Bayesian Meta-Experimental Design: A Recursive Framework for Coordinating Evidence, Selecting Experiment Portfolios, and Updating Scientific Inquiry

Scientific experimentation is increasingly undertaken within complex systems where multiple experiments, interacting variables, evolving hypotheses, and heterogeneous evidence must be interpreted simultaneously. Conventional Bayesian experimental design provides rigorous methods for selecting individual experiments, while meta-analysis synthesises completed studies. However, comparatively less attention has been given to methodologies that recursively coordinate portfolios of ongoing experiments across changing system states. This paper introduces Bayesian Meta-Experimental Design (BMED), a systems-oriented methodological framework that extends Bayesian reasoning from individual experimental decisions to recursive management of experimental portfolios. The framework integrates Bayesian updating, expected information gain, model discrimination, recursive evidence accumulation, experimental sequencing, latent relationship investigation, and adaptive portfolio selection within a unified decision architecture. Rather than replacing established Bayesian statistics, Bayesian experimental design, causal inference, or meta-analysis, BMED provides a higher-level organisational methodology for coordinating scientific inquiry across multiple concurrent investigations. Bayesian inference is treated as one component within a broader recursive research process that incorporates system state estimation, uncertainty management, relationship discovery, experiment scheduling, and iterative refinement of competing scientific explanations. The framework introduces concepts including Bayesian meta-states, recursive evidence ledgers, experiment dependency networks, portfolio optimisation, model discrimination utility, transportability assessment, replication planning, adaptive stopping criteria, and recursive hypothesis revision. Emphasis is placed on transparent uncertainty management, reproducibility, falsification, and disciplined interpretation of evolving scientific evidence. Potential applications include engineering, artificial intelligence, medicine, agriculture, environmental science, economics, organisational research, defence science, and other domains characterised by interacting variables, incomplete knowledge, and adaptive experimentation.