What results can I expect form IMU+compass attitude estimation?

[kvoit]

The title could also be "I am totally lost and don't know what to think" :)
We are trying to get attitude from a lightweight sensor platform for behavior research. We are currently testing with a Pololu AltIMU10 v5, featureing ST's LSM6DS33 IMU and LIS3MDL magnetometer.

Magnetic field is calibrated with the algorithm described in Teslabs' article, gyroscope gets an offset correction. Accelerometer for now is uncalibrated.

To estimate accuracy, we first did a very simple test where we put the dev board into a number of static positions.
For these, we compare the absolute angle of rotation to the initial pose for different filters among each other and with the trivial values derived just from gravity and magnetic field. Looking at the raw data, I would assume the latter to be roughly true.
We also tested the magnetic field elevation, which is for all poses within a few degree of the expected value in our location.

From the 7 filters that accept mag, acc and gyr, aqua, complementary, ekf, fourati, madgwick, mahony and roleq, only complementary is within a few degree difference to the trivially derived orientation. The rest differs notably from it and from each other.

I also noted that results hardly change even if I e.g. multiply angular velocities with a factor of 1000.

Data and analysis can be found here:
https://github.com/bbo-lab/IMU-test/blob/main/IMUDevTest.ipynb
or
https://colab.research.google.com/github/bbo-lab/IMU-test/blob/main/IMUDevTest.ipynb

Are we doing something fundamentally wrong in how we are using the filters or how we judge results?
I can't believe this is the expected performance of these reasonably sophisticated filters.
Or is this what would be expected and AHRS is not for us? Do we need more sophisticated calibration procedures for any sensible results even in these simple cases, and would that still yield sensible results in more dynamic environments?

And if it is the calibration ... which I am starting to strongly suspect ... how long can I expect one to live? Changing the magnetic configuration is clear (will a body have a strong influence, because we cannot calibrate on the body), but will powering the IMU down invalidate it? Body heat? Time?