It's true that python has support for typing, and I do find that mypy helps produce better python code. But I think a meaningful issue is that python's rich ec

It's true that python has support for typing, and I do find that mypy helps produce better python code. But I think a meaningful issue is that python's rich ecosystem of datascience and ML tooling often is at odds with meaningful type annotations. Roughly, you can end up with signatures that indicate that a value is a DataFrame or an ndarray or whatever, but there are a bunch of implicit assumptions on what columns are defined, or what how the shapes of two ndarrays line up, etc. It's easy for a codebase to end up paying the upfront cost of providing annotations, but without actually getting an improved ability to reason about or refactor code.