A Large Eddy-Simulation code, CUFLOW, was improved to simulate fluid flow in a full-scale commercial steel caster including the effects of ruler EMBr with a realistic conducting steel shell. The model was successfully validated with measurements made in a GaInSn physical model 7 and then was applied to explore the flow behaviors in greater detail. The corresponding full sized caster was studied at conditions similar to industrial operations; however the submergence depth was deep to match the GaInSn model in order to assess the model scaling criterion using the Stuart number. Large scale jet wobble and transient asymmetric flow in the mold, which was found with insulated walls, did not occur with a realistic conducting shell for otherwise identical conditions. With conducting walls, the flow was stable and quickly achieved a symmetrical flow pattern, which featured three counter-rotating loops in the upper region and top surface flow towards the SEN. Thus, it is essential to include the effect of the conducting shell when studying transient mold flow with a magnetic field. Relative to no EMBr, the ruler magnetic brake across the nozzle deflects the jets upwards, from ∼30° down to only ∼10° down. This strengthens the flow in the upper region and increases the top surface velocity from NF to SEN, from 0.07m/s to 0.25 m/s in the real caster. The weak upper recirculation region without EMBr becomes more complex, with three distinct recirculation loops, which features upward flows along both the NF and the SEN. The momentum from these flows raises the surface level near both the NF and SEN, and generates more level fluctuations in these two regions. The lower recirculation region becomes a very small elongated loop just below the jet (which matches the small loop just above the jet). Flow below this small recirculation loop aligns quickly to the casting direction. The lower velocities should be beneficial for lessening the penetration and entrapment of bubbles and inclusions. The Stuart number similarity criterion enables a close match of both the time-averaged mold flow pattern (qualitative) and velocities (quantitative) between the 1/6-scale model and the real caster. However, the scaled surface level profile and its time fluctuations were both larger in the scale model, as expected. Thus, it is better to maintain both Froude number and Stuart number similarity conditions.