Atomic chess engine

I feel like positions like those could maybe even be adjudicated by lichess, since the only possible outcome is a draw anyway. I imagine that would be really difficult to do, though.

Why would Lichess allow this unlimited time? Is this not considered wasting its resources?

Actually, I've seen this before. In an engine test match I ran between Houdini 4 and Komodo 8 which I posted here:

en.lichess.org/qJHFx2CA#200
en.lichess.org/RyIjEJHc#200

Stockfish analysis ceases after move 100. So it does not surprise me that this has shown up elsewhere (Stockfish ceases play).

Both, avoiding the 50 moves rule and stopping after 150 moves, is expected and not specific to the Atomic engine. If an engine thinks it is ahead (no matter how much or whether it is right) it will always avoid draw by repetition or 50 move rule. This frequently happens in engine vs. engine matches. For the problem after 150 moves see lichess.org/forum/lichess-feedback/horde-ai-bug.

So both problems might be annoying, but are no bugs. Sorry, but as far as I know the only way to finish the game is to resign.

Well, from now I will not play unlimited games anymore.
en.lichess.org/VsMIsiGT

Two more strange things.
1. Position evaluation. 4 not-blocked pawns + bishop vs 4 non-blocked pawns: it's a draw if side without bishop will not blunder.
2. Final position. A cooperative checkmate was impossible but there was no automatic draw. Maybe it should be fixed?

1. So far we haven't implemented any atomic specific endgame knowledge, so the evaluation is not very credible in endgames.
2. It is difficult to automatically identify positions with non-trivial material configurations that are immediate draws.

@ixdon2 2 hours ago #26

" 1. Position evaluation. 4 not-blocked pawns + bishop vs 4 non-blocked pawns: it's a draw if side without bishop will not blunder. "

The knowledge I implemented to get a score close to zero is:
1. There are no passers
2. The pawns of the side without the bishop are on the squares whose colors are opposite to the color of the opponent's bishop.

Bigbang's analysis since depth 2 is already drawish.

8/1p1Bp3/8/p3k3/4P2p/7P/PP6/6K1 w - - 0 1

dep score time
21 +0.38 0:33.28 g1f2 e5d4 e4e5 d4e3 d7b5 b7b6 e5e6 e3d4 f2f3 d4e3 f3e4 e3d
20 +0.37 0:20.40 g1f2 e5d4 e4e5 d4e3 d7b5 b7b6 e5e6 e3d4 a2a4 d4d5 b2b4 d5d4
19 +0.37 0:12.36 g1f2 e5d4 e4e5 d4e3 d7b5 b7b6 e5e6 e3d4 a2a4 d4d5 b2b4
18 +0.37 0:09.06 g1f2 e5d4 e4e5 d4e3 d7b5 b7b6 e5e6 e3d4 a2a4 d4d5 b2b4 d5d4
17 +0.39 0:06.05 g1f2 e5d4 e4e5 d4e3 d7b5 b7b6 e5e6 e3d4 a2a4 d4d5 b2b4
16 +0.36 0:02.54 g1f2 e5d4 d7e6 b7b6 f2e3 d4d3 e4e5 d3e4 a2a4 e4d4 e6c4 d4d3
15 +0.35 0:01.60 g1f2 e5d4 d7e6 b7b6 f2e3 d4d3 e4e5 d3e4 e6d7 e4d4 b2b4 d4d3
14 +0.34 0:01.06 g1f2 e5d4 d7e6 b7b6 f2e3 d4d3 e4e5 d3e4 b2b4 a5b4 a2a4 e4d4
13 +0.35 0:00.61 g1f2 e5d4 d7e6 b7b6 f2e3 d4d3 e4e5 d3d4 b2b4 a5b4 a2a3 d4d3
12 +0.35 0:00.34 g1f2 e5d4 d7e6 b7b6 f2e3 d4d3 e4e5 d3d4 b2b4 a5b4 a2a3
11 +0.35 0:00.20 g1f2 e5d4 d7e6 b7b6 f2e3 d4d3 e4e5 d3d4 b2b4 d4d3 a2a3
10 +0.33 0:00.12 g1f2 e5d4 d7e6 b7b6 f2e3 b6b5 e4e5 b5b4 a2a3 b4a3
9 +0.33 0:00.08 g1f2 e5d4 d7e6 b7b6 f2e3 d4e5 a2a4 b6b5 b2b4 b5a4
8 +0.31 0:00.04 b2b4 b7b6 g1f2 e5d4 d7e6 d4e3 a2a4 e3d4
7 +0.32 0:00.01 g1f2 b7b6 d7e6 e5d4 f2e3 d4d3 e4e5
6 +0.29 0:00.00 g1f2 b7b6 d7e6 e5d4 f2e3 d4d3
5 +0.29 0:00.00 d7e6 b7b6 g1f2 e5d4 b2b4
4 +0.27 0:00.00 d7e6 b7b6 g1f2 e5d4
3 +0.28 0:00.00 d7e6 b7b6 g1f2
2 +0.24 0:00.00 d7e6 b7b6
1 +1.98 0:00.00 d7e6

@duropo thanks, I meant exactly this. I have explained it less accurately than you did.

So there is a question for devs, why is Stockfish's estimation non-drawish.

Wow. Now Stockfish resigns at move 151 instead of quitting. Thanks, Lichess!