I was typing up the same comment when your hit.

Kelly criterion is a great mental framework, but the math obviously doesn't work out for VC placements. The best alternative I've been able to come up with is to brute force / simulate thousands of different portfolio outcomes and optimize from there.