EKF_YAW_REST then Roll\Pitch are inverted

,

Quadcopter, flying compass-less but with GPS, there’s a single valid IMU on the aircraft.
Since we want to enable PosHold, we opted to use GSF.

We’re aware that the heading needs to be ‘learned’ - either by flight or by walking with the drone on the ground before takeoff.

Strange thing - we took off, and rather quickly after that it felt like the controls of the aircraft switched between roll\pitch. We quickly landed the aircraft with minimal damage.

Looking at the logs - there’s a obvious “EKF3 IMU0 yaw aligned using GPS” and “EKF_YAW_RESET” message right before this happened.

Is it possible that the GSF alignment is at fault - and it actually re-aligned the IMU to the aircraft frame?
Or there’s something else going on here?

Adding log and paramers here:
LOG
PARAMS

Recent question I had, same drone:
Here

Actually I’m unable to find GSF messages in the logs (of this flight previous ones).

Although, as seen in here, our masking parameters were not the same:
EK3_GSF_RUN
EK3_GSF_USE
EK3_IMU_MASK

Will report after a test flight.

EDIT:
Writing to myself here, but any way:
we reversed to a previous set of parameters where we did not have this issue during flight.
Main differences in the parameters:

Param No Problem Problem
EK3_GSF_RUN_MASK 3 1
EK3_GSF_USE_MASK 3 1
EK3_IMU_MASK 2 1
INS_ENABLE_MASK 127 2
INS_USE 0 1

Flight log

This is strange, does someone have an answer on the actual parameters we should set for our use case?