Michael,
Of course, you are welcome to use anything I say on this board or elsewhere as you
wish. This is a public forum, I can't control what happens to statements I (or anyone else)
makes in a public forum.
I think your ideas of allowing predictions to be made public AFTER the window
has past is a good middle ground and should protect the EPR from litigation and
other problems. This is essentially our position - that if a predictor requests a
prediction be made public, we will wait until the window has past and will only
be able to verify or not verify anything the predictor says to the media.
If the predictor asks that we open his prediction file, information about any
prior prediction is available to anyone who requests it - including the press.
We do not act as a publicist for any predictor, all we can do is verify and state
our evaluation of a prediction or set of predictions. Fortunately, since we keep
a low profile, this has not yet been tested in any real way.
As to your request regarding how we evaluate predictions - here goes:

I will illustrate our process by using a prediction that Don posted on this board
On April 10. I assume Don will not mind since everyone who reads this board
KNOWS that Don has a perfect record with his predictions for earthquakes
in San Francisco (none right). So let's see what our evaluation of this prediction
says about what we know. The forecast as posted was:

San Francisco Earthquake 04/23/01
Posted by Don in Hollister on April 10, 2001 at 12:25:03:

Magnitude: 2.5 to 4.5+
From: 04/21/01PST - To: 04/25/01
Location: San Francisco
Lat: 37.9N - Long: 122.7W - Range: 32

Hi All. Let's just call it a gut feeling. It's been to quiet for my liking.
Take Care...Don in creepy town.

The prediction is evaluated on each of it's particulars. First
we look at the location information. In this case, Don has done
a good job of specifying the expected location of this event. The
exception is that the range is given as 32. 32 whats? km, miles, degrees?
Well, we will forgive Don this oversight and assume he meant km.
This gives roughly an area bounded by 38.2N-37.6N and 122.35-123.05W.
The next step is to define the CURRENT seismicity of this region without
regard to forecast, with only the data available at the time that the
forecast was made. We generally use the 7-month period (arbitrary) prior to
the forecast as a background period to determine the current seismicity
rate at the time of the forecast.
From the local catalog (NCSN) we find that there were 20 earthquakes
recorded in the NCSN catalog between 10 October, 2000 and 09 April, 2001
in the prescribed region. These occurred on 18 different days (2 occurred
on March 30, 2001 and 2 on January 3, 2001. The background rate per day
(i.e the probability an earthquake will be recorded in this region on
a given day) is then 18/182 = 0.099. In other words about 1 such earthquake
is recorded in this region every 10 days in the background period. For
this check there was no magnitude limit given - that will be checked
separately. The window Don gave included 5 days. Since the probability
of an earthquake happening on any given day is 0.099 the probability
of an event in a 5-day period (assuming independence) is 0.494 (5*0.099).
Now we can check this against what the NCSN catalog shows actually
did occur during this 5-day period in his defined region. The catalog
shows that two earthquakes occurred in his region during this time -
a Ml 1.5 on April 24 and a Ml 1.7 on April 25. The earthquake rate
during this period was 2/5 = 0.40, a rate more than 4 times the background.
So, what can we state about this portion of the forecast - Don should
get good marks for identifying a time and location where the daily
seismicity rate more than quadrupled at the expected time. It should
also be noted that following Don's posting of his forecast, there were
8 earthquakes on 7 different days between 10 April and 26 April. So the
seismicity rate for the region actually went from a background rate of
about 0.1 per day to 0.44 per day after Don's forecast (for the succeeding
2 weeks 2 days). So it appears that at least part of Don's gut was
feeling something as he said.

Onto the second part of the forecast - magnitude. Don expected
a magnitude for this event of 2.5 or greater. No magnitude scale
was given, so we assume he means Ml, since that is the only one
NCSN uses. During the 6-month background period, the largest
earthquake in this area was a Ml 2.2 on March 18, 2001. There
had been none in the range Don described. Had one occurred within
this range of magnitude, it would have been a spectacularly successful
forecast. The two earthquakes which did occur had Ml 1.7 and 1.5.
The Ml 1.7 was the 9th largest earthquake in the area during the
period October 10- April 26. The odds of choosing a 5-day period
during which one of the 9 largest earthquakes to occur in the
area in a 6-month period is about 0.25 (or 1 in 4). Incidentally,
4 of the 9 largest events occurred in the two week period April 10-25
after Don made his forecast. Can we say that Don was unsuccessful
in his forecast magnitude? In a strict sense, yes he was, the
magnitude of the two events which did occur in his time frame were
not within his forecast magnitude range. On the other hand the
size of the larger of these two events was among the 9 largest
earthquakes in the region during more than a 6-month period and
the odds of choosing a 5-day window that included one of these
is relatively small (1 in 4). Don has to get some credit for
choosing to state that a large earthquake (for the area) would
occur during this time.
We do not like to put numbers on a forecast because this does
not properly describe anything to most people. I am more
inclined to give a letter rating to the forecast. In this
case here are the facts:

1) The seismicity did increase in the area (over a 6-month backgound)
during the specified time and
2) An earthquake with magnitude larger than would be expected
did occur in the area during the specified time
3) The forecast did not correctly specify the magnitude limits

We use the following letter grades:

A: Seismicity increase in specified region is sudden and at least
400% of background rates. An earthquake with specified magnitude and
location parameters did occur within the specified time interval.

B: Seismicity increase in specified region is at least
300-400% of background rates. An earthquake with
location parameters did occur within the specified time interval.
Observed magnitude was lower or higher than expected, but larger than would
be expected from background rates and magnitudes.

C: Seismicity increase in specified region is
150-300% background rates. An earthquake with
location parameters did occur within the specified time interval.
Observed magnitude was lower or higher than expected.

D: Seismicity increase in specified region is
100-150% background rates. An earthquake with
location parameters did occur within the specified time interval.
Observed magnitude was lower or higher than expected.

F: Seismicity rate is at or below background rates.
No earthquake occurred in the region in question during
the interval predicted.

Using these criteria, I am sorry to say, Don, you receive
a "B" rating on your prediction. This is actually quite good,
and suggests you ought to listen to your gut more often.

I realize that many of you would like to see some kind of
numerical rating system for predictions. In a sense this is
more like the Modified Mercalli Intensity scale than like
a magnitude scale. A set of conditions are required to meet
any specific level, but I feel the letters are more understandable
than a numerical rating system would be to the general public.
Of course, a numerical rating is easier to use when a number
of predictions are to be considered, and like magnitude vs
intensity scales, I believe there is a place for both types
of rating systems.