Nature quake paper accused of boosting accuracy by mixing training, testing data (theregister.co.uk)