Aerodynamic stability and control modelling for high-agility configurations relies on heterogeneous data originating from wind-tunnel experiments, computational fluid dynamics simulations of varying fidelity, and flight-test measurements. These data sources differ not only in accuracy and resolution, but also in their uncertainty characteristics, which are often heteroscedastic, regime-dependent, and sample-specific. Classical multifidelity modelling approaches in aerodynamics predominantly rely on hierarchical formulations with predefined fidelity levels and fixed correlation assumptions, which can become restrictive for such heterogeneous datasets. This work investigates non-hierarchical Bayesian data fusion strategies for probabilistic aerodynamic modelling that operate on a sample-wise basis and explicitly account for uncertainty. Two complementary approaches are considered. Product-of-Experts models are employed as a principled Bayesian mechanism for combining independent predictive distributions without requiring an explicit fidelity hierarchy. In addition, weighted Bayesian neural networks are explored as a flexible alternative in which fidelity is not prescribed a priori, but emerges implicitly as a latent, continuous, and sample-dependent property learned from data. This formulation allows the model to adapt locally to varying data quality and to distinguish between uncertainty due to data sparsity and epistemic disagreement across data sources. A central focus of the study is calibrating predictive uncertainties to physically meaningful scales, which is essential for high-agility aerodynamic applications. Elevated uncertainty levels are interpreted as indicators of local model inadequacy or regime transitions rather than mere statistical noise. Although the proposed methods are model-agnostic, they are particularly relevant for CFD-based aerodynamic datasets, where uncertainties arise from discretization effects, turbulence modelling assumptions, and numerical approximations across different simulation setups. The work aims to provide practical insights into uncertainty-aware multifidelity fusion beyond classical hierarchical assumptions, with relevance for safety-critical aerodynamic analysis and decision-making.