Unexpected AGC Behavior

I'm not sure I understand the following AGC versus RF Gain position behavior. It may be the way it should work but it seems strange to me.

In a nutshell, the change in output audio voltage level, as measured by the AFV and dBV measurement capability, is not what I expected. When the RF Gain control is fully clockwise (maximum), audio voltages measured when AGC-F is on and AGC off are shown in Figure 1.  Turning the AGC off results in higher (louder) audio signal levels. This is what I would expect.

Figure 1.  AGC operation with RF Gain fully clockwise.

When the RF Gain is reduced to about the 1 o'clock position (first dot after 12 o'clock), turning AGC off results in lower (weaker) audio signal levels until you get to about -50 dBm input signal level as shown in Figure 2.

Figure 2.  AGC operation with RF Gain at 1 o'clock.

Figure 3 shows all data plotted on the same graph.

In addition to this behavior, when turning the RF Gain through the 9 o'clock position there is a definite 10 dB change in audio level.

The input signals were produced with the XG3 and are:

-117 (-107 with ATT in)
-97 (-107 with PRE in)
-83 (-73 with ATT in)
-63 (-73 with PRE in)
-43 (-33 with ATT in)
-23 (-33 with PRE in) dBm.