As for the "2 Latin Squares" property, which both dyitto and HATMAN guessed, I will explain how it gets to appear in the solution using my correct rule.
First, I have to stress that the correct mathematical term for this is "Euler Square". Please google it if you have never heard of it before.
Now let's use the following notation for the symbols:
1=1-sq
2=2-sq
3=3x3-sq
N=no-X
S=small-X
L=large-X
So the original puzzle can be shown this way:
Code:
.. .. .. | .. .. .. | .. .. ..
.. .. .. | .. 1S 2L | .. .. ..
.. .. .. | .. 2N 3S | .. .. ..
---------+----------+---------
.. 1S 2L | .. .. .. | .. .. ..
.. 2N 3S | .. .. .. | 1S 2L ..
.. .. .. | .. .. .. | 2N 3S ..
---------+----------+---------
.. .. .. | 1S 2L .. | .. .. ..
.. .. .. | 2N 3S .. | .. .. ..
.. .. .. | .. .. .. | .. .. ..
Now, without loss of generality, assume a 3x3 nonet (say N5) has 2S in the central cell (R5C5):
Code:
.. .. ..
.. 2S ..
.. .. ..
Using the "non-connective" rule, R46C5+R5C46={1N,1L,3N,3L}. Without loss of generality, assume R4C5=1N:
Code:
.. 1N ..
.. 2S ..
.. .. ..
Now look at R5C46. They must not be 3L, or it will force one of R4C46 to be 2S, same as R5C5.
--> Hidden single: R6C5=3L
Code:
.. 1N ..
.. 2S ..
.. 3L ..
Now R5C46={1L,3N}. Without loss of generality, assume R5C4=1L:
Code:
.. 1N ..
1L 2S 3N
.. 3L ..
Using the "non-connective" rule, the 4 other cells can be easily solved now:
Code:
3S 1N 2L
1L 2S 3N
2N 3L 1S
As you can see, the nonet automatically becomes an "Euler Square".