This is surprising. I don't know if there's any harm to it aside from making it harder to precisely test BFCache restore's interaction with AX - ideally we would be able to test that there are no events between loadStop (prior to entering bfcache) and loadStart on restore. Or maybe we expect some focusChanged and loadStart and should test for that.