Just out of curiosity, I've coded the "weak" version upto depth 2.
At least, we agree on level 1.
blue wrote:2nd 100
-------
level 2 : 100 (100.00%)
But even on level 2, I find 12 puzzles in weak-T3
So, let's start again from the basics (they should be the same in both approaches):
1) Occurrence of template[1] = 1 digit nb1 + 9 cells that don't "see" each other; nb1 must still be a candidate or a decided value in each of the 9 cells; no other condition.
2) Occurrence of template[k] = k different digits and k templates[1] resp. for the k digits + only compatibility condition: the k sets of cells don't overlap.
Now, in the "weak" approach:
3) A candidate with number nb is eliminated at depth k>0 is there is a set of k digits containing nb such that there is no template[k] for this set.
4) No template[k] is ever eliminated, except indirectly via the elimination of a candidate or of a template[1] that is part of it.
Do we agree on these four points?
.