Plain English
How we measure disagreement
Experts vs. betting markets (the lead list)
A toss-up rating means the forecaster sees no clear favorite. A market price of 75% or more means traders do see one (about 3-to-1). We list a race when the Cook Political Report rates it a toss-up and every betting market on it (Kalshi, Polymarket, or both) gives the same candidate at least 75%. The reverse also counts: forecasters on average call it Likely or Safe, but every market gives the favorite 60% or less, or favors the other party. Recalculated on every update. Both sides of the comparison answer the same question, whether there is a clear favorite, so we compare a rating category with a win chance, never a poll lead with a price.
Where sources pick different winners: 1. Each source picks a side
For every race we ask three questions. Who do the forecasters favor, on average? Who leads our polling average? Who is the favorite on Kalshi and Polymarket? Each answer is Democrat, Republican or too close to call.
2. When a source counts as too close to call
Forecasters: their average rating is a toss-up. Polls: the leader is ahead by less than 2 points. Markets: the two parties' prices are within 5 percentage points of each other (for example 52% vs. 48%).
3. A split: they pick different winners
Forecasters and markets favor opposite parties: 50 points. Polls and markets favor opposite parties: 50 points. Kalshi and Polymarket favor different parties: 25 points. Races are ranked by the total.
4. Why we never compare the sizes
A poll lead is a share of the vote. A market price is a chance of winning. They are different scales: a candidate up 8 points in the polls can fairly be priced at 95% to win. So we only flag a race when the sources favor different winners, never because one number is bigger than the other.
What it is not
It is not our forecast and not a probability. It is simple arithmetic on figures we quote, each with its source and the time we pulled it. Markets reflect where traders put money, and polls have error, so a split is a reason to look closer, not a prediction.