“The 1% to 3% of outlier data points that are not mistakes can lead to new scientific discoveries.”