The endgame training seems useful, but occasionally the software recommends one move over another that leads to checkmate in fewer moves.
Consider the following position
White to move
The pawns stand on the same squares as in a key position in Dvoretsky's Endgame Manual, 2nd ed. (2008). However, Dvoretsky has the kings on b5 and b7. Black is on move. White has the opposition, but it is insufficient to win. With the kings beside the pawns, as in the Chessimo exercise, White wins by seizing the opposition, executing an outflanking, and exchanging the a-pawn for Black's pawn. The king is then in position to escort to the c-pawn to its promotion square.
Chess engines accessing tablebases reveal that 1.Kd7 is the best move. Chessimo prefers 1.Kd6, which I played. After 1...Ka7 (Rybka played 1...Kb7 when I tried the position against the engine), I attempted 2.Kc7.
Chessimo pops up with the message, "This move is very good and leads to the same conclusion. Please continue with the highlighted move."
My move leads to checkmate in two fewer moves. Moreover, Chessimo's solution leads to the intended position two moves later. I understood the position, and attempted to play the most precise move. Chessimo steered me towards a weaker move that still wins.
On their website, the developers of Chessimo claim, "If the answer isn't the most accurate one, the software shows the correct move." In this case, their claim is false. A less accurate move is favored over the most accurate one.
Empirical Rabbit made similar observations in his review of Chessimo's endgame training nine months ago.
For $7.99, the tactics and endgame training offered by Chessimo's iPad version is a lot of bang for the buck, but it should not be used to the exclusion of good endgame books. The repetition built into the program would be more beneficial if careful editing improved the favored solutions.