@neauoire yes! this is how co works, the source code is natively a bytecode, and there is even a subset of the instruction set reserved specifically for source-only tokens (like newlines, indents, comments, and different types of label definitions) and then the text representation is really just a transformation layer/interface. and then there is also a symbol database integrated to the language tooling, similar to the “symbol file” you mentioned (i imagine?)
@jakintosh ah! very nice :) You could probably free the formatting stuff from the specs and put it in the symbols file! It's really nice to have it all in the symbols LUT, it simplifies things a lot I think.