A prompt can produce a game that looks like the one a creator requested. That leaves the question of whether someone can move through it, use the controls and play with other people. Roblox says its evaluations for Build follow that whole path, from the prompt to the resulting experience.
Build combines generated code, art, controls and game rules. The company reports that evaluation helped teams prioritize whether a game reflected the creator's request, worked in three dimensions and felt polished. It also tracked outcomes such as whether someone published after a playtest.
Roblox reports improvements after the tool's New Zealand alpha, but says those figures cannot establish how much the evaluation process caused. The framework is worth looking at because it defines what counts as a game that works beyond the quality of each component taken on its own.