Uncertainty in long-term projections
also influenced our proposal. Given the
probabilistic nature of performance
assessments, it is possible that some
combinations of parameter values will
result in very high doses, even if such
combinations have an extremely low
probability of occurring. Although there
may be only a few results that are very
high, extreme results have the potential
to exert a strong influence on the
arithmetic mean, which could make the
mean less representative of all
performance projections. This
possibility may be increased by the
introduction of additional, and possible
excessive, conservatisms as a way to
account for uncertainties. We expressed
a preference for a statistical measure
that would not be strongly affected by
either very high- or low-end estimates,
believing it appropriate to focus on the
‘‘central tendency’’ of the distribution,
where the bulk of the results might be
expected to be found. We proposed the
median of the distribution as being most
representative of central tendency.
Because it is always located at the point
where half the distribution is higher and
half lower, the median depends only on
the relative nature of the distribution,
rather than the absolute calculated
values. Given our concerns about
specifying a peak dose compliance
value against which performance would
be judged for a period up to 1 million
years, we believed the median might
also provide a reasonable test of longterm
performance. Today’s final rule
departs from the proposal by adopting
the arithmetic mean as the statistical
measure of compliance to be applied at
all times, as explained in section III.A.9
of this document (‘‘How Will NRC Judge
Compliance?’’).
Our consideration of FEPs also was
affected to some extent by uncertainty,
as well as by conclusions of the NAS
committee. In our proposal, the overall
probability threshold for inclusion of
FEPs remained the same as in the 2001
rule, which we believe provides a very
inclusive initial screen that captures
both major and minor factors potentially
affecting performance. Uncertainty
plays a role in the sense that very
gradual or infrequent processes and
events may begin to influence
performance only at times in the
hundreds of thousands of years, when
the overall uncertainty of assessments is
increasing. The additional uncertainty
introduced by these slow-acting FEPs
led us to propose the exclusion of FEPs
if they were not significant to the
assessments in the initial 10,000 years.
We believed this would still provide for
robust assessments that would address
the factors of most importance over the
entire 1 million-year period. We did
consider in our proposal whether
significant FEPs might not be captured
using this approach. In evaluating
whether excluded FEPs might become
more probable or more significant after
10,000 years, and therefore should not
be eliminated, we identified general
corrosion as a FEP that is certain to
occur and represents a significant
failure mechanism at longer times, even
though it is less significant in the initial
10,000 years.