Sorry for the 3 posts in a row, but just wanted to confirm that the signature I posted in image form above is related to the start of dataflash logging. If I log while disarmed and run the same test, that signature still exists just as the log begins. I will attach the same arming only log gathered while log while disarmed is enabled. you can see that arming still exhibits a much smaller error in vertical velocity and position.
arming takes place at about 00:30 on the time axis. you can see that there is still a small error in vertical pos and vel but significantly smaller than when the log starts at arming. The signature seen here will also get smaller/non-existent if all telem/serial ports are disabled (except gps and usb ports). But you can see that the same 1.5 meter vertical pos error exists at the point where the log starts. It also seems to get larger as parameters are downloaded upon connecting to mission planner(between 58:00 and 59:00). Once again, this cannot be replicated in SITL. There is definitely some strangeness related to uart load and dataflash adversely affecting the vertical ekf estimates.
I have attempted to attach the log but it is too large to include in this post.