The key thing about the Suffx Substitution Closure property is that it does not make any explicit reference to the automaton that recognizes the language.
While the argument that establishes it is based on the properties of a Myhill graph that we know must exist, those properties are properties of Myhill graphs in general and don't depend on the speci?cs of that particular graph. This lets us reason about the strings in an SL2 language without having to actually produce the automaton that recognizes it. Perhaps more importantly, it lets us establish that a particular language is not SL2 by supposing (counterfactually) that it was SL2 and showing that Suffx Substitution Closure would then imply that it included strings that it should not.