Multi-cluster models that may introduce new terms are definitely okay. Honeycrisp apples are an example of a modern ag type that became noticed by consumers and commanded a premium market price. Bulk foods may come from a mix of genus – in a meal or even in an individual farmer’s field. So mixture models can be literally causal. On the other hand, human participants will invevitably apply regularization to their comprehension of results, so simpler is better for that non-stat reason too.