Notice that the dump shows that your RMS error is already down in the 0.15 pixel to 0.11 pixel (for RA and declination, respectively). I.e., PHD2 is already guiding to 1/8 of a pixel, with a plate scale of 4.98 arc second/pixel. One can really cannot estimate a centroid to better than that. If you want to do better, try a guide system with a better plate scale (longer focal length or smaller sensor pixels).
Your RA graph (the one with the larger error) also shows an almost sinusoidal curve (oscillatory). Possibly caused by having an aggressiveness (loop gain) that is too large. It is also possible that PHD2 is fighting against an encoder (both PHD2 and a mount's encoder are feedback loops and there should only be one that attempts to correct for gear errors).
If you have a mount with a shaft encoder, turn it off. Let autoguiding determine the feedback, and use the encoder only when you are not autoguiding.
You may be able to do better if you were to (1) use a guide scope with a longer focal length to see if it helps your pixel scale, but more importantly (2) start with very low RA aggressiveness (say 10%) and only turn it up 5% at a time, and waiting 10x or 20x the guide exposure time to wait to see if the change improves the RA error -- and only bump it up again (and again, only a small amount at a time) to see if it improves. Never bump it so much that you start to see oscillations. Back off a lot when you see oscillations. A guide graph that also shows correction pulses will also tell you if the loop gain ("aggressiveness") is too large (i.e., bunches of positive pulses followed by bunches of negative pulses, etc).
You also may be able to improve the decoupling between the RA and declination by changing your camera angle. The log and the red/blue vectors shows a camera angle of 148.5º for RA. See if you can change the camera angle so it is closer to one of the cardinal points (0º, 90º, 180º or 270º) -- i.e., make the red/blue vectors to line up with tRA/dclination axes. Once you are within a few degrees, (because of the sine/cosine behavior), the coupling visually vanishes; so you don't need to be super accurate. I glue a small bubble to my guide cameras so I can do that in the daytime, and it gets me to within a degree or two. This way, a true RA error will not be mistaken for a small declination error.
Chen
