Analyzing past records to predict maximum speeds for dogs, thoroughbred horses, and humans. The headline plateau record (the 'Denny Limit', from the paper's author Mark Denny): 9.48s for the 100m sprint, which is both very impressive and -as all explicit limits- bothersome. Why not 9.3s? What do we have to do to make it a cool 9s?
On one hand, we are not the first epoch that attempts to improve physical performance (or any other sort), nor are we the first ones to use then-current science and technology to do it. The appropriate null hypothesis might very well be that increasing scientific resources provide decreasing gains in performance, so while performance will keep improving, it'll do so with an eventually clear lower bound.
Nonetheless, I can't but think that this might not necessarily be the case. Engineering progressed in almost a linear fashion for thousands of years, but eventually knowledge of physics (most importantly, quantitative, mathematical knowledge of physics) opened the doors for a completely different rate of progress. We seem to be getting close to the point where human performance across the board becomes a matter of engineering, which would make models like the one in the article inapplicable.
Most of the time, the null hypothesis of sustained long-term tendencies is the best analytical framework. But 'most of the time' is becoming rarer these days, isn't it?