How are accuracies of runs with different shuffles distributed?

Question

I set up a network to learn Fashion MNIST in the style of Hands On Machine Learning, page 298. When I ran the code multiple times, the accuracy was slightly different each t