20 test positions that I included some time ago for a test set called Tricky for Stockfish Code. On version 6 now with some updates and replacements. Solve times will very subject to difficulty, engine and hardware. 4 minutes per position was given using my i7-10700 CPU machine with 4 Gb hash allocated.
Chessbase and pgn files with solution and comment available from ...
https://www.mediafire.com/file/3gwinp4i9f9si7m/TFSFCv6.zip/file
Position 1. Nigel Short's Immortal King Walk. 31.Kh2!!
The original BS2830 position begins with 32.Kg3 but the King walk starts a move earlier with 31.Kh2. On finding the game in my database, 31.Kh2 had an infinity sign allocated to it suggesting that at the time the experts could not decide if the move was favourable or not. Today though, SugaR 120523 not only confirms the move is winning but also indicates #19. Harder for engines to find, I was about to replace the original position but have now kept the same game with the key move one move earlier being satisfied it is the correct move.
Position 2. Gosta Stoltz vs Georg Loven, Helsinki 1933 16.Nd5
Having looked through the games of Gosta Stoltz, he was the type of player you would not want to face after a bad day at the office! In the game that provides the position, 16.Nd5!! would fray the nerves of most people. Spectacular and gets the vote. Many engines choose 16.Qe2 that also appearss to be winning but the game move provides a faster route and humans prefer the spectacular, me included!
Position 3. O'Donel Alexander Conel, Hugh vs Marshal E.T. 31.Na4
Another spectacular move 21.Na4 that quickly settles the game when even Zappa Mexico II x64 announces #12 but today's Stockfish code based engines either fail or struggle to find the quick finish key move and strangely, older engines have no trouble as indicated by Zappa finding the mate!
Position 4. Asauskas vs Malisauskas. Vilnius 1964 19.Qf6
Variants on the position have been seen elsewhere, but here taken from the game, an unlikely looking Queen sacrifice forces the issue. The game moves included but Black can prolong the outcome by giving back the Queen, however it is to no avail!
Position 5. Gusev, Yuri vs Auerbach, E. Molniya Sporting Society R5, 29.05.1946 Qxe5
The position has been used in other test sets but provides for another unexpected and spectacular Queen sacrifice that game engines can struggle with. Analysis engines tend to fair better.
Position 6. Summerscale, Aaron vs Jones, Gawain C . GBR-ch 96th, Torquay 31.07.2009 28..Rxh2
Included because the Rook sacrifice looks typically human. More recently, engines have caught up and find the human mind's thinking easier to grasp.
Position 7. First inclusion taken from an engine book match, 30..Rf7!
played shortly after the introduction of nnue with the Stockfish 12 engine. 30..a3?? was played by the engine immediately taking a losing route as seen by the game scores. 20..Rf7 seems to save the day.
Positions 8. The second engine game from a book match leads to a quick demise for Black, 17..f5
with the later Stockfish 220621 engine being completely blind to the threat to its Queen. Today's Stockfish code may still fall for it in bullet games.
Positions 9-10 Book game mistakes.
18..Kxg6!
An obvious choice from two moves was completely missed by the engine. More recent engines have improved here.
Position 10 from the same game and another key move miss that may have saved the game. 14..h6
The Game scores record the White book had been here before!
Position 11. Another book match some six months later and that Tricky Herald book had clearly played the line before. 24..Rac8
Only two moves worth considering and Stockfish 020122 chose the wrong one! Today's engines have improved with this one.
Position 12. Final Book game. 32.Nxe6!!
I did post this game shortly after the game was played. The match engine chose 32.Rd6?? and the game score suggested a win for White but no win was forthcoming with the game finishing as a 222 move draw! White had completely missed the winning 32.Nxe6 that may have been down to the engine's reluctance to sacrifice further material already being down R+P for Knight. Interestingly though, switch off nnue and the Classical only engine finds the move instantly as do most pre-nnue engines!?
Position 13. Zugszwang, ERET 032 from Walter Eigenmann's test set. 1.Ng3
Probably easier for the human eye in this relatively simple endgame position when it quickly becomes clear after the solution move 1.Ng3 then Black is damned if the Knight is not captured and damned if it is! If the Knight is not captured there follows Nf5 and Black's King can no longer stop the F pawn advance and promotion. Capturing leads to the Black bishop being trapped and subsequent forcing of the King to capture the h5 pawn allowing the F pawn to advance and promote.
Position 14. lech Engine IQ Test P42 1.Rh1!
The rook sacrifice is not easy to find for the untrained eye and many engines struggle to find Rh1 but now current Stockfish code finds the move fairly quickly.
15. Mate in 9. Hard CCC 2006 (Vinvin) 1.h7 #9
Highlights a position where there are several mates of different length that distract today's engines but eventually they end up finding the correct lead in move 1.h7 that is #9. Note Chessbase Fritz GUIs "Engine\ Process Test Set" feature will give a correct score when a mate is found whether or not it is the correct length or move! Native Fritz engines tended to stop altogether once a mate was found irrespective of the correct solution and length.
Position 16. Tricky Bishop Hard CCC 2006 (Vinvin) 1.Bd5!
Not so hard now in 2023 but still a nice unexpected move where the unlikely looking 1..Bd5! initially causes rapid blinking of the eyes but the engines confirm advantage to Black with the exposed White King under threat.
Position 17. Mate in 27 (Easy for Engines)
1.Rg4!
Any difficulty is associated with identifying the initial move that requires the rook sacrifice Rg4+. Plain sailing to mate after the capture.
Position 18. Creating the unstoppable passed pawn. 1..Ba6!!
First seen in the Bill Reid post in Selective Search Chess Magazine that was published by Eric Hallsworth. The lead in move is not obvious until seen and would bring a smile to the face if played in a game. Tricky for engines back then but generally found fairly quickly these days.
Position 19 ERET 093 1.Nd6+!!
Despite being a composed position, it certainly has the look and feel of a game that could be encountered. Very tricky for engines though because it requires White to give up material to block off any potential threats from Black starting with the very nice Nd6+. Failure to capture is not an option for Black. The three pawns on the D file prevent the Bishop from contributing and White is able to set up a defence tp prevent Black's Rook and King breaking through.
Position 20. Grandelius, N vs Raznikov, D. Eu-ch U18, Albena 2011 50.Nd3!
Engines are initially distracted by material capture or exchange, Nxh5 or fxg5 but eventually identify the game move Nd3 as being winning and correct. I believe the position has been published in a test set or two but interesting to see the game position was from an Under 18 youth game.
Peter
20 Interesting positions
Moderators: Elijah, Igbo, timetraveller
-
Peter Grayson
- Forum Contributions
- Points: 31 638,00
- Posts: 658
- Joined: 07/11/2019, 17:19
- Status: Offline (Active 2 Days, 23 Hours, 19 Minutes ago)
- Medals: 1
- Topics: 170
- Reputation: 3317
- Location: Newport, South Wales, UK
- Has thanked: 2808 times
- Been thanked: 2389 times