I think you are right, though in my experience in these situations is that defining the exact behavior is very hard. In not many cases was the messy code written messy from the start, it's usually an accumulation of changes (to behavior) - which are very hard to figure out.

I've found it's also hard to test code that isn't modular and split up: before you can test anything you need to mock 2 database connections, one external API endpoint and reading/writing from 3 different folder locations. Either that or you split them up yourself and risk breaking something.