Well, statistics do have that degree of randomness in them. If your measured values are way off from what the code suggests it should be, it's probably just a matter of you having much too small of a sample space to get and accurate reading of the real probabilities.
Consider the error of a sample space of 1. Event A happens 90% of the time, and event B happens 10% of the time, according to the code. If you try once, and get B, and then conclude that B happens 100% of the time, based on your findings, well..., maybe you need to try a larger sample space. In fact, since 9/10 is a reduced fraction, you can't actually get the real answer without a sample space of at least 10, and even then it's unlikely to be exact for most trials. You'd need a much larger sample space to get an approximation to within 1/10 of the real answer consistently.
