Analysis of the intrinsic variance of the time-delay point estimators, found by running the three curve-shifting techniques 200 times on the light curves shown in Fig. 4, starting the optimizations from random initial time shifts. The widths of these distributions reflect the failure of the methods to converge to a single optimal solution given the data. We stress that these distributions do not represent probability density functions of the time delays. A small intrinsic variance does not imply that a time-delay estimation is precise or accurate, only that the error surface is relatively smooth.