The SDG, by design, produces random output on each run. Because of this random behavior, you may find that the output doesn't precisely match the population characteristics you have defined in your configuration files. The probabilistic model is part of the fundamental design of the SDG, and it actually makes configuration (and data generation) simpler. The alternative is a very prescriptive approach where one would have to configure precise counts on every last configurable item to model a target population. The result would be a combinatoric explosion of configuration on any complex population.
Generally you can expect this "imprecision" to affect any configuration element called "Probability" or "Frequency."