GDMP – Graph-Dynamic Mapping Paradigm for Scientific Knowledge

OSF Registration / Preprint Draft Title: Tandem Rule-Based Structural Formalization of Multilevel Theoretical Models Using Large Language Models: the GDMP Case Subtitle: A methodology of tandem human–AI interaction for iterative construction, validation, and formalization of complex multilevel theoretical models Abstract: This work formalizes a methodology for tandem structural formalization of multilevel theoretical models based on an explicit system of rules, meta-rules, and iterative extrapolation using Large Language Models (LLMs) as instruments of analytical validation. The method employs tensor-graph representation, hierarchical organization of nodes and edges across abstraction levels, recursive redundancy elimination, and accumulation of rules as the model’s memory. A strict role separation is central: humans perform synthetic functions (generating new concepts, hypotheses, paradigms, paradoxes), while AI performs analytical functions (checking consistency, structural validation, redundancy elimination). The process is iterative and becomes possible only through persistent storage and accumulation of rules and meta-rules. The GDMP model serves as a demonstration. The publication fixes the methodology, not the detailed rules or internal structure, which remain part of a separate methodological document.