I'm not sure to fully understand your idea.
For the 4th part in your whitepaper, the difficulty function f(l) must be sensitive enough to l to satisfy E2(m)<E1(m), which means that f(max) must be much larger than f(min). How is f(l) designed?
I really don't understand the selfish mining attack in your whitepaper, and I don’t understand how the 1/m^2 was derived?
I think it can be understood like this.
when two miners do not cooperate in mining, their luck l1 will be better, and the mining difficulty f(l1) will be lower;
when they cooperate, their luck l2 will be worse, The mining difficulty f(l2) is higher.
Although the power of their cooperation is *2, when f(l2)>2*f(l1), it can satisfy E2<E1.
From another perspective, the variance of cooperation is greater than that of non-cooperation. one-cpu-one-vote is the result.