After more general studies about compilers for statically-typed languages, I'd say that arrow-meta largely resides hooking into the front/middle end of the compiler. I hope I can share an infographic on how I mean in a bit

That is true. I think most people no longer write Compiler Backend Code in the traditional sense, LLVM can do it better anyway. And lexers can even be generated automatically from an LBNF grammar. So all the interesting parts are in the middle section where type checking and language specific optimizations happen.

The question I have is how do we know what grammar is being used for the lex token generation?