It's late and I was able only to skim your story here, but as I did, I kept nodding and (virtually) shouting "Yes" and "Of course!" If I have time, I'll return to your essay and add more ideas, but here's a preliminary impression.

Disclaimer: I'm not a data scientist (though I have subscribed to "Towards Data Science" because I'm so fascinated by the field). I know more about the general principles of screenwriting than I do about statistics. But here's one principle for creating characters and conflicts when writing a screenplay that I believe is relevant to your discussion: Everyone has an agenda.

This assumption is valid not only for characters in a script, but individuals in real life. If you watch a group of people at a restaurant without being able to hear their conversation, you can quickly see that there's a power dynamic among those diners.

The same principle applies to data scientists. One cannot initiate an investigation without having an agenda (even if the scientific method precludes incorporating that agenda into the research itself).

So anyone seeking to predict or determine criminality without taking into account the various complexities that you bring up here is not only seeking to establish criminality but to create a simplistic system in order to punish those whom the researcher assumes are a priori guilty of some sort of infraction. And that oversimplification is (I believe) more criminal than breaking any law.