We put forward an adaptive alpha which changes with the amount of sample information. This calibration may be interpreted as a Bayes–non-Bayes compromise, and leads to statistical consistency. The calibration can also be ...