When a model's internal representation is structurally identical to the mechanism generating the data, enabling exact inference.