Contextual and Global Falsification of Scientific Models: An Integrated Theory of Epistemic Validity

This paper develops an integrated theory of the epistemic validity of scientific models by unifying central insights from Popper, Kuhn, Lakatos, model theory, and contemporary approaches to approximate truth. The framework introduces a two-level structure of falsification that distinguishes contextual from global model failure, formalizes approximate truth as a domain-specific similarity metric, and embeds model evaluation into an explicit utility function combining approximate truth, explanatory power, and model costs. A further contribution is the concept of the epistemic opportunity space E(t), which specifies the methodological, technical, and institutional conditions under which models can be formulated, evaluated, and stabilized.The theory reconstructs model practice in modern simulation-based sciences and is illustrated using climate-model ensembles. It provides a unified structure for understanding why models persist despite partial falsifications and under which conditions global elimination becomes epistemically justified.