I believe if you change the representation of puzzles from [(pos,range)] to an Array, you'll get a significant speedup
yet because I only recently removed a logic bug that slowed down the search instead of speading it up; ..). so the more interesting bit is that our solvers disagree on which are the most difficult puzzles (requiring the largest number of guesses):
df puzzles involving guesses: 5319 If that's not a typo, I'm baffled. My original needed to guess in 5309
Rot! Typo in _my_ previous message, 5319 is correct.
After posting my cleaned up dancing links solver, I went back to my logical solver. I sent the 36628 line sudoku17 puzzle through it and it could solve 31322 of the puzzles, leaving 5306 resistant. I haven't check my algorithms against your code, but it does not seem like I do any spectacularly more clever. I don't have any time to clean up my code, but I may try turning off some steps to see if one of them gives my the same 5319 number you are seeing. -- Chris Kuklewicz