Scorpio EGBB bug?

[ymatioun]

I have been using 5-piece Scorpio EGBB, and it works fine. But i just came across the following position:FEN=8/4kp2/8/5KP1/4P3/8/8/8 b - - 0 60. Scorpio (called through "probe_egbb_fen") calls it a draw. But this is an obvious loss for black - as confirmed by deep search by Stockfish and online version of Nalimov tablebases.

Looks like a bug to me. Does anybody else get the same result for this position?

Thanks, Youri.

[syzygy]

Did you try with other engines that use the Scorpio EGBB?

[Ferdy]

Did you try with other engines that use the Scorpio EGBB?

Just checking, looks right here, white is winning. However this does not use the probe_egbb_fen().
[d]8/4kp2/8/5KP1/4P3/8/8/8 b - - 0 60

Code: Select all

FEN: 8/4kp2/8/5KP1/4P3/8/8/8 b - - 0 60

Deuterium v14.3.34.130:
 1/3	00:00	 16	0	-1.30	Kd6
 2/4	00:00	 59	0	-1.29	Kd6 Kf4
 3/5	00:00	 210	0	-1.76	Kd8 Kf6 Ke8
[...]
 25/32	00:07	 10,943k	1,475k	-50.63	Ke8
 25/33	00:07	 11,126k	1,476k	-50.28	Ke8 Kf6 Kf8 e5 Ke8 Kg7 Ke7 Kg8 Ke8 e6 fxe6 g6 Ke7 Kh7 Kf6 g7 Kf5 g8Q Kf6 Qf8+ Ke5 Kg7 Kd5 Kg6 Kc6 Kf6 Kd7 Ke5
 26/34+	00:07	 11,220k	1,475k	-49.98	Ke8
 26/33	00:07	 11,492k	1,471k	-50.41	Ke8 Kf6 Kf8 e5 Kg8 Ke7 Kg7 Ke8 Kg8 g6 fxg6 e6 g5 e7 Kg7 Kd7 Kg6 e8Q+ Kf5 Qe3 Kg6 Ke6 g4 Qe5 Kh6 Kf6 Kh7 Kf5

[Ferdy]

Looks like a bug to me. Does anybody else get the same result for this position?

Thanks, Youri.

Implementing the probe_egbb_fen(), I got a draw score too.

[Daniel Shawul]

This is not a bug but a feature Enpassant squares are not encoded in egbbs. , neither is castling or other 1-in-a-million stuff. Hence, that is why the position is hand-picked by the OP.

[Dirt]

Enpassant squares are not encoded in egbbs.

How do en passant squares enter into this? I can't figure it out.

[ymatioun]

Is this because black pawn is on rank 7 and could be moved 2 squares, leading to possible EP move by a white pawn?

Youri.

[Ferdy]

I stucked with Arena during implementation for hours, but when I tried in console, it did not crashed. Then tried the other gui's. Got some interesting results. In item 2 Winboard and Aquarium are displaying drawish scores in less than 3 minutes. However both Fritz 13 and Scid were showing winning scores for white even after more than 10 minutes of infinite analysis time for each. Item 1 remarks, there is a bug in my implementation. What would be the possible consequences to this?

[Ferdy]

Everything works fine now. It has to do with memory issues during implementation of probe_egbb_fen(). And UCI_analyze mode where I do a different thing when this mode is set to on compared to play mode. Then there are gui's that sets UCI_Analyse to on in analysis mode, and there are gui's that just send go infinite, and not bother enabling the UCI_Analyse feature.