For me, I find that ideas evolve best when I can experiment with their implications immediately. I am starting the production of a map against which I can experiment with low-level AI agents to which basic military operations can be delegated.
There are plenty of good data online,. Google Earth and Google Maps (and others) are well known. For raw GIS data, there are truly computer-readable alternatives. Note also that while rail, canal, and metaled roads impact transportation routes, the backbone of the low-level grid should be constant enough to make most of the data re-usable from the 18th century to now.
The map will not be directly (at least initially) a cartographic product. Instead, it will of course be a graph. Graph theory is good fun (and one of the few bits of math I learned at school for fun and have never forgotten), and there are lots of algorithms and analysis tools available to work with them.
For initial visualization I will probably use a tool like GraphViz, which draws graphs based on textual descriptions.