This one I can understand, but still insane, and both Stockfish and Leila had trouble.
Studies to Troll your engines with
Sort:
Apr 9, 2023
0
#63
EndgameStudier wrote:
This one I can understand, but still insane, and both Stockfish and Leila had trouble.
This one by C. Behting is a well known classic and it certainly did give Stockfish trouble in the past, however Stockfish is able to solve it now after all those years of not solving it 
.
Here is Stockfish development version dated April 5, 2023 running on 18 cores with an 8GB hash table solving it in 1 minute 26 seconds:
FEN: n2Bqk2/5p1p/Q4KP1/p7/8/8/8/8 w - - 0 1
Stockfish_23040506_x64_bmi2:
NNUE evaluation using nn-dabb1ed23026.nnue enabled
71/37 01:26 2,656,027k 30,632k +M13 1.Qa6-c8 Kf8-g8 2.Bd8-c7 Qe8xc8 3.g6xf7+ Kg8-h8 4.Bc7-e5 Qc8-c5 5.Be5-b2 Na8-c7 6.Bb2-a1 a5-a4 7.Ba1-b2 a4-a3 8.Bb2-a1 a3-a2 9.Ba1-b2 a2-a1Q 10.Bb2xa1 Qc5-e5+ 11.Ba1xe5 Nc7-d5+ 12.Kf6-e6+ Nd5-f6 13.Be5xf6+
I screwed up the notation on that one, mobile devices have trouble with inserting variations. But yeah, stockfish actually does not solve that one after a while at high depth.
Apr 9, 2023
0
#65
EndgameStudier wrote:
I screwed up the notation on that one, mobile devices have trouble with inserting variations. But yeah, stockfish actually does not solve that one after a while at high depth.
I'm letting it think on the other 2 right now, those are classics also that I haven't seen for a long time.
Many of these I got from the "Chess With Suren" channel on youtube. But yeah a couple of them are well known. There are also many variations of that Chinese wall problem with locked pawns.
Apr 9, 2023
0
#67
It seems to have the first move correct on this one, but it's a matter of resolving the iteration...it's taking forever.
Here is what it has so far:
Stockfish_23040506_x64_bmi2:
NNUE evaluation using nn-dabb1ed23026.nnue enabled
67/86 00:09 331,037k 34,638k +7.66 1.Kc1-d1 Kb5-c6 2.Kd1-e1 Kc6-d7 3.Rg1-g2 Kd7-e6 4.Rh1-g1 Ke6-f6 5.Rg2-e2 Kf6-f7 6.Ke1-f2 Kf7-e7 7.Re2-e1 Ke7-f7 8.Re1-c1 Kf7-g6 9.Kf2-e1 Kg6-f7 10.Rg1-g2 Kf7-f8 11.Rg2-f2 Kf8-f7 12.Rf2-e2 Kf7-e7 13.Ke1-d1 Ke7-d6 14.Re2-e1 Kd6-c6 15.Rc1-c2 Kc6-b7 16.Kd1-c1 Kb7-a6 17.Qa1-a2 b3xa2 18.b2-b4 a2-a1Q 19.Rc2-b2 Qa1xb2+ 20.Kc1xb2 Ka6-b5 21.Re1-d1 Kb5-c6 22.b4-b5+ Kc6xb5 23.Rd1-c1 Kb5-c5 24.Rc1-c2 d3xc2 25.Kb2xc2 Kc5-d6 26.d2-d3 c4xd3+ 27.Kc2-b2 d3-d2 28.Nb1xd2 Kd6-c6 29.Kb2-c2 Kc6-d5 30.c3-c4+ Kd5-c5 31.Kc2-b2 Kc5-b6 32.Kb2-c3 Kb6-c7 33.Kc3-d4 Kc7-d6 34.c4-c5+ Kd6-c6 35.Kd4-c4 Kc6-b7 36.Kc4-d5 Kb7-a6 37.Kd5-e5 Ka6-b5
68/68+ 00:17 601,803k 35,220k +8.87 1.Kc1-d1
68/80+ 00:31 1,130,824k 35,588k +10.24 1.Kc1-d1
68/82+ 12:15 26,507,380k 36,017k +11.95 1.Kc1-d1
It looks like the variation at the 9 second mark is showing the correct idea of bringing the rooks over to the Queen side and sacrificing the Queen at a2 and allowing black to queen the pawn at a1, I think it will solve this one but I just want to make sure. If you were interested here is where that study is from: https://www.yacpdb.org/#search/ODg4MWszcDJwMXAxcFBwMVBwUHBQMVBwMVAxUDNQUU5LMk5SUi8vLy8vLy8vLy8vLy8vMS8xLzEvMA==/1
Yep it did end up solving this one:
Here's a forced mate in 67 that Stockfish couldn't solve. And the concept is actually easy to understand, it just takes alot of moves:
