Exactly. This is why I use nearby fixed objects whenever possible, but of course, always for Start, Finish, and TAs.
With the GPS I use, I usually have to wait 30 to 60 seconds for the device to gather a good complement of satellite signals and produce a stable reading. And, then, naturally, the consequent reading will be accurate only to a radius of several feet or more. By going back to Google Earth at my desk, I am then often able to refine some field point locations thus captured, because I can determine the course path from the aerial imagery in combination with my field notes and my knowledge of the terrain.
Oscar, you make a good point. This is why, when I am measuring off-road, I measure straight line points to straight lines. No "sweeping curves", for the reasons you mention.
I am not commenting to make a hard push against the ban on certifying off-road courses today. Yet, my experience informs me that, an experienced measurer using the right tools and careful practices can indeed create a course that is reproducible with sufficient accuracy. I am convinced that any good measurer could reproduce one of my off-road courses from my data and get within .08% of my measurement. Maybe I will figure out how to test this. One requirement would be that each inflection point would be fixed some small distance within the expected margin of error 90 degrees away from the recorded SPR. In my very humble opinion, this practice could add some distance over the nominal, and probably in the same range, on average, as in the case in which we measure a full tangent when the race nominally divides the road with traffic cones.
As I mentally reconcile the fact that we certify that our courses won't prove to be short, but probably isn't as accurate as a steel-taped course; and that we are satisfied with certifying courses that may be long by many feet or yards, it seems to me that there should be no prima facie reason not to try to establish criteria for accurately measuring and documenting off-road courses.