Fixing what breaks in production.

Reading stack traces, triaging what Sentry sends, and fixing the errors and failing checks that land on a pull request, one problem at a time.