What changed in the forecast

On Point Polling now carries the Gold Standard grade in every polling average, alongside Big Data Poll, The Honest Poll, Emerson, Cygnal, Quantus Insights, The New York Times and The Wall Street Journal. The methodology page and every average now explain the fit in full: each average is a robust LOWESS fit, a local line through time read at the day itself, with the poll weights as its kernel and three bisquare robustness passes that weigh down polls far from the rest.

The polls page is now the whiteboard room, filling the screen, with tabs for Senate, governor, House, national and the archive. The new House tab maps every one of the 101 House districts with public polling and lists them by state, each with the district polling average the House forecast uses. Every race average page is now written on the board too, from the headline to the chart, the poll table and the weight bars, with sticky notes and tape in place of pins, and the hover note now sits on the cursor.

Watchdog and the Ledger records now look like an investigation terminal: green phosphor type, a scan line grid, caution tape across the top and a live status line. The research behind it grew to 2,523 sourced endorsements, 151 campaign stops, 296 podcast and media appearances and 140 media figures. Endorsers are now coloured by their own side, so Nick Fuentes shows as a Republican figure backing Amy Acton, and crowded states on the trail and money maps no longer pile their figures on top of each other.

The home globe drops the On Point HQ arcs and adds aurora over the pole, satellites orbiting with the closest margins, pings on states with new polls, a radar sweep, Pointy's route drawn between the closest races and odds that count up on load. Motion stops for readers who ask for reduced motion.

Watchdog now carries the full Ledger desk: the money desk with totals, where the money went and alerts, the live tape of every reported gift, outside spending report, payment and lobbying report, and pages for every race, candidate, committee, donor, vendor and group, with campaign health and the Ledger Score side by side, outside groups, top donors, timelines and voting records. Signals tracks 2028 prospects, races heating up, money trouble, shared vendors and donors on the move, and there are pages for foreign agents and enforcement cases, lobbying, political ads, events, states and the money web. The pages read a snapshot exported from Ledger and fill in with each new export.

The polling averages now open on an analyst's room: a dry erase whiteboard under a picture light, with tabs across the top for national, Senate, governor and the archive races, a marker map of every Senate and governor average, magnets you can tap, the closest races held up as index cards, and a list on the board that jumps to any average. Watchdog adds an endorsements board with 1,243 sourced endorsements across all 71 Senate and governor races, including Nick Fuentes urging his followers to vote for Amy Acton in Ohio, a media watch of 44 podcasters, hosts and reporters from Mark Halperin and Tucker Carlson to Nate Silver and Larry Sabato, and an endorsements section on every Senate and governor forecast page.

Big Data Poll, The Honest Poll, Emerson, Cygnal, Quantus Insights, The New York Times and The Wall Street Journal are now the Gold Standard pollsters, with the top grade in every polling average. Rasmussen Reports and ActiVote are no longer counted in any average. Trafalgar and InsiderAdvantage are labelled Republican leaning: each of their polls is set against the other polls of the same race ended within three weeks, and the pooled gap, about 3.8 points for Trafalgar and 4.5 for InsiderAdvantage, comes off their margin before they are averaged. The national generic ballot moves from D+7.1 to D+9.5 and Trump net approval from minus 17.5 to minus 21.1. These changes apply to the polling averages; the forecast model keeps its own pollster settings.

PollVault and Emerson College released likely voter polls of ten House districts, fielded October 12 to 14: Florida 14, Pennsylvania 8, Arkansas 2, Texas 35, Iowa 2, Arizona 2, Michigan 7, North Carolina 11, Montana 1 and Missouri 2. Each was averaged evenly with that district's earlier polling and the ten states' House races were re-run. Democrats gain about one seat on average and their chance of a House majority goes from 90.8 to 91.9 percent; North Carolina 11 moves most, to D+3.6. The statewide questions asked inside these districts are district samples and were not added to any statewide average.

The state map on the Senate and governor average pages is now drawn in chalk on a slate, with a push pin in every state that has a public poll, sized by how many polls it has. The four closest polled races are pinned as index cards with red string to their state, each with a small chart of its last six weeks, the small northeastern states hang as paper tags, and the race you pick is circled in chalk. Averages also now start at the first poll of their own matchup, which removes a flat line at even that appeared before it on some races.

The polling averages no longer show the LOWESS trend as a second chart. The trend fit is now the average itself: for each day, every poll keeps its weight for sample size, voter screen, pollster grade and age, and those weights drive a robust local trend fit with three passes that weigh down odd polls, in place of the old outlier clipping. The recency half life stays at about 16 days. Every average page and every Senate and governor forecast page now has the same interactive chart: read any day, switch between the margin and each side, zoom from 30 days to the whole cycle, and see each poll's share of today's average. Also added: 15 national generic ballot polls and 11 Trump approval polls through October 20, including New York Times/Siena, CNN/SSRS, Fox News, ABC/Washington Post/Ipsos, Reuters/Ipsos and Echelon, plus Franklin and Marshall in Pennsylvania and co/efficient in Florida for governor.

A sweep of every race since August added polls the averages were missing in Senate, governor, House, statewide office, ballot measure and state legislative races, and every Senate, governor and House race was re-run. Every forecast page now lists the projected vote by candidate and, for statewide races, by county. Ballot measures carry projected Yes and No votes by county on a new ballot measures map, and every state house and state senate district carries its projected vote.

Every Wedgewood Polls survey has been taken out of the polling averages and the forecast after the firm admitted fabricating data: 18 polls across 17 Senate and governor races in Georgia, Iowa, Kansas, Maine, Michigan, Mississippi, Nebraska, Nevada, Ohio and Texas. Each of those races was re-run without them, and the model now rejects the pollster from every source it reads. Earlier entries in this log that mention Wedgewood describe polls that are no longer counted.

Every Senate, governor and House race now carries a turnout correction from the early vote. In the 38 states reporting ballots, the party mix of the 8.6 million mail and in person ballots cast so far is set against the mix the forecast expected of them, moved for how early in the window it is, compared with the same day in 2022 and 2024 where states published it, and weighted by how many ballots are in. Each state also reads the states most like it and its primary turnout, so states with no early vote yet move with their neighbours. The correction changes each party's turnout among the forecast's simulated voters and recounts every race, so it moves races that lean on low propensity voters more, and House seats in proportion to how close they are. Nebraska moves most toward Osborn, Maine and Colorado toward Democrats, and Florida, Idaho and Maryland toward Republicans. Democratic Senate control goes from 78.6 to 80.6 percent; the House barely moves. Each race page shows the correction as its own step.

Two Alaska governor polls had been read as first choice when they were final rounds, and the model's fundamentals were counted as first choice before the ranked choice count, which gave Kreiss-Tomkins the transfers twice. Both are fixed: Kreiss-Tomkins leads Wilson by 7.2 in the final round, against final round polls of 4 to 10, down from 17.1. Candidate lists in Alaska now show each finalist's own first choice share.

The early vote tracker now opens with a plain language read on how early voting compares with past years. Every state is set against its own count on the same number of days before Election Day in 2022, the last midterm, and in 2024, with a verdict from well ahead to well behind and a chart of the ballots cast day by day in each year. Each state also shows how much of its whole 2022 and 2024 vote is already in, and, where it registers voters by party, how the party split of ballots cast compares with the same point in past years. The past counts come from 20 states' election offices and news reports of their data; total votes come from the US Elections Project and ballots cast by mail and early in person from the Election Assistance Commission.

The swingometer now has its own tab at the top of the site. It adds four new ways to move voters, built from the forecast's simulated voter types: turnout propensity, the model's own chance that each kind of voter votes, with high propensity at 80% or more; 2024 Trump and Harris voters split into high and low propensity; party split the same way; and Trump approval. There are 38 ready made scenarios across turnout, persuasion and whole election environments, from low propensity Trump voters staying home to a 2010 style red wave, and the map is now the real map: Senate and governor races by state and every House district.

A new swingometer sits on the forecast. Shift the vote margin or the turnout of any group of voters, by race, age, education, race and education together, gender, income, party, 2024 vote or how often they vote, or move every voter at once, and all 35 Senate races, 36 governor races and 435 House races update with their vote counts, their odds and each chamber's control odds. Every slider starts at zero, which is the published forecast. Senate and governor races use the forecast's own simulated electorate; House districts start from their state's electorate tilted toward their own partisanship. Scenarios can be shared by link.

