METHODOLOGY-GOVERNED AI FOR REGENERATIVE URBAN PLANNING

Abstract

This study tests methodology-governed LLM execution for regenerative public-realm planning in De Marken, Almere. Under a frozen municipal corpus, it compares a structured non-SCM baseline (A), single-pass Sustainable Community Masterplan (SCM) execution (B), and a staged, role-decomposed SCM workflow (C). Explicit SCM specification strengthened relational territorial reasoning and regenerative direction. Staged execution better preserved that direction through portfolio re-validation, delivery structuring and rendering. Research scores ranked C>B>A; a blind municipal evaluation ranked C>A>B. Methodological gains do not automatically become decision usefulness: overall usefulness stayed low in both single-pass conditions and rose only under staged execution. The study distinguishes specification from execution architecture and treats methodology-governed AI as municipal decision support, not autonomous planning.