[en] To define the requirements placed by the evaluation of nuclear cross sections upon the reporting of experimental results, a model of part of the evaluation process is presented. The model is a straightforward application of nondiagonal weighted least-squares estimation to average cross sections in the energy regions where the shape of the cross section is not given by theory. To combine in a logical way the existing evaluated information with one or more new sets of experimental results, the estimated covariance matrix of each experimenter's results needs to be known on an appropriate mesh. The likelihood that each experimenter may underestimate the uncertainties in his results does not remove the need for him to record for users the estimated magnitudes and correlation patterns of these uncertainties