I've been thinking about what happens when AI systems start depending on data captured in the physical world. Getting images and videos is relatively straightforward. Knowing where they came from, when they were recorded, and whether their records can be independently checked is much harder. That's the problem @vangrid_io is working on. Every capture receives a digital fingerprint containing a content hash, approximate location, and timestamp. Those records are grouped into a Merkle tree, with its root anchored on Base using the Ethereum Attestation Service. Another detail I appreciate is the privacy approach. Faces and license plates are blurred on the device before the footage gets uploaded. But there's an important distinction: verification of a recording's existence and integrity doesn't automatically establish that everything depicted is factual. It gives other parties a traceable starting point for further verification. With its web and Android applications live since July, Vangrid is already testing this approach in practice. For me, the interesting question is how reliable and useful these records become when the network starts handling much larger volumes of real-world data.