“If the election were held today, which party would you vote for?” — the single best early signal for House control. Average computed live from VoteHub polls; weekly points are means of polls ended that week.

• Presidential approval — the president's party has lost House seats in 20 of the last 23 midterms. Loading live approval…
• The Senate map — Republicans defend 22 of the 35 seats up, so the exposure is mostly theirs.
• Retirements — open seats flip far more often than seats with an incumbent running.
• Money & ads — battleground ad spending is the earliest hard signal of where parties think it's close. PRO

A race is called only when it has recent general-election polling: latest poll per pollster, polls ended within 30 days, averaged as a two-party Democratic share. Unpolled and open-seat races are shown as no call — the map and the totals say so instead of guessing.

The generic ballot is the best live signal for House control, but a national margin is not a seat count — VoteWire does not convert it into one. District polling, when available, is what calls individual House seats.

Every state colored by its predicted delegation after Nov 3: solid = both seats one party, split = one seat each. Seats on the ballot are decided by live polling below; unpolled seats are left uncolored as no call; the other seat in each state is the sitting senator.

• Source: VoteHub's polling API (CC BY 4.0) — every poll links to the pollster's own release. No other site's average is copied.
• Rule: general-election polls only (primary polls excluded), ended within 30 days, latest poll per pollster, simple mean of the two-party Democratic share. Each race shows its poll count and latest poll date.
• When the feed fails: the page says "Live polling unavailable" and shows no numbers at all — a missing average is honest, a placeholder is not.
• What is not here yet: live prediction-market odds and live fundraising (FEC) feeds — those connect next; until then they simply do not appear, rather than appearing as estimates.

• Start point: the race's live polling average (50–50 when unpolled), converted to log-odds. Changing the race pulls its live average in as the polling-lead default.
• Adjust: each input adds its weight × value to the log-odds; the result converts back to a probability.
• Uncertainty: the range widens the further out the election is and the closer the race — a 52% forecast is a coin flip, not a call.
• What it is not: not a poll, not a guarantee, and not trained on secret data. Weights are VoteWire's published v1 model.

Inputs are live: polling and approval from VoteHub (CC BY 4.0). Fundraising feeds connect in Pro — until then the fundraising slider is your own scenario input, not data.

Called races are decided by the ensemble below. Uncalled races have no poll or market signal yet — the page says so instead of guessing. The market line is Polymarket's chamber-control market, not VoteWire's model.

House control is read from the live generic ballot plus the chamber-control market. VoteWire does not convert a national margin into a seat count.

The AI Signal Briefing lands each morning, grounded strictly in the numbers on this page — what moved, where the signals disagree, and what would flip the closest races.

• Weights (published): polls 0.50, markets 0.35, social 0.15, renormalized over whichever signals exist. Incumbency adds 1.5 points to the holder's party after averaging. News mood is displayed, never averaged in — coverage tone is not vote intention.
• Uncertainty is shown, not hidden: base range ±4 points, +3 with a single signal, +2 when signals disagree by more than 10 points, +2 when the latest poll is over 21 days old. A 52% signal is a coin flip, not a call.
• Disagreement is the story: races where polls and markets pick different winners carry a "signals disagree" flag. When the crowd and the polls diverge, the page shows both — it does not average the disagreement away.
• What it is not: not a poll, not a guarantee, and not trained on secret data. Social tone comes from Bluesky's public feed — an activist-heavy audience, not the electorate — so it carries the smallest weight by design. No X/Twitter, Facebook, or TikTok data: those feeds are paid or closed, and nothing here is scraped.
• Freshness: signals refresh hourly from the same pipeline that refreshes live polling. The AI briefing updates each morning and quotes only numbers already on this page.

Democrats seized both chambers in a wave that ran through states no one thought were in play — and the voters who delivered it were clear about what they want checked, and what they want lowered.

Read the analysis →

Eight Republican-held seats flip, all thirteen Democratic seats hold, and Georgia leads the way — the full 35-race calls table from final polls, live markets, social tone, a record early vote, and tonight's exit-poll mood. Projections, not official results; South Carolina now AP-confirmed for Andrews (D), correcting VoteWire's earlier R-hold projection.

Read the calls →

Polls are open across the country in a midterm that will decide control of Congress for the final two years of this presidency — and whether the next two years check power or clear its path.

Read the analysis →

The Supreme Court refused the Justice Department's emergency request to set aside ballots from about 20,000 California voters it calls likely noncitizens — and none of the justices publicly dissented.

Read the analysis →

El-Sayed leads Rogers 53–42 into Election Day, and the Stevens endorsement meant to break him instead raised him $2.4 million in a day.

