The battery is a nominal 90kWh and the usable range (i.e., the 0-100% range) is supposedly 84.7 kWh. I think it's actually less but not 6 kWh less. I agree that there are confounders but they can't explain the results.
Regardless of SoC, I recommend the following to convince yourself that something weird is going on with the consumption calculation: drive hard for a few miles, watch the real time kWh/100miles figure in the Eco Data view. Then get off the freeway and watch the kWh/100miles drop precipitously even though you've only covered a tiny fraction of the total distance driven on this 'journey'. Don't need SoC to see that something is awry.
When 0-100% is used like this a lot of people tend to misunderstand. The net capacity of 84.7% is measured from where the BMS stop any further charging (real net 100%) until the BMS will stop any further draw of current (real net 0%). These levels are not the same as the displayed SOC 0 and 100%. There is about 3kWh between a full battery and when the meter starts showing 99%. In other words the car can show 100%, but still lack 3kWh of charge compared to a full battery.
Also, we all know that the car does not come to a halt as soon as the display show 0%. There is more energy there to use.
The usable capacity and the EPA and WLTP ranges are measured between these two extremes.
In addition, there is a brick protection ofr 90-84.7 = 5.3kWh which you cannot use. This buffer is there to prevent damage to the battery.
(disclaimer : It seems to be possible to add some energy through the regen system into this brick protection buffer)
You might be correct that there are weird short term fluctations. I really haven't done much analysis over short distances. I have made a lot of observations over longer trips though, and the average cnsumption displayed has looked fairly reasonable. Also the 100%(full battery) to low number SOC looks OK and do not conflict with average consumption. It's only when I perform measurement in the mid range of SOC I get into this discrepancy.