It's more of an experiment. You can detect bodies entering Areas but Areas contain no provision for saying where in its volume the intruder is. So I played a little with using kinematic rigid bodies for that purpose which is more of a dead end at least for my purposes. (They do detect intruders, but the location can be on the 'far' side wrt to the rigid-body's center, reflecting the purpose of collision detection in bullet. KinematicBody has move and collide but the call is SLOW, worse is test_move.
However, if the modality of a rigid-body is kinematic, there "should be" every reason why it be like any other dumb spatial and have its global transform be a product of its parent and local transforms.
And I can conjure an odd use case where you want your parent body to be physics driven but want it to control kinematics in order to "break" pure physics relations. The kinematics can do collision-y things to the world but not effect the parent. Call it ninja magic.