[bug report] Game incorrectly classified as draw.

Hi,

I have a bug report concerning this game:

en.lichess.org/qd8FTyKR#117

I played White, a friend of mine played Black.

After White's move 59, Black ran out of time. The game was then classified as a draw -- presumably because I only had a King and Knight left (my opponent had King, Bishop, and Pawn). Here's the end position as FEN:

8/5N2/8/8/4k1p1/6K1/8/3b4 b - - 6 59

The adjucation as a draw stems from what I feel is an incorrect application of Article 9.7 of the Laws of Chess. Here's the full text:

"The game is drawn when a position is reached from which a checkmate cannot occur by any possible series of legal moves. This immediately ends the game, provided that the move producing this position was in accordance with Article 3 and Articles 4.2 – 4.7. "

In the position reached when my opponent ran out of time, there are certainly series of legal moves left that will result in White checkmating Black. I shall grant that those require a fair bit of assistance by Black, but that's irrelevant to the correct application of the rule. And since this rule does not apply, I feel that the correct result should have been 1-0 for White.

I appreciate that this is not a simple matter to solve programmatically - determining whether there exists a legal sequence of moves that will lead to checkmate from a given position is a hard problem. So I understand that a heuristic based on material count was put into place.

P.S. thanks so much for this fantastic site! It has re-ignited my love for chess.

You can't mate with a King and Knight and I understand it's possible if black were to assist and get it's king in a corner surrounded correctly, but that's not possible from this position. This would be ruled a draw in USCF as well.

And actually with pawn being a g-pawn there are no possible mates with the current on board material. The King would have to be in a corner i.e. King on a1/h1 with a pawn on a2/h2 to even have a possibility of a mate.

For example here with black to play lichess.org/editor/8/8/8/8/8/4n3/P7/K1k2B2_b_-_-
or
lichess.org/editor/2B5/8/8/8/8/8/7P/3n1k1K_b_-_-

But with g-pawn this is not possible as king can escape.
lichess.org/editor/2B5/8/8/8/3n4/8/6P1/4k1K1_b_-_-

Many reachable mate positions exist, e.g. with the black king on h8, black bishop on g8, white king on h6, and white knight on g6.

Personally I agree with the FIDE Laws of Chess, although there are a couple wrinkles here (IMHO rule 10.2a is not enforceable, and 14E ruins online play, although some strong players passionately disagree):

FIDE 10.2a:
If the arbiter agrees the opponent is making no effort to win the game by normal means, or that it is not possible to win by normal means, then he shall declare the game drawn.

USCF 14E: Insufficient material to win on time
The game is drawn even when a player exceeds the time limit if one of the following conditions exists:
14E1: Lone king
14E2: King and bishop or king and knight
Opponent only have a king and bishop or king and knight, and does not have a forced win.
14E3: King and two knights
Opponent has only king and two knights, the player has no pawns, and opponent does not have a forced win

Hi Toadofsky,

What you refer to as FIDE 10.2a has moved to Appendix G in the latest edition of the rules (Specifically, it's G.5a now).

However, G.3 states that the appendix only applies to "standard play and rapidplay games without increment and not to blitz games".

I think this is because the idea of running one's opponent through the clock is integral to very fast games - once both players are down to 5 seconds or so, often times both players will cease any attempts to do good moves, and rather choose to do merely legal moves. It's just an integral part those kinds of games.

I also don't think a computer should make that call, unless it can prove that mating positions sequences are impossible (eg KN vs K). One could think of situations where a player ends up with just KN in the end and mates deliberately, eg. as part of a game-ending combination. It would be wrong to invoke either G.5a there, so I'd rather not have the computer make the judgement call about what "by normal means" means. I think it should err on the side of not applying G.5a, unless it can prove that it applies.

I have no access to the USCF rules (I cannot easily find an online version), so I cannot judge the context there. I am unsure which rule set lichess aims to enforce - FIDE or USCF?

IMO it's a bit silly to count that as a win for white but technically it should be. Imagine one of these positions where you can actually force mate with N vs some weird position. Would be bad if that got adjucated to a draw.

Hi reddish, thanks for the G.5 information and I agree with you that FIDE (non-appendix) rules make the most sense for online play, although again this isn't my call.

#8 A reason to have unambiguous rules is so Lichess doesn't need to arbitrarily decide what's silly and what isn't. For example:
lichess.org/editor/7k/6pP/6Pb/8/8/8/8/4K2B_w_-_-

By FIDE rules White cannot win by a legal sequence of moves. Now what about:
lichess.org/editor/7k/5Pp1/6P1/8/8/1K6/8/2b4B_b_-_-

A smart Black player would play 1... Ba3 and hold a draw here, however it cannot be assumed that Black sees that... again, which is the point of having rules. And another example:
lichess.org/editor/7k/6p1/6P1/8/8/1K6/8/2b4B_b_-_-

Here, Black certainly can lose by a legal sequence of moves (regardless how surprising such an outcome would be).

Hi Vetinari_Computer,

Of course I agree, the draw is very much _just_, in the game presented.

But at any point it should be clear what the rules are; I was assuming FIDE rules; and then the ruling is not correct. In this situation: who cares, but there may come situations where this actually leads to unjust adjuction.

Hi Toadofsky,

> A reason to have unambiguous rules is so Lichess doesn't need to arbitrarily decide what's silly and what isn't.

Technically, the FIDE rule is unambiguous - but it is hard (perhaps impossible) to implement as an algorithm that needs to run quickly.

According to the FIDE rule, the first example you give is an immediate draw -- players don't even have to claim -- the game is over, and continuing playing is actually a violation of the rules.

It may be possible for a computer to see that in that situation; there are 'only' about 100000 positions to consider to prove that there is no possibility for mate.

I would argue that the current heuristic ("if a player has only K+N material he can never win") is not a good choice, hence the bug report.

In any case, that's arguable. The most important issue here is that the player has the right to know which rules and which heuristics are implemented that affect game play; and this should include corner cases like these. I am not sure if the lichess site provides that level of detail somewhere in the documentation?