📊 How we measure it

• Time frames: Every report is checked 1, 3 and 7 days after publication. The figures above use the 1-day check; the filter (12h, 1 day, 1 week, 1 month) selects how recent the checks are.
• Price Data: Uses yfinance for historical prices (closing prices).
• Classification: AI assigns each article to a market and rates its tone from −1 to +1. Above +0.25 counts as "up", below −0.25 as "down", in between as "sideways". A price move counts as sideways when it stays within ±0.2 % on one day (scaled with the square root of the days for longer horizons).
• Right or wrong: A report is right when the direction of its tone matches the price direction, including sideways to sideways.
• Yardsticks: A hit rate alone says little. We show two comparisons: chance — how often the same mix of up, down and sideways calls would be right if shuffled against the same price moves — and "always up", the share of moves that actually went up. Unless the hit rate is clearly above both, the tone does not predict the direction.
• Calibration: The curve groups reports by the strength of their tone and shows how often each group was right. If it followed the diagonal, a stronger tone would be more reliable.
• What this is not: Not a forecast and not investment advice from AIamond. The figures describe how well headlines matched past price moves; they say nothing about future prices.
• How often each news source is right, against chance →