I think the "learn from your mistakes" feature is great, however, I think it can be made even better as a separate training mode by simply taking your past games (perhaps sorted according to time-control, etc.) which have been analyzed by a computer and generate puzzles based on them.
One way to accomplish this is to add a button after the last "mistake" has been solved (i.e. the small area currently used by the feature) which says "learn from mistakes in next game" which simply launch the "learn from mistakes" screen anew on the most recent game which has not been "solved", going directly into the computer analysis.
I realize this requires a database flag for when a game has been "solved for mistakes" and so might not be entirely trivial to implement...
One way to accomplish this is to add a button after the last "mistake" has been solved (i.e. the small area currently used by the feature) which says "learn from mistakes in next game" which simply launch the "learn from mistakes" screen anew on the most recent game which has not been "solved", going directly into the computer analysis.
I realize this requires a database flag for when a game has been "solved for mistakes" and so might not be entirely trivial to implement...