This is a video I made of how the idle modifications work together to produce a solid idle in my car. http://www.youtube.com/watch?v=8DN8OMbmMqY
Important factors, in my opinion are:
AC feedforward for AC loads
Idle Valve Voltage Correction for dealing with electrical loads
Adaptive Idle Advance takes care of keeping RPM close to target
PID reduction window- minimized idle valve action near target (so it doesnt fight AIA), yet aggressive when outside the window
This uses 3.3.0b gslender 2.4.5 (prerelease version), which has another small tweak I'm testing- adaptive idle VE, which does for idle VE what adaptive idle advance did for spark. It has a much subtler effect that AIA, but works the same way. There is a VE adder vs RPM target error curve, which adds to gVE- richer when below target, leaner when above target, in tenths resolution. This gives a faster but subtle fuel response, again like AIA.
I know, it does add yet another layer of complexity, but the data does support it. This is a scatter graph without AIV:
With AIV turned on, note that the RPM clusters around target more. There is a mild trend showing AIV working, with the AFRs richer below target. Very subtle.
I also found AIV has a beneficial effect on WUE, correcting for less than optimal WUE curves.
I do suggest not having EGO correction on for the idle area. Needless to say the basic spark, VE tables and PID should be tuned before even attempting to mess with the advanced idle modifications