Huh. Based on a cursory understanding of type roles, I would also say that its role should be representational. However, I'm hesitant to change the role manually if the compile inferred nominal for a good reason — why did that happen? 🤔

I think it happened because way down in what a Behavior actually is is something with a GADT or something that forces us to be nominal. This means we'll probably need to employ a trick to get this to be representational (usually I do this with a Coyoneda-like existential mapping function). I'll give this a shot and open up a PR!