This document is very informative to read, and is a history lesson for people in the field who were not there. In review, this document appears well-written. I cannot judge upon its correctness. A question -- which may be offtopic -- I had while reading the document is how software "looked" during this time, and how software adapted during these changes. In other words, as we are aware, the root hints built into (or configured for) DNS clients such as resolvers are the entry points into the DNS system today. This is where DNS clients learn where to start. There was a time when hosts.txt files were in use. At the time when DNS appeared, and root servers and the concept of resolution was introduced, how did software in those days react to these changes? This was a significant deviation from a file-based database. It is difficult to think of a change of this scale happening today. Although there are clear boundaries of function such as stub, resolver, and authoritative service which have allowed changes such as use of secure transports, various DNS aspects such as DNS names, hostnames, RDATA syntax, etc. are used directly by applications in ways that make replacement of DNS unthinkable. There is some commentary on how change from hosts.txt to the DNS should happen in RFC 881, but what really did happen? How long did it take for the DNS to become dominant? How many hosts were connected during that time? What resolver implementations were available in those days, and how were they configured to adapt to the changes in the list of root servers mentioned in this document (outside the root hints query a.k.a. priming)? Was the hosts.txt file reduced to hints for TLDs and somehow involved in the resolution, or were hints configured separately? Postel wrote in RFC 881:
In the long run the Internet will become too complex and change too fast to keep a master table of all the hosts. At some point the master table will be reduced to simply the entries for the domain servers for the top level domains. By this time all normal translation of host names into addresses should take place by consulting domain servers.
By this, can the root zone be interpreted as what is left of the master table? Was a statically distributed root zone ever in use within the new DNS protocol during the transition? I guess not.