I don't think it's currently worth burning the time to fix, but I want to note that I'm still not happy with how often the benchmarking gives incorrect results. It might just not be possible to test the scaling the way I am, since one operation transitions from $\mathcal{O}(n^m)$ to $\mathcal{O}(n^{m+1})$ depending whether the graph is purely depth-like or purely breadth-like, and the reality is that the test case is neither.
I don't think it's currently worth burning the time to fix, but I want to note that I'm still not happy with how often the benchmarking gives incorrect results. It might just not be possible to test the scaling the way I am, since one operation transitions from$\mathcal{O}(n^m)$ to $\mathcal{O}(n^{m+1})$ depending whether the graph is purely depth-like or purely breadth-like, and the reality is that the test case is neither.