Atomic chess engine

One more interesting thing about post-game analysis.
This game's analysis says that black loses in 10 after 1. Nf3 f6 2. e3 d6 3. Nd4 c6? 4. Nb5 cxb5 5. Qh5+ g6 6. Qb5+ Nc6 7. Qb6 Kf7 8. Qc7 Bd7?? (loses in 10 - I didn't see this in the game)
en.lichess.org/6FkDJ17c
And this game's analysis says that 8. ... Bd7 in the same position was the best move.
en.lichess.org/Vx3GLaS0/white#16

It's strange, isn't it?

@ixdon2: Since the analysis starts from the last move, it knows your winning moves after Bd7 in the first game, but doesn't in the second game. And since the search time is limited, Stockfish does not always find the best move.

Anyway, there apparently is a lot of room for improvements of Stockfish. I will have a look at its output in these positions to find out why it does not find the best moves and hopefully will come up with ideas to improve. Thanks for your reports.

@ubdip
Thank you. I don't mean that Stockfish must have a brilliant soultion for each position, I just give some feedback for a possible improvement.

I have one more question about stockfish analysis. When I look at Stockfish's suggested sequences, I often see repeated moves and their sense is really unclear for me.
For example: this game, move 19.
en.lichess.org/ZupBpByo/white#36
1. Nf3 f6 2. e3 e6 3. Nd4 c6 4. Nb5 cxb5 5. Qh5+ g6 6. Qb5 Nc6 7. Qb6 axb6 8. Bb5 Kf7 9. Bxd7 Rxa2 10. b4 h5 11. Bb2 Ke8 12. O-O Nh6 13. h3 Ng4 14. hxg4 h4 15. Ra1 h3 16. g4 h2+ 17. Kh1 Kd7 18. Ra8 b5 19. Ra7+

Stockfish says 19. Ra7+ is a blunder... hm. OK, let's have a look at the suggested sequence:
19. Rd8+ Kc7 20. Rc8+ Kd7 21. Rc7+ Kd8 22. Rc8+(?) Kd7 23. Rc7+(?) Kd8 24. c4 e5 25. Rc8+ Kd7

If 24. c4 is a good move, why not to play it after 21. Rc7+ (at the same position) instead of playing it after 23. Rc7+? Does it make sense to spend two moves without changing the position?

@ixdon2: Thanks. Actually, this is a very interesting report, since 2-fold repetitions like in the 19.Rd8+ variation should be evaluated as an immediate draw (0.00) internally to avoid them, so there probably is something wrong with hash keys. Edit: The problem I suspected does not seem to be the there, which is confusing.

@ubdip

And one more strange thing about analysis. Stockfish has a very strange endgame tactics for side which has 2 pawns for knight or bishop.

en.lichess.org/CHdCTpeQ/white
After move 26. Rxe2 the position is won for white but the position is estimated as a draw.
- white has two pawns for a knight (it means that white will promote a pawn to queen, sooner or later)
- there will be blocked pawns at f-h lines (it means that there will be no draw like KQvK withs kings.

But Stockfish says that move 26 is a blunder and position is +0.6, meaning it's a draw.

The position immediately turns from this +0.6 to +6.6 after "26. ... h5": Stockfish says it's a blunder (meaning that black could make a draw).

OK, let's have a look at the suggested sequence.
(26... Ne3 27. Kg1(?) Nd1 28. Kf1(?) Kd7 29. c4 f5 30. b4 Kc6 31. Ke2(??) Nc3+ 32. Kd3 Ne4 33. Ke3(?))

(?) marks useless king moves. White should push pawns and keep some blocked pawns at lines f-h to win.
(??) marks bad king move under a check Nc3+ - it doesn't change really the position, it just loses two moves for white.
But in suggested sequence by Stockfish white moves king around instead of pushing pawns, promoting a queen and checkmating black king. Why?

P. S. Sorry for such a long post, but I don't know how to make it shorter.

Thanks. Pawn and minor piece endgames should be rather easy to evaluate correctly, so I will have a look at it.

@ubdip

And one more thing about repetitions.
A suggested sequence in this game is even more strange: there's a 3-fold repetition, but engine doesn't notice that.

en.lichess.org/o5vxVM5T

8. Nxc7?? { (2.87 → -4.04) Blunder. Best move was e4. } (8. e4 f4(1) 9. Ng7+ Kf8 10. Ne6+ Ke8(2) 11. Ng7+ Kf8 12. Ne6+ Ke8(3) 13. a3 d5 14. Ng7+ Kf8)

After 12. Ke8 there is a draw due to 3-fold repetition, isn't it? Or maybe I misunderstand 3-fold repetition rule?

#47 I had to dig out the rule book for this one! According to rule 9.2.2, the answer is no. As for whether the rule makes sense, I don't know.
http://www.fide.com/fide/handbook.html?id=171&view=article

9.2
The game is drawn, upon a correct claim by a player having the move, when the same position for at least the third time (not necessarily by a repetition of moves):
a. is about to appear, if he first writes his move, which cannot be changed, on his scoresheet and declares to the arbiter his intention to make this move, or
b. has just appeared, and the player claiming the draw has the move.

Positions are considered the same if and only if the same player has the move, pieces of the same kind and colour occupy the same squares and the possible moves of all the pieces of both players are the same. Thus positions are not the same if:
1. at the start of the sequence a pawn could have been captured en passant.
2. a king or rook had castling rights, but forfeited these after moving. The castling rights are lost only after the king or rook is moved.

@Toadofsky
Thank you. I didn't dig it so deep.
As far as I understood it at lichess, both players have a button "clain draw", when current position is a 3-fold repetition (meaning exactly this position was before twice).

But it means that white could claim a draw instead of playing 13. a3... OK, never mind :)

#49 I'm being argumentative here because if I'm wrong, Stockfish has a bug I need to fix. I already fixed a similar bug in that "normal" Stockfish does not support 3-fold repetition (instead supporting 2-fold). That said...

According to rule 9.2.2, that position only occurred twice:
"Positions are not the same if a king or rook had castling rights, but forfeited these after moving. The castling rights are lost only after the king or rook is moved."

The positions after 8...f4 and 10...Ke8 are not the same position, because Black had castling rights in the first position (in theory -- but not in practice -- Black could have castled after 9. Ng7+ but chose not to).

So if after 13. a3 White claims a draw, his claim is invalid.