I thought folks might find this interesting. We recently did some PX4 fixed wing testing with firmware based on 1.8.0 where we forced the GPS driver to stop and watched the vehicle behavior. We would restart the GPS driver and watch where the vehicle position jumped to as a measure of how far off the EKF navigation solution had wandered.
Amazingly we had one loiter remain stable for > 6 minutes before the global position failsafe finally kicked in.
Based on this log (and a couple others we've done) I think the EKF position error estimate is optimistic, at least for our hardware setup and flight profiles.
Here is the flight log for those interested in diving into the details. Feel free to post your thoughts or observations in this discussion thread.
FYI @dagar @Paul_Riseborough @Kjkinney