PyBayesDM: Hybrid Framework for Bayesian Adaptive Management under Deep Uncertainty

Decision-making under deep uncertainty presents a persistent challenge for adaptive management in complex systems. Standard Bayesian Decision Theory performs well for prescriptive optimization under confident beliefs but is less robust when uncertainty is profound. Philosophical frameworks, which emphasize resilience and learning, typically lack a formal integration with Bayesian rigor. This paper develops a Hybrid Bayesian Adaptive Management Framework that unifies these prescriptive and philosophical paradigms within a single Bayesian formulation. The framework formally unifies both decision rules as parametric instances of a Bayesian adaptive management problem. It employs an entropy-driven dynamic weighting mechanism to continuously adjust the balance between efficiency and resilience based on posterior uncertainty, and introduces a suite of performance metrics designed to evaluate multi-dimensional outcomes beyond conventional measures. Experiments conducted on simulations inspired by climate-vulnerable agriculture in Vietnam show that the hybrid approach achieves statistically significant improvements (p < 0.05) in resilience and risk-adjusted performance under high uncertainty. The prescriptive framework excels in short-term reward but degrades under increased noise, while the philosophical framework maintains robustness, sometimes at a cost to efficiency. The hybrid strategy delivers balanced performance across uncertainty levels, confirming the utility of entropy-based arbitration. All experiments are reproducible using the developed Python implementations. This work offers a principled approach to adaptive decision-making under deep uncertainty, with implications for climate adaptation, resource management, and policy design