1. Achieve the same position with black to move so he has to advance a pawn (avoid knight checks along the route)
2. Repeat Process 5 times until black has no more pawn moves.
3. Checkmate black depending on which knight he moves.
Apr 9, 2023
0
#69
EndgameStudier wrote:
Here's a forced mate in 67 that Stockfish couldn't solve. And the concept is actually easy to understand, it just takes alot of moves:
1. Achieve the same position with black to move so he has to advance a pawn (avoid knight checks along the route)
2. Repeat Process 5 times until black has no more pawn moves.
3. Checkmate black depending on which knight he moves.
Good one, the old "lose the move" technique, I may have an engine that solves this, but it's not Stockfish, it's called "Slowchess 2.9 blitz", if you are interested in this engine you can get it here:
https://3dkingdoms.com/chess/slow.htm
Apr 9, 2023
0
#70
EndgameStudier wrote:
Here's a forced mate in 67 that Stockfish couldn't solve. And the concept is actually easy to understand, it just takes alot of moves:
1. Achieve the same position with black to move so he has to advance a pawn (avoid knight checks along the route)
2. Repeat Process 5 times until black has no more pawn moves.
3. Checkmate black depending on which knight he moves.
Ok, neither Stockfish nor Slowchess solved that one, BUT Crystal 5 KWK (a Stockfish derivative that modified the Stockfish code to specialize in solving studies) solves this one rather quickly (about 5-6 seconds):
here is the analysis:
FEN: 8/8/1K2N3/3N4/p1p5/rpk1p3/bp1np3/qrb1N3 b - - 0 67
Crystal 5 KWK avx2:
NNUE evaluation using nn-ad9b42354671.nnue enabled
39/141- 00:05 129,911k 22,203k +9.50 1.Kb5-b6 Ra3-a4
39/141+ 00:05 129,981k 22,208k +M123 1.Kb5-b6
39/141+ 00:06 145,232k 22,735k +M67 1.Kb5-b6
39/141 00:06 151,013k 22,975k +M67 1.Kb5-b6 Ra3-a4 2.Kb6-b7 Ra4-a3 3.Kb7-c8 Ra3-a4 4.Kc8-d7 Ra4-a3 5.Kd7-e8 Ra3-a4 6.Ke8-f7 Ra4-a3 7.Kf7-f8 Ra3-a4 8.Kf8-e8 Ra4-a3 9.Ke8-d7 Ra3-a4 10.Kd7-c8 Ra4-a3 11.Kc8-b7 Ra3-a4 12.Kb7-b6 Ra4-a3 13.Kb6-b5 e4-e3 14.Kb5-b6 Ra3-a4 15.Kb6-b7 Ra4-a3 16.Kb7-c8 Ra3-a4 17.Kc8-d7 Ra4-a3 18.Kd7-e8 Ra3-a4 19.Ke8-f8 Ra4-a3 20.Kf8-f7 Ra3-a4 21.Kf7-e8 Ra4-a3 22.Ke8-d7 Ra3-a4 23.Kd7-c8 Ra4-a3 24.Kc8-b7 Ra3-a4 25.Kb7-b6 Ra4-a3 26.Kb6-b5 e3-e2 27.Kb5-b6 Ra3-a4 28.Kb6-b7 Ra4-a3 29.Kb7-c8 Ra3-a4 30.Kc8-d7 Ra4-a3 31.Kd7-e8 Ra3-a4 32.Ke8-f7 Ra4-a3 33.Kf7-f8 Ra3-a4 34.Kf8-e8 Ra4-a3 35.Ke8-d7 Ra3-a4 36.Kd7-c8 Ra4-a3 37.Kc8-b7 Ra3-a4 38.Kb7-b6 Ra4-a3 39.Kb6-b5 e5-e4 40.Kb5-b6 Ra3-a4 41.Kb6-b7 Ra4-a3 42.Kb7-c8 Ra3-a4 43.Kc8-d7 Ra4-a3 44.Kd7-e8 Ra3-a4 45.Ke8-f7 Ra4-a3 46.Kf7-f8 Ra3-a4 47.Kf8-e8 Ra4-a3 48.Ke8-d7 Ra3-a4 49.Kd7-c8 Ra4-a3 50.Kc8-b7 Ra3-a4 51.Kb7-b6 Ra4-a3 52.Kb6-b5 e4-e3 53.Kb5-b6 Ra3-a4 54.Kb6-b7 Ra4-a3 55.Kb7-c8 Ra3-a4 56.Kc8-d7 Ra4-a3 57.Kd7-e8 Ra3-a4 58.Ke8-f7 Ra4-a3 59.Kf7-f8 Ra3-a4 60.Kf8-e8 Ra4-a3 61.Ke8-d7 Ra3-a4 62.Kd7-c8 Ra4-a3 63.Kc8-b7 Ra3-a4 64.Kb7-b6 Ra4-a3 65.Kb6-b5 a5-a4 66.Kb5-b6 Nb4-d5+ 67.Nf6xd5+
