Idea to improve puzzle sets

I've noticed many problems that follow a similar pattern. Here is an example:

lichess.org/training/nEsBn

There is an obvious candidate move (in this case Nf6), but it's not at all obvious that it's winning. In the natural "human" line, it only becomes clearly winning after quite a few more moves. But in the "computer" line where black just gives up the queen, it's obviously winning after the next move.

Since these problems only have one obvious candidate move, many players end up playing it by default and the problem gets an unnaturally low rating, whereas if they had to find all the moves againt the more natural defense, most would fail and it would be rated much higher.

My suggestion would be to do a form of A-B testing by varying the line that's proposed when there are two defenses of similar strength. If players have more trouble with one than the other, than that defense would become the default line for the problem and it would end up rated at its actual human difficulty.

I have no idea how easy or hard this would be to implement, but I thought I'd suggest it anyway.

It would be easy to implement, since the puzzles are derived from actual games. The puzzle could just nab the moves played by the humans who probably don't see weird queen sacs to prevent completely opaque checkmates.

I agree that although most puzzles show the single best move, there are many to me that are dogmatic and ignore viable options, or sometimes just wrong.

I think the puzzle masters should consider a concept Magnus has talked about, the concept of playability. Magnus has talked about making choices based on the ease of playing the position, versus making the suggested best engine move.