Beyond Evaluation: World Notions, Normativity, and the Structural Simulation of Ethics in AI Systems | Synapse