40/133 00:07 175,498k 23,995k +M67 1.Kb5-b6 Ra3-a4 2.Kb6-b7 Ra4-a3 3.Kb7-c8 Ra3-a4 4.Kc8-d7 Ra4-a3 5.Kd7-e8 Ra3-a4 6.Ke8-f7 Ra4-a3 7.Kf7-f8 Ra3-a4 8.Kf8-e8 Ra4-a3 9.Ke8-d7 Ra3-a4 10.Kd7-c8 Ra4-a3 11.Kc8-b7 Ra3-a4 12.Kb7-b6 Ra4-a3 13.Kb6-b5 e4-e3 14.Kb5-b6 Ra3-a4 15.Kb6-b7 Ra4-a3 16.Kb7-c8 Ra3-a4 17.Kc8-d7 Ra4-a3 18.Kd7-e8 Ra3-a4 19.Ke8-f7 Ra4-a3 20.Kf7-f8 Ra3-a4 21.Kf8-e8 Ra4-a3 22.Ke8-d7 Ra3-a4 23.Kd7-c8 Ra4-a3 24.Kc8-b7 Ra3-a4 25.Kb7-b6 Ra4-a3 26.Kb6-b5 e3-e2 27.Kb5-b6 Ra3-a4 28.Kb6-b7 Ra4-a3 29.Kb7-c8 Ra3-a4 30.Kc8-d7 Ra4-a3 31.Kd7-e8 Ra3-a4 32.Ke8-f8 Ra4-a3 33.Kf8-f7 Ra3-a4 34.Kf7-e8 Ra4-a3 35.Ke8-d7 Ra3-a4 36.Kd7-c8 Ra4-a3 37.Kc8-b7 Ra3-a4 38.Kb7-b6 Ra4-a3 39.Kb6-b5 e5-e4 40.Kb5-b6 Ra3-a4 41.Kb6-b7 Ra4-a3 42.Kb7-c8 Ra3-a4 43.Kc8-d7 Ra4-a3 44.Kd7-e8 Ra3-a4 45.Ke8-f7 Ra4-a3 46.Kf7-f8 Ra3-a4 47.Kf8-e8 Ra4-a3 48.Ke8-d7 Ra3-a4 49.Kd7-c8 Ra4-a3 50.Kc8-b7 Ra3-a4 51.Kb7-b6 Ra4-a3 52.Kb6-b5 e4-e3 53.Kb5-b6 Ra3-a4 54.Kb6-b7 Ra4-a3 55.Kb7-c8 Ra3-a4 56.Kc8-d7 Ra4-a3 57.Kd7-e8 Ra3-a4 58.Ke8-f7 Ra4-a3 59.Kf7-f8 Ra3-a4 60.Kf8-e8 Ra4-a3 61.Ke8-d7 Ra3-a4 62.Kd7-c8 Ra4-a3 63.Kc8-b7 Ra3-a4 64.Kb7-b6 Ra4-a3 65.Kb6-b5 a5-a4 66.Kb5-b6 Nb4-d5+ 67.Nf6xd5+
41/133 00:07 178,461k 24,107k +M67 1.Kb5-b6 Ra3-a4 2.Kb6-b7 Ra4-a3 3.Kb7-c8 Ra3-a4 4.Kc8-d7 Ra4-a3 5.Kd7-e8 Ra3-a4 6.Ke8-f7 Ra4-a3 7.Kf7-f8 Ra3-a4 8.Kf8-e8 Ra4-a3 9.Ke8-d7 Ra3-a4 10.Kd7-c8 Ra4-a3 11.Kc8-b7 Ra3-a4 12.Kb7-b6 Ra4-a3 13.Kb6-b5 e4-e3 14.Kb5-b6 Ra3-a4 15.Kb6-b7 Ra4-a3 16.Kb7-c8 Ra3-a4 17.Kc8-d7 Ra4-a3 18.Kd7-e8 Ra3-a4 19.Ke8-f7 Ra4-a3 20.Kf7-f8 Ra3-a4 21.Kf8-e8 Ra4-a3 22.Ke8-d7 Ra3-a4 23.Kd7-c8 Ra4-a3 24.Kc8-b7 Ra3-a4 25.Kb7-b6 Ra4-a3 26.Kb6-b5 e3-e2 27.Kb5-b6 Ra3-a4 28.Kb6-b7 Ra4-a3 29.Kb7-c8 Ra3-a4 30.Kc8-d7 Ra4-a3 31.Kd7-e8 Ra3-a4 32.Ke8-f8 Ra4-a3 33.Kf8-f7 Ra3-a4 34.Kf7-e8 Ra4-a3 35.Ke8-d7 Ra3-a4 36.Kd7-c8 Ra4-a3 37.Kc8-b7 Ra3-a4 38.Kb7-b6 Ra4-a3 39.Kb6-b5 e5-e4 40.Kb5-b6 Ra3-a4 41.Kb6-b7 Ra4-a3 42.Kb7-c8 Ra3-a4 43.Kc8-d7 Ra4-a3 44.Kd7-e8 Ra3-a4 45.Ke8-f7 Ra4-a3 46.Kf7-f8 Ra3-a4 47.Kf8-e8 Ra4-a3 48.Ke8-d7 Ra3-a4 49.Kd7-c8 Ra4-a3 50.Kc8-b7 Ra3-a4 51.Kb7-b6 Ra4-a3 52.Kb6-b5 e4-e3 53.Kb5-b6 Ra3-a4 54.Kb6-b7 Ra4-a3 55.Kb7-c8 Ra3-a4 56.Kc8-d7 Ra4-a3 57.Kd7-e8 Ra3-a4 58.Ke8-f7 Ra4-a3 59.Kf7-f8 Ra3-a4 60.Kf8-e8 Ra4-a3 61.Ke8-d7 Ra3-a4 62.Kd7-c8 Ra4-a3 63.Kc8-b7 Ra3-a4 64.Kb7-b6 Ra4-a3 65.Kb6-b5 a5-a4 66.Kb5-b6 Nb4-d5+ 67.Nf6xd5+
