Heh. It wouldn't be perfect. But if the system of N people has C < N people who will make or approve corrupt decisions, and each person had to have 1 randomly chosen person sign off on each possibly corrupt decision, you go from C/N chance of a randomly chosen decision being a corrupt one to a (C/N)^2 chance. If the system has a relatively low percentage of corrupt people, that could be a big win.

If C is too large, you could improve the scheme by adding another sign-off to increase it to (C/N)^3, as well as by punishing the original requester and anybody who approves along the way when a corrupt decision is caught.