champagne wrote:1)Super candidates are used when other process fail. It is out of the scope of that thread.

2)Super candidates are limited to AALS AC2 for the time being

Super candidates, even limited to AALS, have a size (number of cells). This size is an important parameter, as the number of super candidates increases exponentially with the number of cells. (candidates are super candidates of size 1).

Another parameter is the size (length and/or breadth) of the chains or nets based on these super candidates. Here again, we may have an exponential increase of possibilities when this size increases.

What'd be very surprsising is that your algorithm wouldn't dispaly any increase in computation time and/or memory as a function of these parameters if if was modified to take them into account.

Such problems are in the scope of this thread, even though your algorithm in its present state doesn't allow to obtain the answer.

My idea when I opened this thread was to compare different ratings of the rules and/or puzzles. The question was (and still is): what is the mean complexity of using such rule (or what are the relative mean complexities of several rules)? With a possible corollary: how should we order the rules?

The results above show that the usual ratings are reasonably well correlated - at least for the scale of complexities available in the first 1.000 puzzles in sudogen0 (from L0 to L7).

I agree that this first approach of the problem should be completed with other puzzles, including harder ones. Don't forget this question was raised only a few days ago and we can't have all the answers at once.

I could use the full 10.000 collection to extend the range. Or we could assemble another collection of puzzles with a broader range.