I think a prediction should be a hit or miss depending on the specific parameters stated in the prediction. That's the only way to keep a predictor's record honest. If he/she can't set specific parameters for a particular prediction, then he/she shouldn't be trying to predict earthquakes. The testing of methodology should occur prior to actual prediction. For example, if someone wanted to use animals to predict earthquakes, he/she should establish a correlation of anomalous animal behavior before using that as a basis to predict earthquakes. In establishing correlations, a best-fit analysis could be beneficial in assessing the quality of a correlation record.