Comments: 
From: eub 20161217 10:00 am (UTC)
 (Link)

Is the basic intuition like: a (x, y) can be "rotated a little" to an (x, y)'  subject to grid quantization, sure, but for larger values you don't notice much. So why can't the path be rotatedalittle equally as well as the point can?
Simplicius says: Given a path, if you're willing to compute its sum, you can then tweak one or two steps of the path to get your (x,y)' instead.
But that's not satisfactory, because it doesn't seem like you should have to do that?
Simplicius says: wolog say we start with an E (1, 0) that we're rotating alpha c.c.w. The rotated (cos, sin) should be our expectation, and we want to give probabilities on NESW that yield that expectation. Can't do that by blending just E and N, obviously, that just gives a straight line interpolant that falls inside the circle. But we can do it by blending E, N, and E+N.
Is that allowed, to rotate E to EN sometimes? Because it does seem like you're going to have to.
Right, it is critical, I think, to see it as the path rotating, not merely the point; what I'm trying to get a sense of is why the combinatorics of numberoflongpathsto(x,y) seems to have a rotation symmetry? Why is it the 2norm that falls out, of all things?
Changing the length of the path as it rotates a little bit seems acceptable, as long as the length changes average out to nothing over the long run, and/or is negligible for long paths.
From: eub 20161218 08:09 am (UTC)
 (Link)

My intuition is that is really isn't the paths who have this property, it's their sums. That the sequencing of the paths needs to be erased to create the effect.
Backing up  to me the rotational symmetry is not the basic thing here, the Gaussianness is. (You can twiddle the setup to get a Gaussian ellipsoid.) In one dimension, the Gaussianness is born out of the 2^n paths at the point where you sum them down to get the (n choose k) line, where you diagonally scan the ncube. The multidimensional Gaussian comes together in the usual way.
Of course this assumes some properties of Gaussians which you may be preferring to roll your own intuition of?  