Given the vagueness of their statements about the 143M records which "may" have been affected, it seems likely that they weren't logging the relevant queries in any useful way, and that is definitely an architectural failure.