Consistently inaccurate body composition measurements

Hey!

I was excited about the Body Scan - I had a Fitbit Aria 2 smart scale before, but the body fat % measurements it was giving were inaccurate so I was hoping that Body Scan as a newer device and also having a handle will have better results.

After the first weigh-in, it seemed to show pretty much the same results, but I decided to keep using and maybe it will improve. I have now been consistently using the Body Scan for almost 2 months and the results don't seem to have gotten more realistic.

For reference, I got a full body DEXA scan done on November 25th, which showed total body fat mass at 16kg, which just visually seems about right.

On December 12th, I did a measurement with my Body Scan, that showed fat mass at 12.4kg and immediately after went to my local gym to get a body composition measurement there (Jawon IOI 353) that showed body fat mass at 14.9kg.

It seems that Body Scan consistently measures 2-3kg less fat and 2-3kg more muscle than "industry" machines, which already seem to have a similar error compared with DEXA scan. This is a very big error margin. It would still be OK if it was just offset and at least showed an accurate trend in change of body weight and composition, but this error makes it hard to trust any of the other measurements the Body Scan claims - ECG, nerve health, etc.

I have also searched for reviews on the internet and found many people complaining of similar errors.

I know that the Body Scan also has an "athlete mode", but I still don't understand what it would change and whether I qualify for it. The description says that you should have a resting heart rate below 60bpm, but the Withings app doesn't show resting heart rate (I have the Withings Scanwatch Light), only average HR and since it also includes HR measured during workouts, I don't think that counts? And it also requires working out more than 8 hours a week and while I do go to the gym 3-4 times a week, mostly commute by biking and try to get a run or two in a week, I'm not sure if that applies.

Is this inaccuracy something that can be addressed with a software update and are there any plans to improve these measurements and calculations?