I've always wondered why a 100ms vs 1s startup time mattered when initialization had several longer steps, this is a much more useful metric