Cross sections, benchmarks, etc.: what is data testing all about

In order to determine the consistency of two distinct measurements of a physical quantity, the discrepancy d between the two should be compared with its own standard deviation, σ = √ (σ12+σ22). To properly test a given cross-section library by a set of benchmark (integral) measurements, the quantity corresponding to (d/σ)2 is the quadratic form d+C-1d. Here d is the vector of which the components are the discrepancies between the calculated values of the integral parameters and their corresponding measured values, and C is the uncertainty matrix of these discrepancies. This quadratic form is the only true measure of the joint consistency of the library and benchmarks. On the other hand, the very matrix C is essentially all one needs to adjust the library by the benchmarks. Therefore, any argument against adjustment simultaneously disqualifies all serious attempts to test cross-section libraries against integral benchmarks. (author)