Strategic Test Suite (STS): STS(v7.0) Offer.002

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.

Cooked or Not?

By "cooked" I mean something similar to that term in chess composition. But here we have a strange case, because the intended best move is best, but the idea behind the intended move was to trade the Bishop for the Knight!

According to the website for STS, the Offer of Simplication positions have these traits:

• Placing pieces in direction of trade especially when you're either materially superior or positionally superior.
• Tests your engine's understanding of simplification, trading of one of opponent's active pieces.
• The best moves involve the offer for trade of pieces.

In most cases, the offer has to be accepted but sometimes the offer is best declined.

This makes it clear that the intent was to trade the Bishop for the Knight. However, as Stockfish analysis shows, Black can win material with that Bishop move. The analysis is complex, and it does not involve an immediate capture of the Knight. So the position is cooked because the intended idea of that immediate capture is not best.

The Human Perspective

Let's look at White's threats; that is, pretend it was White's move. (1) What we see is that White can drive the Black Rook away from protecting the Black Bishop with a5, and so threaten to win the Bishop with a fork. Black can get his Bishop to safety in that line, but then White's Queen still comes to c8 where it and the Knight now attack the e6-pawn.

So the active Knight should be pinned to the Queen with Be5. Now there is the idea of "trading of one of the opponent's active pieces", but there is also the idea of "piling on the pinned piece". This is where calculation is required. Black has an attack on the dark squares around the White King. By combining that attack with the attack on the Knight, White eventual trades the Bishop for the Knight and wins a pawn. This requires a 20 half-move look-ahead to verify; though it is mostly forced White responses. Hence, intuitive play and experience would be required.

The immediate exchange of Bishop for Knight after the pin, leads to a Rook and pawn endgame where it is not clear that Black has a win.

Stockfish's Perspective

Stockfish excels at tactics. Stockfish has no problem calculating the 20+ half-move look-ahead variation that results in the win of a pawn and a won endgame. The amusing thing, for which I have no good explanation, is that Stockfish picks 1...Qh2+ as the best move despite the fact that the variation it then gives leads right back to the initial position, and then it picks ...Be5.

EPD Changes Needed

The scores in the EPD should be Be5=10, Qh2=9, Kh7=4, a5=2, Bd6=2, or the first two can be swapped or both be 10.

Notes

(1) It would be really useful if we could enter a null move on Lichess, but that functionality is not there. We can however, click the "show threat" button (or press 'x'). We can then drag the mouse cursor over those moves and a small board will pop up to show the moves.

Links

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