Michigan's Senate map was redrawn by court order in Agee v. Benson, with the new lines approved July 26, 2024 for this year's election, and Wisconsin's legislative maps were redrawn in 2024. Neither state's 2026 Senate seats were on the 2024 ballot, so the forecast had been reading those districts on their 2022 lines. Every Michigan Senate district now starts from its 2024 presidential vote on the new lines, and the 16 redrawn metro Detroit districts no longer carry a candidate term from their old 2022 results. District 13, the open seat from West Bloomfield and Commerce to Milford and Lyon Township, moves from about D+5 on the old lines to a seat Trump carried by 1.7 points on the new ones, and Sean Carlson's projected lead over Ryan Berman narrows by 2 points. Wisconsin Senate districts now take their 2024 presidential vote from the three Assembly districts inside each one. Senate seats in Minnesota, Montana, Iowa, Illinois, Oregon, Washington, Idaho, North Dakota and Alaska use the same sum where their house districts nest. Thanks to the reader who flagged Michigan District 13.

Each race, chamber, ballot measure and polling average page now carries its current numbers in its search title and description, a share card with the leader and the odds, and structured data that marks it as a dataset with its source file, so search engines and AI assistants can quote it and link back. The site publishes a sitemap of nearly 1,000 pages, a robots file that welcomes search and AI crawlers, and a plain text guide at /llms.txt with the headline forecast and the closest races, plus /llms-full.txt with every race listed. The forecast hub ends with an index of every Senate, governor, House, statewide, legislative and ballot measure forecast. Results pages that still pointed search engines at the old On Point Polling address now point at onpointpolitics.com.

The forecast adds every statewide ballot measure on the Nov. 3 ballot, 145 questions in 39 states and the District of Columbia, each with its chance of passing and a county map of the Yes vote. Each measure is put in a policy group and given the side its Yes vote leans to; the group sets how the Yes share moves with the state's partisanship and with each county's 2024 presidential vote, fitted on the statewide measures of 2018, 2022 and 2024 in the MIT Election Lab precinct returns, and legislative referrals and bonds run ahead of citizen initiatives. 47 measures have public polls; undecided voters break two to one for No and polls paid for by one side are moved four points toward the other. The odds use each state's own bar to pass, from 60 percent in Florida and 55 percent for Colorado amendments to two thirds in New Hampshire and a majority of all ballots cast in Hawaii, Minnesota, Tennessee and Wyoming. 122 measures are favored to pass. Close calls include Missouri Amendment 3 and Proposition A, Massachusetts Question 3, California Propositions 4, 5 and 40, Montana CI-133 and Florida Amendment 3.

Every state house and state senate race is now checked against the general election candidate lists, from secretary of state certifications where they could be read and from statewide ballot guides elsewhere. Seats with no Democrat or no Republican on the ballot are called for the other party, candidates who withdrew after the primary are noted on their district, and 76 races where an independent is the main challenger are modeled as independent races: the independent runs about 7 points behind a party nominee unless a poll says otherwise, and a district poll counts for more. Pennsylvania Senate District 46 is one: Democrat Evan Snyder withdrew in June, and independent Jodi Borello is tied with Republican Camera Bartolotta at 43 in State Navigate's Oct. 3 to 5 poll of 1,201 likely voters, in a seat President Trump carried by about 29 points; the forecast moves to Bartolotta +5.2 with Borello at 22 percent. Special elections for vacant seats in Colorado, Illinois, Ohio, Utah, Wyoming, North Dakota and West Virginia are added. New polls: State Navigate in West Virginia Senate 1 and 8 and Iowa Senate 35, the corrected Pennsylvania House 145 poll, Cygnal and Beacon Research for Michigan Enjoyer on the Michigan House, and the October Change Research legislative ballot in North Carolina. Coverage gaps remain in the Illinois, Indiana, Kansas, Michigan and Ohio House lists, where those districts keep the primary results.

Every state legislative chamber now has a district map, and the hub and the legislatures page carry one national map of every state house or state senate district on the 2026 ballot, colored by who is favored, by the forecast margin or by the seats changing hands. Official boundary files could not be loaded into this build, so the shapes are rebuilt county by county from the MIT Election Lab precinct returns: a county one district holds whole goes to it whole, so rural lines follow county lines the way the real maps do, and a county split between districts is divided among them in proportion to their 2024 presidential voters there, each piece on the side of the county that faces the rest of its district. The national map zooms and pans, with quick views of the Northeast, the Great Lakes, the Southeast, Texas and the West. Seats not up in 2026 are hatched in the color of the party that holds them. Chamber pages open on the map with the district picked beside it and the full table below, and the forecast pages now use more of a wide screen, with the ballot hemicycle and the control strip side by side.

The forecast page opens on a new hub that puts every 2026 forecast on one screen. A hemicycle of the whole ballot draws all 6,733 seats as dots in five rings, the Senate at the center, then governors, statewide offices, the House and every state legislative seat, each sorted and colored by who is favored, with hover odds and a click through to the race. A control strip gives the Senate, House, governor, statewide and legislative outlook, a tile map shows the odds of control in every legislative chamber, a knife's edge list ranks the closest races and chambers, and one spectrum lays every forecast margin on a single axis. The race desk with its maps, polls and counties now sits below the hub, and a link that names an office still opens it directly.

The forecast adds every partisan state legislative chamber on the 2026 ballot, 87 chambers and 6,113 seats in 5,704 districts, each with its own page. Each district starts from its 2024 presidential vote, read off the precincts that voted in it from the MIT Election Lab returns, or its 2022 governor or Senate vote moved to 2024 terms where the seat was last up in 2022. The district moves with the state's 2026 House forecast and the national generic ballot at 85 percent, half of what the last legislative winner ran ahead of the district carries over, the 2026 primaries move districts in the 15 states where they are posted, and a seat with no candidate from one party is called for the other. Legislative generic ballot polls in Texas, Wisconsin, North Carolina, Arizona and Georgia and 24 State Navigate district polls in Pennsylvania count by recency and sample. Seats not up in staggered senates keep the party that won them in 2024. Democrats are favored in 43 chambers, from 31 majorities today, with the Arizona, Michigan, Minnesota, New Hampshire and Pennsylvania houses, the Wisconsin Assembly and the Arizona, New Hampshire and Vermont senates among the likely flips.

The forecast adds every 2026 race for lieutenant governor, attorney general, secretary of state, treasurer, auditor, agriculture commissioner and Texas railroad commissioner, 114 races in 33 states, 12 of them uncontested. Each race starts from the state's 2026 Senate and governor county forecasts at 30 percent, the ticket it shares a ballot with, and adds the approval fundamentals at 15 percent, the Gallup party identification leg relaxed to 10 percent, the state's 2014, 2018 and 2022 down ballot results moved to 2026 at 25 percent, and this year's primary vote for the office at 20 percent. Incumbency, primary unity and polls move each county by how loose its voters are in the ticket splitting model. A sweep found 86 polls in 33 of the races; where none exist, the primaries carry more weight than in the Senate and governor races.

The early vote page is rebuilt as one dashboard, nationally and for every state. Four tiles give total ballots cast, early in person, mail returns and the ballots still to come, each against what the electorate model expects. The modeled 2026 vote card reads the ballots cast so far into a vote share range for a race, the Senate, governor or House vote, against the forecast. The final electorate shift sets how the projected 2026 electorate voted in 2024 against the 2024 result, and a forecast check says whether the early vote points to the forecast being about right, too Republican or too Democratic. Beside the map, with bubbles sized by ballots, the party, race, age and gender of the ballots cast are reported where a state publishes them and modeled from the census and the On Point Polling national dataset where it does not, with the reported share shown. The better or worse than expected verdict stays below. The expected split of the 2026 vote now comes from the On Point Polling national dataset, what likely voters say they will do, pooled half and half with each state's 2024 record: 27 percent by mail, 23 percent early in person and 50 percent on Election Day of 128.5 million votes, about 35.2 million, 29.3 million and 64.0 million.

