Each system receives zero to three points for the selected product context, zero to three for implementation fit, and zero to three for the priority. A score of eight does not mean the system is objectively better than one scoring six. It means the documented characteristics align more closely with those three inputs. Changing one input can and should reorder the shortlist.
Product context carries practical constraints. A Shopify embedded app benefits from matching the host admin, so Polaris receives a strong context score there and almost none for an unrelated product. A public-service site values clear forms, accessible defaults, and familiar interaction patterns, which raises USWDS. A distinctive React product gets more value from Radix because the behavior is supplied without a finished brand skin.
Stack fit asks whether the system has a maintained, first-party path for the technology that will ship. Guidelines alone are not implementation. A beautiful Figma library does not remove the engineering cost of missing components, framework adapters, focus behavior, or release maintenance. Treat a community port as a separate dependency with its own maintainer and release risk.
The priority breaks otherwise close decisions. Ant Design rises when speed and dense business workflows matter. Carbon rises when data and system breadth matter. Radix rises when brand control and accessible interaction foundations matter. The score deliberately does not include visual taste because a prototype is a better test of visual fit than a numerical preference.