> Your IDE will do type checking as well. Yes, the lexer stage will parse all identifiers into 'ID', but to provide a contrived counter-example, what if I do want to allow 'pi' to be assigned based on something that can only be processed in a later stage (perhaps an annotation, or a result of compile time function evaluation, etc). It will push the logic of handling 'ID' to later stages anyway. Now you're handling IDs in at least 2 places - that's strictly worse than the status quo.

Typechecking and access control are clearly semantic analysis considerations, which are better handled in a separate pass on an explicit AST. In particular, they may require a symbol table (i.e. an environment of names) that is not possible to compute during the parsing phase.