Isn't that just an argument for always doing open book tests, then? Seems like there's no downside, and as already mentioned, it's closer to how one works in the real world.