If you are interested in getting the free Crystal 5 KWK UCI chess engine you can download it from this page:
https://ipmanchess.yolasite.com/compiles.php you'll have to scroll down a little bit.
drdos7 wrote:
Arisktotle wrote:
And how does that explain that chess.coms engines solve the 5-mover relatively fast on my PC?
It doesn't, but if you had a local version installed it would solve it faster because you could use more than 1 core. The only reason it "solved" the mate in 5 is because it was searching 3 lines and happened to stumble upon the solution in one of the lines. If you had the local version you could run up to 24 lines at the same time.
That would confirm there is indeed a chance element in the search in the same way there is for humans when they assign priorities to search candidates and get "lucky". This can be found out though. When we try solving many of the engine trolling positions we can't always be lucky, so the solving times should vary a lot even among those in the same class of difficulty and between different engines and even between different runs of the same engine. That's what "chance" can do to us!
I don't know about the 3 lines. I usually set my display for 3 lines - though 1 is enough to get a solution which doesn't guarantees there is just 1 solution. The way I interpret it is that the number of lines only affects the display - there is always a top 3 intermediate score list - and not how many lines are actually searched by the engine. This can be witnessed in the occasional changes in the top 3 candidates list during the analysis. The search is apparently broader and candidates come and go in the process.
But basically the whole "game engine" philosophy fails for compositions in general. All problem and endgame solutions are digital - you either found the solution or you didn't - and there is no continuum of scores which are more or less promising as roads to success. Only when engines are perfect relative to the complexity of the challenge, their analyses will meet the composition requirement for solution perfection.
What surprises me so far is that Komodo is doing better than the higher rated StockFish engines. Which makes me reconsider which engine to use when testing compositions on chess.com! But I'll first do some more testing of the engine trollers!
Apr 10, 2023
0
#72
Looks like Stockfish also solve the other "chinese wall" study:
These type of positions used to be impossible for Stockfish to solve, but apparently it has evolved enough to solve them with the exception of the mate in 67.
