They might have implemented a metadata store 1 layer before S3 to guarantee read after write consistency since write to a new object is consistent in S3. Only updates are eventually consistent.