This is true, but the definition of the top of the hierarchy is arbitrary
Not at all true. The top of the hierarchy must be default free.
and is the nexus of the debate about "topological" versus geographical addressing, which I interpret as "ISP at top" versus "exchange point at top" hierarchies. Both are valid topological hierarchies.
True, however, geographic addressing has some rather severe practical problems. The exchange point at the top becomes a single point of failure. So it needs replication. But then, there needs to be interconnect between the exchange points. Who provides it? All this and more has been beaten to death. If you start with the premise of geographic addressing and try to beat it into working, you end up with an ISPAC. See ftp://ftp.juniper.net/pub/users/tli/ispac.txt. Tony