Here is the final analysis after letting it think overnight while I slept:
Stockfish_23040506_x64_bmi2:
NNUE evaluation using nn-dabb1ed23026.nnue enabled
83/85 7:55:44 2,100,719,449k 73,595k +200.00 1.Kc1-d2 Kd7-e6 2.Na2-c1 Ke6-f6 3.Nc1-e2 Kf6-e6 4.Kd2-c2 Ke6-d7 5.Ne2-g1 Kd7-e6 6.Ng1-h3 Ke6-f6 7.Nh3-f2 Kf6-e6 8.Kc2-d2 Ke6-d7 9.Kd2-e2 Kd7-e7 10.Nf2-d1 Ke7-d7 11.Nd1-b2 Kd7-c6 12.Nb2-a4 Kc6-d6 13.Ke2-d1 Kd6-c6 14.Kd1-c2 Kc6-d6 15.Na4-b6 Kd6-e6 16.Nb6-d5 Ke6-f7 17.Kc2-d2 Kf7-e6 18.Kd2-d1 Ke6-f7 19.Nd5-c7 Kf7-e7 20.Nc7-b5 Ke7-d7 21.Nb5-a3 Kd7-d8 22.Na3-c2 Kd8-c7 23.Kd1-e2 Kc7-d7 24.Ke2-f2 Kd7-d6 25.Nc2-e1 Kd6-e7 26.Kf2-g2 Ke7-d7 27.Kg2-h3 Kd7-d8 28.Ne1-g2 Kd8-e7 29.Ng2-h4 g5xh4 30.Kh3xh4 Ke7-f6 31.g4-g5+ Kf6-f7 32.Kh4-g4 Kf7-g7 33.Kg4-f5 h6xg5 34.Kf5xg5 Kg7-h7 35.h5-h6 Kh7-h8 36.Kg5-f5 Kh8-h7 37.Kf5xe5 Kh7-g6 38.h6-h7 Kg6xh7 39.Ke5-f5 Kh7-h6 40.Kf5xf4 Kh6-h7 41.e4-e5 Kh7-g6 42.Kf4-e4
Apr 10, 2023
0
#73
Arisktotle wrote:
drdos7 wrote:
Arisktotle wrote:
And how does that explain that chess.coms engines solve the 5-mover relatively fast on my PC?
It doesn't, but if you had a local version installed it would solve it faster because you could use more than 1 core. The only reason it "solved" the mate in 5 is because it was searching 3 lines and happened to stumble upon the solution in one of the lines. If you had the local version you could run up to 24 lines at the same time.
That would confirm there is indeed a chance element in the search in the same way there is for humans when they assign priorities to search candidates and get "lucky". This can be found out though. When we try solving many of the engine trolling positions we can't always be lucky, so the solving times should vary a lot even among those in the same class of difficulty and between different engines and even between different runs of the same engine. That's what "chance" can do to us!
I don't know about the 3 lines. I usually set my display for 3 lines - though 1 is enough to get a solution which doesn't guarantees there is just 1 solution. The way I interpret it is that the number of lines only affects the display - there is always a top 3 intermediate score list - and not how many lines are actually searched by the engine. This can be witnessed in the occasional changes in the top 3 candidates list during the analysis. The search is apparently broader and candidates come and go in the process.
But basically the whole "game engine" philosophy fails for compositions in general. All problem and endgame solutions are digital - you either found the solution or you didn't - and there is no continuum of scores which are more or less promising as roads to success. Only when engines are perfect relative to the complexity of the challenge, their analyses will meet the composition requirement for solution perfection.
What surprises me so far is that Komodo is doing better than the higher rated StockFish engines. Which makes me reconsider which engine to use when testing compositions on chess.com! But I'll first do some more testing of the engine trollers!
