Sooner or later all SNO’s will face the fact that a node is under performing or making bad hard drive noises or showing other hints of imminent failure. As a SNO the data on that node did take some time and effort to collect, just letting all of it go back onto the network might feel hard for someone. Of course moving the node to new hardware is an option, but also a process with a lot of potential failure modes that will ultimately end in total loss of the data anyway. I bet a good percentage of the data on any node will be eligible for transfer to a node behind the same public IP or operated by the same SNO.
As of now I understand graceful exit will pick the nodes to offload the data of the retiring nodes to.
The idea is that a SNO can list a number of nodes that would be preferred for the data to offload when performing graceful exit (as long as that data is eligible for one of the listed nodes). So that when performing a graceful exit most of the data can be kept by the SNO and not given back to random nodes on the network, unless it is needed due to network storage policies.