Hi Steve,
The comment in the article for a "fully replicated data set" is referring to an individual public folder as part of validating your existing PF replication is healthy and up to date. For example consider a public folder named 'Sales' with replicas on three different Exchange servers in the organization. A "fully replicated data set" for the 'Sales' folder would be what you have if PF replication is healthy across all three servers and all three servers show the same content if you were to look at each one's representation of the 'Sales' folder.
If outbound PF replication was broken on one of the three servers, then it is possible your other two servers would be missing data originally authored on the server with broken outbound PF replication. The result would be two of your servers would not have a "fully replicated data set" due to missing the content authored on the server with outbound PF replication broken. Essentially this simply means we want to see your existing PF replication happy, healthy, and up to date before initiating the migration.