What happens when an AI model encounters something it has never seen before? For Physical AI, this is not an edge case that can simply be ignored. Real environments contain unfamiliar objects, changing conditions, and unexpected interactions. A diverse dataset can help expose models to more of these situations, but collecting useful data at scale also requires clear task requirements and reliable quality checks. That is why I think the infrastructure behind real-world data collection deserves more attention. @vangrid_io is working in a space where connecting data contributors with AI training needs could be important. Would you prioritize collecting rare edge cases or expanding coverage of everyday environments first?