Results after adding 28 Oct 01 measurements. Used the current mathematical Calculation method. Added "Standard

Error" calculated error bars to the source data plots.

Selected clocks representing 86% of the total are plotted. Selected according to the lowest standard deviation of the 2nd
derivative data. Tossed out a few measurements (14%) showing the most deviation of absolute value from zero. Original
curve: (A1), 1st Derivative: (A2), 2nd Derivative: (A3).

Selected clocks representing 60% of the total are plotted. Selected according to the lowest standard deviation of the 2nd
derivative data (less than .0026). Original curve: (B1), 1st Derivative: (B2), 2nd Derivative: (B3).

Selected two of the best clocks TP10 and TP20. Selected according to the lowest standard deviation of the 2nd derivative
data (both have about the same STDev). Original curve: (C1), 1st Derivative: (C2), 2nd Derivative: (C3).

The following is an average of the above three curves (D1). The next curve shows what happens if this average is extended
to estimate an Accumulation of 1 Day Slippage by 15 may 03 (D2). Close up view of the bottom of above curve D2 (D3).

Summary: The trend over the last 3+ years shows a slope of 50 second per year speed up of our clocks when compared to
Atomic Time. This is equivalent to a possible slow down of Atomic time because of added leap seconds by the same
amount. There is always the possibility of this slope being due to battery discharging or other inherent error sources. The
hope is by having lots of measurements (and many different manufactures) over an extended time that most of these errors
are minimized. It is expected, as we get closer to the PS time that the slowing will become easer and easer to measure.

Mike