a realistic
alternative for modernized correspondence chess.
Thus, maybe ICCF correspondence chess maybe
can be saved; of course when it's only a variant it may
take some time to replace the official WC top tournaments (
but something worth-while to try/simulate somewhere anyway.
PS maybe it won't work with zero points (instead of 1/4)
in such a situation i might atart a thread about updating
tournament software (to be able to handle other scores
than only 0.0.5, and 1).

[lkaufman]

about the LSS chess server: thx for the info, found them.
https://www.chess-server.net
So they have some special tournaments (even 'test
tournaments'); as for the Black draw odds (without castling),
looks like an interesting variant, but with strong play it should
have implications for the opening choice. as for engine
analysis, i'm not aware of strong (eg SF) engine which can
play such a variant; there exists a SF derivative, 'Fairychess'
https://fairy-stockfish.github.io/variants/
but i don't see the draw odds variant in the list of variants...

Thus i still advocate a variant without the 3 pos repetition draw rule,
as example, i had a look again at the Moll vs Osipov game in the fianl
(the Bogo-Indian defense against the White Catalan setup).
After some more inspection, to prevent move repetition tih
a move as Nb6d5, he has to play another move, eg. Ba7b8
But then White is slightly better. And in such a way a lot
of draws can be prevented, i suspect.
With such a modified 3 pos rep (loss) rule, the draw
rate probably can be reduced significantly; whereas it
may look a bit drastic to penalize such a 3 x position-repetition
with a loss, the engines can often avoid this (except in
some endgames maybe). Maybe it would lead to a winning rate
of approx. 30 pct for White and 10 pct Black,
and at the highest levels maybe only 10 pct for W and
5 pct for B, don't know, but it would make correspondence
chess competitive again. NB wondering what mr Kaufman
woudl think of this (zero points instead of 1/4 after such
repetition) ? To me it looks worth-wile investigate. I'll send
a PM to Ferdy to ask about his SF version which avoids draws.
Worth a try, i would suggest, first maybe only at this LSS
server, but after more experience with such games, and/or
possibly other modifications could (maybe should) be made.
But then, when we got a solid system, it could be advised
for the ICCF to use as well, i.e. become a realistic
alternative for modernized correspondence chess.
Thus, maybe ICCF correspondence chess maybe
can be saved; of course when it's only a variant it may
take some time to replace the official WC top tournaments (
but something worth-while to try/simulate somewhere anyway.
PS maybe it won't work with zero points (instead of 1/4)
in such a situation i might atart a thread about updating
tournament software (to be able to handle other scores
than only 0.0.5, and 1).

I have been advocating for either forbidding or downscoring (i.e. 1/4 point or so) moves that repeat a third time for a while now, although I think that it works best if combined with making stalemate a loss for the immobile side and/or making bare king a loss (or downscoring it) (either when it first arises or only after the bare king side makes a move that doesn't create K vs K, two variants here). Together these would dramatically increase the percentage of decisive games, although if we are talking about Correspondence chess then even a ten fold increase might still be a tiny percentage. But most of the wins would be by White, it makes double-rr more essential. As for the merits of forbidding (or calling a loss, which is slightly different if stalemate remains a draw) repetition vs. downscoring to something like 1/4, there are arguments both ways, but I'll just say that downscoring is much more likely to be accepted by the chess community than outright loss or forfeit, as it preserves more of the integrity of the game so that historical games will still make some sense.

[jefk]

ok thx for the comments,mr Kaufman,
before a further reaction, I have to think a bit more
about it, as for other stalemate rules, i thought
there were some different proposals in the past, I
have to check that out. Meanwhile Ferdy promised
me to have a look again at his modified SF version.

Acceptance by the larger chess community (incl ICCF(
for such modifications seems hard to achieve, I'm not
going to 'push' anything but hope a useful combination of
modifications can be found and an alternative variant
will gradually evolve (and in some ways 'accepted').

[Ozymandias]

This is how I'd categorize the proposals I've seen, to reduce the number of draws:

- changes to the board, the more drastic and unpopular. Increasing size sure would help but the board is the most ancient part of the game, even shared with other board games.
- changes to the pieces, either in number or mobility. The first option tied to a bigger board. The second not just exclusively. Can be as subtle as castling rights, but would still mean changing the game as we've come to know it for 400 years.
- changes to the initial position. Popular, although it probably wouldn't solve the problem.
- changes to the rules
-- scoring system, generally accepted, negligible gains.
-- draw rules, without changing the game, probably the only way to address the problem of too many draws(*). FIDE says that a repetition of positions, insufficient material or a number of moves without exchanges or pawn advances are reason enough to declare the game a draw, but that's mostly to avoid prolonging human games. Going back to the 16th century, stalemating or leaving your opponent without pieces, wasn't even a draw. For computer chess, this should be reconsidered.
-- win rules, as proposed in another thread, can completely change the dynamics of the game.

* Not saying it would solve it.