Myself and regarding to the Internet many folks are using NPT (network prefix translation) and translate (static) ULA to a dynamic GUA prefix.
Wireguard is a layer-3 device and the original author has some rejections why IPv6 is not being treated well with Wireguard.
If you connect via Wireguard you will establish the connection on the wan IP of the router.
But the address of the tunnel comes into play later.
To avoid a (masquerade) nat for the WG clients, you either needed (static) GUA addresses for the WG Interfaces on the router and the client or prefix translation, which is not really a "classical" nat.
In case you are not aware of it, Linux usually can be addressed via all addresses on all interfaces regarding if the address is assigned on the incoming interface or not. To simplify the behavior. Devil is in the details but usually if you throw a packet to a Linux box and it has the address somewhere assigned the packet gets processed...