Check out what the modified version of Stockfish designed specifically for studies and mates (named Crystal 5 KWK) does with that mate in 5:
This is searching only 1 line and it finds the solution in not even 1 second
FEN: 8/2Nb4/pp6/4rp1p/1Pp1pPkP/PpPpR3/1B1P2N1/1K6 w - - 0 1
Crystal 5 KWK avx2:
NNUE evaluation using nn-ad9b42354671.nnue enabled
1/1 00:00 670 335k +5.82 1.f4xe5
2/2 00:00 1k 687k +7.04 1.f4xe5 f5-f4
3/3 00:00 8k 2,785k +7.12 1.f4xe5 f5-f4 2.e5-e6
4/4 00:00 16k 5,219k +7.60 1.f4xe5 f5-f4 2.e5-e6 f4xe3 3.Ng2xe3+ Kg4xh4 4.e6xd7
5/5 00:00 26k 6,543k +8.00 1.f4xe5 f5-f4 2.e5-e6 f4xe3 3.Ng2xe3+ Kg4xh4 4.e6xd7
6/6 00:00 36k 9,067k +8.37 1.f4xe5 b6-b5 2.e5-e6 Bd7-c6 3.e6-e7 f5-f4
7/8 00:00 57k 9,493k +8.85 1.f4xe5 b6-b5 2.Nc7-d5 f5-f4 3.Nd5-f6+ Kg4-f5
8/9 00:00 104k 12,970k +9.37 1.f4xe5 f5-f4 2.Ng2xf4 Kg4xf4 3.e5-e6 Bd7-c6 4.e6-e7 b6-b5
9/11 00:00 141k 14,148k +9.95 1.f4xe5 f5-f4 2.Ng2xf4 Kg4xf4 3.e5-e6 Bd7-c6 4.e6-e7 Bc6-d7 5.e7-e8R Bd7xe8 6.Nc7xe8
10/19 00:00 338k 16,910k +11.65 1.f4xe5 Bd7-a4 2.Nc7-d5 f5-f4 3.Nd5-f6+ Kg4-f5 4.Re3xe4 Ba4-c6 5.Re4xf4+ Kf5xe5 6.Nf6xh5
11/19 00:00 589k 18,419k +11.75 1.f4xe5 f5-f4 2.Ng2xf4 Kg4xf4 3.e5-e6 Bd7-c6 4.e6-e7 Kf4-e5 5.e7-e8Q+ Bc6xe8 6.Nc7xe8 b6-b5 7.Ne8-g7 Ke5-f4 8.Ng7xh5+ Kf4-g4
12/26 00:00 2,698k 29,976k +M5 1.Kb1-c1 Re5-b5 2.Kc1-d1 Rb5-a5 3.Kd1-e1 b6-b5 4.Ke1-f2 Ra5-a4 5.Re3-g3+
13/9 00:00 3,683k 34,421k +M5 1.Kb1-c1 Re5-b5 2.Kc1-d1 Rb5-a5 3.Kd1-e1 b6-b5 4.Ke1-f2 Ra5-a4 5.Re3-g3+
AND here is the same engine with 10 lines enabled, it finds 10 different mates in 1 second 
:
FEN: 8/2Nb4/pp6/4rp1p/1Pp1pPkP/PpPpR3/1B1P2N1/1K6 w - - 0 1
Crystal 5 KWK avx2:
11/14 00:00 27,357k 31,885k +11.62 1.f4xe5 Bd7-c6 2.e5-e6 f5-f4 3.Ng2xf4 Kg4xf4 4.e6-e7 b6-b5 5.e7-e8Q Bc6xe8 6.Nc7xe8
11/16 00:00 27,357k 31,885k +12.15 1.a3-a4 a6-a5 2.f4xe5 Bd7-c6 3.b4-b5 f5-f4 4.Ng2xf4 Kg4xf4 5.e5-e6 Kf4-f5 6.b5xc6
11/15 00:00 27,357k 31,885k +M9 1.Kb1-a1 Bd7-e8 2.Ka1-b1 a6-a5 3.Kb1-c1 a5xb4 4.a3xb4 Be8-f7 5.Kc1-d1 Re5-a5 6.b4xa5 b6-b5 7.Kd1-e1 b5-b4 8.Ke1-f2
11/15 00:00 27,357k 31,885k +M9 1.Nc7-a8 a6-a5 2.Na8-c7 Re5-e8 3.Kb1-c1 a5xb4 4.a3xb4 Bd7-b5 5.Kc1-d1 Re8-a8 6.Nc7xa8 Bb5-a6 7.Kd1-e1 b6-b5 8.Ke1-f2
11/15 00:00 27,357k 31,885k +M8 1.b4-b5 a6xb5 2.Kb1-c1 b5-b4 3.a3xb4 Re5-b5 4.Kc1-d1 Rb5-a5 5.b4xa5 Bd7-c6 6.Kd1-e1 b6-b5 7.Ke1-f2 b5-b4 8.Re3-g3+
11/13 00:00 27,357k 31,885k +M7 1.Nc7-b5 a6-a5 2.Kb1-c1 a5xb4 3.Kc1-d1 b4xc3 4.Nb5xc3 Re5-e7 5.Kd1-e1 Re7-h7 6.Ke1-f2 b6-b5 7.Re3-g3+
