I measured CCFC error signal for 3 hours with 30 minutes interval (Fig. 1). Fig. 2 shows the CCFC error signals around 0 crossing point. The 0 crossing point of the fitting result (dashed curves) changed by (1.04-0.97)*54 = 3.8 Hz in 3 hours.

I also measured the nonlinear gain and shot noise before/after the whole measurement.

The nonlinear gain changed from 4.7 to 4.3 in 3 hours. This nonlinear gain change causes the CCFC amplitude change. The normalized CCFC amplitude can be written as follows.

Normalized CCFC amplitude = x/(1-x^2)^2 = (1-1/sqrt(g))/(2/sqrt(g)-1/g)^2

Figure 3 shows the normalized CCFC amplitude as a function of nonlinear gain. When the nonlinear gain changes from 4.7 to 4.3, the normalized CCFC amplitude changes from 1.07 to 0.97 by a factor of 0.9. In fact, the CCFC amplitude changed from 132mVpp to 118mVpp by a factor of 0.9 in 3 hours.