Something I keep think about every now and then is that there has to be some proper way to evaluate 'near misses' statistically. I have ideas, but haven't tried to implement it mathematically.

Basically, some sort of falloff gradient from the 100% hit boundary down to some reasonable distance as 0%. It would affect the expected probability. It should work for location, magnitude, and time windows. The 'problem' is determining an appropriate fall off curve.

But I already realize that defining such a curve is not that simple. It changes depending on all the parameters of the prediction and the resulting quake. Thus, the curve is different for every prediction and possible hit. For example, a prediction for a 4+ quake, the resulting curve depends on the magnitude of the quake. A larger quake falls off more slowly since the larger quake has an effect further away.

But it's still more complex than that.

Once all the dust settles after the new years maybe we can tackle this idea?

Brian