> ## Documentation Index > Fetch the complete documentation index at: https://docs.vangrid.io/llms.txt > Use this file to discover all available pages before exploring further. # How verification works > Checks at ingest, the capture fingerprint, anchoring on Base, and the verification layers that are not live yet. Crowd captured data is worth nothing to a buyer who cannot distinguish it from fabrication. Verification is therefore not a feature of Vangrid, it is the condition under which the data has value at all. Vangrid verifies captures in layers, from cheapest and most automatic to most expensive and most adversarial. Three layers are live. The rest are marked as not live yet and written in the future tense. ## Live: checks at ingest Every capture is validated when it arrives, before it can be submitted to a bounty: * Video quality and length * Consistency between the camera and location signals * Duplicate detection against captures already in the network Submissions that fail are rejected with a reason. ## Live: the capture fingerprint Every capture is fingerprinted as it is recorded. The fingerprint contains a hash of the content, a coarse geohash, and a timestamp. The footage itself is never part of it. Only the fingerprint is derived, which means a capture can be proven unaltered without anyone seeing what it shows. Anyone holding a capture can recompute the hash and compare it against the record. Altered or fabricated data fails that check. ## Live: the fingerprint anchored on Base Fingerprints are accumulated over a short window and committed together as a Merkle tree. The root of that tree is written to the Base blockchain as an attestation, and a new window closes roughly every quarter of an hour. Batching is what makes anchoring affordable at the rate captures arrive: one transaction carries a whole window rather than one transaction per capture. It costs a small amount of temporal resolution, since the chain establishes that a fingerprint existed by the end of its window rather than to the second. What a holder gets in return is an inclusion proof. Given a capture and the short sequence of sibling hashes that accompanies its record, anyone can recompute the root and compare it against the one published on chain. The check terminates against the public ledger rather than against an assurance from the party selling the data, and no step of it requires a response from Vangrid. Anchoring proves existence, ordering and integrity. It does not by itself prove that the scene is what it claims to be. The layers below address that. Contributors manage none of this. No wallet pop ups, no gas, no waiting. ## Not live yet: device attestation at capture time Hardware backed attestation will confirm that a capture originated on a genuine, unmodified device before it is accepted. Until it ships, the fingerprint records what was captured and when, and says nothing about the device that produced it. ## Not live yet: sensor cross correlation Camera, GPS and motion signals will be checked against each other, so that a spoofed location fails a physics check rather than a policy check. ## Not live yet: independent verification under collateral Automated checks resolve the cheap attacks. They cannot resolve a capture that is technically well formed and substantively false, and a central review team is both a bottleneck and a single point of trust. Verification of those cases will be delegated to independent operators. Captures will be sampled and assigned pseudorandomly, so neither side chooses the other. Operators will sign attestations that attach to the capture's provenance record, backed by collateral that is forfeited when an attestation is found false. The point is not decentralisation for its own sake. It is that a buyer auditing a provenance record should be reading a claim someone was paid to make and can be penalised for making falsely, rather than an assurance from the party selling the data. See [Token utility](/protocol/token-utility) for the mechanism.