This is funny: puzzle generator prefers a mate in 17 than a mate in 2

http://en.lichess.org/training/2809

After the first two moves: 1.Qh5, Bg7 we have a mate in 2 starting with 2.Re7 that gives the message "puzzle failed" (?!)

And here is the winner: Mate in 17 starting with Qg6. I could not believe it, so I confirmed it with a second engine. So, the puzzle generator can see a mate in 17, but can`t see a mate in two?

Suggestion: Allow users to make the necessary modifications in the puzzles, maybe by voting on different moves. Too complicated?

Not too complicated, but requires man hours to produce.

The problem is that it plays Bg7 instead of Qe7.

The line pointed out by Stockfish in the source game was Qh5 Qe7, so that's curious indeed.

How is this even possible?! Stockfish got seriously confused. Haha.

Doesn't look like stockfish got confused, as the analysis in the source game is correct. Probably just a hard-to-track-down problem with parsing the solution together.

I want to understand how the puzzle generator failed here. Where may I find the source code for the puzzle generator?

This was the link provided in the lichess blog post about puzzle creation:

github.com/clarkerubber/problem-creator

on my pc the right move is Qe2xh5 Qd8-e7 Rf7xe7+ Ke8xe7 Qh5-f7+ Ke7-d8 Qf7-f6+

with different engines : no mat quickly
strange ...