An easy position, for humans

[Rodolfo Leoni]

Here the fact is, engines will never consider Ke4-e5 as they'd start to see a drawish score. But if their search limites the King moves within the area they can intercept the black f4 pawn, score keeps high. Search only avoids the 3 fold repetition.

There should be possible to detect this repetitive score and to set it as a draw score.

In fact, position derives from analysis I made with the Deep Blue-Kasparov game, the game McBrain-Komodo Michael posted. This is the position:

[d]8/3r2p1/3P1p1p/1P2k3/1R2p3/2P4P/6P1/6K1 b - - 2 13

In game, Komodo played Kxd6, and white quickly won. The position I posted derives by Rxd6 which seemingly leads to draw.

[peter]

Hi Rodolfo!

I wonder if there's a way to detect it for having better evaluations...

Here is Deep Shredder 13 with 24 threads of a 12x2,67GHz Intel CPU and 32G hash started empty:

8/8/1P1k3p/3P2pP/4KpP1/8/8/8 w - - 0 1

Analysis by Deep Shredder 13 x64:

71.Kf3 Kd7 72.Ke2
+- (2.02 --) Depth: 7/7 00:00:00 1kN
...
71.Kd3 Kd7 72.Kd4 Kd6 73.Ke4 Kd7 74.Kd3 Kd6 75.Kd4 Kd7 76.Kc4 Kd6
+- (2.03 ++) Depth: 43/108 00:00:44 1874MN
...
71.Kd3 Kd7 72.Kd4 Kd8
+/- (1.32 ++) Depth: 46/108 00:01:17 3323MN
71.Kd3 Kd7 72.Kd4 Kd6 73.Kc3 Kd7 74.Kd3 Kd6 75.Ke4
+/- (0.98 --) Depth: 46/108 00:01:17 3333MN
...
71.Kd3 Kd7 72.Kc3 Kd6 73.Kc2 Kd7 74.Kd1 Kd6 75.Ke2 Kd7 76.Kf3 Kd6 77.Ke4 Kd7 78.Ke5 f3 79.b7 Kc7 80.d6+ Kxb7 81.Ke6 f2 82.d7 f1Q 83.d8Q Qf4 84.Qd5+ Kc7 85.Qd7+ Kb6 86.Ke7 Kc5 87.Qe6 Kb5 88.Qxh6 Qxg4 89.Qg6 Qe2+ 90.Qe6 Qxh5 91.Qe5+ Kc4 92.Qe3 Qg4 93.Qc1+ Kd3 94.Qa3+ Kd2 95.Qa2+ Kd3 96.Qb3+ Kd2 97.Qd5+ Ke3 98.Qc5+ Kf3 99.Qc3+ Kf2 100.Qc5+ Kg2 101.Kd8
= (0.06 --) Depth: 48/108 00:01:56 5179MN
...
71.Kd3 Kd7 72.Kc3 Kd6 73.Kc2 Kd7 74.Kd1 Kd6 75.Ke2 Kd7 76.Kf3 Kd6 77.Ke4 Kd7 78.Ke5 f3 79.b7 Kc7 80.d6+ Kxb7 81.Ke6 f2 82.d7 f1Q 83.d8Q Qf4 84.Qd5+ Kb6 85.Qf5 Kc6 86.Kf6 Kd6 87.Kg7 Ke7 88.Qc5+ Ke6 89.Qc6+ Ke7 90.Kxh6 Qxg4 91.Kg6 Qf4 92.h6 Qf7+ 93.Kxg5 Qg8+ 94.Kf4 Qf7+ 95.Ke3 Qb3+ 96.Ke4 Qf7 97.Qc5+ Ke8 98.Qe5+ Kf8 99.Qb8+ Ke7 100.Qb4+ Kf6 101.Qb6+ Kg5 102.Qd6
= (0.05 --) Depth: 52/111 00:03:07 8633MN
71.Kd3 Kd7 72.Kc3 Kd6 73.Kc2 Kd7 74.Kd1 Kd6 75.Ke2 Kd7 76.Kf3 Kd6 77.Ke4 Kd7 78.Ke5 f3 79.b7 Kc7 80.d6+ Kxb7 81.Ke6 f2 82.d7 f1Q 83.d8Q Qf4 84.Qe7+ Kb6 85.Qd6+ Kb5 86.Qe5+ Kb6 87.Qe2 Kc7 88.Qc2+ Kb8 89.Qf5 Kc7 90.Ke7 Kb6 91.Kf6 Kc7 92.Ke7
= (0.00) Depth: 52/111 00:03:41 10248MN
...
71.Kd3 Kd7 72.Kc3 Kd6 73.Kc2 Kd7 74.Kc3
= (0.00) Depth: 55/111 00:04:25 12276MN
71.Kd3 Kd7 72.Kc3 Kd6 73.Kc2 Kd7 74.Kc3
= (0.00) Depth: 56/111 00:04:32 12663MN

