I wonder here if the confusion is because it sounds like it is less about semantic meaning and more about a kind of type safety. Tully’s story is like this - making absolutely sure you can’t execute if you call the wrong thing. Georg’s statement is also true - the advertised name is just as much a helper in guiding you in correct wiring but it doesn’t give you type safety and is probably more closely aligned to what people think of as semantics than the type name itself.
For some cases, like the image example, the line between relying on type name vs topic/service name is neither black or white. It is definitely half of one and half of the other.