How we verify

Verdiktum keeps score on public predictions. Every verdict follows the same standardized, journalistic process — so a psychic, an analyst and a politician are all measured the same way.

The process

1. Quote & source. We capture the verbatim claim with its source and publication date. Each claim links back to the exact moment in the source video.
2. Verification. The claim is cross-referenced against multiple sources. AI does the heavy lifting at a scale humans couldn't match; the framing stays neutral.
3. Verdict. We issue a rated verdict with a short justification and the sources behind it.

The verdict scale

A resolved claim is scored as a hit (e.g. fulfilled, true, occurred) = 100%, partial = 50%, or a miss = 0%. Pending and unfalsifiable claims are excluded from the average — we do not force a verdict where one isn't possible.

Prediction power

Not every correct prediction is equally impressive. We score prediction power (1–10) for specificity, lead time and how precisely the claim matched what happened. A vague claim that "could have been a coincidence" scores low even when it's marked a hit.

Calibration

Calibration asks a simple question: when a speaker sounds confident, are they actually right more often? We compare stated confidence against real accuracy across a speaker's record.

Contradictions

We also detect when a speaker said contradictory things across different videos — with links to each statement.

AI + journalism, kept neutral

Verdicts are produced with the help of AI and are always labelled as such. A verdict is an analytical assessment, not a legal ruling. We apply the same rubric to everyone, limit quotes of living people to short excerpts, and offer a right of reply to anyone featured.