[Rodolfo Leoni]

Hi Rodolfo!

I wonder if there's a way to detect it for having better evaluations...

Here is Deep Shredder 13 with 24 threads of a 12x2,67GHz Intel CPU and 32G hash started empty:

8/8/1P1k3p/3P2pP/4KpP1/8/8/8 w - - 0 1

Analysis by Deep Shredder 13 x64:

71.Kf3 Kd7 72.Ke2
+- (2.02 --) Depth: 7/7 00:00:00 1kN
...
71.Kd3 Kd7 72.Kd4 Kd6 73.Ke4 Kd7 74.Kd3 Kd6 75.Kd4 Kd7 76.Kc4 Kd6
+- (2.03 ++) Depth: 43/108 00:00:44 1874MN
...
71.Kd3 Kd7 72.Kd4 Kd8
+/- (1.32 ++) Depth: 46/108 00:01:17 3323MN
71.Kd3 Kd7 72.Kd4 Kd6 73.Kc3 Kd7 74.Kd3 Kd6 75.Ke4
+/- (0.98 --) Depth: 46/108 00:01:17 3333MN
...
71.Kd3 Kd7 72.Kc3 Kd6 73.Kc2 Kd7 74.Kd1 Kd6 75.Ke2 Kd7 76.Kf3 Kd6 77.Ke4 Kd7 78.Ke5 f3 79.b7 Kc7 80.d6+ Kxb7 81.Ke6 f2 82.d7 f1Q 83.d8Q Qf4 84.Qd5+ Kc7 85.Qd7+ Kb6 86.Ke7 Kc5 87.Qe6 Kb5 88.Qxh6 Qxg4 89.Qg6 Qe2+ 90.Qe6 Qxh5 91.Qe5+ Kc4 92.Qe3 Qg4 93.Qc1+ Kd3 94.Qa3+ Kd2 95.Qa2+ Kd3 96.Qb3+ Kd2 97.Qd5+ Ke3 98.Qc5+ Kf3 99.Qc3+ Kf2 100.Qc5+ Kg2 101.Kd8
= (0.06 --) Depth: 48/108 00:01:56 5179MN
...
71.Kd3 Kd7 72.Kc3 Kd6 73.Kc2 Kd7 74.Kd1 Kd6 75.Ke2 Kd7 76.Kf3 Kd6 77.Ke4 Kd7 78.Ke5 f3 79.b7 Kc7 80.d6+ Kxb7 81.Ke6 f2 82.d7 f1Q 83.d8Q Qf4 84.Qd5+ Kb6 85.Qf5 Kc6 86.Kf6 Kd6 87.Kg7 Ke7 88.Qc5+ Ke6 89.Qc6+ Ke7 90.Kxh6 Qxg4 91.Kg6 Qf4 92.h6 Qf7+ 93.Kxg5 Qg8+ 94.Kf4 Qf7+ 95.Ke3 Qb3+ 96.Ke4 Qf7 97.Qc5+ Ke8 98.Qe5+ Kf8 99.Qb8+ Ke7 100.Qb4+ Kf6 101.Qb6+ Kg5 102.Qd6
= (0.05 --) Depth: 52/111 00:03:07 8633MN
71.Kd3 Kd7 72.Kc3 Kd6 73.Kc2 Kd7 74.Kd1 Kd6 75.Ke2 Kd7 76.Kf3 Kd6 77.Ke4 Kd7 78.Ke5 f3 79.b7 Kc7 80.d6+ Kxb7 81.Ke6 f2 82.d7 f1Q 83.d8Q Qf4 84.Qe7+ Kb6 85.Qd6+ Kb5 86.Qe5+ Kb6 87.Qe2 Kc7 88.Qc2+ Kb8 89.Qf5 Kc7 90.Ke7 Kb6 91.Kf6 Kc7 92.Ke7
= (0.00) Depth: 52/111 00:03:41 10248MN
...
71.Kd3 Kd7 72.Kc3 Kd6 73.Kc2 Kd7 74.Kc3
= (0.00) Depth: 55/111 00:04:25 12276MN
71.Kd3 Kd7 72.Kc3 Kd6 73.Kc2 Kd7 74.Kc3
= (0.00) Depth: 56/111 00:04:32 12663MN

