At the same time, though, a map error of a few hundred feet seems to be enough for it to skip the waypoint. The illustration I gave above for the missed waypoint was caused by a map discrepancy of only a few hundred feet.

Most software seems to lock the position to the nearest road segment that approximately goes in the right direction. When we have full street level info that might be a solution that will actually fix a lot of these 'off by a hundred feet' issues.

Besides, I thought you'd already changed it so that it used the road intersection data rather than the vehicle bearing for its L/R calculator anyway.

No I was using the current position wrt to the waypoint instead of the bearing from the gps. And because waypoints are never really exactly where the turn is, the resulting directions were almost as good as random.