About JEV and the rules of the site

JEV is a very fast, cheap, typed classifier. It is TypeSafe's System One model. You give it a headline and a set of typed questions. It gives back typed answers, in about 150 ms.

We also run two local models on our own GPUs. They check the official answer. The status page shows how often they agree.

JEV does not write news. It reads it. For each headline it answers four questions:

• Impact. Market moving, notable, routine, or promo and noise.
• Sentiment. The model view of the effect on the asset price: positive, negative, or neutral.
• Category. For example regulation, hack, ETF, stablecoin, or DeFi.
• Sponsored odds. How likely the article is a paid promotion.

1. Collect. A collector reads public feeds of crypto outlets and public market data all day.
2. Read. JEV rates each headline. A story that many outlets carry is grouped, so you read it once.
3. Log. A high-impact story becomes a public signal. We store the price at that moment.
4. Grade. We record the price 1 h, 4 h, and 24 h later. The scoreboard shows the hit rates, and it shows the misses.
5. Publish. Nothing is edited after the fact. A price that arrives too late is not used. The cell stays empty.

The full rules are on the methodology page.

• We show a dash when we do not know a value. We never invent one.
• Every panel shows the age of its data. A stale panel says so, and a banner names it.
• A hit rate needs at least 5 samples. Below that, the page shows a dash.
• Simulated results run privately and are not shown here. See the track record page.
• Headlines link to their outlets. We do not write or edit them, and we do not copy article text or images.
• A token risk label is an automated score from public checks. It is not a finding of fraud. You can dispute a label.
• The site has no ads and no tracking scripts. It loads no font or code from a third party.

General information, not investment advice or a recommendation. Automated ratings can be wrong. Signals and simulated results are historical or simulated and predict nothing. A signal is a model output. A token risk label is an automated hint. Do your own research.

The free JSON API and the RSS feed carry our derived data: news ratings, signals, the scoreboard, token risk blocks, and system health. A link back to cryptonews.org is welcome. Read the terms first.

Found a wrong number, a bad label, or a broken page? Write to desk@cryptonews.org. We answer within 2 business days. See corrections and disputes.