Interesting endgame puzzle

I have never posted anything in this forum before, but I am a little confused about this puzzle: lichess.org/training/XpUvu

After looking at the position for nearly 10 minutes, I found a variation which continued with 44. ... bxa5 45. kxc5 kxf4 46. d4 a4 47. d5 a3 48. d6 a2 49. d7 a1=Q 50. d8=Q after which black can give a few checks and eventually win the g pawn as well. Seeing this, I mentally decided that by being two pawns up in a barren queen v. queen endgame, black has to be winning. But of course, that was not the best solution, and I got the puzzle wrong, since 45. ... a4 first before 46. ... kxf4 is better.

However, when I ran my own line through the engine, the engine said that although not as good, 45. ... kxf4 is winning as well [-1.9]. White's queen will give a bunch of checks, but if both players play perfectly, one of black's extra two pawns will eventually queen and white will lose.

I have never seen the engine make a mistake, and so if a puzzle can only have one winning solution, what the heck is going on over here? Does anybody know?

Yes, your variation wins as well, as one can easily confirm with the 7-men tablebases.

I think it might have been due to limited search depth, or the evaluation gap was huge.

This puzzle is probably borderline incorrect in that way. When you see a4, it is clear that it is a much simpler and cleaner win, but Kxf4 comes still with a very pleasant advantage (yet, with much work ahead).

Would be a candidate for a chessable soft fail. :-)

Ah okay, so that's probably it. Thank you for the clarification :)