I have Fossil G6. I tried to integrate crude “point to specific place” dial, and stumbled upon the fact, that #CMP# returns tilt, instead of heading, making any compass-related stuff impossible. I dug around expressions and found gyroX(), which seems tied to tilt side to side (for example), gyroY(), which is tied with tilt front/back and there is no gyroZ() - which should perhaps return something tied to spinning myself around. Why it is the case?
Is there perhaps correct way to process that?
I used the #CMP# tag in MAG 2085 for a working compass and had no problems. The expression I used is (round(-#CMP#)). If you want to look inside the face and see how it’s done, inspection is open. The element is named “Compass-4-Points.png”.
It SHOULD work. It’s simply has wrong source data. Sadly, your face behaves the same - compass spins around with how I tilt my hand, instead of real heading. This may be watch specific issue.
The first thing you have to do is wave your watch around in a figure 8 pattern so the compass can align with north. After that happens it will be pretty close, but I’m not sure what position the magnetic sensor is inside a watch. You might only get correct readings if you are holding the watch’s face on a horizontal plane or on a vertical plane. I’ve found even after the watch gets the north bearing it’s still a little on the jumpy side. I don’t do hardly any watchfaces with the compass function just due to those drawbacks.
figure 8 pattern
Just like a phone - Yes, that’s correct, and I’ve done it. Again, my watch is feeding tilt data VERY precisely into compass heading. 
So clearly watch firmware issue, not Facer’s.
Well, I have update to WearOS 3 from 2.x ready, I might try to jump at it and see if it fixes things.
And thank you very much for sharing your face here, you helped me out here
Good luck with the Update, let us know how it all works out for you afterwards please.