11/13 00:00 27,357k 31,885k +M7 1.Nc7-e8 Bd7xe8 2.Kb1-c1 Re5-a5 3.Kc1-d1 Be8-a4 4.b4xa5 Ba4-e8 5.Kd1-e1 b6xa5 6.Ke1-f2 Be8-b5 7.Re3-g3+
11/13 00:00 27,357k 31,885k +M7 1.Nc7xa6 Re5-d5 2.Kb1-c1 Rd5-a5 3.b4xa5 Bd7-c6 4.Kc1-d1 b6-b5 5.Kd1-e1 b5-b4 6.Ke1-f2 b4xc3 7.Re3-g3+
11/13 00:00 27,357k 31,885k +M7 1.Nc7-d5 Re5xd5 2.Kb1-c1 Bd7-a4 3.Kc1-d1 Rd5-a5 4.b4xa5 Ba4-e8 5.Kd1-e1 b6xa5 6.Ke1-f2 Be8-b5 7.Re3-g3+
11/9 00:00 27,357k 31,885k +M5 1.Kb1-c1 Bd7-a4 2.Kc1-d1 Re5-a5 3.b4-b5 Ba4xb5 4.Nc7-d5 Bb5-a4 5.Nd5-f6+
---------------------------------------------------------------------------
11/15 00:01 68,748k 44,440k +M9 1.Kb1-a1 Bd7-e8 2.Ka1-b1 a6-a5 3.Kb1-c1 a5xb4 4.a3xb4 Be8-f7 5.Kc1-d1 Re5-a5 6.b4xa5 b6-b5 7.Kd1-e1 b5-b4 8.Ke1-f2
11/15 00:01 68,748k 44,440k +M9 1.Nc7-a8 a6-a5 2.Na8-c7 Re5-e8 3.Kb1-c1 a5xb4 4.a3xb4 Bd7-b5 5.Kc1-d1 Re8-a8 6.Nc7xa8 Bb5-a6 7.Kd1-e1 b6-b5 8.Ke1-f2
12/15 00:01 68,748k 44,440k +M8 1.b4-b5 a6xb5 2.Kb1-c1 b5-b4 3.a3xb4 Re5-b5 4.Kc1-d1 Rb5-a5 5.Kd1-e1 Ra5-a1+ 6.Bb2xa1 b3-b2 7.Ke1-f2 b2xa1Q 8.Re3-g3+
12/15 00:01 68,748k 44,440k +M7 1.Nc7-e6 Bd7xe6 2.Kb1-c1 Re5-a5 3.b4xa5 b6xa5 4.Kc1-d1 Be6-d5 5.Kd1-e1 a5-a4 6.Ke1-f2 a6-a5 7.Re3-g3+
12/15 00:01 68,748k 44,440k +M7 1.a3-a4 a6-a5 2.b4-b5 Re5-e6 3.Kb1-c1 Re6-f6 4.Kc1-d1 Bd7-c6 5.Kd1-e1 Bc6-d5 6.Ke1-f2 Rf6-c6 7.Re3-g3+
12/13 00:01 68,748k 44,440k +M7 1.Nc7-b5 a6-a5 2.Kb1-c1 a5xb4 3.Kc1-d1 b4xc3 4.Nb5xc3 Re5-e7 5.Kd1-e1 Re7-h7 6.Ke1-f2 b6-b5 7.Re3-g3+
12/13 00:01 68,748k 44,440k +M7 1.Nc7-e8 Bd7xe8 2.Kb1-c1 Re5-a5 3.Kc1-d1 Be8-a4 4.b4xa5 Ba4-e8 5.Kd1-e1 b6xa5 6.Ke1-f2 Be8-b5 7.Re3-g3+
12/13 00:01 68,748k 44,440k +M7 1.Nc7xa6 Re5-d5 2.Kb1-c1 Rd5-a5 3.b4xa5 Bd7-c6 4.Kc1-d1 b6-b5 5.Kd1-e1 b5-b4 6.Ke1-f2 b4xc3 7.Re3-g3+
12/12 00:01 68,748k 44,440k +M7 1.Nc7-d5 Re5xd5 2.Kb1-c1 Bd7-a4 3.Kc1-d1 Rd5-a5 4.b4xa5 Ba4-e8 5.Kd1-e1 b6xa5 6.Ke1-f2 Be8-b5
12/9 00:01 68,748k 44,440k +M5 1.Kb1-c1 Bd7-a4 2.Kc1-d1 Re5-a5 3.b4-b5 Ba4xb5 4.Nc7-d5 Bb5-a4 5.Nd5-f6+
Here another position. Does the engine find the Draw? My old desktop PC (Duo core) thinks White is better.
newbie4711 wrote:
Here another position. Does the engine find the Draw? My old desktop PC (Duo core) thinks White is better.
It's well-known but probably not a draw after all. White should not capture the black queen but maneuver its king to a4 and ultimately conquer the pawn on a5.
