A multi-dimensional resilience assessment framework for indoor aging-friendly environments: LLM-assisted indicator construction using Google Gemini Pro

Assessment of aging-friendly housing commonly separates accessibility, functional support, indoor environmental quality, and safety technology, making it difficult to examine how dwellings accommodate changing occupant and environmental conditions. This framework-development study introduces SAFER (Spatial, Adaptive, Functional, Environmental, Responsive), a five-dimensional resilience-oriented indicator system. A four-stage workflow combined a systematic review that retained 186 publications and yielded 312 candidate indicators, Gemini Pro-assisted first-pass screening, expert content-validity assessment, and a preliminary application to six dwellings. Eighteen resilience-framework indicators were added to the review-derived pool, and 43 multidisciplinary experts assessed indicator importance and content validity under research-team oversight. The resulting expert-reviewed framework contains 47 indicators. All 43 final AHP matrices satisfied the prespecified consistency criterion (CR ≤ 0.10) and were included in the group aggregation; the dimension-level AHP weights were Spatial (0.162), Adaptive (0.231), Functional (0.145), Environmental (0.264), and Responsive (0.198), with Environmental and Adaptive highest. A preliminary application to six purposively selected dwellings (n = 6) produced primary AHP-weighted scores from 42.8 to 80.9. Equal, case-derived entropy, and 0.5 AHP–entropy hybrid weighting were used only as secondary sensitivity analyses; the case ordering was unchanged under these alternatives, although the small, non-independent case set means that this sensitivity result does not establish external validity. SAFER should therefore be understood as an expert-reviewed framework with aggregate content-validity evidence and preliminary feasibility results, not as a fully validated assessment instrument. Larger independent samples, preserved assessor-level records, and cross-model LLM comparisons are required before normative thresholds or broad policy use.