To evaluate regardless of whether the resulting network at step four is indeed a winning ticket, train the pruned, untrained network and examine its convergence behavior and accuracy.