To expand on what Jasem said, I suspect that our mapping of coordinates to pixels (not in the center of the image) can be off for certain types of setups.
Specifically images that astronomy.net considers parity=positive, which means they are mirrored relative to the other type of images. 
For example, it looks like images from prime-focus telescopes (e.g. RASA) are mirrored in this way, whereas the more common telescope/camera pairs
do not have this issue.  This is still a suspicion, and not confirmed.  I'm trying to debug with one tester.

Re logging, how about something like this (or more logging):
 

Bottom line, I suspect that log files that contain output with this line:are fine, but ones where the parity is instead detected as "pos" may be problematic right now.
 

Just found this...great explanation. Ignore that request in the message I sent you minutes ago, and thanks again!!!
Jon

Without chasing down the problem, I can confirm that the new polar alignment scheme does not work with my RASA/CGX-L config. The symptom is as previously described; the motion in each mount axis does not correlate to the drawn vectors. Worse, the results are not sufficiently repeatable to warrant trust without verification. After confirming many times that the results were different each time invoked, (verified with drift aligns after each attempt over several nights), I gave up on the feature and returned to other align methods. Just a heads-up to folks to not blindly trust this feature without doing verification by drift align if it matters to you..... CS Doug

Are you using 3.5.3 or later?
I made a change in that release that might fix the issue

Good morning Hy

I'm as well facing different results from the new polar alignment. One time it worked like magic and didn't touch the scope for 3 nights, and it was one of my best images, but a few days ago (in a different location up the north US), I spent 2 hours trying to perform polar alignment and I was getting different results each time I'm close to the match, and I had to stop with 50" error. To be fair, I will try again tonight and will share my results.

PS: During the last session I also tried the PHD polar drift align, but also it has variable inconsistent results

Forgot to mention that both sessions were done using my portable setup WO Z61, iOptron SmartEQ Pro+, Pi 4 Kstars/Indi, MyFocuserPro2, ZWO OAG with ASI120MM guide, ASI183MM Pro, ZWO EFW (LRGBHaS2O3)

Mohamed,

My understanding is that 50" error is very good. I agree that there is some noise in the polar alignment system, possibly due to the software, mount, or seeing. In my experience, with my setup, this noise seems to be less than an arc minute. I stop adjusting once I have the error less than an arc minute or so.

Hy

Hy

I know and was happy to reach the 50" total error, but guiding, later on, was giving me 3-4 degrees of error. This didn't happen the time before (it was flawless)

I have a question that could e basic ut worth checking, as I use OAG, in both PHD & Internal guider, I enter my scope parameters as guide scope parameters ()and the camera is the guide camera), but I was thinking that I only look into a small portion of the scope FOV periphery and not the whole aperture. Does this affect the guider calculations?

I believe that only your focal length, and the guide-camera's pixel size matter.
I don't believe the aperture matters, though I also believe the aperture you used is the correct one.
Hy

Question. Is the all-sky polar alignment process sensitive to cone error?

my understanding of the principle of operation is, that the algorithm is not depending on a perfect alignment of the scope or camera in relation to the mount axis. The algorithm will always find the point in the sky around which the axis of the mount revolves, i.e. the MOUNT will be polar aligned. Whether your scope is parallel to the mount axis or not is a different question. So even with a pefectly polar aligned mount axis you could still have a cone error. But the PA algorithm should not be affected by this.
Probably Hy can comment on this as I just derive this from my theoretical understanding.

A  quick question, apologies if it has been covered earlier in this long thread:

My CEM70 is often persnickety about doing the rotations in the polar alignment assistant, so I usually check "Manual rotation". When I get the nice triangle in the PAA, the alt-az adjustments don't move the star along the legs of the triangle, but in some other directions. Is it expecting that the mount has returned to the zero position before clicking "Refresh"? That would be a bit odd, since it's plotting the triangle on the third image, which is at 60° RA from zero, and I'm prompted to select a star there. (I don't know if clicking on a new star even works once you've clicked "Refresh".)

My CEM70 has <em>also</em>  been rather persnickety about parking and returning to home position in the past, I know that Jasem looked into this for iOptron mounts at least once but I'm rather chary about letting it smack into the tripod again to discover if the problem still exists so I have the "park afterwards" checkbox off too.

@Dirk, @Andrew:  I would agree with Dirk that is should be unaffected by cone error.