Two decades ago, the United States issued the first patent claiming the idea of an automated iris biometrics system. Today, multiple companies offer commercial biometrics products. However, there are still many unanswered questions about the workings of the iris biometrics algorithms. Before iris recognition systems are more widely employed, we must ask, ``when do these algorithms fail?' Previous research has assumed that all parts of an iris code are equally valuable. Alternatively, some researchers claim that parts of the iris are more valuable, but they still use the same portions of the iris for all subjects. No previous researcher has attempted to experimentally determine how different parts of a particular subject's iris code may be more or less valuable. I obtained multiple images of 24 subjects' eyes to study the reliability of individual bits in the iris codes. I develop a theoretical explanation of the sources of inconsistencies, based on the coarse quantization of complex coefficients in creating the iris code. Another source of inconsistency in the iris code is dilation of the iris. The majority of iris research ignores the degree of dilation in processing iris images for biometric purposes. I experimentally quantify how much the Hamming distance is affected by iris dilation.