He's changed the stats by taking a look at the articles to verify things like disambig status, so you'd have to be quite careful how you did your analysis to ensure no reflexivity.

Personally, I think the reflexivity is part of the fun if you do a followup analysis. For example, I recently scraped WP to find 'the first unused acronym on Wikipedia': https://gwern.net/tla - it turns out to be 'CQK', and I'm looking forward to checking back in 10 years or so to see if anyone wound up using 'CQK' for a company or something, precisely because I wrote it up. We'll see!

He is talking about its uniqueness in the 2021 dataset, which is time bound and unaffected by current phenomena.