Comment by cheesecakegood

In theory Jev “calibrates” the probabilities, meaning a probability of 20% is optimized to happen near 20% of the time, which as you point out traditional models with schemas do NOT optimize for at all