RESPONSIBILITY-WEIGHTED AGGREGATION OF QUALITY CRITERIA IN MULTI-LAYER IOT SOFTWARE ARCHITECTURES

Abstract

Additive quality models are widely used in architectural evaluation because they are transparent, computationally simple, and suitable for integration into ranking and optimization procedures. However, in multi-layer Internet-of-Things systems, directly summing the Edge, Fog, and Cloud contributions introduces a structural bias: the resulting score depends not only on criterion fulfillment but also on the number of layers in which the criterion is realized. This leads to inter-layer double-counting, destroys the unified interpretation scale of criteria, and complicates cross-criterion and cross-scenario comparison. To address this problem, the paper introduces the layer-responsibility matrix W, which distributes the total responsibility for each quality criterion across architectural layers. A corrected aggregation formula is derived as a weighted sum of normalized layer-level contributions under row normalization of W. The paper also provides a lightweight elicitation procedure that allows architects to instantiate W from scenario characteristics, control logic, dominant risks, and computational placement. The basic properties of the proposed formalism are established, including non-negativity, boundedness, invariance with respect to the number of layers, scenario adaptivity, and interpretability. A numerical example demonstrates how the proposed mechanism eliminates inflated criterion values while preserving the linearity of aggregation. A decision-level numerical example further shows that responsibility-weighted aggregation can reverse the ranking of candidate portfolios and thereby change the architectural decision outcome. The approach is further illustrated through two case studies, infrastructure monitoring and control, and bionic prosthesis software, showing that the same aggregation rule remains valid across domains, whereas responsibility distributions vary according to domain logic. The results justify treating W as an independent component of formal decision support for IoT software architecture.