I enquired with the current owner of the CCS811 product concerning what comprises the typical indoor volatile organic compound gases and sensor responses to them, and received the following reply: (abbreviated)
========start of snipped text======
Below is a typical list of gases that build up and exposure at high levels for long durations can cause sick building syndrome (you can find a lot of material online on sick building syndrome):
CCS811 is sensitive to combustible gases in general with substance specific sensitivities. To name some of the gases which the sensor reacts to, there is Acetone, Ethanol, Acetaldehyde, hydrogen, methane and carbon monoxide, as well as other reducing/oxidizing contaminant gases. Because CCS811 is not a selective sensor the sum of all VOCs present in the ambient air will contribute to a sum signal, which is a lot of the time used as an indication of indoor air quality.
You can find material on the CCS811 product page here. We are working on new white paper which also may be helpful and I will share once they are available. It is also worth looking at theENS160 which is our latest gas sensor as it has superior performance and a lot of great features. You will find a lot of useful information in the ENS160 datasheet which also show international as well as national air quality standards. I would recommend using the ENS160 for any new design or for migration to take advantage of the superior performance and many new features.
=======end of snipped text=====
Before you ask, as far as I know, adding ENS160 to SCK is not currently on the SCK road map,(and the team has lots of other work to do) however that’s sensible since that device is only in pre production status.
Thus It is unclear whether the SC team will in future consider an upgrade or simply add support for the device if it is added as an extra sensor (eg as a breakout board). I guess they might if there is enough interest.
I note the the ENS160 device includes similar auto calibration feature that is used by CCS811, so there is no guarantee the new model would act any better than the old model on the behaviour noted by yourself.