> From my perspective, the main problem with using type annotations is that there's nothing checking them in CI.

Yeah, fair concern. In fact I think I'm on video (from PyCon 2018) warning everyone not to do that in their codebases, because what you really don't want is a bunch of annotations that have gradually accumulated falsehoods as the code has changed around them.

Still, I think from "some annotations + no checking" the good equilibrium to land in "some annotations + checking", not "no annotations + no checking". (I do mean "some" -- I don't predict we'll ever go sweep all over adding them.) And I think the highest-probability way to get there is to let them continue to accumulate where people occasionally add them in new/revised code... because that holds a door open for someone to step up to start checking them, and then to do the work to make that part of CI. (That someone might even be me! But I can think of plenty of other likely folks to do it.)

If we accumulated quite a lot of them and nobody had yet stepped up to make checking happen, I'd worry.  But with the smattering we currently have, I think that point is far off.