Hi Peter.

It looks SMK found a way to get a better evaluation. I tried both McBrain and AsmFishW but none of them had a fail low in about 1 hour...

Rodolfo

[Lyudmil Tsvetkov]

W King is tied to black f4 pawn. B King will always move Kd7-d8-d7-d8.... A clear and easy draw. But engines cannot see it.

[d]8/8/1P1k3p/3P2pP/4KpP1/8/8/8 w - - 1 71

If black king always move Kd7-d8 then b7 wins.

practically 1.b7 Kc7 2.Kf5 and it is a draw but I do not think that it is simple for humans to be sure about it.

A type error, Uri...
Kd7-d6-d7-d6...

engines are nuts.

sometimes, even more than humans.

[Rodolfo Leoni]

engines are nuts.

sometimes, even more than humans.

Agreed.

[Nay Lin Tun]

Evaluations are just imaginary scores for engines to lead to the right path. I dont mind seeing an engine's scores of +9 or +3 or +1 in a draw position unless these evaluations lead the engine from the right to the wrong path. (I mean from win to draw or from draw to lose etc). In this position my sf without szygy did not do a mistake and did the right moves that lead to draw. So my sf did the perfect job. The thing is that sf moving white king from h1 , f3,d1 triangle is that all moves within these triangle lead to the same draw result.

How often engines do mistake in a standard 2 hour game in common 4 cores desktop? SF will do more than 99% accuracy, while super GM will do 95-98% accuracy at most. In fact there is no human being who can legitimately claim a win against stockfish in a standard setting. (2 hour game).There may be one in Million game where human might win. Do u still think who is better? Judge by your common sense.

In term of positional evaluations, there are lots of human mistakes in 500 ECO opening lines where 10-20 % of those ECO popular lines are unplayable and forced lose to one side(eg,some gambits variationns , some variations of King Indian defense and some variations of King Gambit accepted. ). Meanwhile , how often stockfish lead to wrong or inferior variation in opening ?Just show me that rarity. May be one in hundred game or one in thousand game, I wanna see that opening line. Same to middle games and endgames as well. In endgame , engines do more mistakes but still far more accurate statistically than GM.

Now these days, GM or IM commentators do not dare to comment on live game without computer assist as they know that they will likely make a funny joke if they express their own opinion.

[Rodolfo Leoni]

Hi Nay,

(hey, "Hi Nay" sounds nice!)

........... I dont mind seeing an engine's scores of +9 or +3 or +1 in a draw position unless these evaluations lead the engine from the right to the wrong path. (I mean from win to draw or from draw to lose etc).

