A limit-computable, self-reflective distribution — AI Alignment Forum