Slippery indeed. How about this concept. Given a prediction, we take the prediction's minimum magnitude (Martin's would be 3.0) and we apply it to each entire candidate catalog (or maybe a 100 mile radius). The first occurance in that catalog of the minimum predicted Magnitude becomes the starting historical date for the analysis. The ending historical date is the current date (or as close as the catalog gets to it due to update lag), or if the catalog states that it had an ending date, then use the stated ending date. The idea behind this approach is that if a 3.0 is found within 100 miles from what your predicting in a given catalog, then it pretty much proves that the method used to build that catalog was able to detect an eq of the magnitude your trying to predict.

What do you think?