Gas is fair and adds no tilt: every county is read against the national price rise both ways, so a county whose rise ran above the nation's moves toward Democrats and one whose rise ran below it moves the same distance toward Republicans, the national total is zero, and an extra dollar of rise counts what an extra dollar of price does in the On Point Polling fit. Local conditions, data centers and gas are now built into the legs rather than laid over the finished forecast: their county pattern is part of every leg's county shape, and their statewide part enters only the approval, Gallup and seat history legs, since the polls already carry each state's conditions. The polls weigh more as they pile up: the Gallup floor is 40 percent with four or fewer polls in the last 60 days and falls to 15 percent at 24, the poll weight can reach 90 percent, and House district polls count at 0.80 with an incumbent running and 0.88 in an open district. Across the competitive races the forecast now sits 0.8 points from its polling averages on average, from 1.7. Every race was re-run at 2,000 simulations. Senate control 77 percent counting Osborn and 73 percent without; governor control 91 percent; House control 89 percent with a median of 239 Democratic seats. Michigan Senate El-Sayed +4.1, from +9.1; Iowa Senate Turek +1.6, from +5.3; Minnesota Senate Flanagan +7.2, from +9.5; Georgia Senate Ossoff +10.2, from +11.5; Maine Senate Jackson +0.8, from +1.6; Texas Senate Talarico +1.3, from +2.1; Nebraska Senate Osborn +1.0, from +2.3; Iowa governor Sand +11.2, from +13.5; Wisconsin governor Crowley +2.3, from +4.1; Kansas governor Holscher +0.1, from +1.5.

The early vote party estimate now reads every state the way the states that report party do. Every state that registers voters by party but sends its ballots without it is set to its own published registration, county by county, measured against the forecast's registered voters, so mail voters keep their own lean. Version 3 set Kentucky, Maryland and Idaho requests to the statewide registration mix and gave Alaska, Delaware, Kansas, Louisiana, Massachusetts, New Hampshire, New York, Oregon, Utah and West Virginia one pooled gap. States that report party on requests are now matched to those requests exactly and spread across counties by their registration. States without party registration keep the primary history model. Tested on the 16 states that report party on requests, the registration step alone lands 8.3 points from the reported request margin on average, against 11.9 for one pooled gap. Kentucky, Louisiana and West Virginia move toward Democrats, where old Democratic registrations run deep, and Massachusetts, New Hampshire, Oregon and Utah show far more unaffiliated voters.

Added InsiderAdvantage, Oct. 13 to 14, likely voters: Texas Paxton 49, Talarico 48; Maine Jackson 49, Collins 47; Kansas Marshall 48, Hamilton 47; the release gives no sample sizes, so 800 is a placeholder, and the Kansas field dates are taken from the same release. Added Wedgewood in Mississippi, Oct. 12 to 14, 511 likely voters: Hyde-Smith 52, Colom 48; Alliance for a Better Future, a group running ads for Rogers, in Michigan, Oct. 6 to 8, 600 likely voters: Rogers 46, El-Sayed 44; Change Research in Florida governor, Oct. 8 to 10, 920 likely voters: Donalds 47, Jolly 47; YouGov for City Forward Collective in Wisconsin governor, Sept. 22 to Oct. 9, 1,715 adults: Crowley 46, Tiffany 42. The Pollfinity New York governor poll now enters at its published 499 registered voters. House: Emerson in RI-1 and RI-2, Research & Polling in NM-2, campaign polls in GA-3 and GA-7, and National Public Affairs in NE-2. Texas Senate Talarico +2.1, from +2.8; Kansas Senate Marshall +1.8, from +0.6; Mississippi Senate Hyde-Smith +6.0, from +7.0; Michigan Senate El-Sayed +9.1, from +9.8; Maine Senate Jackson +1.6, from +1.2; Florida governor Donalds +1.7, from +3.0; Wisconsin governor Crowley +4.1, from +3.7. Senate control 87 percent counting Osborn and 83 percent without; governor control 94 percent; House control 91 percent.

The gas term no longer pulls any county toward Republicans. Before tonight it was measured against the national rise, so counties and states where prices rose less than the nation's $1.27, Georgia, North Carolina and Ohio among them, were pushed toward Republicans. Now a county whose rise ran above the national rise moves toward Democrats through voters who call their finances difficult, a county at or below it gets no gas effect at all, and the state level takes the state's own average of those moves, county and state both at half the fitted size. The local conditions caps are loosened: the compounding multiplier is held at six times, from three, a county's combined term only within about 46 points of margin, from 22, and a county may run 0.15 in log odds past the edge of its own historical lean, from 0.05. Every race was re-run at 2,000 simulations. Senate control 87 percent counting Osborn and 84 percent without; governor control 94 percent; House control 91 percent with a median of 242 Democratic seats. Georgia Senate Ossoff +11.5, from +8.1; Michigan Senate El-Sayed +9.8, from +7.1; North Carolina Senate Cooper +11.5, from +9.3; Ohio Senate Brown +4.7, from +3.2; Iowa Senate Turek +5.3, from +3.8; Georgia governor Bottoms +0.4, from Jackson +1.6.

Local conditions now compound. Each county's terms, jobs, wages, tariff exposure, population growth, migration, faith, rurality, news deserts, data centers and the gas rise through financial strain, keep the direction their fit gave them, and when several of them push the same way the county's whole local term grows: one factor alone moves as before, five lined up one way move about 1.8 times as far, up to three times. Counties with more local news per resident feel their conditions more, about 0.6 times in a news desert and about 1.65 times in the best covered places. The county term now counts 2.0 times, from 1.5, and the statewide part 1.25 times, from 1.0, so the average county spread from local conditions grows from about 2.1 to 3.7 points. Where no incumbent is running the race leans on its polls and on President Trump's approval: the seat history share falls to 0.03, the approval leg rises to 0.16, the statewide history pull runs at 0.15 of its usual size, and open House districts move toward the approval leg with their polls at 0.82 of the blend. Every race was re-run at 2,000 simulations. Senate control 80 percent counting Osborn and 75 percent without; governor control 90 percent; House control 90 percent with a median of 240 Democratic seats. Georgia Senate Ossoff +8.1, from +9.5; Iowa Senate Turek +3.8, from +3.3; Alaska Senate Peltola +1.9, from +1.4; Iowa governor Sand +12.5, from +11.6; Kansas governor Holscher +0.8, from +1.5; Georgia governor Jackson +1.6, from +0.9.

The national race map and the county atlas are now one map card: the office switch at the top drives both, and a Map switch flips between states or districts and every county with its layers. The seat bar, odds, crosstabs and each race's candidate totals are unchanged. Every map now prints the same tooltip, each candidate's projected votes and share with the total and the margin, and the national map adds each side's odds to win. A new spending tracker shows where the money is going in every Senate, House and governor race: money working for each side, outside money, money per 2024 voter, and where each party's outside money is landing, from FEC filings through the latest reports and state filings for governors. Each race page adds its own money card with the biggest outside groups and, in Senate battlegrounds, AdImpact aired spending through Sept. 30.

The early vote page opens on a Total cast view that adds mail ballots returned and ballots cast in person in every state, with the party estimate carried through both. A new electorate tracker, built from On Point Polling likely voters and the 2024 Trump vote, sets out how many votes the forecast expects by mail, early in person and on Election Day in every state and county, about 31, 33 and 37 percent of 128.5 million nationally, calibrated to each state's 2024 EAVS mode shares. It reads each place's ballots so far as returning Trump voters, returning Harris voters and new voters, says whether the early vote is better or worse than expected for each party against what the forecast expects of those same kinds of ballots, and splits that into where the ballots come from and who is voting inside each place. A demographics section shows every breakdown civicAPI publishes for the ballots so far, with age and race against the On Point Polling expectation; states that publish no Hispanic category have Hispanic voters counted with Asian and other voters.

Added KSTP/SurveyUSA in Minnesota, Oct. 8 to 12, likely voters: Flanagan 45, Tafoya 42 for Senate and Klobuchar 48, Demuth 41 for governor; the release gives no sample size, so 650 is a placeholder. Added Emerson College for Nexstar in Connecticut, Oct. 10 to 12, 800 likely voters: Lamont 51, Fazio 38. Pollster house effects were refit with the new polls. Minnesota Senate Flanagan +8.1, from +10.0; Minnesota governor Klobuchar +16.2, from +16.9; Connecticut governor Lamont +15.1, from +17.2; Alaska Senate unchanged at Peltola +1.4. Senate control 81 percent counting Osborn and 77 percent without.

