Lila: Wrong results for some endgames

[Assios]

When having only a bishop or knight left, and your opponent runs out of time, you win (according to FIDE rules) if it is possible to win by any legal move-order. In this game, f.ex., black can avoid losing by letting his time run out: lichess.org/QptnrYGIG37H

Note that this applies to all positions with you have a bishop/knight against one or more pawns, since it doesn't have to be a forced win, like in the previous example.

[Vinvin20]

1 lone B or N = at more draw when win on time !
That avoids 99.9999999% of stupid results !

[Unihedro]

For study, both KB v KP and KN v KP can lead to white checkmating if black so plays:

If such a rule is to be implemented, there is no need to check for edge cases because black has a corresponding promotion (albeit stupid) that leads to getting checkmated.

[Unihedro]

Also, note that I replaced the pawns with pieces because a pawn can't change the file it's on, but a promoted knight / bishop can - and even though bishops comes in both dark and light, the four corners provide enough breathing room for a mirror case.

[arex1337]

@Unihedro did you consider most people are probably not familiar with that style of chess pieces? :p

[Assios]

These bishop/knight endgames are possible to win, and should be a win instead of draw when the flag falls:

Knight against pawn, rook, bishop or knight.
Bishop against pawn, knight and opposite-colored bishop.

[ddugovic]

Regarding the FIDE rule, does "any legal move ordering" also include move orderings which cause threefold repetition or 50 moves without a pawn move or capture? I understand that in recent years TDs have been able to claim 5-fold repetition, so it's not as if all move orderings would win.

IMHO time out should either result in a loss for the defender (the player who timed out) or add increment to the clock & force the defender to demonstrate a win/draw (and in either case, only 1/2 point be awarded).

[Assios]

It just means that if you can produce a win by doing legal moves (doesn't matter if the moves are good or bad), then it's a win if the flag falls. The 50 move rule and threefold repetition isn't a problem, since you can do ANY moves to produce a win as long as the moves are legal (if you could win by repeating the position before mating, of course you can win without repeating the position as well).

If there is NO way that the position could have been won when the flag falls, then it's a draw.

[Unihedro]

By Assios:

"Knight against pawn, rook, bishop or knight.
Bishop against pawn, knight and opposite-colored bishop.

those should be won"

The new draw conditions would be the old draw conditions minus the rules he listed, which is clear enough and pretty much covers it all.

[ornicar]

I can work with that.

[Vinvin20]

A knight against about everything (except only queen(s) ) wins.
But IMHO, it's a bad thing to encourage wins on time with a lone knight (or bishop).

For the material KBKN there's only very few positions with a mate (and only mate in 1) : example http://fr.lichess.org/analysis/k7/n1K5/B7/8/8/8/8/8_w_-_-
Distance to mate by material is here : http://kirill-kryukov.com/chess/tablebases-online/tbs.4-men/all-4-men-tbs.zip
Page here : http://kirill-kryukov.com/chess/tablebases-online/

[Assios]

@Vinvin20 That's the rules. Isn't it equally bad to encourage wins on time in positions like this? http://lichess.org/x2PyYB1swky3

[Unihedro]

White... missed... mate in one...

[Vinvin20]

@Assios : with lone knight or bishop, with more time, the game is draw for 99.99% of the positions.

[Assios]

@Vinvin20 So is 1 pawn against queens and rooks and pawns etc... Point is, we should stick to the rules, there are an infinite amount of positions where a drawn og lost position wins by flag.

[Vinvin20]

I don't like this rule. Other chess sites have set the rule too : with a lone minor -> no win on time.

[Unihedro]

Thanks for sharing.

[Assios]

That's not because they don't like the rule, but because they think it's hard to implement, because they think they need computer eval (at least on ICC).

[Vinvin20]

https://www.chess.com/forum/view/livechess/wrong-insufficient-material-rule

http://chess.stackexchange.com/questions/663/is-it-a-draw-by-insufficient-material-if-the-board-is-k-vs-k-n-n

[Unihedro]

Feel free to write to FIDE regarding your concerns. You have my support.

[Vinvin20]

ICC and chess.com didn't write to FIDE.

[Vinvin20]

@Assios : that's called a "practical decision". I hope Lichess will continue like this too.

[Assios]

But you don't need computer eval, and it should be quite easy to implement. Do you think this should be a draw as well? It is highly unlikely that black would ever win the game if it went on: http://lichess.org/x2PyYB1s

Or what about this? http://lichess.org/PtTNLOSlTXen

[Vinvin20]

As long as you won't understand that having a queen on the board is not the same power as having a lone N or a lone B, I'd wrote tons of text for nothing.

[Assios]

It is not the queen that matters, but the fact that it is highly unlikely to win the games.

[ddugovic]

Also there are some positions where the opponent cannot win by any series of legal moves:
http://lichess.org/editor/4k3/6p1/6P1/8/8/8/8/4KB2_w_-_- (edited and Vinvin is still correct; either side can win this; what I really meant was a fortress like http://lichess.org/editor/4k3/8/8/8/8/4p1p1/4P1P1/5BK1_w_-_- where White cannot selfmate)
http://lichess.org/editor/7k/6pP/6Pb/8/8/8/8/4KB2_w_-_- (double fortress)

I favor any rule suggested by @Assios although I think it's silly that anyone play chess without at least a 1-second increment.

[Vinvin20]

@ddugovic : in the first position, white can win with legal moves : black takes the pawn, goes to make an underpromotion in B or N and then go to selfmate in a corner with his K+minor. That shows how stupid the rule is ;-)

[Assios]

Yes, there are some positions that can't be won that are almost impossible for computers to find. Like this: http://lichess.org/LcQLFdk46B8a

That's why I suggest we focus on positions that are possible to judge, like knight against pawn, rook, bishop or knight and bishop against pawn, knight and opposite-colored bishop.

[Unihedro]

I'm open to the idea that the new criteria will introduce erroneous cases, but as long as it cuts down on bad cases I'm all ears, and the rule set is promising and already excludes cases where K + N is battling against a rook.

While I agree that FIDE rules suck, and that most chess servers are inventing their own, Lichess is running in gold standards and even _links to the wikipedia page of FIDE chess_ in place of "standard" (in the same format with other chess variants), so it's kinda awkward not to follow it.

[ddugovic]

I fully agree with @Assios and have coded and written tests for what he requests.

I apologize for my earlier behavior.

[niklasf]

Done via https://github.com/ornicar/scalachess/pull/97