This is really poor test though, of course the most recently trained model knows the newest libraries or knows that a library was renamed.

Not disputing it's best at reasoning but you need a different test for that.