Issues with baselines, spread, inclusion of irrelevant models, and differing levels of smoothing on the chart shown compound each other and make the purported differences between the models and cherry-picked observations far wider than they are in reality. Tricky.
