The hardest dataset to collect might be the one sitting right outside the lab. Physical AI needs to understand actual environments, not just simulated ones. But traditional mapping has obvious limits. Satellites can't see inside buildings. Road-mapping fleets aren't walking through warehouses. And text datasets tell you almost nothing about the geometry of a loading dock. This is the gap @vangrid_io is trying to fill. Instead of waiting for a mapping fleet, a requester can commission a specific site and fund the capture. A contributor uses a smartphone to record the environment. Privacy is handled at the edge, while capture provenance is recorded through cryptographic proofs and Base-based attestations. I think that's the more interesting part of the model. The network isn't just collecting clips. It's creating a way to request, verify and settle real-world spatial data for a specific place. For Physical AI, that could become a very different kind of data rail.