Discretized probability distribution of the log-odds score under the null (background) model. Built once per motif, then queried for each observed score to obtain a p-value.