Another bug. The Pot 3 value is bleeding into Analog Input when read. [There may be other bleeds, but that's the one I've focused on].
So if you do a=analogRead(A0) [pot 3] and then b=analogRead(A4), you'll often get a's value in b instead of the correct value.
As I understand it, the primary reason for this is that the problematic input (in this case audio-in) has too much impedance, and so can't charge the ADC capacitor in time for the read, and in turn the read generally reflects the *previous* read, which in this case was pot 3.
My workaround for this is to do a throwaway analogRead(A4) immediately after reading pot 3. Sometimes I have to do two of them. That's pretty slow. :-(
A high impedance is a problem -- it's going to impair sampling from the audio input. :-( I'm guessing it's coming from the lowpass filter. Can we lower the impedance in audio in? Or are we just stuck? Would this be a problem for other inputs too?