It's possible to pseudo-anonymise the ratings: Have a central database, Alice, where every registered rater and rating agency are assigned random alphanumeric codes, another database, Beatrice, where every report is assigned tied to the alphanumeric codes representing the author(s) and their affiliation(s).

N months after a rating has been issued (where N possibly depends on the type of rating), the rating is compared to the actual reality. This information is then input into the Beatrice database.

The people who run the Beatrice database can then mine the now blinded data for noteworthy patterns. Any suspicious patterns discovered can be analysed independently, and only if they are actually damning is the Alice database contacted for retrieval of the identity of the suspicious author(s) and/or institution(s).

That's roughly analogous to how you'd blind a medical trial. It's not idiot-proof (no system is, and even if it were, the universe is continually working on inventing more creative idiots), and it's certainly not fraud-proof either (no institution in this universe can be made completely fraud-proof), but it would be head and shoulders above what we currently have.

(Assuming you even need rating agencies at all...)

- Jake

Friends come and go. Enemies accumulate.