That's only true for greenfield projects where people start a new project with types in mind. It's absolutely a nightmare for old projects because, without the need to write types, people write all kinds of code that cannot fit within what's possible in python's type annotations.

In my opinion it's a bad practice, and rewriting code to be typeable is a good idea for refactoring.

But I write Python for some time now, and I know what you mean. I have nightmares about codebases with dynamically generated class fields for example (though I heard ruby is even worse)

It would be nice if python would show types in the documentation. Not only do I need that all the time, it would show python was taking type safety serious.

For example, knowing the return type of a function is Union[DataFrame,Series] rather than simply DataFrame would save a lot of bad errors.

Starlark, unfortunately, does not really support (Python style) types yet. Facebook's version has some kind of types, but ideally Starlark would just learn to do mypy types.