From Poka-Yoke to Decision Stability: Why AI-Supported Decision Architectures Require Diagnostic Methods Beyond Traditional Quality Assurance

Artificial intelligence is increasingly embedded into operational decision processes where correctness alone may no longer provide sufficient governance evidence. Building on principles from Lean Manufacturing, system safety, and emerging AI governance research, this paper introduces Decision Stability as a conceptual framework for understanding how AI-supported decision architectures behave under operational uncertainty. Rather than focusing solely on correctness, compliance, or model performance, the paper argues that future governance capabilities may increasingly depend on the stability, observability, and governability of decision architectures. Replay-based governance diagnostics are discussed as one possible architectural response for making emerging instability observable before visible operational failures occur. The paper is conceptual and proposes a research agenda for empirical validation.