Computer-Generated Chess Problem 03111

Published online for the first time, consider this KQNN vs krpp chess problem generated by a computer program, Chesthetica, using the 'Digital Synaptic Neural Substrate' computational creativity approach which does not use any kind of deep learning. There is no known limit to the quantity or type of compositions that can be generated. The largest endgame tablebase in existence today is for 7 pieces (Lomonosov) which contains over 500 trillion positions, most of which have not been seen by human eyes. This problem with 8 pieces goes even beyond that and was therefore composed without any such help.

8/7p/2K3pk/6rN/3NQ3/8/8/8 w - - 0 1
White to Play and Mate in 3
Chesthetica v12.06 (Selangor, Malaysia)
Generated on 24 Nov 2020 at 4:50:36 AMSolvability Estimate = Difficult
Even with the same version number, each copy of Chesthetica 'evolves' and performs somewhat differently over time. White is over a rook's worth in material but the precise win in this position still needs to be found. If this one is too easy or too difficult for you, try out some of the others. Feel free to copy the position into a chess engine and discover even more variations of the solution. If you're bored of standard chess, though, why not try this?
Solution