I actually wrote a post about this, using a metric for expected outcomes instead of ev probability

https://tompollak.me/blog/ev-v-eg/