Agent pattern catalogs describe many possible architectures, but engineers still face the question of which pattern to pick and how to combine them. The paper argues that catalogs alone are not enough; what is missing is evidence to support those decisions. To address this, the authors extend a cognitive-function–execution-topology framework with an explicit, evidence-driven selection procedure.

The proposed procedure is grounded in two kinds of evidence: industrial case studies and controlled reference-system experiments. The former provides real-world context, while the latter offers reproducible, head-to-head comparisons under controlled conditions. The combination is intended to give engineers a more systematic basis for choosing among alternatives than relying on pattern descriptions alone.

The work is positioned as a practical step toward closing the gap between architectural knowledge and engineering decision-making. By embedding selection directly into an existing framework, it aims to make pattern composition a more rigorous and defensible process.