In a recent Chess Life article, the author mused that sometimes chess engines make moves that really puzzle humans.
That got me wondering about having one or more grandmasters (sans computers) analyzing a chess engine game where they look at a single position in the game, in which the most obtuse move(s) are made by the chess engines. (I say move(s), as the engine making the puzzling move might be replied by an equal puzzling move).
So the grandmasters look at the position of the first puzzling move and move what they think is a better, or at least less puzzling move. Then they let the chess engine game resume from that point. Whatever time was left is what you still allow the engines to finish the game in.
Then we can look to see if the game ended similar or not, plus we can see if the game was shortened, or lengthened from the limited human help.