That's a really interesting question, and it's one I'd love to answer in a future work. This blog mostly focuses on characterizing context depth limits.