Automated detection of false information is crucial for combating “fake news” on online social media networks (OSMN), reducing the need for manual discernment. Existing solutions often use content or context features of OSMN documents in isolation, ignoring temporal and dynamic changes, which limits model robustness. Furthermore, the quality and impact of OSMN documents’ features on prediction trustworthiness are rarely considered. This paper introduces MAPX, a model-agnostic framework for evidence-based aggregation of predictions from existing models in an explainable, adaptive, and dynamic manner; assessing the quality of OSMN document features. Extensive experiments on benchmarked fake news datasets in real-world data quality scenarios show MAPX’s effectiveness, consistently outperforming state-of-the-art models. For reproducibility, a demo of MAPX is available at this link .