Most programming languages are defined in terms of program text, Lisp is not. Lisp is defined in terms of in-memory structures and what happens if you pass these in-memory structures to a hypothetic of actually existing function "eval".

With this way of viewing things your "( is actually a command" no longer is useful/ no longer makes sense. (I wouldn't say you're completely wrong, though, just that another view of things seems simpler to me.)

You could probably implement a Lisp with a scanner and a parser but things are a lot easier if you follow the traditional read-eval route (or preferably read-macroexpand-eval).