Added Marist in Iowa, Oct. 8 to 11, likely voters: Turek 48, Hinson 45 for Senate and Sand 52, Lahn 45 for governor; Marist gives no likely voter count, so 512 is estimated from its margins of error. The YouGov Iowa poll is now entered at its published 2,245 likely voters, Oct. 6 to 12, instead of a 1,000 placeholder. Iowa Senate Turek +3.3, from +3.0, 74 percent; Iowa governor Sand +11.6, from +11.7. Senate control 82 percent counting Osborn and 77 percent without.

Three governor rules. Where the incumbent is running, the Gallup leg follows the national environment at 0.6, like the other legs, since incumbents hold up better against a wave than the open seats in the Virginia and New Jersey backtest. In a governor race with 10 or more polls the Gallup floor is 30 percent rather than 40. The statewide data center and gas pushes together are held within 2 points. Nevada governor Ford +2.3, from +6.4, against a polling average of Lombardo +1.2; Arizona Hobbs +9.2, from +12.3; Iowa Sand +11.7, from +13.8; Nebraska Pillen +3.6, from +1.7; Texas Abbott +2.7, from +1.3. Alaska Survey Research's Oct. 1 to 3 Alaska governor poll now enters on its first round, 48 to 52, and the Alaska poll average is shown on the final round basis of the forecast. Governor control 90 percent, from 95, with a median of 28 Democratic governors.

Gas now counts by how much prices have risen since October 2025, not by how high they are: about $1.27 a gallon nationally, from about 91 to 93 cents in Oregon, North Carolina and Georgia to $1.71 in Michigan, from AAA state and metro averages against a year earlier. Counties where the rise is steepest move toward Democrats through voters who call their finances difficult. The data center pull is scaled to three quarters of the 6.3 point gap the Economist and YouGov now measure within 2 kilometers of a data center, about 2.8 times the earlier size. The 2026 county primary results confirm the readings in use, and Ken Paxton and both Alabama Senate nominees now take their runoff results while Annie Andrews takes her contested primary. Primary candidate and party unity terms are up about ten percent and the coalition model is stronger. Every race was re-run at 2,000 simulations. Senate control 81 percent counting Osborn and 77 percent without; governor control 95 percent; House control 90 percent with a median of 241 Democratic seats. Iowa Senate Turek +3.0, from +1.8; Nebraska Osborn +1.4, from +0.1; Alaska Peltola +1.4, from +3.6; Georgia governor Jackson +1.0, from even; Oregon governor Kotek +7.7, from +11.4.

The county map adds gas price rise, each county's increase since October 2025, gas price rise pull, how far that rise moves the margin through voters who call their finances difficult, and data center pull, how far nearby and planned data centers move each county's margin.

Added Quinnipiac in Connecticut, Oct. 8 to 12, 1,225 likely voters: Lamont 58, Fazio 38. Connecticut governor Lamont +18.2, from +15.9. Governor control 95 percent.

The Gallup leg, each county's 2024 vote moved by the On Point Polling demographic swing and its own elasticity to Gallup's 50 Democratic to 39 Republican, then the 2026 electorate, local conditions, data center and gas terms, now carries at least 40 percent of every Senate and governor forecast and 60 percent of each state's House fundamentals level, and governor races take the national environment in full. Applied the same way, Gallup's third quarter 2025 reading of 48 to 41 called Virginia's 2025 governor race at D+15.7 against an actual D+15.4 and New Jersey's at D+15.8 against about D+14.7, where the final polling averages had D+10.6 and D+4.6. Every race was re-run at 2,000 simulations. Senate control 80 percent counting Osborn and 76 percent without; governor control 95 percent; House control 89 percent with a median of 240 Democratic seats.

Added YouGov in Iowa, Oct. 8 to 12, likely voters: Turek 47, Hinson 46 for Senate and Sand 50, Lahn 44 for governor, sample size not yet published; and InsiderAdvantage in Georgia governor, Oct. 12 to 13, 800 likely voters: Jackson 49, Bottoms 47.

Every state with ballots reported as Unspecified is now estimated from the same inputs on the national map, the national cards and its own state page: its county counts, its reported ages and races, and reported party where it has any, so a state shows the same rating wherever it is opened. Margins now count every ballot with a party label, third party and unaffiliated included, the way the states and L2 report them. The party estimate moves from party identification to party registration: states that publish party on their requests use their own measured gap, California, Kentucky, Maryland and Idaho are set to their published registration, and states without party registration are modeled the way L2 assigns party, from the primary ballot a voter pulled, county by county, with 2026 counting for half and 2018 to 2024 for the other half. That moves Michigan, Wisconsin, Illinois, Minnesota and Texas toward Democrats and Idaho and Maryland toward Republicans.

Every Senate, governor and House race was re-run at 10,000 simulations. The statewide top out is removed and every effect, from the candidate and primary terms to past results, is applied county by county and summed to the state. The 2026 primaries weigh more, a party term of 0.50 and a candidate term of 0.55, and each county's primary reading is now shrunk toward its state's by its own primary vote, n over n plus 300, so small counties do not swing on a few hundred ballots. Previous results weigh more, 0.32 in Senate races and 0.40 in governor races. New polls: Public Policy Polling in the Illinois Senate race, Stratton 49, Tracy 36, the first public poll there, and Lake Research in South Dakota governor, Rhoden 51, Ahlers 44, plus every poll from the October 12 sweep, with polling averages rebuilt. Illinois Senate Stratton +21.3, from +26.7; South Dakota governor Rhoden +9.6, from +7.1; Kansas Senate Marshall +0.2; Kansas governor Masterson +0.3; Nevada governor Lombardo +3.5. Senate control 87 percent counting Osborn and 82 percent without; governor control 92 percent; House control 95 percent with a median of 246 Democratic seats.

Added Mason-Dixon for the Star Tribune and Morning Consult in Minnesota; Trafalgar in North Carolina, Georgia and Texas; InsiderAdvantage in Michigan, Alaska and Kansas; Wedgewood in Michigan; Rasmussen in Maine and South Carolina; GrayHouse for Norton in Arizona; Research & Polling for the Albuquerque Journal in New Mexico; the Manhattan Institute in California; and the full Washington Post and Schar School Maine poll, now 800 likely voters with Bennett at 11. House polls added in Pennsylvania 1, South Carolina 1, Colorado 4, North Carolina 1 and Kentucky 6. Kansas Senate is now Hamilton +0.2, 50 percent, from +1.4 and 59 percent; South Carolina Senate Graham +1.6, from +0.7; Minnesota Senate Flanagan +8.9; Texas Senate Talarico +3.4, 74 percent; Alaska Senate Peltola +3.1; Michigan governor Benson +9.8; Arizona governor Hobbs +11.1. Only races with new polls were re-run. Senate control 88 percent counting Osborn and 85 percent without; governor control 93 percent; House control 95 percent with a median of 247 Democratic seats.

Added Targoz Market Research in Florida, Sept. 28 to Oct. 4, for a sponsor connected to the Heritage Foundation and so weighted as a Republican poll: Donalds 52, Jolly 42, sample size not yet published; and Z to A Research for Rust Belt Rising, a Democratic aligned group, Sept. 28 to Oct. 1: Michigan El-Sayed 49, Rogers 42 and Benson 50, James 40; Pennsylvania Shapiro 53, Garrity 34; Wisconsin Crowley 49, Tiffany 44. Florida governor Donalds +1.7, Jolly 37 percent, from +0.8 and 43 percent; Michigan governor Benson +7.8; Michigan Senate El-Sayed +5.2; Wisconsin governor Crowley +5.2; Pennsylvania governor Shapiro +25.4. Senate control 88 percent counting Osborn and 86 percent without; governor control 92 percent.

