Scraping doesn’t even have to be hourly, scraping a thread or account once then `diff`ing the two between weeks or months or years would easily reveal any interventions.

Which would actually expose users who you should retroactively investigate. Canaries.

I presume you are aware of “fuzzed” users. With unassociated comments, submissions, and possibly even changing names. I only recently discovered this myself after a little digging, but you may be able to do this and continue posting.

I’ve been hesitant to mention this anywhere, as discussing how you would red team a site you like publicly might not be the best for the site.

There are other clever possibilities, such as using some combination of artificial & organic content to wash users identities. Like banned or shadowbanned accounts mixed with real users. An acid bath if you will. If you dissolved 100 people into each other, who can really say who is who?

I don't think changing usernames is a strong action either. This is because the fingerprint is your language, not the name you're associated with. See the Enron dataset and the project in Ng's intro AI course, but we'd need to scale that quite a bit (which is very hard).

I'm not aware of the fuzzed users. But I think you should mention it in the open if they are problematic.

I have thought of AI to rewrite my language but that's difficult and only fuzzing too.