That was a false intuition I had at that time because... this approach passed the tests (which are pretty insufficient). I've been proved wrong about that with other problems.

@Blind4Basics I am also interested in why checking one subsquares is enough as well.

@Blind4Basics: any explanation why checking only one subsquare is enough ?

lol

That's a very good observation! I assumed that the input is well formed, but I agree that nothing guaranties that. We can change the comparison:
if len(np.unique(view)) != 9:

to :
if not np.array_equal(np.sort(view), np.arange(1,10)):

so that now it checks only for the array [1,2,...,9]