Ni_Stylized Grey Space Modalities as a Design Language_IEEEAccess

We recast urban grey space\u2014the threshold between building and street\u2014as a controllable language of form. A stylized 15 dimension \u00d7 5 intensity codebook (75 stimuli) isolates modal levers for canopy, edges, seating, shading, signage, commerce, crowd, facilities, and more. Two cohorts evaluated each scene with 5 point instruments aligned to eight perceived affordance constructs and multiple outcomes; N=701 public and N=220 experts passed quality control. Regularized models using the 15 D modality vector alone predicted composite outcomes with R\u00b2\u202f=\u202f0.722 (public) and 0.665 (experts), establishing strong, held out fidelity. Reliability and correlation structures revealed a shared affordance backbone\u2014safety & control, legibility & wayfinding, boundary usefulness\u2014that organizes perception across cohorts, while publics weighted microclimate\/comfort and experts weighted cleanliness\/operability. A representation level adversarial deletion (linear residualization of the latent with respect to affordances) yielded R\u00b2_full\u2192R\u00b2_noM = 0.722\u21920.756 (public) and 0.665\u21920.757 (experts), indicating no additional unique variance is required from measured affordances at this resolution; the stylized morphology already encodes their predictive content. The findings justify decision ready rules: prioritize moves that consistently raise outcomes in both cohorts\u2014e.g., shaded seating with clear sightlines and porous edges\u2014while tuning cohort specific modifiers. The approach turns grey space into a reproducible design language that links form, affordance, and outcome, and tests mediation necessity with an explicit adversarial stress test.