Example :
lichess.org/training/C09cKStarts ok.
white rook to F3 to force exchange
Pawn to G4
RookxRookG3
PxRookG3
KxP G3
PxP H3
Then it has the effrontery to claim that PxP H3 is not the best move and that it should be KxP H3.
Ok, both lead to a win, but PxP is clearly better as the pawn is then one space further forward and already on the side protected by it's own king.
After 47. gxh3 black king can easily reach g8 and white can do nothing to prevent it so that the resulting endgame is a simple draw. Which is something you (a) should know or (b) would have learned if you tried to evaluate the resulting position with an engine.
E.g. 47. gxh3 Kd7 48. Kg4 Ke7 49. Kg5 Kf7 50. Kh6 Kg8 and then black can simply alternate between g8 and h8 and there is no way white can make any progress.
No the pawn on the h file will draw. Black king manages to reach f7/f8 quick enough.
Ok, fair enough I hadn't counted the moves out to see that the Black king could get across, which of course it can, but AFAICS it'll be a draw for KxP as well so it isn't better than PxP.
Wrong again. After 47. Kxh3, the resulting endgame is easily winning for white. White is left with a g-pawn, the king in front of it and with enough space to control the path forward.
Suggested study material:
lichess.org/practice/pawn-endgames/key-squares/It turns out these computer-generated puzzles are usually correct indeed :-)
Oh righto you can win with a g-pawn but not an h-pawn in that position, so the puzzle was completely correct. Thanks, every day's a schoolday. :-)
Pretty much even when the engine says the puzzle is wrong, if you just let the engine run a little longer, it will see that it isn't wrong.