Added Emerson College and Nexstar in Iowa, Oct. 5 to 6, 650 likely voters: Hinson 47, Turek 45 for Senate and Sand 49, Lahn 45 for governor; and The Washington Post in Maine, Sept. 30 to Oct. 5: Jackson 50, Collins 46 for Senate and Pingree 49, Charles 36 for governor, sample size not yet published. The St. Pete Polls Florida sample is now 836 likely voters. Iowa Senate Turek +1.3, 59 percent, from +2.3 and 66 percent; Iowa governor Sand +8.1, from +10.6; Maine Senate Jackson +4.0, 79 percent, from +3.4; Maine governor Pingree +12.2. Senate control 88 percent counting Osborn and 85 percent without; governor control 92 percent; House control 95 percent with a median of 247 Democratic seats.

Added East Carolina University in North Carolina, Sept. 30 to Oct. 3, Cooper 50, Whatley 43; co/efficient in Ohio, Oct. 5 to 6, Husted 44, Brown 43 for Senate and Ramaswamy 45, Acton 43 for governor; St. Pete Polls in Florida, Oct. 5 to 7, Moody 47, Nixon 45 and Donalds 45, Jolly 45, sample size not yet published; DHM Research for The Oregonian, Kotek 46, Drazan 44; Guidant for Public First Action governor results in Texas, Abbott 46, Hinojosa 44, and Michigan, Benson 50, James 45; the Alaska Survey Research governor final round, Kreiss-Tomkins 54, Wilson 46; the Maine People's Resource Center governor poll; and PPP in New York 18, Ryan 52, Auringer 35. Ohio governor Acton +3.9, 78 percent, from +5.6 and 87 percent; Ohio Senate Brown +4.4, 81 percent, from +4.9; Florida Senate Moody +3.1, from +4.3; Florida governor Donalds +0.9, Jolly 43 percent; Oregon governor Kotek +9.9; Michigan governor Benson +7.3. Senate control 88 percent counting Osborn and 85 percent without; governor control 93 percent; House control 95 percent with a median of 247 Democratic seats.

New fundraising and cash on hand figures checked October 8 replace the September finance records. Governor races use the average of each side's share of money raised and of cash on hand; Florida counts all accounts, including the candidates' political committees. Senate races use money raised plus outside spending for each side. Races whose figures are not comparable keep their previous records. The 24 races whose finance term moved were re-run. South Carolina Senate Graham +0.8, from +0.1; Minnesota governor Klobuchar +12.5; Michigan governor Benson +7.9; Maine governor Pingree +11.3; Texas governor Abbott +2.5. Senate control 89 percent counting Osborn and 86 percent without; governor control 93 percent.

Hendrix College's statewide Arkansas poll from the same week and method had Shoffner ahead of Cotton by 3, about 17 points more Democratic than every other poll of the race, and its governor poll ran about 9 points more Democratic than the race average. Its Arkansas 2 poll, Jones 48 to Hill 44, is now corrected by that measured miss instead of the smaller national house effect, which reads it as about Hill +10. Arkansas 2 moves to Hill +4.6, Lean Republican, Jones 23 percent. House control 95 percent with a median of 247 Democratic seats.

A second sweep covered the other 368 House districts and found 67 more general election polls, so 190 district polls in 89 districts are now on record. Nonpartisan district polls now carry the same pollster house effect the Senate and governor averages use, and Hendrix College polls are pooled under one name. A sample size that its own stated margin of error contradicts now enters at the size the margin implies: the Hendrix poll of Arkansas 2 lists 1,112 likely voters with a margin of error of 5.3 points, which fits about 340. Arkansas 2 Jones +2.1, 63 percent, from +3.1. Nebraska 2, California 48, California 22 and Arizona 1 tightened; Wisconsin 6 widened. House control 95 percent with a median of 248 Democratic seats.

A full sweep of every 2026 general election poll of the current nominees in the 67 Lean, Tilt and Lean Republican House seats found 123 district polls in 48 districts, most of them sponsored by a campaign, party or partisan group. Each district's polling entry was rebuilt from the full record: a 45 day recency half life, sample size weighting, partisan sponsors weighted at 0.7 and moved 1.5 points against the sponsor, informed ballots and polls at 20 percent undecided or more left out. Seven Florida districts, Colorado 5, New Jersey 2 and Texas 23 have polling in the model for the first time. The biggest moves: Texas 23 Herrera +3.3, Nebraska 1 Backemeyer +1.0, New Jersey 2 Van Drew +6.4, Missouri 2 and Minnesota 1 now near even. House control holds at 96 percent with a median of 248 Democratic seats.

Added UMass Lowell and YouGov in Texas, September 18 to 28, 850 likely voters: Talarico 45, Paxton 45, Brown 2 for Senate and Abbott 48, Hinojosa 44, Dixon 1 for governor. Abbott's favorable rating of 48 to 43 and job approval of 52 to 48 from the same survey join the candidate ratings. Texas Senate Talarico +2.8, 69 percent, from +2.8 and 71 percent; Texas governor Abbott +2.7, 72 percent, from +2.5. Senate control 89 percent counting Osborn and 86 percent without; governor control 93 percent.

Added GQR for the Hobbs campaign in Arizona governor, September 30 to October 4, 600 likely voters: Hobbs 49, Biggs 45, Other 3, entered as a Democratic internal. It is the first Arizona governor poll since early September, so it carries most of the recent weight. Arizona governor Hobbs +9.9, 95 percent, from +13.7 and 99 percent. Governor control 93 percent; Senate and House unchanged. The Emerson New York poll for PIX11 was already in the morning update.

Added CNN/SSRS in Ohio, September 29 to October 5, 760 likely voters: Brown 49, Husted 43 for Senate and Acton 50, Ramaswamy 42 for governor; CNN/SSRS in Iowa, 801 likely voters: Hinson 45, Turek 43 for Senate and Sand 51, Lahn 42 for governor; Quinnipiac in Pennsylvania, 1,159 likely voters: Shapiro 59, Garrity 34, and Susquehanna, 700 likely voters: Shapiro 52, Garrity 31; Emerson College for PIX11 in New York, 900 likely voters: Hochul 52, Blakeman 40; Guidant Polling for Public First Action in Texas, Talarico 47, Paxton 44, and Michigan, El-Sayed 48.5, Rogers 44 as the average of two ballots, both with field dates and sample size not published so 600 and an October 5 end date are placeholders; GBAO for the Hamilton campaign in Kansas, 600 likely voters, Hamilton 47.5, Marshall 42.5 as the average of two published figures; and FITSNews in South Carolina governor, 638 registered voters: Wilson 45, Johnson 35. House: NC-1 GQR for Davis and Neighborhood Research, TN-5 Impact Research for Molder, TX-2 Impact Research for Finnie. Iowa Senate Turek +2.2, 65 percent, from +3.4; Ohio Senate Brown +4.9, 84 percent; Ohio governor Acton +5.7, 87 percent, from +3.2; Kansas Senate Hamilton +1.6, 59 percent; South Carolina governor Wilson +6.1. Senate control 89 percent counting Osborn and 87 percent without; governor control 94 percent; House control 96 percent.

Added Alaska Survey Research in the Alaska Senate race, October 1 to 3, 1,512 likely voters: Peltola 47, Sullivan 45, Dan J. Sullivan 5, Heikes 3 in the first round, Peltola 51, Sullivan 49 in the final round; and Quantus Insights in Michigan, 762 likely voters released October 7: El-Sayed 48, Rogers 47, with field dates not published so the end date is a placeholder. Alaska Senate Peltola +1.5 in the final round, 60 percent, from +2.6; Michigan Senate El-Sayed +4.7, 82 percent. Senate control 88 percent counting Osborn and 86 percent without; governor control 93 percent; House control 96 percent.

The estimated 2026 midterm electorate, built from On Point Polling vote history and turnout, used to move county vote shares only in the census leg. It now also moves the approval leg and the polling leg county by county, with each leg keeping its statewide level so nothing is counted twice; the polling leg takes the pattern only in proportion to the presidential year races in its history. Eight polls entered twice as two ballot versions were merged into one row each, in Florida and New York governor, Iowa, Idaho, Michigan governor, North Carolina and Texas governor. Every Senate, governor and House race was re-run at 10,000 simulations. Statewide margins moved by less than half a point; Senate control 89 percent counting Osborn and 87 percent without, governor control 93 percent, House control 96 percent with a median of 248 Democratic seats.

