I now have a notion of a game world, consisting of both the actual state of cells and their channel arrangements, as well as the player’s guessed arrangements. (Not that there’s yet any interface to let the player guess.) Here’s the latest screenshot; click for a full-size version.
To be honest, I’m surprised that there aren’t too many small cycles. Pure randomness already does a decent job of generating long meandering paths. I was worried that I might need to massage the random data some to get it into a more useful form, but I might be able to work with this as is.