My understanding is that the ratings have never(?) been global ones; they are their estimate of what you will think.

It is possible they A-B test aspects of the process used to generate these.