The copy doesn't make it clear enough, but we're actually doing our best to adjust for small samples with a little empirical Bayesian magic. As the number of samples (inspections) goes down, the result is closer to the mean for the whole city.