Generic Iterative Refinement Funnel: a parameterized architectural pattern for adaptive interactive diagnostic systems

Abstract

The article is devoted to the development of a parameterized version of the architectural pattern *Iterative Refinement Funnel* (Generic IRF v3.0), aimed at enhancing the flexibility, scalability, and domain independence of interactive diagnostic systems. The proposed approach eliminates the limitations of earlier implementations, in which key parameters were hard-coded, thereby complicating adaptation to new domains and necessitating modifications to the program code. In Generic IRF v3.0, the principle of full configurability is implemented, whereby system behavior is altered through parameter settings without intervention in the algorithmic core. The paper introduces a multi-level parameter system comprising strategic, threshold, functional, and domain levels. Strategic parameters define the logic of iterative hypothesis space reduction, including the selection of heuristics, response processing, and termination criteria. Threshold parameters establish numerical constraints to ensure robustness, while functional parameters describe computational procedures, including evaluation metrics and informativeness functions. Domain parametersfacilitate the integration of domain-specific knowledge. The proposed architectural solution is the *Configuration Layer*, which centralizes the management of all variability points and ensures a separation between invariant logic and configurable components. Within this layer, a system of domain presets is implemented for typical application scenarios, including medical diagnostics, technical support, legal classification, and recommendation systems, thereby accelerating deployment and reducing adaptation costs. Additionally, a mechanism for automatic parameter optimization based on Grid Search with support for A/B testing of configurations is proposed. This enables empirically grounded selection of optimal settings, taking into account data characteristics and diagnostic quality requirements. Experimental validation across four domains confirms the preservation of diagnostic accuracy (83.3–84.1%) while maintaining full compatibility with existing code and enabling adaptation to new domains through configuration alone, without code modifications. The obtained results demonstrate the feasibility of using the proposed approach as a foundation for building universal intelligent decision support systems.

Keywords: architectural patterns, software parameterization, interactive diagnostics, configuration flexibility, domain-specific adaptation, Strategy Pattern, decision support systems.