Why the mix of late and nullable?

platform and processManager are nullable because I set them in createRunner. I set them in createRunner since they depend on variables that can change between tests. I then explicitly null it out in the tearDown (a similar test had a bug where test assets were re-used between tests, so I started explicitly nulling out fields after tests that were not guaranteed to be initialized in setUp()).

I've had a lot of problems with using setup helpers that close over part of the test declarations themselves, it gets very hard to track the active state. I would suggest some modifications:

move createRunner to be a top level, so that you need to explicitly pass it all values its using. You could also return some sort of test context object which has fields initialized to what you want them to be. That should let you remove the teardown logic and nullables.

I do not think you should be defensively programming around test configuration problems by nulling things out

that makes sense to me.

removed all side effects from createRunner(), and instead initialized processManager and platform directly in the test, PTAL