coloin wrote:I think I searched for these by random generation of an overwhelming amount of 9plus13s with {-1} search -> 8plus13s / 9plus12s
Is that [ somehow] an exhaustive search ? or how ?
It wasn't an exhaustive search. I did more or less what you did.
I kept a record of the 9plus13s, and made sure that each one got {-1,+1} and {-2,+1} tested. The 9plus12s got {-1,+1},{-2,+2} and {-1,+2} tested.
I also did a thing where I look for small unavoidable sets in the solution grids that include two or more puzzle clues, and try changing the clues to match the alternate fill for the UA set, to see if that produces a valid puzzle. It doesn't have a large yield.
I did look into an exhaustive search for the "9c box" case, but it was looking like best I could hope for was (
~15) ~21 days to complete it, running 8 (physical) cores at 4.5 GHz. (15 days if I throw my other machine at it too).
I liked the way you opti-morph the puzzles.. keeping the first row/ box 5 as 123/456/789 -> a resultant minimal morph.
Thanks