gsf wrote:tarek wrote:The long awaited update.....great.....

I've noticed that some older ratings have changed (slashed down

) when I re-ran them through this updated version, was that intentional ???

on nov 6 in this thread RW found a bug in my solver's proposition constraint logic

fixing that bug had the effect of lowering the ratings for puzzles previously affected by the bug

also, to accomodate the latest hardest some of the 993?? and 994?? ratings may have changed

stop finding harder puzzles and the ratings will stabilize

Now the distribution of values is less uniform than previously

we would expect finding much more 994xx than 995xx

I find 10 times more 995xx than 994xx

Probably others observed that

Also many values are like attractors : in particular around 9994X

I have no 9993X,9992X,9991X but plenty 9994X !!

Maybe this is normal, just a chaotic behavior

Any idea on that?

From a ranking program I would expect some logarithmic behavior

But I understand the challenge

If I find hard sudoku , what value should I reach to publish in this list

above 99800 or 99900 or 99950 or 99960

for Explainer above 10 or 10.5 or 10.6 or 11.0?

or combination of 2 values

can somebody explain in english terms what is expected from a sudoku to be considered very hard

For exemple: if I have 20 contradictory chains of length 20 , is it harder of 1 contradictory chain of length 60?

Probably statistics could be a good evaluator of the difficulties to find chain of length 30 vs length 50

I would be interested by the equation used to rank the sudoku

Should not all developers collaborate on this equation , I assume no single individual can find the best ranking program, and it will be even hard for a team of experts. If the implementation can be hidden I consider that

the ranking equation should be open and voted by others experts.

thanks