It seems to me that the predictor needs to use some common sense here.
The method of evaluation that Roger uses is quite legitimate and allows for
near-misses. In this case, Don would have been much better off (at least on
paper) to have started out with a 10 km radius if he thought the event would
occur at Santa Rosa. Then, if the event occurred say 25 km from Santa Rosa,
the predictor would still get a lower probability than if he had made a forecast
for 35 km. Likewise with magnitude. If the predictor intends to predict an event
of Ml 2.5-3.0 including the limits from 2.0-2.5 merely serves to increase the
chance that the event will occur randomly. In other words, a narrower window
in each field will allow for lower probabilities and since the evaluation allows for
near-misses as you go away from the forecast (but does not decrease th
probability as the event occurs nearer the stated parameters), the predictor
loses nothing and gains much by making a very tightly constrained forecast -
provided an earthquake does occur.
It seems it is up to the predictor to adjust.