Although large-scale language models (LLMs) and coding agents are commonly used in user interface (UI) development, developers find it difficult to reliably assess their creativity in visual design and interaction. Existing tests rely on human experts, who can accurately assess usability through critical flow tests but are slow and expensive, or on automated judges, who can measure but are imprecise and invisible. We present FlowEval, an index-based framework that measures whether a generated UI supports interaction flow by comparing navigation traces from real websites to traces from generated analogs using index-based similarity metrics (eg, dynamic time warping). In a small study with professional UI testers, we show that index-based metrics are highly correlated with human judgment, suggesting that they can provide a harmless but reliable test for UI production systems.
- † Purdue University
- ** Work done while at Apple
