The New Sudoku Players' Forum

[champagne]

It seems that a common property of such puzzles is to have an important number of small UAs (more than 50 UAs with less than 10 cells)
And the output in "non easy puzzles" is in my sample over 200 per entry.

You might want to look at:
1) how may (valid) puzzles come out of the front end and into the minimalizer.
2) how may come out of the minimalizer and into the "easy or not" tester.

Hi Blue,

Thank for advices, Tracking the process with my current debugging code, I have an idea of this, but i'll add counts to see more precisely what happens.

As primary findings, I have

a small "quasi redundancy" when a small ua hit the given only once. In this case, the "out field" "-1" gives "similar" new puzzles.
a high cost to pay when 2 clues are added.

I think that the most important is to hit new solution grids if possible, but that when one hit has been seen, the process can be skipped to a next case. I'll think of shortcuts in this direction.

The main factor compared to the run in the 20-23 area seems to be this high number of small UAs.
Reversely, I see several steps where only one additional very big UA comes. This is one case where the previous remark can play a role.

[champagne]

The bad case came after a +2 clues (to get a valid puzzle) with a significant number of minimal puzzles to find.
I put a filter to stop the process if the non minimal puzzle is "very easy to solve".
The global output is reduced, but likely for puzzles mainly below the final cutoff skfr 10.5
The run time is divided by 10 in my sample.
I do the next vicinity steps using this new code.