The Nuclear data Evaluation Pipeline of Uppsala University (NEPU) has been developed to perform reproducible data driven nuclear data evaluations. Until now, the pipeline has been used to perform evaluation above the resonance regions for structural materials. Cross-sections for neutron reactions of neutron energies in the range 100 keV to 5-10 MeV in structural materials fluctuate due to resonances in the excited states of the compound system.In this work, we describe the changes we have made to NEPU to allow it also to include energy ranges of cross-sections for neutron energies ranging from 1 to 5 MeV for Fe-56. NEPU uses TALYS. TALYS relies on the assumption that resonance fluctuations are averaged out. This assumption can be described as a model defect in the intermediate incident neutron energy region, where a large number of overlapping resonances causes the cross-section to fluctuate around the energy averaged value. We show how we have used Gaussian process regression to describe the model defects of TALYS to include the cross-section fluctuations in our evaluation pipeline.In addition to co-variance matrixes, NEPU produces random files. Each set of random files incorporate information of for example cross-sections and angular dependencies of the various reaction channels. The distribution of the sets of random files is capable of reflecting more complex covariances than the pure covariance matrixes. In addition, the random files can be used in the Total Monte Carlo technique.We will demonstrate the results of this work by showing how the models agree with nuclear data from the EXFOR database, after the model defect correction has been applied.