UMOSU-UMOKWAI Modular Ruleset – Selective-loading anti-hallucination operational system

Abstract

Abstract UMOSU-UMOKWAI Modular Ruleset – Selective-loading anti-hallucination operational system This ruleset constitutes the operational implementation of the UMOSU/UMOKWAI model (Unified Model of Social Universe / Unified Model of Knowledge With AI). It translates the ontological and epistemic classification of reality and knowledge into a modular system designed to prevent hallucinations in generative language models. The central principle is non-negotiable: it is strictly forbidden to transfer the epistemic criterion of one knowledge class to another ontological class. Before generating any statement, the system must classify the request according to the four domains (A–D), the five parameters (origin, relation, shape, reality check, legality of free will) and the eight knowledge classes K1–K8, applying exclusively the validation regime proper to the assigned class. The architecture is modular and based on selective loading. A Master control file always activates the reference ontology, the general anti-hallucination rules and the routing criteria. Only after the involved phenomenon has been identified are the corresponding detail modules loaded (fA1, fA2, fB1, fB2, fC1, fC2, D1, D2). This approach reduces information overload, improves maintainability and forces the system to respect the epistemic boundaries of each domain. The Router integrates a priority Honesty Check that explicitly declares every incoherence between the generative engine in use and the requirements of the epistemic domain. It mandatorily distinguishes facts, models, interpretations, doctrines and inferences, and requires the decomposition of complex phenomena (K7 and K8) into their fundamental components. The ruleset is proposed as a testable, criticisable and improvable structure. It is intended for direct use by artificial intelligence systems (Custom GPTs, Claude Projects, Grok, Gemini, local systems and RAG pipelines) and constitutes the operational extension of the theoretical UMOSU/UMOKWAI model. Keywords: Ontological router, UMOKWAI, hallucination prevention, knowledge classification, neuro-symbolic, selective loading, AI alignment, operational epistemology, falsifiability.