Such outputs and especially these steps/jumps don't look like expected behavior. A few things I would check in such conditions:
Do all timestamps look as expected?
Does BSEC return any warning/error message?
How does the raw data (esp. gas resistance) look like?
If one was previously stored, did you try clearing (or temporarily disabling) your state file?
Do you still see the same behavior if you uncomment this line from the sensor API (i.e. enabling floating-point compensation)?
Monitoring the IAQ accuracy output could also provide further hints on the behavior/status of the library. I would suggest to confirm the points above are all working as expected first, then I would also recommend letting the library run for some time/making sure it had time to calibrate itself before comparing the output to your reference sensor.
... View more