Two signals, one number. Prediction markets (Kalshi, Polymarket) put a price on each outcome — a 60¢ contract that pays $1 if Democrats win means traders think it's about a 60% chance. Polls tell us how voters say they'll vote; we convert the polling lead into a chance of winning. The blend is a weighted average of the two, and you can move the slider to trust one more than the other.
Where to look. The big number is the chance Democrats control each chamber. “Where polls and markets disagree” lists races where the two signals point different ways — the most interesting races to watch. Click any race for its own blend, ☆ it to follow it, and use Copy link or Save image to share what you see.
Updated every 6 hours — market prices and polls are both collected automatically. Every number is linked to its source.
Every 2026 Senate race starts at its market favorite. Click a state to hand it to the other party and see who controls the chamber.
Tuesday, November 3, 2026 — every Senate seat, the House battleground, and the hour-by-hour march of poll closings.
Before results: the market favorite and its chance. After: the race call.
Blend. The headline number is a weighted average of two independent estimates of the chance Democrats win control: the prediction-market price and a polling-implied probability. Default weights are 40% polling and 60% markets; the slider shows other weightings.
Market price. For each market we use the midpoint of the best bid and best ask for the Democratic contract. The chamber figure averages Kalshi and Polymarket. Race-level chances come from Kalshi (Polymarket where Kalshi has no market), with the two sides normalized to sum to 100%. Races whose bid–ask gap is 4¢ or more are marked thin market: few traders, so the price is less reliable.
Polling-implied probability — House. We average two public generic-ballot averages (Silver Bulletin and FiftyPlusOne) and convert the Democratic margin m to a probability with a normal curve. Two parameters are our assumptions, not sourced facts: Democrats need to win the national vote by about t = 2 points to win a majority, and late-September generic-ballot polling misses the final result by about σ = 4 points.
P(D House) = Φ((m − t) / σ) = Φ((m − 2) / 4) = —
Polling-implied probability — Senate. For each modeled race we take the polling margin — our own average of that race's individual public polls (see Race polling) — add a shared national polling error (σ = 3) and a race-specific error (σ = 4), and simulate 20,000 elections. Seats not modeled are assumed to stay with the party that holds them. Democrats need 51 seats because the vice president breaks ties for Republicans. Nebraska's independent, Dan Osborn, is not counted toward either caucus.
Race polling. Every race's polling margin is our own weighted average of its individual public polls, collected automatically every 6 hours from VoteHub's polling database. We use polls that ended in the last 30 days (60 if a race has fewer than two pollsters), count only each pollster's most recent poll, and prefer the likely-voter version when both are released. Newer polls count more (a poll loses half its weight every 14 days), as do larger samples; registered-voter polls count 85% as much as likely-voter polls. Polls sponsored by a campaign, party or PAC count half; pollsters with a known partisan lean count three-quarters. Where the Cook Political Report rates a race, its rating is converted to a margin (Toss-up 0, Lean 5, Likely 8, Solid 13 points) and blended in: 25% of the polling number for well-polled races, 50% for thinly polled ones, and all of it where there are no recent polls. A race's number never moves more than 10 points in a single update without review.
Race-level blend. Click any race to blend its own market price with its own polling. The race's polling margin is converted to a win probability with Φ(margin ÷ 5), using the same combined polling error as the Senate model, then weighted with the slider. District polls are sparse and often sponsored by a campaign or party committee; those are flagged P. Poll dates are shown; polling more than three weeks old is marked as aging, more than 60 days as old.
Market rating. Derived from the race-level market chance: Toss-up 40–60%, Lean 60–75%, Likely 75–90%, Solid 90%+. It is not a Cook Political Report rating. Likely flips are races whose market favorite is not the party that holds the seat.
Divergence. The chamber alert appears when polling-implied and market probabilities differ by more than 5 points. “Where polls and markets disagree” ranks races by the gap between the two.
Trend and what changed. Every update saves the inputs it used; the trend chart and the “what changed” summary recompute the blend from those saved inputs, so past points use today's methodology.
Path to 51. Seats not on the 2026 ballot (34 Democratic-caucus, 31 Republican) are fixed. Each 2026 race starts at its market favorite; the tipping-point race is the one that gives Democrats their 51st seat when races are ranked by Democratic market chance. Louisiana has no market and is shown as a Republican hold.
Seats. Market-implied seat counts come from Kalshi's seat-bracket markets, normalized to sum to 100%. The range shown covers roughly the middle two-thirds of the probability.
Following. Races you follow are stored only in this browser; the “change since your last visit” compares against what this browser last saw.
For information only — not trading, betting or investment advice. Market prices and polls are collected automatically every 6 hours and pass an automated safety check before they are published. Markets and polls move between updates.