ixsetf wrote:I understand analysis of a pattern is very difficult, so searching all of them will likely be impossible. However, by generating a small sample for a few of these, it might be possible to focus the search to patterns with particularly high potential for hard puzzles. I know this might not be something you want to invest your time into, but it's something that I would like to invest my time into once I get the required tools and programming skills for preforming this search myself. That said it may be a while before I actually get to that point, so everything you're doing is extremely helpful.
Working on patterns with a supposed high potential has been the start for the search of "hardest puzzles".The true problem is that we have no clear idea of what is a pattern with high potential.
That's why the search turned to an extended vicinity analysis
- starting from a random basis as explained eleven some posts earlier
- starting from already known puzzles for me.
After 2 or 3 years, we have a collection of more than one million puzzles with very high SE ratings.
Trying to solve these puzzles, it appeared that several special properties can be identified? The most common is the JExocet.
I selected that property to go in the direction you indicate.
My first step has been to re use existing patterns where the JExocet has been found and to generate puzzles with a similar property. It's not so easy and the code I have does not accept all cases.
When and if this is really positive, I intend to generate new patterns "of interest" but I am already 72 years old and one day I'll be forced to stop.
If you are young enough, you can try a wider scope.