Predicting soot production in industrial systems using an LES approach represents a great challenge. Besides the complexity in modeling the multi-scale physicochemical soot processes and their interaction with turbulence, the validation of newly developed models is critical under turbulent conditions. This work illustrates the difficulties in evaluating model performances specific to soot prediction in turbulent flames by considering soot production in an aero-engine combustor. It is proven that soot production occurs only for scarce local gaseous conditions. Therefore, to obtain a statistical representation of such rare soot events, massive CPU resources would be required. For this reason, evaluating soot model performances based on parametric studies, i.e., multiple simulations, as classically done for purely gaseous flames, is CPU high-demanding for sooting flames. Then, a new strategy to investigate modeling impact on the solid phase is proposed. It is based on a unique simulation, where the set of equations describing the solid phase are duplicated. One set accounts for the reference model, while the other set is treated with the model under the scope. Assuming neglected solid phase retro-coupling on the gas phase, the soot scalars from both sets experience the same unique temporal and spatial gas phase evolution isolating the soot model effects from the uncertainties on gaseous models and numerical sensitivities. Finally, the strategy capability is proven by investigating the contribution of the soot subgrid intermittency model to the prediction of soot production in the DLR burner.