Too difficult for computers

The signing of the scoresheet is not part of the game space but of the competition management. The game end position is the final goodbye to legality and illegality of game events. After it, stuffing 32 weird looking wooden objects in an undersized box is perfectly "legal". @jetoba (FIDE arbiter) has described how this even applies to "dead positions". When the players continue playing unaware that their position is dead, all their moves, resignations or agreements are cancelled in retrospect. Even signed scoresheets may be overturned by the TD - when it's clearly a mistake and detected before the next round. But that's competition, not game law. Note that higher management can also declare a complete competition "illegal". For instance, I bet that the players never got ELO-points for the Fischer-Spasski return match.

Btw, the scenario of incorrect reconstruction (or illegal moves) followed by continuation of the game only exists under Rapid and Blitz conditions. That the players are allowed to continue "the game" (there is no game) serves only to establish a competition score to fill the void of a missing game score! But to keep things simple, why would we solve any puzzle by Rapid rules?

Here's a nice white to play and win study by Romanian Mihai Neghina that computers don't seem to understand:

I'll post a PGN of the solution tomorrow. In the meantime let me know if an engine sees the win here.

It's a great display of domination! I think you posted this one before. Or was it just similar?

Yes, I've posted it before, the strange thing is that I have an engine that sees the first few moves but doesn't demonstrate a winning score, it shows 0.00. I tried out the new one here named "Torch" and it does the same thing, it sees the first few moves but gives the same 0.00 score. They need to offer the Torch engine as a downloadable UCI engine that you can install in a GUI either free or as a paid product like Komodo Dragon.

Here is the solution to this study:

I must admit, that after letting Torch think about this one for a several hours it sees the win here with a correct continuation, but just about every other engine i've tried is clueless.

A fairly recent Stockfish from git (I used an I. Ivec compile) initially thinks that 1.Kf3 is slightly better that 1.Nd4, then after about one minute thinks that 1.Nd4 is the best, and around 50-ply does find the winning mechanism.

Thanks for the info. Here is the Torch eval I was talking about:

ja schade,das die kaputt ist aber das ist ja mit manchen passiert im vor computer zeitalter.

I've had some good and some bad experiences with the Torch version offered by chess.com. And indeed the engines I tried settle on 0.00 which is rational. If you can't let the queen out of the cage and it is easy to keep it in without making progress what else is there to do? Until you finally do make real progress of course!

Sting and huntsman (without nnue) solves this one almost instantly

Also, heres the puzzle created by me that none of the engines i tried seems to be able to solve. Black to play and draw. Huntsman is the only one that instantly

finds the first move but doesnt seem like it shows the correct continuation (at least not in first few minutes)...

It looks like the engine "Crystal 7" finds the draw instantly (less than 1 second):

If you would like to get this engine you can download it here for free.

What??????????

why do people cap their engine depth at 20 and say the puzzle is impossible for engines

Perhaps they don't understand how to change the depth or even what the depth means, another thing that people do is let the engine search 3 lines instead of one which also degrades the search.

Is this really the best move for black?

The commentary says blacks is "defending a pawn", but it looks to me like black is just giving up his queen for a rook.

Grandmasters play this very stuff all the time when there is a "no draw before move 30" rule in the tournament, and they don't want to waste much time to get their pre-arranged draw: they are copying the game Hamppe- Meitner which was played 154 years ago, but of course they don't brag that they have created something.

The best move that Black can make in this position is to resign. If we put the Black Queen back on g6 Stockfish says that White can force checkmate in 9 no matter what Black does.