[en] It is argued that a global calculation scheme for accelerator-driven systems should consist of a combination of intranuclear cascade codes above 150 MeV and evaluated data libraries for energies up to 150 MeV. A recent promising development is the integration of the two nuclear data methods (intranuclear cascade + data files/transport) into one code system (MCNPX). This will allow maximal flexibility concerning the choice of intranuclear cascade models and data files. It is evident that the total nuclear data task is quite sizeable: one has to keep track of nuclear processes starting from 1-1.5 GeV incident protons down to the final reaction stages with thermalizing neutrons and gamma-ray cascades for a large number of residual nuclides. Along this trajectory, a whole panoply of different appropriate nuclear models applies, each with their own validity range. In addition, the required physical methods may differ from nuclide to nuclide. With regard to the tractability of the work, it is important to observe that certain parts of the accelerator-driven system may depend more critically on the quality of nuclear data, and therefore deserve more attention, than other parts. In addition, the data requirements may differ per material. It can easily be imagined that for a target material like lead, the number of produced neutrons per incident proton is a key quantity, whereas for a shielding material like iron, the secondary particle energy-angle distribution is more relevant (for economical aspects of the shielding) as well as activation cross sections. These simple considerations show that sensitivity studies for the different parts of the whole device are indispensable as a parallel area of research in the field of intermediate energy nuclear data. They should provide a valuable guideline of isotopes and reactions to be measured and evaluated. 53 refs