I like the idea of levels of trust, perhaps determined by algo and then manually adjustable.