Pazz
If you read through all of the three major threads related to this magnetic declination issue you will find...
People effected are in areas with generally more than 10 degrees of declination. There is almost nobody effected with less than 10 degrees declination. People have flown their Phantom in areas of both high and low declination and have attested to the difference in flight performance. DJI is in a low declination area and probably never sees this in their testing.
Almost all of the people affected have observed that the TBE, drift, JHook, and Course Lock error are very noticeable in the first few minutes of flight and then virtually disappear. How long it takes to disappear varies with declination. If you land but keep it powered, then takeoff it "remembers" the refinement and doesn't have a problem. If you land, change the battery, start it up and takeoff again it goes through the whole process again from scratch. There is no doubt that it is learning. In a video I posted recently you can watch the problem go away (fairly quickly with the Beta 3.05) as it "learns" when the motion it commands based on the magnetic compass doesn't match the motion it gets from GPS data.
Geometrically GPS requires three satellites to get a position but four is considered a practical minimum. DJI requires a minimum of six to be in GPS mode. Each satellite adds reliability and accuracy. Often I see 8 to 12 in flight which further reduces errors and fluctuations. Certainly there are GPS errors and they have an effect but that is not what is causing TBE, drift, JHook, and Course Lock error. If it was it would happen randomly throughout the flight - but that is not what we are seeing.
There is a subset of us (not me) who rotate their compass, usually twice the declination. They claim that reduces the problem greatly - probably because it reduces the mag/GPS discrepancy. Most feel this doesn't solve all the problem and is not a good long term fix.
That's why I think it is declination, it is learning, and it's not GPS errors.