Shouldn't this be a draw?

[hgm]

What is the point in adjudicating tablebase positions? If both engines have tablebases, they should move instantly, so no time can be saved by it. If their EGTB support is buggy, they will bungle the game in this process. If they do not have EGTB support, they will need time to produce a move, but this is exactly what you test them for, as they might now err. If you adjudicate in such a case you might as well adjudicate every game as draw in the opening...

[hgm]

OK, that could be an argument. But it would only be valid if the EGTB was correct, and the implementation of the engine using it would be correct. Which for Nalimov EGTBs is certainly not the case, as the example in the leading post of this thread shows. Even if the GUI would have access to correct EGTBs (e.g. DTZ50), it could not assume that the engine in the formally won position would also have correct EGTBs, and would not blunder.

A more reasonable action (than adjudication) in this case would be to have the GUI strongly reduce the available time for both engines, by a factor 100, say. Than the engine supposed to have certainly won would still have to prove it can. And it would provide a stronger incentive for engines not having tablebases to start doing so.

[bob]

Many computer chess tournaments state in the rules something to
the effect of "Once a tablebase endgame is reached, the game is over
and the tablebase evaluation is used to decide the game assuming
that one or both programs have the necessary tablebases."

As far as I am concerned, this should apply to freestyle tournaments
as well. I became rather bored in a few freestyle matches where my
opponent kept on playing. I guess he thought I'd crumble due to time
pressure, but I kept gaining time due to fast moves that I obtained from
EGTBs. Once it became clear to him that I would not lose on time
he accepted my draw offer.

I think it better to follow the official FIDE rules of chess, which makes that game a real draw. We are trying to play by the rules of the game that everyone else has to use. It makes no sense to modify them for computer events. If you don't like the draw, then (a) use DTC/DTZ tablebases, or else (b) modify your program so that you realize that the above mate is really a draw and avoid it. To not play according to FIDE rules makes no sense...

[bob]

What is the point in adjudicating tablebase positions? If both engines have tablebases, they should move instantly, so no time can be saved by it. If their EGTB support is buggy, they will bungle the game in this process. If they do not have EGTB support, they will need time to produce a move, but this is exactly what you test them for, as they might now err. If you adjudicate in such a case you might as well adjudicate every game as draw in the opening...

The point is that if the losing side does not have EGTB support it is going to play a lot of moves in vain, thereby wasting time.

But it isn't "wasting time" if the "winning side" doesn't understand that the 50 move rule trumps a mate in 100 declaration by an EGTB probe. The FIDE rules of chess make no mention of _any_ game violating the 50 move rule because a tablebase has proven that the mate is longer than 50 moves. For a while this was done, but the rule was rescinded and has not come back in any form.

[bob]

Shouldn't this be a draw?

Rybka shows a Tablebase mate in 70 moves here. However does not the 50 move rule apply here? Yet after Black's 63rd move White resigned.

Why should the 50 moves rule apply here? White still has a knight.

Only with the lone king, 50 moves rule countdown starts.

That's wrong. "A player may claim a draw when both sides have made 50 consecutive moves without advancing a pawn or capturing any piece..." Note that you do not have to make a claim and both sides can _choose_ to play on. But for computers, the "losing side" would certainly make the claim after 50 moves, and be correct in doing so.

[BubbaTough]

This argument has convinced me, and at the same time it gave me an idea. What if we replace the 50-move rule with the '200-move rule' which would mean that any game reaching move 201 is drawn.

My counter proposal is that we use a 10-move draw rule...add some excitement. After 10 moves with no pawn move or capture, the side to move has to do a pawn move or capture, or he loses. Should help reduce draws. Players can alternatively stand on their head for 10 seconds, which works up until the 12th move, after which the normal forfeit applies. This should help humans even the odds against their computer counterparts (humans with good balance anyway).

-Sam

[bob]

The point is that if the losing side does not have EGTB support it is going to play a lot of moves in vain, thereby wasting time.

But it isn't "wasting time" if the "winning side" doesn't understand that the 50 move rule trumps a mate in 100 declaration by an EGTB probe.

This argument has convinced me, and at the same time it gave me an idea. What if we replace the 50-move rule with the '200-move rule' which would mean that any game reaching move 201 is drawn. This has some advantages:

1. Winning EGTB lines can be longer than 50 moves.
2. Easier to detect (particularly for humans).
3. No more games where the better side is approaching the 50-move rule then sacrifices a pawn just to make the game yet another 50 moves longer. I saw a lot of games reach 250 moves even with the use of the 50-move rule.

This rule is based on the record that the longest human game not ending with a draw took 193 moves. I think 200 is a good enough limit.

Of course resigning and agreeing to a draw is still encouraged.

I don't like artificially changing the official rules. Now you would want to develop a chess engine that is slightly different than the one you would use for real chess. Because now all you need to do is get to move 200 if you want a draw, and in the worst case, you might at move 195 sac your queen and two rooks to force your opponent to move 200 where you can draw, but you can't use that in a real game and you have to then know how to actually play that kind of ending out safely...

[bob]

Here is the pgn from the endgame. This is the longest path but there
were alternate long paths. After 59 moves the Knight is captured.


[D] 7k/8/4KN2/8/8/8/1bb5/8 w - - 0 1

1.Ng4 Bb1 2.Nf2 Kg7 3.Nd1 Ba1 4.Ne3 Ba2+ 5.Kf5 Kf7 6.Ke4 Ke6 7.Kd3 Bg7 8.Ke4 Bb1+ 9.Kf4 Bf8 10.Nc4 Bg6 11.Kf3 Bg7 12.Ne3 Ke5 13.Nc4+ Kd5 14.Ne3+ Kd4 15.Ng2 Be5 16.Ke2 Bh5+ 17.Kf2 Kd3 18.Ne1+ Ke4 19.Ng2 Bc7 20.Ne1 Kf4 21.Nd3+ Kg4 22.Ke3 Bb6+ 23.Kd2 Bd4 24.Nb4 Kf3 25.Kd3 Be3 26.Kc3 Ke4 27.Nd3 Bd1 28.Nb2 Bd4+ 29.Kd2 Bb3 30.Nd1 Bg7 31.Nc3+ Kd4 32.Ne2+ Kd5 33.Nf4+ Ke4 34.Ne2 Bh6+ 35.Kc3 Bf7 36.Ng3+ Kd5 37.Kd3 Be8 38.Ke2 Ke5 39.Kf3 Bf4 40.Kg4 Bc6 41.Nh5 Be3 42.Ng3 Bc5 43.Ne2 Ba4 44.Nf4 Ke4 45.Ng6 Bb6 46.Kg3 Ke3 47.Nh4 Bc7+ 48.Kg4 Bd7+ 49.Kh5 Bd8 50.Ng2+ Kf3 51.Nh4+ Ke4 52.Ng6 Be8 53.Kh6 Kf5 54.Nf8 Bf6 55.Nh7 Bb2 56.Nf8 Kf6 57.Nh7+ Kf7 58.Kg5 Kg7 59.Nf6 Bxf6+ *

I have never studied this particular endings with DTC tables, but I wonder if it is still won with correct DTC play as opposed to perfect DTM play? That is, can you force the win of the knight within 50 moves, even if it extends the depth at which mate is forced, to take the 50-move rule out of the equation?