All scores that say that white is better or black is better are pretty meaningless. Engines assign those values on the basis of material balance - which clearly goes white's way once the black queen has disappeared. Only when the analysis ends in "+Mx" you can be "sure" that white wins. And you need to let your engine run like 24 hours to have any confidence that it is a draw on the basis that the engine found no mate. Not by any other evaluation score. You can also run "partial analyses" by mixing clever player choices between analysis sections. For instance you can start an analysis part after the a5-capture which focuses the engine completely on following the critical path section and discards all other options.
The only part where the human still beats the machine is heuristic logic which is what we use when we detect the backdoor entry a5 into the black position. We can use the combined strengths of humans and engines to crack challenges such as this one.
Did any engines solve this yet? It's only a few moves (I had the black king on the wrong square last time sorry) It has to understand when it's reached a drawn endgame position.
EndgameStudier wrote:
Did any engines solve this yet? It's only a few moves (I had the black king on the wrong square last time sorry) It has to understand when it's reached a drawn endgame position.
It's simple. The SF engines are quickly back at 0.00. NNUE even lower. It probably sees that black has a forced draw, therefore black can only win otherwise. Before it has refuted all black winning attempts it gives a small advantage to black. Which stands to reason.
The basic error you can make with the chess.com engines is to assume that the display lines show an end result. You never know. Even when it shows a "mate in 4" it can change its mind to "mate in 3" ten minutes later. So let it run freely before drawing conclusions.
EndgameStudier wrote in post #61:
I don't remember the solution, but engines cannot solve these "Chinese wall" type of problems. Another more complicated one:
White has a forced win here that requires alot of back and forth maneuvering with multiple pieces, until the pieces are re-arranged in such a way that white can finally break through by sacking a piece in the right position and push a pawn. Studies like this are why I chose my username as it is, these positions are the most fascinating aspects of chess in my opinion.
Yep, this one was posted and discussed extensively on chess.com some years ago. I solved it in one minute 
because the poster originally misplaced the black king on c5. See how easy it then is? After repair it took me about 2 hours. Not that hard due to very restricted play options.
This one is also very accessible for partial analysis. As a human, first rearrange the pieces in a way you know is achievable and which is near the exit "Qa2" Then start the engine to find the final breakthrough moves in the right order. Not sure the engine finds the whole solution alone from the starting diagram in a reasonable time span. Almost every move is a piece sac and it appears to me that engines resist that - which translates into repeats and delays. Btw, the 50-move rule is disabled for endgame studies by law but would greatly affect the process if SF enables it and the solution comes close to the 50M line. As is also true for a lot of tablebase endgames.
drdos7 wrote in #73:
Check out what the modified version of Stockfish designed specifically for studies and mates (named Crystal 5 KWK) does with that mate in 5:
This is searching only 1 line and it finds the solution in not even 1 second
FEN: 8/2Nb4/pp6/4rp1p/1Pp1pPkP/PpPpR3/1B1P2N1/1K6 w - - 0 1
Crystal 5 KWK avx2:
NNUE evaluation using nn-ad9b42354671.nnue enabled
AND here is the same engine with 10 lines enabled, it finds 10 different mates in 1 second :
Right! Those are the results I was looking for. There is a massive difference between a game engine and a composition engine as there is a massive amount of meta-logic connected to solving compositions. Notably knowing the goal is achievable and knowing maximum duration prunes the search tree by 99.9999%. Obviously those additions change the "general purpose" game environment of a chess engine into a "dedicated" problem solving environment. There are already a number of dedicated solvers on the market like Popeye and WinChloe. I don't have one because I stopped installing software on my desktop. There is a similar addition for solving "fortresses" which may also be part of the Crystal package. I was told though that the fortress checks slow down the game analysis functions and fortresses are rare in chess games. Perhaps the composition solving module would do the same to game SF?
What is strange here is that chess.com offers us a Puzzle forum and loads of Puzzling options but did not implement software like Crystal to help us analyze and design puzzles. On top of the numerous errors in the Puzzle interface, it shows that chess.com is completely uninterested in puzzles. To them, puzzles are merely game fragments used by players in training sessions for tomorrow's game. It is disgusting 
I don't remember the solution, but engines cannot solve these "Chinese wall" type of problems. Another more complicated one:
White has a forced win here that requires alot of back and forth maneuvering with multiple pieces, until the pieces are re-arranged in such a way that white can finally break through by sacking a piece in the right position and push a pawn. Studies like this are why I chose my username as it is, these positions are the most fascinating aspects of chess in my opinion.