@leblitzer,
1...Rxd5 2.Nxd5? is a blunder, defeated by the repeated Distraction tactics you demonstrate.
But white can use tactics too. Consider a Distraction tactic 2.f5+ by the white pawn.
[FEN "5b2/5p2/3pk3/pprR3p/4KP2/P1P1N1P1/1P6/8 b - - 0 1"] 1...Rxd5 2.f5+ Rxf5 ( 2...Kf6 { ? } 3.Nxd5+ { 0.00 } ) 3.Nxf5 d5+ 4.Kf4 Bd6+ 5.Kg5
Now black is only up one pawn. That typically is not good enough to be a ChessTempo solution. The Correct solution is up a rook-for-a-pawn.
An endgame up one pawn, with master level play, is advantageous and can often be won. Stockfish16, at depth30, believes black can force the win in this inferior move order variation, via a precise long endgame, but ChessTempo "typically" does not give ALTs for long endgames (one pawn advantage) if you missed a short-term, material winning tactic.
The Toga eval of 1...Rxd5 2.f5+ is 0.81 depth20. Toga's choice of 6...Kd6 is inaccurate and that likely draws against best defense. The only winning move 6 is trading pieces and winning the one pawn advantage king-pawn endgame because black also has the centrally placed king as a positional resource.
Stockfish, looking deeper into the one pawn advantage endgame, found 6...Bxd4 and gives eval +3 at depth 21 and +4 at depth30. But careful, accurate, sometimes sharp play is needed to convert the win.
[FEN "5b2/5p2/3pk3/pprR3p/4KP2/P1P1N1P1/1P6/8 b - - 0 1"] 1...Rxd5 { ? second best } 2.f5+ Rxf5 3.Nxf5 d5+ 4.Kf4 Bd6+ 5.Kg5 Be5 6.Nd4+ Bxd4 { sharp } ( 6...Kd6 { ? Toga's mistake } 7.Nxb5+ { drawish } ) 7.cxd4 { 7...a4 and f6+ can win, b4 draws, f5 or king moves lose. }
The above analysis shows how ChessTempo continues to evolve using the "two pawn" threshold. For problems rated above 2200 or so, CT is occasionally disabling problems which experts and masters identify as being "clearly winning" with one pawn advantages supported by dominant positions. This problem is not rated that high. And I would assert that if the computer engine Toga limited to depth20 can blunder the endgame, then that one-pawn endgame is not "clearly winning."
So I do not suggest disabling this problem, even though the dominant error is 1...Rxd5. I also suspect many players who chose 1...Rxd5 miscalculated as @leblitzer did.
Current ratings
Blitz 1699 on 2/5 successes.
Standard 1749 on 8/21 successes.
3 star quality.
The Correct solution is choosing to play the Distraction tactic first. 1...f5+ 2.Kf3 Rxd5, winning a rook. White fights back, the best she can, winning one pawn of compensation with 2.Rxf5 Rxf5.
It appears at the end of the problem that white can regain more compensation by taking the rook. Both Toga at depth19, and Stockfish at depth30, see 3.Nxf5 as a worse idea. It reduces the material balance to B-for-2P, but white's pawns are far from promotion. The evals are +5 for black, and mate around move 21 when I try some variations. Some of those variations, however, involve both sides promoting a queen and more endgame tactics. So if I am white against club level and expert players, I am taking 3.Nxf5 and making my opponent demonstrate his endgame skills. The "computer best" 3.Kf3 is going out with a wimper. The 3.Kf3 endgame is so easy and robust, a caveman can do it.
[FEN "5b2/5p2/3pk3/pprR3p/4KP2/P1P1N1P1/1P6/8 b - - 0 1"] 1...f5+ 2.Rxf5 ( 2.Nxf5 Rxd5 3.Ne3 Rd2 ) ( 2.Kf3 Rxd5 ) Rxf5 3.Kf3 { wimpy } ( 3.Nxf5 { eval -5.5 instead of -4.8, but much sharper to convert. } d5+ 4.Kd4 Kxf5 5.Kxd5 )