If it was done the initial position/direction in and current position/direction out, then I guess you could always use vector math to add offsets using let’s say keyframed vector inputs. You could either add the keyframed value before it goes into initial position, to animate the gizmo before it gets transformed by the user, or you could add it to the current position, to animate it after it was transformed by user.
I still just can’t figure out what technically necessitates the output being on the input side and requiring reverse direction flow.
EDIT: Visual example:
This:
VS something like this:
And I guess the node would eventually also take single value bool to toggle whether gizmo is visible/active.