VERIFICATION OF THE NEAR REAL TIME ACCOUNTANCY DATA

In previous papers it has been shown that under appropriate assumptions the optimal procedure for verifying the data and establishing the material balance of one inventory period is based on the MUF-D statistic, where MUF is the usual material balance test statistic and D ist the algebraic sum of the differences of the operator's and of the inspector's measurements. In this paper we analyse the problem of verifying the data a.nd establishing the material balances of a sequence of inventory periods. We determine that test procedure which is optimal in the sense of the overall probability of detecting a diversion of nuclear material. In so doing, we account for the internal consistencies of data falsification strategies, as they provide additional detection possibilities. The result of the analysis shows that the MUF-D statistic is no longer optimal, instead one arrives at a linear combination of MUF and D, weighted with appropriate variances, i.e. the Z-statistic introduced some years ago by J. Jaech. This is important for all Near-Real-Time Accountancy (NRTA) studies: Even though NRTA evaluation procedures are primarily developed in order to shorten the detection time, it is necessary to know the best test statistic from the probablity of detection point of view in order to find an evaluation procedure which satisfies an an acceptable compromise between these two criteria.