Repeated Bernoulli trials give rise to Gaussian distributions which is where the e exponential comes from.

This an assumption and an approximation and is not necessarily a good fit. Pulling from actual probabilities would generally perform better.

The rest is massaging to better fit the different objectives.