Residential floor-plan evaluation requires both computable geometric measures and experience-based judgments of layout quality, including lighting and ventilation, functional zoning, circulation design, and privacy. Existing automated methods often provide limited traceability when qualitative architectural knowledge is involved. This study presents an auditable and modular workflow for interpretable residential floor-plan evaluation. The workflow first converts parametric floor-plan geometry into structured spatial evidence describing room functions, daylight conditions, adjacency relations, entrance relations, and selected inter-room distances. A residential-design knowledge graph is then used to retrieve relevant architectural knowledge. A multimodal large language model jointly interprets the structured evidence, the floor-plan image, and the retrieved knowledge to identify which predefined rules are triggered and select the corresponding score bands, rather than generating unconstrained scores. Programmatic checks validate the output format, rule names, permitted score bands, and duplicate items before the validated rule-based scores aggregated with quantitative geometric indicators to produce dimension-level and comprehensive scores. For validation, among the 200 collected floor-plan pairs, 50 pairs with unanimous expert preferences were retained for the primary validation set. Under this validation protocol, the final configuration achieves an Expert Preference Agreement Rate (EPAR) of 84%. Component ablations demonstrate the effects of structured spatial evidence, explicit scoring rules, and knowledge retrieval on agreement with expert preferences. The modular design also provides a basis for replacing the underlying model, retrieval settings, or scoring rules within the same auditable workflow. Disagreement cases identify directions for further refinement. The workflow therefore provides an auditable and interpretable basis for residential floor-plan evaluation.
No takes yet. Share an insight, caveat, or question.
Bu et al. (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: