Game drawn when it should be a win!

Hi everybody,

First off, I don't need a fix on this match result but would like to start a discussion if chess.com should work on implementing an improvement on their engine and specifically on whether a game is drawn or not.

What am i talking about? Well when a player times out and the other player has not enough pieces to mate it isn't a win but a draw. The Fide rules state:

(source: https://www.fide.com/fide/handbook.html?id=171&view=article)

The focus of this will be the last line and mainly the words "any possible series of legal moves."

Let's take this game: https://www.chess.com/daily/game/169293170

In this event my opponent's' account was closed due to fair play. I knew at the last move I would have won.  But somehow the chess.com engine gave it as a draw, This would be correct if I only had king and bishop. However there are plenty of (possible but unlikely) ways I could have checkmated my opponent. For those who doubt this I'll give the following example:

The site has decided to implement a simplified version of the USCF rules for insufficient material checks; it does not use FIDE rules. It only looks at the material the side with time has. K+B, K+N and K+N+N are draws.

 https://support.chess.com/customer/en/portal/articles/1444798-how-do-i-claim-a-draw-

It is not trivial to program a system to determine if mate is possible in a time loss situation, in all given conditions. The site decided on a trade-off and in most cases, the ruling would be the same under FIDE and USCF rules.

There are a ton of topics on it, and a few where @erik (the site CEO) discussed the implementation.

giving you a win could be appealed by black because he can claim he is being called a complete idiot.

 The thing is off course that the rule state any legal ways. It doesnt state likely ways. I know K+B against a lone K is a draw but in this case K+B vs. K and 2 pawns should have been a white win due to the timeout. Should the site change from the simplified version to a more elaborate version? I think it can be done easily if the engine just counts the pawns it should be able to determine the possibilities.

Engines are designed to find the best moves, not to find the worst possible moves to see if a mate is possible. Based on the previous topics, that staff has been involved in, it is highly unlikely they will ever implement FIDE's rules on it.

if a timeout for black  occurs and there is a K+B for white and K+B and 2 connected pawns for black. Do we agree that according to Fide rules (i have no idea why Uscf rules would apply on this site in the first place, we all play according to Fide rules except for one country, but this all is besides the point) white should be granted the win? What other combinations do we know that is always a win? (It is this way I would like this discussion to go forward, what combinations can we find to make the engine more precise)

It is trivially easy for a human to look at a position and determine if mate is possible with a material combination.  Pretty much any combination of two or more pawns on the side without time would allow mate to be possible, by various under promotions.

Things get much trickier when you look at other material combinations, where it may not be possible to blockade the king in a manner to allow mate to be delivered (unblockable and the other king not in danger).

Sure, the site could decide if any material surplus existed, on the side out of time, that as long as the other side had one piece other than the king, that it would be a win. It probably would have about as many false wins as the site gets draws that would be wins under FIDE now.

But the site, while worldwide, started as primarily a US-based one. They went with the easy implementation and it mostly matches USCF rules with the exception of 2 knights when the side without time having a pawn on the board (loss under USCF, draw here). USCF also does not do the mate by any possible series of moves; helpmates don't count.

In no way shape or form do i require engines like stockfish. What i need is the chess.com program to determine the game winner according to the rules of chess. we can achieve that by finding as much possible combinations that win. I do not need an engine. I do need combinations that win in this event. Which combination would win according to you?

Lots of them. About as many that wouldn't and would be draws.

 If you want a good history of why the implementation was chosen, take a look through this topic: https://www.chess.com/forum/view/help-support/no-mating-material--draw

Thanks for bringing this up. It's challenging to come up with rules when there is no human arbiter to look. We have done our best to combine some rules from FIDE, and some from USCF, and made the most sense of them for what works for an auto-arbitrated online server. There are certainly some compromises to be made and no easy answers. 

Are there plans to comfirm to FIDE rules and how can the users help to improve the engine. Because currently they don't comply fully with FIDE rules on chess which given that this is a worldwide chess site might be a good idea.

If Chess.com were to conform to FIDE rules, then they would have to consider implementing FIDE Guideline III.5  This rule, which is equivalent to the USCF's "Insufficient losing chances," rule allows a player who is short on time to stop the clock and request that the game be adjudicated a draw.  It's hard to implement this rule when there is no way to stop the clock and there is no arbiter available.

There is no single set of chess rules.  FIDE has its rules, USCF and others have their rules, and Chess.com have their rules.  When you play on Chess.com, Chess.com rules apply.

Unfortunately this was a daily game. So that would be similar to correspondence chess. no live time clock to stop to begin with. You would be right if this was live chess. But it isn't. The thing is that chess.com isn't a trade-off at the moment. It's just USCF rules. So it hugely favors the americans. My advice for this site and for Erik is to adapt the FIDE rules so this will be a worldwide chess site and not one with the US national federation quirks nobody else in the world has any care for.

I understand what the FIDE rules are and they are implemented in a way to remove the need for subjectivity but the USCF rules on insufficient material make a whole heck of a lot more sense.

I live by the rules of the site/rating system I play under but I don't see how winning under those conditions really makes any sense. The player with the material was completely winning and a draw is the best you could ever hope for had they not timed out and you would likely have lost barring a stalemate.

You do realize the player was cheating to get into this position right?

I'm not specifically speaking to how one got to the circumstances in that game but the idea that a position like that could be decided as a win on timeout of the stronger side (materially) when the game was otherwise completely lost. That's why I think the USCF implementation is objectively better but it does require more subjectivity on the part of a TD (or arbiter) and I understand why the rule exists as it does for FIDE.

For an online, automated process like Chess.com, I think it makes a whole lot more logical sense to not go by the "mate possible by any legal series of moves" ruling and go with the simpler, more likely scenario like chess.com uses.

The thing you can program the software to recognize the positions that lead to a mate. We all can help. So again, what position would you like to propose to give to the chess.com programmers to make their software more in line with the rules of chess most of the world follows (everybody besides the US)? Because that's what this topic is about.

I very seriously doubt that they will change anything with the insufficient material code. erik has posted here already and the topic I linked to goes through a lot of the discussion of why the current code exists as it does.

The problem you run into is that there will likely be a lot of positions where mate is not possible by any legal series of moves but it isn't possible to define such positions by simple material checks.  I'm sure the majority of cases can be hit easily but you are very likely to miss just as many positions with that method as the existing one where the code will give a win where it really should be a draw under FIDE.

Maybe I'm wrong and the implementation would be super simple and there are no edge cases or there are very few edge cases. But have at it. If you can come up with a set of easy checks and can guarantee that they won't wrongly give a win to someone that should have a draw, maybe erik will have the team code an implementation.

There are other related FIDE rules the site doesn't code for. Such as dead positions or ones where mate is not possible by any legal series of moves but from a material standpoint there is sufficient material. I don't see any coding effort being made to fix that either, since the number of games they impact is extremely small.