Abstract

Artificial systems increasingly exercise authority without direct answerability. The Constitutional Architecture of Hybrid Societies (CAHS) posits that legitimacy in human–machine governance depends on procedural couplings of agency, authority, and civic learning. This study empirically validates the CAHS mechanism of answerability with bite, operationalised as a Challenge Membrane granting pause rights, bounded response windows, and independent escalation during algorithmic decision processes. Using a stochastic agent simulation of a municipal resource allocator, we compare systems with symbolic recourse (explanations without effect) to those with an operational Challenge Membrane. Across 400 replications per condition, the membrane produced faster trust recovery after error, fewer legitimacy complaints, and markedly higher trace & escalation integrity. Results support CAHS theses on reversibility and contestable authority. All code and data are openly released.