physical ai conversations keep focusing on better models. the harder constraint is the training data. robots and world models need fresh, ground-level 3d captures from real places that cannot be scraped. @vangrid_io built a network where a buyer can request a specific site, a civilian captures it with a phone, and the result is anchored on base with onchain provenance. usdc settles through the protocol. over a million captures are already on the grid. the difference is that the data layer is live and queryable while most models are still waiting for usable real-world inputs. the grid is how that gap starts closing.