Time to completion (in appendix A9) should be treated as log-normally distributed, or by some other one-sided distribution because one cannot complete the task

Time to completion (in appendix A9) should be treated as log-normally distributed, or by some other one-sided distribution because one cannot complete the task faster than 0 seconds. This transformation will rule out confidence ranges with negative time. BTW, log-normal distribution tend to produces events P(x>E(X)+d) more frequently than events P(x<E(X)-d). If one needs reasons why software projects often late, this is one of them.