I have a detached garage with its own subpanel. The garage has an exterior light fed from this panel, controlled by 3-way switches in the garage and house; the 3-way in the house shares a plastic 2-gang box with another switch that controls the hallway lighting. My question is whether or not this 3-way switch in the house is properly grounded:
The garage sub has its own grounding electrode and there are no metallic connections between the house and garage (other than the sub feeder and the wires for this switch, which are buried in separate PVC conduits). The feeder ground is bonded to the ground in the main panel, and is properly isolated from neutral in the sub. There is an EGC pulled in the switch conduit which is bonded directly to the ground bar in the sub (both conduits enter directly into the panel, the switch wiring is spliced inside) - the other end is tied to the metallic J-box that terminates the conduit inside the house. From that box, 14/3 Romex completes the run to the switch, and the EGC in that cable bonds the switch yoke to the J-box. This switch is NOT bonded to the other switch in the wall box, which is fed from the main panel within the house. It seems to me that this creates the possibility for these two switches to be on different ground planes if the subpanel or the feeder wiring were to become damaged. I believe the correct thing to do would be to tie the adjacent switch yokes together and break the bond with the garage panel at the J-box (such that the EGC coming from the garage would be left "dead-ended" to protect the wiring in the conduit) but I'm having a hard time finding anything that either supports or prohibits this approach in the NEC. Garage subs are a common "add-on", so I'd imagine this issue must come up frequently along with them; can anyone confirm the correct approach to this?
I cannot give a code reference at the moment, but it has always been my practice to tie all grounds in every box together. The goal is to establish a grounding "network", a level of grounding redundancy. Someone here will be along to state the code article I'm sure, but in any case, you'd be right to tie the grounds together.