Sensors work to select the same node/element from one load case to the next (as long as there is no geometry change the mesh should remain the same) within the same study - especially useful if you need pretty pictures. However, if the mesh changes the old sensors are useless (there's a drop-down box to pick the most up to date set - IDK how to get rid of the useless ones).
One 'best-practice' that I utilize is to ONLY copy the primary 'tab' to run a new LC. (I usually do a 1G 'does it behave as I expect' load case initially.) It seems that as with a part-from-a-part-from-a-part-from-a-part gets corrupted so do LCs.
If you learn something better I would be interested.