Whether or not you're doing science doesn't depend on which software tools you use, it depends on whether you're advancing the frontier of human knowledge. If you're producing publishable scientific research, you're doing science. If you aren't, you're probably not doing science.

I don't think the article is saying otherwise, it's just highlighting a similarity between what type systems do, and the sort of reasoning we associate with science.

I prefer to call them guild navigators.

This isn't a reply to your substantive point (of course I agree that fortran/numpy programmers do science), but certainly when I'm coding with numpy I have a mental model of the shapes of the arrays, very akin to the mental model I have of the types of objects in e.g. Haskell or Python. Sure, the shapes are runtime properties and they're not enforced anywhere, but that doesn't matter to my mental model of what's going on.