Strategic Test Suite (STS): STS(v6.0) Recapturing.024

A continuation of the discussion started here:
jomega's Blog • Strategic Test Suite (STS): The EPD file's best and alternate best moves. • lichess.org

Another interesting position for which Stockfish 14 failed to find the best move in the 3 second (depth 24) test.

The Human Perspective

This position is in the "Recapturing" category; suggesting that White needs to decide on which piece to recapture with on f4. However, if one has the time, one should look around for a possibly better move than a reflexive recapture.

White is currently down 2 pawns; hence White will be down a pawn when/if he recaptures on f4. However, White has a good outpost at d5. The half-open f-file and weak dark-squares around Black's King on the kingside may give White good kingside attack prospects. Black's d6-pawn is isolated as is White's e4-pawn. However, if Black captures a Knight on d5, then White might get rid of his isolated e-pawn.

The 'natural' move is Nxf4 and then soon Ncd5. Trying that Ncd5 move first, White wonders what would happen if Black tries to hold the f-pawn with g5. White realizes that this would allow sacrificing the e2-Knight for 2 pawns and a kingside attack. Intuition/experience or calculation shows that the sacrifice is sound. If Black does not play g5, then he cannot defend the f-pawn and White will soon play Nxf4. This is an example of a trap that gives the opponent a chance to go wrong, and if he does not then you end up in the same position you would have had anyway.

Stockfish's Perspective

Stockfish gives 1.Nxf4 and 1.Nd5 essentially the same evaluation. Stockfish has no internal model of an opponent like a human might (1). Stockfish uses the same code to evaluate positions in its search regardless of such things. Stockfish specifically does not have code to set traps! We might look at Stockfish's suggestion and label a move a trap, but that is not the same thing. It would be very interesting to find any chess engine that in this position has code that does the equivalent logic I put above in the human perspective; that 1.Nd5 could be played with the intention of trapping the opponent into playing 1...g5.

(1) Stockfish use to have a draw contempt option, but that was removed.

EPD Changes Needed

Whatever of the two moves 1.Nxf4 and 1.Nd5 are taken to be the best move in the EPD, the other move should be a score of 9. For reasons I point out in the study, the other two moves should both have a score of 5; half the best score.

Links

- The Strategic Test Suite (STS) home page.
https://sites.google.com/site/strategictestsuite/
- The STS-rating code.
https://github.com/fsmosca/STS-Rating