Why does this puzzle offer f8=R as better than f8=Q?

Question

Here's a chess puzzle that gave a confusing response. https://chesspuzzle.net/Puzzle/52112

[FEN "5R2/1p6/p4Ppk/3r4/7p/1P6/P7/1K6 w - - 0 1"]

1. Rh8+ Kg5 2. f7 Rf5 3. f8=R

If you play f8=Q in this final position, you will not fail the puzzle, but be told "There is a better move..." (f8=R).

It seems inevitable that the black captures the promoted piece, and the result is the same after white recaptures.

Incidentally, f8=Q was played in the game. https://www.chess.com/games/view/15563079

What is more curious, is that I have played other puzzles on this site that allow multiple moves in a position for continuation, and my assumption therefore was that it only suggests a "better move" when your move doesn't change the theoretical outcome of the game, but gives significantly lower advantage (i.e. still decisively winning, but slower). This would suggest that chesspuzzle.net has evaluated f8=Q to be an inaccuracy of some sort.

Is there any reasoning to be found for this? Or is it just a strange computer evaluation?

I have tried inputting the position on the analysis board on lichess.org, and it flip flops between the two moves as the depth builds, but eventually starts showing f8=Q with a consistent advantage. (The opposite of chesspuzzle.net)

This seems strange also, as in my perhaps naïve reasoning explained initially, both moves should be identical.

Answer 1

If anything, promoting to a queen is the better choice, as it forces Black to exchange their rook. For promoting to a rook, Stockfish NNUE actually slightly prefers moving the f5-rook away instead of playing ...Rxf8 (although any continuation gives White a large winning advantage). So at the very least Black is given a choice.

But even if we were certain it's best for Black to play ...Rxf8 if White promotes to a rook, there would still be no reason to say that promoting to a rook is better than promoting to a queen - both lead to the same outcome. What possibly happened is that the site's computer calculated one line further/differently, and arrived at an evaluation that was higher in the line for promoting to a rook. Although since you said the site allows multiple solutions to sometimes be accepted, this seems like a bug in their software.

Answer 2

It seems as if the puzzle is based on an actual game played between Jorden van Foreest and Martin Petrov in April 2019.

If you click on the button to show the solution, it shows the full game, of which the puzzle is but a fragment. It looks to me as they want you to solve the puzzle by following the game as played, not by coming up with another solution.

The curious thing is that 365Chess.com has the promotion as a Queen, while ChessPuzzle.net has it as a Rook. Analysis on Chess.com puts promotion to a Queen at 0.94, while promotion to a Rook is 0.96, even though Black's next move is Rxf8 in both cases.

Answer 3

For an engine, it makes no difference if you promote to a rook or a queen. In both cases, your advantage is huge. There is no real resistance for black, but his best chances, if any, are to give his rook for the promoted piece. 3.f8R Rxf8 and 3.f8Q Rxf8 result in the same position. No difference.

But chess puzzles have their own etiquette. And one of the unwritten rules: the less the value of the promoted piece, the more beauty. Underpromotion to a knight or a bishop would not add more beauty but are a blunder, because the black passed pawns are strong now. So a rook or queen it has to be. Both are winning, and the underpromotion is in puzzle terms the „better“ move.

Answer 4

It is a quirk/bug of the software used by that site to pick moves expected in puzzles.