The Michigan State compound-gene model is a clean example of how AI science stories get made: a real result, a strippable caveat layer, and a coverage template that has been run enough times to feel like journalism. The result is not false — the model does what the preprint says it does. What the coverage omits is the question of what that result means inside a field where dozens of similar models are being generated monthly, each arriving with its own press release, none arriving with the peer-community friction that…