We present a practical characterization of statistical consistency for differentiable surrogate losses