Comment by _zoltan_

I think the distinction is when the model decides to read code top to bottom vs when the model chooses to parse code indirectly to save on tokens, there's also AST tooling to let the model see project structure.