I read it as after requirements and before code. This is how I would expect it to work too. It reads the existing codebase, which also makes sense for non greenfield projects as in the example.