Added Suffolk University in Iowa, October 1 to 4, 500 likely voters: Turek 48, Hinson 44 for Senate and Sand 55, Lahn 43 for governor; Marist in New York, October 1 to 4, 1,361 registered voters: Hochul 53, Blakeman 42; and Arizona State University for Arizona governor, Hobbs 47, Biggs 34, among 1,300 registered voters. Iowa Senate Turek +3.4, 74 percent, from +2.6; Iowa governor Sand +11.2; New York governor Hochul +13.4; Arizona governor Hobbs +13.7, from +11.1. Senate control 89 percent counting Osborn and 86 percent without; governor control 92 percent; House unchanged at 96 percent.

Added Saint Anselm College for AARP in New Hampshire, September 28 to 29, 1,042 likely voters: Pappas 51, Sununu 44 and Ayotte 52, Warmington 45; the Independent Voter Project and a Hilton campaign poll in California; Berkeley IGS from September; UMBC in Maryland; Public Policy Polling in Illinois; Mitchell Research for governor in Michigan; co/efficient in New York; and catch up polls in Georgia and North Carolina. Field dates and samples were corrected for Big Data Poll in Minnesota, the Maine People's Resource Center, SurveyUSA in New Mexico and Noble in Nevada. House polls added in KY 6, NY 17, VA 1, VA 5, VA 6, WA 3 and SC 1. California governor Becerra +15.6, from +23.3; New Hampshire Senate Pappas +11.9; New Hampshire governor Ayotte +8.1; Kentucky 6 moves to Tilt D. Senate control 89 percent counting Osborn and 86 percent without; governor control 92 percent; House control 96 percent with a median of 248 Democratic seats.

The stated registered voter net approval of minus 24 is removed. Every county's Trump approval is now leveled to the On Point Polling likely voter electorate itself: 40.3 percent approve and 57.1 percent disapprove, net minus 16.9, across 4,721 respondents through the October Benchmark. Every Senate, governor and House race was re-run at 10,000 simulations. Because the approval leg's national level stays on the generic ballot, statewide margins moved by less than a point outside the newly polled races, and House control holds at 96 percent with a median of 248 Democratic seats.

Added YouGov likely voter polls of the Senate and governor races in Texas, Florida, Michigan, Ohio, Kansas, Nebraska and New Hampshire, from 1,520 to 3,949 voters each, and Mason-Dixon for Telemundo in Florida, September 28 to October 1, 625 likely voters: Moody 50, Nixon 40 and Donalds 48, Jolly 41. Texas Senate Talarico +2.8, 70 percent; Texas governor Abbott +2.6, 71 percent; Michigan governor Benson +7.2; Nebraska governor Pillen +5.9; Florida governor Donalds +1.4, 61 percent, from a toss up; Florida Senate Moody +4.3, 77 percent. Senate control 89 percent counting Osborn and 85 percent without; governor control 92 percent.

Re-run with Marist in Maine, October 1 to 4, 1,505 registered voters: Jackson 51, Collins 47 for Senate and Pingree 52, Charles 40, Bennett 6 for governor; SurveyUSA for KOB 4 in the New Mexico Senate race, Lujan 54, Marker 36; and Noble Predictive Insights in Nevada, Lombardo 42, Ford 40. The New Mexico and Nevada polls were released without field dates, and Nevada without a sample size, so those are placeholders. Maine Senate Jackson +3.2, 73 percent, from +3.7; Maine governor Pingree +11.4; New Mexico Senate Lujan +21.4; Nevada governor Lombardo +2.4, 64 percent, from a toss up. Senate control 88 percent counting Osborn and 85 percent without; governor control 92 percent; House control 96 percent.

Re-run with Mitchell Research for CBS Detroit in Michigan, October 1, likely voters: El-Sayed 45.6, Rogers 41.4, the four minor party candidates 7 together. The sample size was not published, so 650 is entered to match its 3.8 point margin of error. Michigan Senate El-Sayed +5.3, 85 percent, from +4.8. Senate control 88 percent counting Osborn and 85 percent without; governor control 93 percent; House control 96 percent.

Every Quantus Insights poll in every state now counts as an independent poll rather than a Republican sponsored one. A partisan tag was meant for polls paid for by a campaign, party or partisan group, and Quantus publishes its own polls; its lean is now measured from its results like any other firm's, at about 1.9 points toward Republicans across 14 polls. Its polls now count toward each race's polling depth and are no longer moved by the sponsor adjustment. Nine races were re-run: Florida Senate Moody +3.9, 75 percent, from +3.4; Minnesota Senate Flanagan +7.8, from +8.6; Iowa Senate Turek +2.5, 68 percent, from +2.8; Alaska Senate Peltola +2.5, 68 percent, from +2.2; Arizona governor Hobbs +11.1. Senate control 88 percent counting Osborn and 85 percent without; governor control 93 percent; House control 96 percent.

Gallup party identification of 50 Democratic to 39 Republican among adults now anchors the census leg in every Senate, governor and House race. It is converted to a vote through the On Point Polling respondent database: every ACS cell's On Point Polling party mix is moved until the nation's adults identify 50 D, 39 R and 11 independent, and each cell and party then turns out at its DSMeridian likely voter propensity, which gives a likely voter vote of D+15.4. The approval and previous results legs stay on the national generic ballot at D+12. The House now uses the correlation calibration the Senate and governor races took on October 3: the shared national environment runs at half strength, each district takes its region's shock, and the spread removed goes back to the state shock, so every district is exactly as uncertain as before but states and regions move less as one. The House also runs at 10,000 simulations for the first time, on the October respondent database. Races moved toward Democrats by 0.4 points on average and at most 1.5, in safe seats; no race within 15 points moved more than 0.9. House control rises from 89 to 96 percent: about 2 points from Gallup and about 5 from the correlation calibration, with the median at 248 Democratic seats. Senate control 88 percent counting Osborn and 85 percent without; governor control 94 percent.

Re-run with Rasmussen in New Hampshire, October 2 to 4, 979 likely voters: Pappas 49, Sununu 44; the Maine People's Resource Center poll released October 5, 684 registered voters: Jackson 49, Collins 47; the New York Times/Siena Kansas governor poll, September 22 to 30, 605 likely voters: Masterson 49, Holscher 44; and SurveyUSA for KOB 4 in New Mexico, released October 5, 567 likely voters: Haaland 47, Hull 41. The Maine and New Mexico polls were released without field dates, so their dates are placeholders. New Hampshire Senate Pappas +12.9; Maine Senate Jackson +3.6, 77 percent; Kansas governor Masterson +0.9, 56 percent; New Mexico governor Haaland +11.6. The House takes Emerson for PIX11 in NY-17, September 27 to 29: Conley 48, Lawler 46.

Re-run with Wedgewood Polls in Georgia, October 2 to 3, 600 likely voters: Ossoff 55, Collins 45 and Jackson 51, Lance Bottoms 49; and the Zenith Research poll for the Bengs campaign in South Dakota, released without field dates or a sample size: Rounds 44, Bengs 38. Georgia Senate Ossoff +10.8, 99 percent; Georgia governor Lance Bottoms +2.4, 69 percent; South Dakota Senate Rounds +13.0. The House takes campaign polls in TX-9, Mealer 45, Gutierrez 39; TX-10, Gober 48, Rourk 41; and VA-5, McGuire 46, Perriello 45. TX-9 moves to Mealer +7.3, 85 percent. Senate control 87 percent counting Osborn and 84 percent without; governor control 92 percent; House control 89 percent.

Every Senate and governor race was re-run at 10,000 simulations with the October On Point Polling Benchmark added to the respondent database: 529 analysis ready respondents, 4,767 in all. October is the most Democratic wave so far, D 56 to R 38 after the undecided allocation, and the pooled database moves from D+9.9 to D+10.7. Every October only question, from public figure favorables and twenty policy items to the 2028 vote and the generic ballot effect items, is now mapped onto every earlier respondent from their demographics and core political answers, each model checked against October respondents it did not see. Races moved toward Democrats by 0.1 points on average and at most 0.7, in the Iowa governor race. Senate control 87 percent counting Osborn and 84 percent without; governor control 93 percent; House control 90 percent.

