What I mean is: this culture or place trains this model, that culture trains that one; they don’t have to be squeezed into one big model like MoE. At inference time, each turn you know this turn should be computed by this one, that one plans, another verifies.