@therealkuu @mcfly This array is 10,000 positions and they analyzed 3.4m codes.
That's an average of "only" 340 per square, but as we can see, humans are weak, so there's an over-representation in some areas, so the mid orange is less than 340. Probably about half.
My guess is it's not unimaginable that those are just random numbers that are not found in the sample. No particular reason to filter them out.