Yeah, it's NP-complete to decide whether a cell in Minesweeper must be a mine: https://logic.pku.edu.cn/ann_attachments/np.pdf.
In practice I suspect a SAT solver would make quick work of the positions that actually appear in games.
In practice I suspect a SAT solver would make quick work of the positions that actually appear in games.
(Edit: Here it is! https://pwmarcz.pl/kaboom/ And the write-up: https://pwmarcz.pl/blog/kaboom/ )
This is similar in spirit to my take on the game: https://magnushoff.com/articles/minesweeper/
Unfortunately, not being familiar with SAT solvers, my implementation can grind to a halt in some configurations :)
I find in a lot of repetitive learning, you have a very noisy signal, you don't know if you succeeded because of luck or you did something right.
This variant takes out the luck part.