An event that, given the current model, is a surprise at the time of its occurrence and has a big impact, but is subsequently rationalized by a reformulation of the model (e.g. by assuming errors with a heavy-tailed distribution in place of a normal distribution).