Tried looking at what happens when I tack on some extra information to the translation of the atomic-normal coercion into vertex-edge form; it apparently gets me back injectivity on all normal linear terms, but I can't make any sense of what it's telling me.
