A useful way to think about conversation maps, doubling as topic maps, is that each node in this map is its own topic. Topics are uniquely addressable. Consider this example, in which I sketch a small structured conversation:

Question: What are the causes of climate change?

Answer: CO2 in the upper atmosphere
Answer: Methane and refrigerator magnets

Notice that the first answer gives a single response: CO2 in the upper atmosphere, a topic all by itself.

Notice that the second answer gives two responses: Methane, and refrigerator magnets.

Why do we care about unique addressability? Precisely because, while we might raise no debate about methane (well, many of us have no quibble with that claim), refrigerator magnets as causally related to climate change is worthy of a good defense in the face of predictable skepticism. Thus, an important heuristic in any structured information system is:

One topic per node.

We can then respond to a node which answers with "refrigerator magnets" with whatever evidence we have to support or refute the claim; it's confusing to have to single out that topic from a multi-topic node.