Proposal for an Implementable Toy Model of Informed Oversight — AI Alignment Forum