I wonder about the resolution: was snorm() reimplemented correctly, or was an RNG with the 'wrong' interval supplied?