No poll was released overnight, but a check against the aggregators found sixteen September polls the model did not carry, and they are now in: in North Carolina, High Point, YouGov, Opinion Diagnostics for Common Cause, Quantus Insights and Rasmussen; in Ohio, Rasmussen and Quantus Insights; Trafalgar in Alaska; Big Data Poll in Michigan; InsiderAdvantage in the Georgia Senate race; UMass Lowell/YouGov in both Maine races; and the New York Times/Siena governor questions in Maine, Michigan and New Hampshire. The New York Times/Siena polls of October 3 now carry their published sample sizes and field dates. Alaska Senate Peltola +1.8, 63 percent, from +2.9; Ohio Senate Brown +3.5, 77 percent; North Carolina Senate Cooper +11.0. The House takes American Pulse in MT-1, Flint 45.9, Forstag 45.4, which moves to Flint +0.7. Senate control 86 percent counting Osborn and 83 percent without; governor control 93 percent; House control 90 percent.

Three statewide polls released in late September were missing and are now in: American Pulse for NonStop Local in Montana, September 19 to 24, about 600 Montanans: Alme 45.1, Bodnar 26.1, Bankhead 18.9; and MassINC for CommonWealth Beacon in Massachusetts, September 14 to 22, 800 likely voters: Healey 54, Minogue 32 and Markey 54, Deaton 28. Montana Senate Alme +19.5; the house effect refit also nudged both New Mexico races. The House takes NYT/Siena in Alaska at large, Begich 47, Hill 35; the Tulchin poll for the Dunlap campaign in ME-2, Dunlap 50, LePage 48; DCCC Analytics in OH-9 and WI-1, both tied at 48; and Lighthouse Research in UT-1, McAdams 63, Owen 21. ME-2 stays a toss up at 52 percent for Dunlap and OH-9 moves to Kaptur at 65 percent. No new statewide poll was released October 2 to 4 that the model did not already carry. Senate control 88 percent counting Osborn and 85 percent without; governor control 93 percent; House control 90 percent.

Every Senate and governor race was re-run after a backtest on 139 Senate and governor races from 2018 and 2022, each fit on one cycle and scored on the other. The spread now depends on how well a race is polled: races with eight or more polling houses missed by less than the old spread said and thinly polled races by more, so the spread runs from 0.90 to 1.08 times its old size, and each race's own shock has the fatter tails real misses showed. In those cycles races in different regions shared no measurable miss, so the national vote shocks are halved and a small regional shock is added: Senate races now move together at about 0.17 instead of 0.62, and each race is exactly as uncertain as before. Each pollster's house effect is measured across all 59 polled races, shrunk and centered on zero so the average as a whole never moves, which cut the typical miss 31 days out from 6.3 to 6.1 points in the backtest; registered voter polls take the measured likely voter gap, 0.3 points this year. Inside the last three weeks the polls' floor rises to 0.64 on election day. No undecided break rule beat an even split, so undecided voters still split like decided ones. A data check now runs on every sweep and fixed three entries: the AARP poll in North Carolina at 1,115 likely voters, Stratus in the Texas governor race at 856, and DHM in Oregon. Each race is now simulated 10,000 times and each day's forecast is archived. Senate control 88 percent counting Osborn and 85 percent without; about 10 points of that rise comes from the lower correlation between races. Governor control 93 percent.

Each candidate's 2026 statewide favorable and unfavorable rating, or job approval for officeholders, was researched from public polls: 192 readings, with both candidates rated in 34 races. Across those races the Democratic minus Republican net rating explains how far the polls run from the model's fundamentals, correlation 0.57, and predicting each race from the others cut the typical miss by a quarter. The candidate term is now pulled toward what the ratings predict instead of toward zero, about 3.3 points of margin per 10 net points, so heavily polled races are still set by their polls and thinly polled races lean on the ratings. Campaign and partisan polls count 0.7 and move 1.5 net points against their sponsor. The largest moves are 0.7 points: California governor, Mississippi Senate, New Hampshire governor and South Carolina Senate. Senate control 72 percent counting Osborn and 71 percent without; governor control 89 percent.

A county can now run past the most or least Democratic it has been relative to its state, in the 2016, 2020 and 2024 presidential races and past races for the same office, only as far as counties actually did in 2022: 0.09 in log odds plus a quarter of its last presidential trend, the narrowest room that held 95 percent of the 2022 vote. The old hand set room was 0.15 plus half the eight year trend; 38 counties now top out instead of 3. A county's own past Senate and governor lean and a returning candidate's own county pattern were both tested on 2022 and made the county map worse, so neither was added. The re-run also takes this morning's polls: New York Times/Siena in Texas, Alaska, Ohio, Kansas and Iowa, CBS News/YouGov and Trafalgar in Maine. Senate control 72 percent counting Osborn and 71 percent without; governor control 89 percent.

Re-run with Quantus Insights in Alaska, released October 2, 758 likely voters: Sullivan 46.1, Peltola 45.5. Alaska Senate Peltola +0.9, 56 percent. The House takes the Cygnal poll for the Langworthy campaign in NY-23, Langworthy 54, Gies 33. Every other poll released since October 1 was already in. Senate control 73 percent counting Osborn and 72 percent without; governor control 90 percent; House control 90 percent.

Both poll lists were checked against the full public record. Kansas adds Trafalgar, September 29 to October 1, 1,095 likely voters: Hamilton 44.4, Marshall 43.5, and Global Strategy Group, September 13 to 16: Hamilton 47, Marshall 43. Texas adds Pulse Decision Science, September 24 to 27: Paxton 48, Talarico 45, fills in Stratus Intelligence at 856 likely voters, and counts the Overton Insights poll once instead of twice. Texas Senate Talarico +1.9, 62 percent; Kansas Senate Marshall +0.9, 57 percent. Senate control 74 percent counting Osborn and 73 percent without.

Re-run with co/efficient in Nebraska, September 28 to 30, 1,113 likely voters: Ricketts 45, Osborn 40 for Senate and Pillen 43, Walz 38 for governor; and Rasmussen in Iowa, September 28 to 30, 1,028 likely voters: Turek 47, Hinson 43. Nebraska Senate moves to Ricketts +1.1, 58 percent; Nebraska governor Pillen +9.4; Iowa Senate Turek +2.2, 65 percent. The House takes co/efficient in SC-1, the Cline internal in VA-6 and the Dembo internal in KY-6; KY-6 moves to Alvarado +0.7, 53 percent. Senate control 74 percent counting Osborn and 73 percent without; governor control 90 percent; House control 90 percent.

AdImpact aired spending through September 30 is now stored with each battleground Senate race: Republicans lead aired advertising by $61 million in Texas, $41 million in Ohio, $27 million in Maine, $17 million in Iowa, $15 million in Michigan and $12 million in Alaska, and hold 61 percent of future reservations. The model does not read ad spending, so the ten races re-run moved by 0.2 points or less. Senate control 74 percent counting Osborn and 72 percent without.

Re-run with State Navigate: Scott 45.7, Janoo 43.6, in a sample that recalls Scott +61 in 2024. Vermont governor Scott +10.6, 96 percent, from Scott +12.2, 98 percent. Governor control unchanged at 90 percent.

Re-run with Big Data Poll, likely voters with leaners: Flanagan 45.8, Tafoya 43.6, Whiting 5.2. Minnesota Senate Flanagan +8.3, 93 percent, from Flanagan +9.3, 95 percent. Senate control 74 percent counting Osborn and 71 percent without.

Re-run with InsiderAdvantage, September 28 to 29, 1,200 likely voters: Graham 46, Andrews 45, other 6. South Carolina Senate Graham +1.9, 64 percent, from Graham +2.3, 68 percent. Senate control 75 percent counting Osborn and 71 percent without.

Re-run with Wedgewood Polls, September 27 to October 1, 400 likely voters: Jackson 52, Collins 48. Maine Senate Jackson +3.3, 73 percent; Maine governor Pingree +13.1, unchanged. Senate control 74 percent counting Osborn and 71 percent without.

