We tracked 1,472 claims from 420+ public figures and 300+ prediction market questions. Our 4-model AI consensus beats human markets by 6.7 points. The data speaks for itself.
If you follow high-credibility sources — the people Crene scores in the top quartile — you're getting information that's right nearly 80% of the time on market and economic predictions. Follow the bottom quartile, and you're barely better than a coin flip.
This isn't about who's famous or who has the most followers. Warren Buffett scores 95% not because he's Warren Buffett, but because his public predictions about markets and the economy are verified correct at that rate. Cathie Wood scores 0% on market calls — not because we dislike her, but because the data says so.
The gap compounds. Over a year of following high-credibility sources vs. low-credibility ones, the informational advantage is enormous. This is the problem Crene solves: separating signal from noise, systematically.
We run 4 frontier AI models against every active prediction market question on Polymarket and Kalshi. The consensus outperforms human traders.
Sources with 2+ verified market/economy claims · Updated daily
Full dataset: 420+ sources
Get the complete leaderboard with historical accuracy trends, model-by-model breakdowns, and CSV export.
Unlock Full Rankings — $29/moHow we verify predictions
We continuously crawl news sources, interviews, social media, and public statements from 420+ tracked figures. When a source makes a verifiable prediction — "inflation will fall below 3% by Q2" or "Tesla will hit $300 by year end" — our system extracts and logs the claim with its date, context, and specifics.
Each claim is independently evaluated by four frontier AI models: Claude (Anthropic), GPT-4 (OpenAI), Gemini (Google), and Grok (xAI). Each model receives the original claim, the current state of the world, and relevant data. They independently assess whether the prediction came true.
A prediction is marked "correct" only when 3 or more models agree. This consensus mechanism reduces individual model bias and catches edge cases. Claims where models disagree are flagged for review. The system produces three outcomes: correct, wrong, or partially correct.
Each source receives a Crene Credibility Rating (CCR) based on their verified prediction track record. The score weights recent predictions more heavily and adjusts for claim volume. Sources with more verified claims have more statistically robust scores. Scores update daily as new verifications complete.
The data compounds daily. Every prediction verified makes the signal stronger.
Crene, Inc. · San Francisco · crene.com