Or can we sum more accurately by taking the stable_mean and multiplying by n?