Scoring the news is easy. Measuring what it did is the point.
Most news tools stop at "bullish" or "bearish". Evident scores each headline, then measures what the price actually did 1, 4 and 24 hours later against the market — and only shows an expectation when enough history exists.
Most news tools for traders stop at a label: this headline is bullish, that one is bearish. A label is an opinion. What a trader needs to know is whether headlines like this one have actually moved the price — and by how much, compared with everything else that was moving at the time.
Step 1: collect
Evident collects headlines from 13 publishers, including CoinDesk, Cointelegraph, The Block, CNBC and MarketWatch, plus per-ticker feeds. Each headline is stored once (duplicates across publishers are merged), tagged with the assets it mentions, and linked to the original article. We keep the headline, the link and a one-line summary of our own — the article itself stays with its publisher.
Step 2: score, carefully
A language model scores each headline for tone (from very negative to very positive), importance and category, such as regulation, listing, hack, macro or earnings. Two details matter:
- Calibration. A small model we tested called almost everything important. We compared models on the same set of headlines and now score with a larger model (Llama 3.3 70B) and an explicit scale for what "important" means.
- Opinion is capped. Commentary and price predictions ("analyst says BTC could reach…") are limited in importance. An opinion about the market is not news about it.
Step 3: measure
This is the part that labels skip. 1, 4 and 24 hours after publication, Evident measures what the asset's price did and subtracts what the market did in the same window (the benchmark: Bitcoin for crypto, the S&P 500 for US stocks, the BIST 100 for Borsa Istanbul). The remainder is the headline's abnormal move, and it is compared with the asset's normal volatility, so a 2% move on a quiet stock and on a volatile coin are not treated the same.
Step 4: expect — only with evidence
"After similar headlines" shows the median measured move of past headlines with the same category and tone. It appears only when at least 8 past cases exist. With fewer, Evident says there is not enough history instead of making a number up.
Where it shows up
The scored and measured feed is in Markets → News, and the AI agent receives important, scored news for its asset as part of every decision — as one input among many, not as a trigger.
Measured news impact describes the past. It is not a forecast and not investment advice.
Evident does not give investment advice. Past performance does not indicate future returns. Leveraged and automated trading carries the risk of losing capital.