It is interesting that we (and our solvers) can get into a rut in terms of which techniques we look for/apply and in what order. Whether using my solver or trying by hand, I'll always look for patterns before locked sets/claiming before fish... unless (by hand) one happens to just jump out at me. More common when I solve by hand is to see a pattern of 2 or 3 while scanning for singles.

Of course my solver nails all singles first. One of the many kewl things about HoDoKu is the configurability that lets you change the order of things.

I finally settled on an "absolute" scoring scheme when I started adding many different puzzle sources into my solver (i.e. for direct downloading) and found the ratings varied so much it was hard to explain what to expect when users downloaded different puzzles (UClick and SudokuPuzz are especially bad). It gave me a way let users have a semi-consistent idea of how one puzzle compares to another.

Once I did that, it wasn't too long before I started applying that to searching for a better chain/net (a la Daj's SIN/chain search); my search is rewards the largest "reduction" in the difficulty score (when there is a tie there are other criteria). It made a huge diference in the number of chains/nets I needed to crack a puzzle that exhausted my solver's other techniques.

Then I used the scoring code as a coarse (and fine grained, as desired) measure to generate puzzles of specific difficulty overall or (akin to HoDoKu) with specific features present (or not) regardless of level. Much more fun than just randomly generating puzzles.

This discussion happened along just as I started thinking about whether or not it would make sense to apply this scoring to SSTS moves instead of taking them blindly (as I do now). But that would mean for any given grid state looking at each potential pattern, locked set, fish, etc and scoring the result (scoring is expensive), then taking the move that reduced the score the most. It would not guarantee the shortest path, but if my experience with chains/nets is any indication, it will be a shorter path than counting eliminations.

Slow? Probably though not as much as you might think for single moves (i.e. what is my next, best move?).

Not trying to convert you, Hobiwan. Just looking for input on the idea

Cheers...

- drac