I got that position by analyzing a McBrain_Dev-Komodo11.2 game:

[pgn][Event "Mac Pro x5690 3.46 Ghz 18 CPU"]
[Site "Mac-Pro.local"]
[Date "2017.07.20"]
[Round "-"]
[White "SF-McBrain 2.7 64 POPCNT"]
[Black "Komodo 11.2 64-bit"]
[Result "1-0"]
[TimeControl "360+6"]
[FEN "Rr6/5kp1/1qQb1p1p/1p1PpP2/1Pp1B3/2P4P/6P1/5K2 w - - 0 1"]
[SetUp "1"]


1. Ra6 Qe3 2. Qxd6 {+0.76/31 +347} Re8 {-0.77/26 +351} 3. Qd7+
{+1.16/35 +324} Re7 {-0.73/27 +350} 4. Qc6 {+1.20/32 +318} Qxe4
{-0.60/28 +338} 5. d6 {+1.08/29 +322} Qxf5+ {-0.68/31 +299} 6. Kg1
{+0.93/37 +306} Rd7 {-0.75/29 +298} 7. Qxb5 {+0.66/27 +310} Qe6
{-0.75/31 +298} 8. Qc5 {+0.90/34 +298} Kg6 {-0.46/30 +294} 9. Rc6
{+0.63/28 +302} e4 {-0.45/28 +293} 10. Qxc4 {+0.69/35 +296} Qxc4
{-0.55/29 +288} 11. Rxc4 {+0.93/31 +299} Kf5 {-0.43/31 +283} 12. b5
{+0.83/34 +289} Ke5 {-0.56/30 +283} 13. Rb4 {+0.89/35 +288} Kxd6
{-0.64/32 +274} 14. Rd4+ {+1.29/31 +288} Kc7 {-0.88/32 +269} 15. Rxe4
{+1.16/32 +291} Kb6 {-0.70/33 +259} 16. c4 {+0.97/41 +237} Kc5
{-0.69/31 +258} 17. Kh2 {+0.89/39 +212} Rf7 {-0.68/35 +235} 18. Rf4
{+0.52/47 +198} Re7 {-0.97/35 +229} 19. Kg3 {+1.52/34 +194} Re6
{-1.23/36 +201} 20. Kf3 {+1.88/34 +186} g6 {-0.75/34 +197} 21. h4
{+2.76/33 +185} h5 {-1.14/34 +197} 22. g4 {+4.23/31 +185} hxg4+
{-3.03/35 +192} 23. Kxg4 {+4.72/28 +189} Kd6 {-4.41/36 +188} 24. h5
{+5.91/31 +186} gxh5+ {-5.56/37 +176} 25. Kxh5 {+7.27/31 +186} Ke5
{-6.34/38 +154} 26. Kg4 {+9.92/30 +176} Rb6 {-6.57/34 +156} 27. Kf3
{+10.51/30 +171} f5 {-6.64/35 +157} 28. Rh4 {+14.22/29 +160} Rb8
{-6.85/34 +155} 29. Ke2 {+15.14/27 +162} Rg8 {-6.80/33 +155} 30. Rh7
{+16.83/29 +162} Kd4 {-250.00/36 +142} 31. Rc7 {+107.91/30 +152} Rg2+
{-250.00/43 +147} 32. Kf3 {+3.19/29 +156} Rb2 {-250.00/42 +152} 33. Rc6
{+128.23/34 +149} Ke5 {-250.00/32 +156} 34. b6 {+128.28/30 +149}
{User adjudication} 1-0[/pgn]

I was investigating why Komodo played the weak move 13. ..Kxd6 instead of 13. ..Rxd6. It's possibly due to the fact Komodo search found the position I posted as alternative main line, and it scored wrong. I don't remember the move sequence which lead to that position, but Komodo probably discarded a draw line because of that wrong evaluation.

