Can an LLM measure UI thresholds from screenshots? 164 crops vs DOM gold labels A developer benchmarked whether large language models can judge UI style-rule compliance from screenshots by comparing GPT and Gemini verdicts against DOM-derived gold labels across 60 contrastive rule pairs. GPT scored 54/60 (90%) while Gemini scored 41/60 (68%), with both models performing far better on visual rules like contrast than on numeric thresholds such as font size and line height. The developer concluded that image review is reliable for heading sizes, line counts, and line length, but pixel-adjacent threshold checks should stay in the DOM. People paste screenshots into an LLM and ask "does this page follow the rules?". I wanted a number for how far that works, with the answer key coming from somewhere the model can't see: the DOM. Ten rules, frozen from my own site's style rules the thresholds are mine, not a standard : I built the site at the commit before the fixes 7bec67f , opened 12 pages at 1280 and 375 in headless Chrome via CDP, and measured every rule with getComputedStyle / getBoundingClientRect . That gives 12 × 11 = 132 judgments R2 counts twice , of which 36 are "element not present on this page". 96 pairs remain. From the same render , the script clips 1:1 crops of the elements in question — 164 PNGs, SHA-256 of each in a manifest. That matters: my first attempt measured positions in an iframe and shot with headless Chrome, and at 375px the two disagreed by 300–700px, so six footer crops were blank white. // R1 = smallest font-size among text nodes inside