The physical AI data bottleneck is often treated as one problem. It isn't. There are at least two very different data requirements. Manipulation data can be collected in controlled environments. Spatial ground truth is different. The physical world changes continuously. A centralized lab can record thousands of hours of a robot manipulating objects. It can't continuously refresh ground truth across a changing physical environment at global scale. That's the structural problem @vangrid_io is targeting. The interesting part isn't just the distributed capture. It's where the computation happens. Raw video doesn't need to become blockchain data. The phone can process the capture at the edge, turn it into structured spatial data, while the protocol anchors its provenance onchain through hashes and EAS. The blockchain handles verification. The edge handles the heavy data. That architecture makes sense for physical AI. But it creates a different bottleneck. Can consumer devices process high-fidelity spatial data fast enough to maintain the refresh rate that autonomous systems will actually need?