Some fitnesses can be measured on individuals and used to infer trait fitnesses. The latter reflect actual/possible future organisms, other elements in population-environment systems. I argue that researcher interests specify what count as real population-environment systems and thus selection. 8/9