This research investigates the soil cutting problem with a flat blade excavating soil. Experiments identify and characterize oscillatory soil failure patterns, based on data including forces on the robotically-controlled excavation tool, as well as granular soil flows tracked via a high-speed camera and a novel computer vision-based technique. In the experimental results we observed that the soil failure surface oscillates between being curved concave down and concave up; this may be attributable to transient soil hardening and softening behaviors. The oscillations are more prominent in denser soil, and there is apparent correspondence between shifts in failure surfaces with rises and falls in force data. Numerical simulations are also presented, based on a hybrid model that combines a modified McKyes soil cutting model with a position based method (PBM) to model surcharge particles, and which achieves real-time simulation performance. We incorporated the observed dynamic soil failure patterns in this simulation model to improve its realism for the investigated soil/machine interaction scenario. The parameters of these simulations, including internal and external friction angle, an empirical surcharge factor, as well as simulation hyper-parameters, are optimized using a gradient-free Downhill Simplex method to find a best fit to the average trend in the experimental data. Mean absolute percentage errors below 10% are typically achieved, demonstrating accurate and efficient performance appropriate for engineering applications.