How often engines do mistake in a standard 2 hour game in common 4 cores desktop? SF will do more than 99% accuracy, while super GM will do 95-98% accuracy at most. In fact there is no human being who can legitimately claim a win against stockfish in a standard setting. (2 hour game).There may be one in Million game where human might win. Do u still think who is better? Judge by your common sense.

In term of positional evaluations, there are lots of human mistakes in 500 ECO opening lines where 10-20 % of those ECO popular lines are unplayable and forced lose to one side(eg,some gambits variationns , some variations of King Indian defense and some variations of King Gambit accepted. ). Meanwhile , how often stockfish lead to wrong or inferior variation in opening ?Just show me that rarity. May be one in hundred game or one in thousand game, I wanna see that opening line. Same to middle games and endgames as well. In endgame , engines do more mistakes but still far more accurate statistically than GM.

Now these days, GM or IM commentators do not dare to comment on live game without computer assist as they know that they will likely make a funny joke if they express their own opinion.

Don't mind at Lyudmil's provocation. He intended nuts are better than me, and I returned him the comment...

Engines can do much less things than humans, but what they can, they do much better.

Regards,
Rodolfo

[Lyudmil Tsvetkov]

Evaluations are just imaginary scores for engines to lead to the right path. I dont mind seeing an engine's scores of +9 or +3 or +1 in a draw position unless these evaluations lead the engine from the right to the wrong path. (I mean from win to draw or from draw to lose etc). In this position my sf without szygy did not do a mistake and did the right moves that lead to draw. So my sf did the perfect job. The thing is that sf moving white king from h1 , f3,d1 triangle is that all moves within these triangle lead to the same draw result.

How often engines do mistake in a standard 2 hour game in common 4 cores desktop? SF will do more than 99% accuracy, while super GM will do 95-98% accuracy at most. In fact there is no human being who can legitimately claim a win against stockfish in a standard setting. (2 hour game).There may be one in Million game where human might win. Do u still think who is better? Judge by your common sense.

In term of positional evaluations, there are lots of human mistakes in 500 ECO opening lines where 10-20 % of those ECO popular lines are unplayable and forced lose to one side(eg,some gambits variationns , some variations of King Indian defense and some variations of King Gambit accepted. ). Meanwhile , how often stockfish lead to wrong or inferior variation in opening ?Just show me that rarity. May be one in hundred game or one in thousand game, I wanna see that opening line. Same to middle games and endgames as well. In endgame , engines do more mistakes but still far more accurate statistically than GM.

Now these days, GM or IM commentators do not dare to comment on live game without computer assist as they know that they will likely make a funny joke if they express their own opinion.

don't listen to Rodolfino, there is no provocation on my side whatsoever.

yes, evals are imaginary scores, but without those scores, an engine can not play at all!
no matter how advanced its search is.

claiming that engines play openings better than top humans is very unrealistic, of course.

in terms of opening play, SF is somewhere the level of a 1800 elo player, not more, sometimes even 1500.

actually, I can not think of a position, where SF would pick the right opening move, maybe there are couple of those, but I still can not promptly think of one.

if you know one, please post.

[Lyudmil Tsvetkov]

SF:

[d]rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1

does not find the best move for white

[d]rnbqkbnr/pppppppp/8/8/4P3/8/PPPP1PPP/RNBQKBNR b KQkq e3 0 1

does not find the best black reply either

[d]rnbqkbnr/pppppppp/8/8/3P4/8/PPP1PPPP/RNBQKBNR b KQkq d3 0 1

does not find the best move here too

[d]rnbqkbnr/pppppppp/8/8/2P5/8/PP1PPPPP/RNBQKBNR b KQkq c3 0 1

does not find it here too

so, 4 openings positions, 4 separate tries, and SF fails in all 4 to pick the best move.

that is more than sufficient for me, 100% failure rate, it can hardly get any worse.

[Spliffjiffer]

maybe ure not from this world but id suppose im not the first one taking this into account