My thoughts are similar to yours and to Michael's on this Roger.
It seems that some credit ought to be given for near-misses. Perhaps some
sort of exponentially decaying function outside the targeted region would be
better than a linear function of score with distance.
After all, isn't it the responsibility of the predictor to set parameters (including
closeness) for the prediction. If a predictor says his event is likely to occur
within 2 degrees (222.222 km), fine, the probability of occurrence can
be determined from that. How will it benefit the predictor to get 10 points
for an event that occurs in New Madrid when the prediction is for an earthquake
in Mexico. How does this benefit anyone for that matter. All it does is
muddy the waters.
I was under the impression that "syzygy" folks claim to be able to tell when an earthquake
will occur, but not where. I don't see how this discussion is very relevant to
their particular set of problems.