Read the analysis →

Proposition 40 would levy a one-time tax of up to 5% on wealth over $1 billion, but two rival measures could void it even if it passes.

Read the analysis →

More than 51 million ballots are already in, the president's approval has hit a historic pre-midterm low, and the courts have now rejected two of the administration's election lawsuits. The closing argument is written. What remains is who shows up on Tuesday.

Read the analysis →

Two days before voters decide who controls Congress, shoppers open HealthCare.gov to find benchmark premiums up 13% for 2027 — the second steep rise since Congress let the enhanced subsidies expire.

Read the analysis →

Three days before the election, the Justice Department escalated its legal pressure on election officials — and a federal judge told it to stop. The ballots are already in; the question is whether anyone will be allowed to second-guess them.

Read the analysis →

Four days out, the lines at early-voting sites are telling a story the horse-race polls keep missing: voters are not waiting to be persuaded. They are already deciding — in record numbers, in the states that will settle Congress.

Read the analysis →

Senate — 51 seats wins. Today it is 53 Republican, 47 Democratic-allied. Democrats need a net gain of 4 seats for a majority. Live polling consensus loading…

House — 218 seats wins. Today it is 218 Republican, 214 Democratic, 1 independent, 2 vacant, with all 435 seats on the ballot. Live generic-ballot average loading…

On election night this card becomes the live count: seats called, seats still out, and the net change each party needs from the remaining races.

• Call only when the trailing candidate has no path. No "apparent winner" or "likely winner" language — a race is called, projected, or too close to call, and the three look different on the board.
• Every call gets a Decision Note. Why it was called: the margin, the share of vote counted, and what kind of vote is still out (mail, Election Day, provisional) and where.
• Official sources only. Results come from state and county election officials — never modeled, never imputed, never estimated from partial returns.
• Context with every lead. Every margin shows % reporting and which counties and ballot types remain, because a raw lead without that context is how early leads mislead.
• Results are unofficial until certification. Mail, provisional, and military ballots follow different state rules and different timelines; the board says so wherever it matters.

The board is built light on purpose: tile map, called-vs-projected-vs-too-close states, and the path-to-power count — no auto-play video, no heavy geographic files — because election-night traffic spikes late and a slow page loses the night. Updates run at least every 15 minutes, and if any automated feed fails, results are updated by hand under the same call standard above — the fallback is written down before it is needed, not improvised during it.

An average is vote share right now: if the polls say 51–49, the average is 51–49. A forecast is a win probability: a 51–49 average in a volatile race might be a 60% chance, or a 52% chance, depending on uncertainty. Every number on VoteWire is labeled as one or the other, and a 51% average is never presented as a 51% chance.

Live averages, computed on every visit. VoteWire fetches current polls from the VoteHub API (CC BY 4.0), keeps general-election polls ended within the last 30 days, takes each pollster's latest poll per race, and averages the two-party Democratic share. Every average on the site shows its poll count and latest poll date, and every poll links back to the pollster's own release through VoteHub.

Each Senate race's consensus percentage is its live polling average — no added tilt, no stored estimate. Races within 5 points of 50% are labeled Toss-up — genuine coin flips, not leans dressed up as calls. Open seats show the latest matchup rather than a party call, because the feed does not label candidate parties.

Uncertainty is shown, not hidden. The single-race Predictor tool (Forecast page) publishes its full weights — polling, fundraising, environment, momentum, incumbency, approval — plus a plausible range and the exact polling lead needed to flip the race to 50%. Its baseline is the race's live average (50–50 when unpolled); toss-ups are dominated by the polls, exactly as they should be.

VoteWire does not use polls that major aggregators have retracted, and retractions are logged here rather than silently dropped. The October 2026 fake-pollster episode — in which fabricated polls passed aggregators' checks for months before being pulled — is why this log exists at all. Only general-election polls feed the averages; primary and hypothetical-matchup polls are excluded by rule.

Corrections & retractions log: none to date for VoteWire's own averages. Industry retractions noted in October 2026 (polls removed by major aggregators after the fake-pollster findings) are excluded from every VoteWire average.

• Structural facts are current: chamber compositions, seats up, election dates — from Ballotpedia, the Federal Election Commission, and state election officials.
• Polling figures are live, not stored. Every poll number on the site is computed from VoteHub's current feed when the page opens. If the feed fails, the page says so and shows no numbers — never placeholders.
• Prediction markets and fundraising are not shown yet because their live feeds are not connected — market odds and FEC money data arrive as the next build, and until then they simply do not appear rather than appearing as estimates.
• VoteWire is independently operated. Its paid tiers are Pro and Teams subscriptions; the free topline — averages, race explorer, maps, calendar — stays free, because the topline is how readers find the deeper work.