STS(v15.0) AT.006

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

The intent of the position is that the engines should avoid the pointless exchange of Queens.

Stockfish 14 failed to find the best move in the 3 second test. Stockfish 14 suggested 1.Qb6/23. That is an alternative in the EPD with a score of 9 out of 10.

My study on this position is here:

The Human Perspective

Material is equal. Black has a passed a-pawn and a well-posted Knight. White has a weak d-pawn and a bad Bishop. White's King is more exposed than Black's. White has offered a Queen exchange. While White is attacking the a5-pawn three times, and it is only defended twice, the tactics do not support White trying to win that pawn at the moment.

Exchanging Queens is actually very bad. As I show in the study, Black will end up losing his passed a-pawn. Though Black can get the d4-pawn in exchange, the analysis shows that all the pieces, except a Rook for each side, will get exchanged. The result is a book drawn Rook and pawn endgame.

So the question becomes, where to move the Black Queen. Now ...Qd3 obviously threatens the White Bishop, and just as obviously, White can play Rd1 in response. Now Black would have to move the Queen again, or counter-attack the White Queen (because of the threatened discovered attack). The latter is complex and the former becomes a tempo gaining move if White plays the Rook back to a1.

The move ...Qb6 relies on analysis that shows that White cannot win the a-pawn with Bxa5. In many of the lines after ...Qb6, Black gets his Bishop to b4.

Some of the other alternative places to move the Black Queen can just end up repeating the initial position. A cat-and-mouse idea that Black might try to grind White down, or gain moves for the next time control.

It seems from the human perspective that ...Qd3 goes in for complications while ...Qb6 avoids those complications and goes for strengthening the Black position with moves like ...h5 and later ...Bb4.

Stockfish's Perspective

From a centipawn view, Stockfish dislikes ...Qd3 by almost 100 cp. Stockfish chooses ...Qb6 as best. On ...Qd3, Stockfish calculates that the attention will change to the d-file, but White has resources.

So why does the EPD have ...Qd3 as best, and most of the alternatives as 9 out of 10?

As an experiment, I gave the original position to Fritz-12; a program that was around during 2009 when the STS was being created and moves for the EPD originally chosen. Sure enough, Fritz-12 chooses 1...Qd3 with -0.58/19. Qb7, Qb2, and Qb6 are about the same score according to Fritz-12. At the time, many so called 'alpha/beta' engines were being used to vet the EPD. Most probably gave answers similar to Fritz-12 here.

EPD Changes Needed

Qb2 and Qb7 can end up back in the initial position; so not that great. Qb8 gives White too much counter-play. I'd change all of these to 4. Qd3 and Qb6 are both reasonable alternatives, so I'd make them both 10.
Hence; Qd3=10, Qb6=10, Qb8=4, Qb7=4, Qb2=4.

Next Steps

I've been looking at EPD positions in which, even at much longer times than the STS-rating allows, Stockfish did not pick the intended best move. It would be interesting to look at the positions for which Stockfish did not pick the intended best more or any of the alternative moves in the EPD. I'll leave that for a later time.

I hope you have enjoyed these blog posts on the STS.

Links

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