While the example above shows our program correctly giving higher probability to the correct user over other users, these probabilities are still not far enough apart to be conclusive for many applications. We strongly suspect that the main limitation on the effectiveness of this program in distinguishing users unambiguously lies in the gathering of the raw data from the X server. The X server reports keystroke latencies to a granularity of only 10ms, and when typing is very rapid, it truncates them to zero. A granularity of 1ms would provide more accurate results.