How We Forecast

Where the numbers come from, and how to read them.

Every number on this site is a market-consensus forecast: the aggregated judgment of thousands of people who back their opinions with real money on regulated prediction exchanges. When new information arrives — a poll, an injury, a data release — the consensus moves within minutes, long before the story is written up. We read those moves every morning and publish the ones that matter.

Why consensus odds beat pundits

Decades of forecasting research point the same direction: aggregated, incentivized crowds outperform individual experts at predicting near-term events. A pundit pays no price for being wrong. The consensus does — every day, in public. That makes it the most honest forecaster available, and it is the only source we use.

What each section means

• Today’s Number — the single most newsworthy probability of the morning: the biggest overnight move that is still genuinely undecided.
• What Moved — forecasts that shifted at least 4 points since yesterday, ranked by how much money is behind them. A 4-point overnight move in a liquid forecast almost always means real news.
• Where Forecasts Split — the rare mornings when two independent consensus sources disagree meaningfully on the same question. One of them is mispricing reality; watching which one corrects is its own story.
• Up in the Air — the highest-stakes open questions: forecasts with the most money behind them that still sit between 10% and 90%. These are the stories the world genuinely hasn’t decided yet.
• Basically Decided — outcomes at 93%+ that settle within a week. The consensus has effectively called it; the headline just hasn’t been printed yet.

What we deliberately leave out

Thinly-traded questions (not enough money behind them to mean anything), near-certainties that were already near-certain yesterday (99% → 99.5% is not news), and hourly financial-noise brackets. If a number is on this site, it cleared every one of those bars.

The honest caveat

A 70% forecast fails three times in ten — that is what 70% means. We publish probabilities, not predictions, and we keep every past edition permanently archived so you can judge the record for yourself.