I agree that we'll eventually want to formalize munging. However, whenever that happens, it should still happen after the AST is constructed. Munging shouldn't happen at all if you're doing refactoring or the like, where you query the AST. Munging is code generation concern.
Prior to my patch, the AST needs to support both :name and :name-sym keys to get around the fact that the :name key loses fidelity of the original symbol. This complects the Clojure-specific parse phase with target-specific needs. The patch renames :name-sym to replace :name.
I read Raphael's notes and he and I have a email thread going on too. With that in mind, this is a step towards untangling the architectural layers of the compiler. By better segregating parse from emit, it will be easier to break the analyzer out from the code generation backend.