I thought this might be of general interest. I leveraged pyright's ability to generate type coverage scores of the most downloaded PyPI packages (I used August top 1000). A few are missing, because they either failed to pip install on my MacBook or they are native code. I also looked for type stubs, looking first in typeshed, then for a '<package>-stubs' package, then 'types-<package>', and included the version number of that package, if found. Overall, here was the distribution of scores for packages with py.typed: [image: image.png] And without: [image: image.png] You can see the detailed results at typescore/latest-scores.md at main · gramster/typescore (github.com) <https://github.com/gramster/typescore/blob/main/latest-scores.md> Thanks Graham