Hi, not sure I understand what the end result should be but two things:
(-) The use of the vertex normal branch results in everything where the normal is horizontal or pointing downwards will always have the RVT material (negative B component of the normal, zeroed by the Saturate). Thats why the lower half of the sphere is always using the RVT material. And from what I understood what the end result should be, this Vertex normal branch may not be necessary.
(-) Using the difference between the RVT height and the pixel position would result in the “height part” being relative to the landscape height. If it should only be object dependent and not dependent on the height of the landscape, then you could remove the RVT world height sample and only use the pixel position and the object position/scale/bounds