GENERALIZED EVIDENCE-TO-SYSTEM PROJECTION THEORY (GESPT): Reference-Conditioned Accuracy, Latent-Target Underdetermination, and System-Level Generalization

AbstractThe Generalized Evidence-to-System Projection Theory (GESPT) is proposed as an integrative framework for analyzing the relationship between available evidence and claims about a broader target system. Its central concern is the distinction between the accuracy of a description within an observed or otherwise documented reference set and the validity of extending that description to a larger, partially unknown, or temporally extended system.The framework is motivated by a criminal-case observation: records of detected offenders may support accurate descriptions of detected cases, while the same records alone may be insufficient to establish that every offender will eventually be detected. The absence of comparable records for undetected cases does not prove that those cases have any particular outcome. It establishes a limitation on what can be inferred from the available evidence without additional assumptions or information.The proposed framework distinguishes reference-set accuracy, target-system validity, evidence availability, selection, measurement, transformation, inference, projection, labeling, decision-making, and feedback. Its central Evidence-Boundary Non-Entailment Principle (EBNP) states that truth within a reference set does not, by itself, logically entail universal truth within a broader target set. The Reference-Accurate, Target-Underdetermined Mechanism (RATUM) describes the coexistence of an accurate analysis of available data and unresolved uncertainty about the broader system.A formal model is provided using reference and target sets, selection probabilities, estimands, projection gaps, and identifiability conditions. Several propositions follow under explicitly stated assumptions. Applications to statistics, causal inference, systems theory, information processing, cybernetics, network analysis, criminology, sociology, and machine learning are discussed as conditional extensions rather than presumed universal results.The document is a proposed theoretical synthesis, not an already validated universal law. Its originality, empirical performance, and domain-specific scope require systematic comparison with established methods and independent testing.