After the discussion and reactions yesterday, I still think that
a propagation graph is the right way to visualize a traversal.

The reason is that in 80% or more of the cases, the propagation graph
gives the correct information. If the propagation graph is inaccurate,
the tool should raise a flag.

The propagation graph is the folded traversal graph.

If after practical use, this solution is not satisfactory, we can later
attempt to visualize the traversal graph directly.

Binoy and Kedar: do you agree with this? Do you know where to find
the algorithm for flagging when the propagation graph is inaccuarate?
It has been implemented in Demeter/C++ by Cun Xiao.

-- Karl