Your summary of SweeperBot's current strategies are correct, although by default cheating is only done as a last resort (if we're guessing, and would have stepped on a mine, guess again).

I agree that the current logic is very sub-optimal, most of it was hacked together during coffee breaks at OSDC a few years ago, when I was preparing for talks, and hasn't been touched since. A better AI and faster image-handling are the two areas I think have the greatest scope for improvement.

I would love contributions towards the AI, and would be very happy to accept submissions and suggestions. My current goal is to try and get the code cleaned up and properly documented with a decent test suite; the original code from Matt Sparks used a lot of package variables, and many of them are still present in the code I've released.

Patches are extremely welcome, and I'll see if I can find a suitable way to make my git repository accessible to assist that. However please don't let that hold you up; any patches against what's released on the CPAN will apply just fine (I'll make sure of it).

Best regards,