We derive a Bayesian model to forecast the continuation of a cumulative progression of records. We show that the model compares favourably to past least-square prediction models for records in running events. We also check the validity of the model for the case where we have access to the underlying attempts, using data from athletic events in the Olympics.