Traning puzzles - add variants/improvements with human defences?

Is it possible to create better tactics training puzzles by allowings users to create the "human version" starting from the computer-generated example? Here's an example where the computer defence misses the beautiful point:

http://en.lichess.org/training/57679

1.Ng5 Bxg5 2.Bxg5 f6 human defence (2...Kg8 computer defence 3.Bxd8) 3.Bxg6 Rf7 4.Bxf7 fxg5 5.Re8+! Beautiful interference/deflection Qxe8 (5...Bxe8 6.Qf8#) 6.Qf6#

if there is a group of strong players like you who can differantiate between human and comp response, and select puzzles into one awesome pool, that would be awesome!!!

After 2. Bxg5, stockfish responds not with f6 but with 2 ... Nd4.

I don't really get the suggestion here. You want to arbitrarily declare some line other than the best one to be the "human" line? Who determines this and how do they determine it?

If the "human line" were as predictable as you seem to believe, then the humans in question would be playing ... like computers. You're essentially saying you want puzzles where the object is to find some arbitrary move that isn't as good as the best move, how is this helpful?

it s not arbitrary #3 with time as you become stronger you ll comprehend what a computer move looks like

@Uso: Nonsense. What the comment suggested is the very definition of arbitrary. What makes his proposal the "human line" instead of any other move that's not the best one? He pulled it out of thin air because it happens to be the one he looked at, that's what arbitrary is.

MrBug's line not only involves white playing something other than the best move, but also requires assuming black will play something other than the best move (according to Stockfish) in response. How does this help you learn what the best move is? All it does is reinforce bad moves.

The "best move" according a computer isn't necessarily the most testing move from a human point of view. For example, the main line of an endgame study is typically the most beautiful/instructive variation, which isn't necessarily the computer's principal variation. Or, if you've beaten a computer at chess, you'll find that it sometimes gives away material when it pull out all stops to prevent a mating attack that you might not have managed to calculate yourself.

@Little_Bobby_Tables: In my line, all of White's moves were best according to Stockfish. My post wasn't about the attacking moves that the solver must find, but the defensive responses that get presented the the solver. I also have no idea why your Stockfish suggested 2...Nd4, when mine claims m7 starting with 3.Bxd8 Raxd8 4.Bxg6! fxg6 5.Re7.

I think you misunderstand chess puzzles:

What you are supposed to do is calculate the whole line, then play it. You are supposed to have a solution to both moves already prepared in you mind so no matter what is played, your response should be prepared.