The data is unreliable. If we knew how much of the data was faked we could compensate for it, but we don’t. We could discard the outliers, but we don’t know if we’re discarding valid data, and someone who is deliberately tainting the dataset would submit a bunch of samples that are only a little bit off as well.

And while some of the numbers must be from trolls, manufacturers (and shady investors) are heavily incentvized to sway the listings.