Im afraid to say, gentlemen.
The nature of the UKF (the main position attitude and velocity estimation filter) means that when a GPS fix is aquired in flight, it will immediately begin to correct the heading and velocity estimates based on the GPS info that just became available.
If the UKF decides that the heading or velocitity estimates is off, it will correct it. This can manifest in manual flight as a slight twitching in the attitude control. Under normal conditions, this goes away as the filter converges, but it can be a bit unnerving. However, I have never had anything like what you describe happen in manual mode.
In very rare cases, if a GPS fix is aquired during while in PH mode, the position can get screwed and the craft goes for a new position with the full limit of what the nav controller allows. Its usually associated with limited or poor GPS reception close to the 3m threshold of the 3D lock state...which is something that can happen if your antenna or groundplane is not up to par.
However, if GPS lock is lost, you should not get any adverse behavior when its regained...but if you cant keep a lock in flight with a reasonable view of the sky, it points to a problem with the antenna that should be solved.
But looking at your logs, I definately see something odd going on in the 3rd flight. The PosN and VelN estimates makes a huge jump in the latter of that flight, and I bet that has something to do with the subsequent "flip". which may just have been the nav controller trying to go to a new position 20 meters away:
The fact that you went from inside to outside without a reboot and that you carried it around before taking off could also have set up the filter with a hard dataset to converge. Combined with a poor GPS reception this could have tipped the scale as the GPS accuracy decreased towards the end of the last flight...but its a guess.
A few recommandations:
Check the antenna and increase the size of the ground plane. The one you have on there is very small...probably that factors into the poor reception that your logs seem to indicate.
If you intend to use Position hold, you should always make sure you have a decent lock on the ground before taking off - give it some time after the blue LED comes on to improve the accuracy and check it in telemetry.
You should be able to get down to around 1m HACC on the ground - even with some trees around. But many trees obscuring the view of the sky will impede GPS signal reception, specially this time of year when there is leaves on the trees.
After GPS lock is aquired and settled, then takeoff and fly some straight lines back and forth and right/left about 10 meters i manual mode to allow the UKF to converge the heading and velocity estimates before trying a position hold.
You should make sure that your heading reported by QGC is as close to true as possible when the craft is on the ground. The compass is only used to estimate the initial heading of the craft before takeoff, so if its massively off, the UKF will correct the heading when GPS signals become available.
If the heading is off, redo the compass calibration routine and make sure the generated parameters are stored in the Flash or Eerprom and that they give a good heading estimate on bootup. Check that inclination and declination settings are correct for your location before doing the compass calibration.
Id like to see a copy of your params, pls.
On the upside, I think that your vibration levels and the overall balance of the craft is looking good. I think it should be something that could be made to fly good in any mode once we get to the bottom of this...