Statistical analysis may help answer some intriguing questions in athletics, such as when the current world records will be improved. Sport records are extreme observations, which can be analyzed through extreme value theory. However, modeling is only one part of the problem, since estimation is also troubled by small sample issues. Here, we present some improved estimates of the expected time to break the record. The property needed for this task is probabilistic calibration. Bootstrap-based approaches can help assess and recover this property to improve predictions. We show that, thanks to improved estimates, the near future is richer in new records than suggested by the classical estimates.
