 Natural frequencies in probabilistic reasoning

Counts preserving how cases from a common population fall into nested classes and evidence subsets. Bayesian inference can then use a ratio of joint counts, $n(H\cap E)/n(E)$. For uniform sampling from those cases, conditional probability is the number satisfying both class and evidence divided by the number satisfying the evidence, so the ratio follows directly without reconstructing separately normalized rates. Separately normalizing each class to the same artificial denominator does not preserve the original prevalence automatically.