I can explain "get position in geometry" and how you derive displacement from it. The output from the "pos" node is a single [x, 0, z] value
but repeated for each point (or micro triangle) seen by the camera. Tre node tree below is called repeatly for each of these values.
In principle you first make some vector operation and then you make a transformation to a scalar (e.g a noise function, or a length function).
After this you make some tranformations and feed the resulting scalar into the second input of a displacement node.
What is important to note is that none of these oprerations changes the original point obtained from the original "get pos" point, and that
the length operationn always from the [0, 0, 0] of the current coordinate system.
In other words the process is:
1) select one or several points (fixpoints) in the X,Z space
2) among these points select the fixpoint which is nearest the point you have from the get operation.
3) Map that fixpoint onto [0, 0, 0]
4) Convert to scalar
Conceptually the mappings that you do to the selected fixpoint, is reverse so the length you measure is mapped back (measured relative) to the
fixpoint.
By mapping I means a transformation of the the coordinate system:
1) Scaling. (multiply a vector by a constant)
2) Moving the origin (subtract two vectors)
3) Rotating the vector about y.
Have fun