What “verified”
actually means at SVN.

We confirm who shot the video and that the file we have is the file they sent.

Before anything else, we anchor the file. The moment a submission lands at SVN, we cryptographically hash it. Writing those hashes to a permissioned blockchain ledger (VeriFact) is implemented in our private beta environment and is a planned roadmap feature for public production — it is not yet anchoring every public submission on-chain. We also bind every file to a verified contributor account so we can answer one of journalism's hardest questions: who actually filmed this, and is what we're looking at the same file they sent us?

We read the file's own metadata and check it against the contributor's story.

Every video file carries hidden context — embedded creation timestamps, device model, codec signatures, and (sometimes) GPS coordinates. We extract all of it and compare it against what the contributor told us. If they say it was shot yesterday at noon in Kyiv on an iPhone 14, the file should agree. When platforms have stripped metadata (Instagram, WhatsApp, and most social uploads do this), we flag it openly rather than guessing.

We screen for synthetic generation, face-swaps, and obvious manipulation.

We run every video through a panel of detectors that flag signs of AI generation, face-swapping, and frame-level tampering. We prioritize free, open-source, and locally-run models — both for cost and because we want to be transparent about what we're using. No detector is perfect. The field is an arms race. We treat AI Detection as a signal that contributes to a final human decision, not as a verdict on its own.

We confirm the video shows what, where, and when the contributor says it does.

This is the work most people don't realize verification requires. We look at the scene itself. Do the shadows match the time of day? Is the architecture consistent with the claimed city? Do the signs, license plates, uniforms, and weather agree with the date? When GPS is available, we cross-reference it against satellite imagery. When it isn't, our editors use OSINT techniques — landmark matching, street-view comparison, weather records — to independently confirm location and time.

We check whether other credible sources independently report the same event.

A single video, no matter how convincing, is one data point. We look for independent corroboration — other footage of the same event from different angles, wire reports, local journalists on the ground, official statements. When we can corroborate, we say so and link to what we found. When we can't, we say so plainly and explain why we still published (or didn't).

A senior editor reads every signal above and makes the final call. No exceptions.

Automation gets us most of the way. It does not finish the job. Every video that reaches publication is reviewed by a Senior Editor (a role; named reviewer rosters are still under legal and editorial review) who reads every signal from the first five steps, weighs ambiguities the system can't resolve, and signs off on the decision. Editors can reject content that passed every automated check if context, ethics, or completeness demand it. No video publishes without this sign-off.