To predict the afternoon temperature of a greenhouse, a researcher models sunlight, ventilation, and the insulating cover. She uses measured material properties without explaining the molecular origin of each one. This does not make the model arbitrary: predictions are checked against temperature records, and the limits of the model are stated. Another researcher studies those molecular origins, but that work answers a different question. A useful explanation need not reproduce every explanation on which its inputs ultimately depend. The greenhouse model shows that a model can be useful when it ________. For example, the researcher checks whether the predictions remain accurate when ventilation increases. If they fail under those conditions, she revises the model instead of claiming that its previous success guarantees accuracy in every situation.