I think musings is the best place to put this.
Anyway, this is me rambling on how I think the ratings system works.
The official descriptions are a bit too "so you already understand machine learning" and are not quite willing to discribe the nature of what controls the rating system.
Sure, there's a possibility of the Civitai running a Panopticon but they don't have the staff for it and computers can be dumb. (withholds rants on generated script gets mispronounced by the same service that made up the script mostly whole cloth.)
Anyway, I think it comes up with a rating based on some preliminary findings thst aren't directly related to tags, which is noticeable when there is a mincemeat Lora as the daily challenge, (so much red, but it isn't quite gory)
However, the system isn't constantly rating everything, it only issues relating when someone asks.
I think Knights of New processing may account for changes in rating. And you can def force a recalc by inputting a different rating vote, and that can be pretty timely, particularly for images.
in short, Civitai takes a lot of shortcuts to generally be more high rating then low rating, but a given work can be re-rated at basically any time.