Background: In neutrino oscillation experiments, the hadrons created in neutrino-nucleus collisions are becoming important observables. The description of final-state interactions (FSI) of hadrons with nuclei in the large phase space probed in these experiments poses a great challenge. In the analysis of neutrino experiments, which operate under semi-inclusive conditions, cascade models are commonly used for this task. The description of FSI under exclusive conditions on the other hand can be treated successfully by using relativistic optical potentials (ROPs). Purpose: We formulate conditions under which the ROP approach and cascade model can be directly compared. Through this comparison the treatment of FSI in cascade models is studied and benchmarked. Method: We study single proton knockout with data from the T2K experiment's near-detector muon neutrino flux. We feed the neut cascade model with events distributed according to the cross section of a relativistic distorted-wave impulse approximation (RDWIA) calculation that uses the real part of an optical potential (rROP). We impose cuts on the missing energy of the resulting events to define a set of events which undergo only elastic FSI; these can be compared to RDWIA calculations with the full optical potential. Results: The neut cascade and ROP give similar cross sections for proton kinetic energies Tp>150MeV for carbon, oxygen, and calcium nuclei. A necessary condition is that a realistic nuclear density is used to introduce events in the cascade. For Tp<100MeV the ROP and neut cross sections diverge strongly in shape, and differences in magnitude are larger than 50%. Data of transverse kinematic imbalance allow us to discriminate between different approaches to FSI, in particular the large δαT region is sensitive to the presence of nonelastic FSI. Due to experimental errors and a large nonquasielastic contribution, the comparison to T2K data does not give an unambiguous view of FSI. We discuss electron scattering data and provide results for kinematics covered in the e4ν analysis. We argue that with a simple cut in missing energy FSI can be studied with minimal confounding factors. Conclusions: The agreement of the ROP and neut cascade under T2K conditions lends confidence to these models as a tool in neutrino oscillation analyses for sufficiently large nucleon kinetic energies. Our results urge caution when applying a cascade model for small nucleon energies. The assessment of model assumptions relevant to this region is strongly encouraged. The approach presented in this paper provides novel constraints on cascade models from proton-nucleus scattering and can be easily applied to other neutrino event generators.