Every poll, one honest average.

The average. Every general-election poll for a race goes into one weighted average. A poll counts for more when it is newer (its weight halves every 14 days), when it interviewed more people, and when it screened for likely voters. A pollster's second and third recent surveys of the same race count for less, so nobody floods the average. Polls paid for by a party or a campaign count half and are moved 2.5 points away from their sponsor.

The estimate. For Senate and governor races the polling average is blended with a fundamentals number: how the state voted for president in 2024, shifted by today's national generic ballot. Races with many recent polls lean on the polls. Races with one poll or none lean on the fundamentals. The blended margin and its uncertainty give each candidate a chance of winning, and that chance sets the rating: Toss-up under 55 percent, Tilt to 65, Lean to 85, Likely to 97, Safe above.

The balance of power. The map is simulated 20,000 times. Each run adds one shared national polling miss to every race plus a separate miss per race, because polls tend to be wrong in the same direction at once. The share of runs a party wins a chamber is its chance of control. Republicans need 50 Senate seats because they hold the vice presidency; Democrats need 51.

The House. There is no reliable poll for most of the 435 districts, and several states redrew their maps mid-decade, so the House number is a national estimate built from the generic ballot and the 2024 result, with a wide range. District polls are listed where they exist and are averaged the same way as statewide races.

What this is not. A win estimate is a probability, not a call. A candidate with a 70 percent chance loses three times in ten. Frothier does not take a side and does not adjust any number by hand.

Results. After polls close, a race is marked called only when major news organizations have projected a winner, and marked verified once state election officials certify the count.