There seems to be some confusion here between 'reliability' in terms of sampling reliability (the degree to which the characteristics measured in the test articles represent those characteristics in the population on the who), versus reliability describing the probability of successful operation (unity minus the probability of failure).
These are two different definitions of reliability, and should not be confused.
A risk analysis can require a component to have very high functional reliability (i.e.: a very low probability of failure) to reduce the risk of harm to an acceptably low number, but the verification test requirements can have a 90% or 99% sample reliability that dictates test sample sizes, or multiples of standard deviations to ensure the population's characteristics are sufficient to deliver that high reliability/ low probability of failure.
There seems to be some confusion here between 'reliability' in terms of sampling reliability (the degree to which the characteristics measured in the test articles represent those characteristics in the population on the who), versus reliability describing the probability of successful operation (unity minus the probability of failure).
These are two different definitions of reliability, and should not be confused.
A risk analysis can require a component to have very high functional reliability (i.e.: a very low probability of failure) to reduce the risk of harm to an acceptably low number, but the verification test requirements can have a 90% or 99% sample reliability that dictates test sample sizes, or multiples of standard deviations to ensure the population's characteristics are sufficient to deliver that high reliability/ low probability of failure.