Earlier quoted context omitted.
There's a similar phenomenon you might encounter in language: if you have a text, and you sample one random letter at a time from it, it turns out not to be very difficult to determine what language the text is written in. Does the letter 'i' exist in French? Does it exist outside French? Sure, but nevertheless you can rule French in or out based on how often you sample a letter and get 'i'. French uses the same 'i'…
Same kind of thing works with counting the frequency of particular digits (n-digit analysis) in a given position for an integer or decimal. Especially for the higher order decimal digits there's a far-from-random distribution that you can use to infer whether there's been any tampering, given enough values to study.
Inversely, the lowest order decimals tend to be evenly distributed, which is also a useful way to detect tampering (trying to make the numbers look "random" does anything but).