Silly idiotic rule: K+N vs. K+p is 1-0 when timeout

If you manage to fix that problem, feel free to add it to: github.com/ornicar/lila

> So it narrows it down from a huge amount of positions to two. I'd call that more consistent.

or three? lichess.org/forum/general-chess-discussion/funny-draw-position



> If you manage to fix that problem, feel free to add it to:

It is fixable, but with too much effort I think. That is, to correctly implement FIDE rules one need a possibility to re-consider the result of a game by computer and human intervention. That is, full-blown arbitration after the end of a game.

In an ideal world the game would end when it is no longer possibility to deliver mate. But we cannot check every game with tablebase or an engine.

I think there may be an algorithm for most "fortress" positions (for lack of a better term):

1. Flood-fill whether either player can capture a piece (or a king); if so, it is very likely the attacker can checkmate
2. If flood-fill finds a capture, do a helpmate search

Separately I'm developing helpmate-search capability into Stockfish, although progress is slow:
github.com/ddugovic/Stockfish/tree/helpmate

@ In an ideal world the game would end when it is no longer possibility to deliver mate. But we cannot check every game with tablebase or an engine.

The position above was not considered to be draw by stockfish. (But it may be possible to be considered draw, because search space is very limited.)

It realizes its a draw almost instantly:

@Toadofsky

> Flood-fill whether either player can capture a piece; if so, it is very likely the attacker can checkmate

Flood-fill whether either player can capture a piece; if not so, try to exaust all the positions (there will be small number of such real game positions, like 64*64 if KPPPP+KPPPP or 64*64*64*64 KBPPPP+KBPPPP); if no mate, then draw.

But this did not prevent a flaw from the start of discussion: trying to win for KN side by cutting a flag of KP side in KN vs. KP.

Increment does prevent.
If the KN player times out that is not more just.
Likewise KN vs KN, KR vs KR and so on and so forth.

Whether that's a flaw is a highly subjective thing.

You agree to play with some time limit, and overstep it. What you claim would happen if you got to play the position out is irrelevant; you don't get to play it out.

If you want the "just" result for these sorts of trivial positions, then play a time control where you get to actually play it out.

There should be a penalty imposed for overstepping the agreed upon time limit; the fact that someone doesn't like the penalty because the result would be different if they didn't violate the constraint in the first place is a bit immaterial :)

People want to win on time but not to lose on time. Winning on time possible but with full-comprehensive insurance.

-If you play without increment: fine, you can win on time. Draw-back: losing is possible as well.
-If you play with increment: fine, you will not lose on time. But: Oops, you can't rescue lost positions over the time method.