Re-run with Fox News in Texas, September 24 to 28, Talarico 51, Paxton 49 and Abbott 52, Hinojosa 47; Rasmussen in Texas, September 28 to 29; Trafalgar in Michigan, September 28 to 30; InsiderAdvantage in Ohio, September 28 to 29; Bravo Group for PennLive in Pennsylvania; and Cygnal in Alaska. Texas Senate Talarico +2.2, 65 percent; Michigan Senate El-Sayed +4.8, 79 percent; Ohio Senate Brown +5.0, 83 percent. The House takes the Washington Post and Schar School polls of VA-1 and VA-5 and DHM for OPB in WA-3; VA-1 moves to the Democrat at +1.1. Senate control 73 percent counting Osborn and 70 percent without; governor control 90 percent; House control 90 percent.

Every Senate, governor and House race was re-run with refreshed campaign finance. Governor money is updated from the newest state filings in Florida, Ohio, Wisconsin, Pennsylvania, Minnesota, Maine and Arkansas, and six records were corrected: Arizona and Kansas no longer count a pledge and an ad reservation their sources excluded, Vermont and Rhode Island now compare both candidates on the same basis and date, Tennessee takes exact filed totals, and South Dakota's money term is off until the Democratic nominee files. Ohio now compares 2026 donations on both sides, Acton $16.5 million and Ramaswamy $14.4 million before his own $25 million, and moves to Acton +5.8, 87 percent. Kansas governor moves to Masterson +2.1. No Senate or House nominee has filed with the FEC since the last update; quarterly reports are due October 15. Senate control 75 percent counting Osborn and 72 percent without; governor control 90 percent; House control 90 percent.

Re-run with Fox News in Michigan and Iowa, September 24 to 28; co/efficient in Maine, September 28 to 30; Suffolk for USA TODAY in Ohio, September 23 to 27; Trafalgar in New Hampshire, September 26 to 27; DHM for OPB in Oregon, September 17 to 27; and RMG for Alabama Policy Action, September 21 to 25. Maine Senate Jackson +1.7, 62 percent; Ohio governor Acton +5.3, 84 percent; Ohio Senate Brown +6.2; Michigan Senate El-Sayed +5.7; Iowa Senate Turek +1.1. NY-17 takes Emerson for PIX11, Democrat 48, Republican 46. Senate control 75 percent counting Osborn and 72 percent without; governor control 89 percent; House control 90 percent.

Every Senate, governor and House race was re-run after a 2024 backtest of the method in eleven states. Each county now starts from its 2024 presidential lean and moves the way counties actually moved in the 2018 and 2022 midterms, with the rest of the model's county pattern held to the spread those midterms showed; the assumed trend carry is replaced by the measured one, a small reversal. Each state's turnout now blends the vote history model with its own midterm record, which predicted 2022 better than either alone. No poll was adjusted and no statewide margin moved more than 0.3 points, but the county maps did: Miami Dade goes from D+11.7 to D+2.3 for Nixon and Starr from D+27.0 to D+20.4 for Talarico, and Florida's Hispanic counties now run about 5 points ahead of the state's swing instead of 11 to 14. Senate control 77 percent counting Osborn and 75 percent without; governor control 88 percent; House control 90 percent.

The Early Vote page's party estimate for the fourteen states that report no party now starts from the forecast's own D+12 electorate, voter type by voter type, and gives each type its chance of voting by mail or early in person under its state's ballot rules. Vermont and Hawaii mail every voter a ballot, so their requests are now read as the voter file: Vermont moves from D+59 to D+35 and Hawaii from D+53 to D+20. The mail skew is checked against Florida, North Carolina, Pennsylvania and New Jersey, and Michigan's published ages now reweight its estimate. Nationwide requests read D+19.4 counting reported party and the estimate.

Every Senate, governor and House race was re-run. Each county now carries forward its 2016 to 2024 trend relative to its state, weighted toward the latest cycle and scaled up or down by whether its 2026 primary electorate moved the same way, and the office history leg now carries more of the state's own trend for the office. The trend moves votes between counties, not the statewide level: Miami Dade moves to the right and Duval to the left, the Rio Grande Valley moves right and the Dallas exurbs left, and statewide margins moved by at most 0.3 points. Senate control 78 percent counting Osborn and 76 percent without; governor control 89 percent; House control 89 percent.

Re-run with Big Data Poll's Tar Heel State Poll, September 27 to 29, 642 likely voters with leaners: Cooper 50.8, Whatley 38.5. Cooper +12.7, 99 percent.

Every Senate, governor and House race was re-run on a D+12 generic ballot, now the one national anchor for both the approval leg and the census leg, and on a net Trump approval of -24 among registered voters. On Point Polling respondents put likely voters about a point more disapproving, and the simulated 2026 electorates come out at about -26. Senate control 77 percent counting Osborn and 74 percent without; governor control 87 percent; House control 89 percent.

Michigan Senate and governor took NPR/Marist: El-Sayed 51, Rogers 44 and Benson 53, James 41. Iowa Senate took Quantus Insights: Hinson 48.5, Turek 46.6, now a toss up. Wisconsin governor took Marquette: Crowley 49, Tiffany 46. Nebraska governor took Lake Research for the Walz campaign: Pillen 40, Walz 40.

Every Senate, governor and House race was re-run. Each county now counts its own citizen adults by race group from the Census Bureau's 2020 to 2024 citizen voting age tables, in place of one national rate. Counties can no longer run past the most or least Democratic they have ever been relative to their state, turnout stays within what each county has actually cast, and Black, Latino and Asian voters stay inside a realistic statewide range built from Pew's validated 2020 and 2024 vote and On Point Polling respondents. Senate control 78 percent counting Osborn and 76 percent without; governor control 87 percent; House control 89 percent.

Re-run with InsiderAdvantage, September 28 to 29, 1,200 likely voters: Flanagan 46, Tafoya 45. Flanagan +9.5, 95 percent.

Every Senate, governor and House race was re-run with a voter behavior layer built from the On Point Polling respondent file. Each simulated voter type now carries its own reasons to turn out, read from stated intent, ballot plans, past voting, friends who vote, approval intensity, economic strain and the niche questions, and every simulation moves those motivations up or down, so a low commitment year or soft Trump approvers staying home changes who votes. States with both a Senate and a governor race share a state level shock set by how often On Point Polling respondents switch parties, and the model estimates each race's split ticket voters. The House also takes 36 new district polls in 29 districts, including a Republican poll of CO-8 by McLaughlin and Associates with Evans 48, Rutinel 46. Senate control 79 percent counting Osborn and 77 percent without; governor control 86 percent; House control 89 percent.

Re-run with the governor ballot from Big Data Poll's Buckeye State Poll, September 26 to 27: Ramaswamy 47.9, Acton 46.8. Acton +2.5, 68 percent.

Every Senate, governor and House race was re-run. Third party candidates no longer fall to zero when a race's polls skip the question: their share now blends what a candidate of that ballot label usually draws, the state's own past third party vote for the office, and only the polls that asked. The census leg now reads the unified On Point Polling respondent file, counts citizen adults, and adds a small pooled effect for each state. Senate control 78 percent counting Osborn and 77 percent without; governor control 88 percent; House control 90 percent.

Every Senate, governor and House race was re-run. Five points of weight moved from the polls to the fundamentals, the 2026 primaries now carry more of the candidate signal, and Senate and governor races read the state's own last two results for that office, moved to this year's environment. Senate control 78 percent counting Osborn and 75 percent without; governor control 87 percent; House control 90 percent.

Both Michigan races were re-run with Cygnal (R) and Beacon Research (D) for Michigan Enjoyer, September 18 to 21, 600 likely voters: El-Sayed 45, Rogers 40 and Benson 45, James 38. Senate: El-Sayed +4.7, 80 percent. Governor: Benson +7.3, 90 percent. Senate control 79 percent counting Osborn; governor control 86 percent.

Both Ohio races were re-run with Big Data Poll, 682 likely voters, Brown 46.9, Husted 42.5, and NPR/Marist, September 24 to 27, 1,298 registered voters: Brown 51, Husted 43 and Acton 50, Ramaswamy 44. Senate: Brown +6.6, 88 percent. Governor: Acton +4.0, 77 percent. Senate control 80 percent counting Osborn; governor control 86 percent.

Both Florida races were re-run with the Change Research poll of September 26 to 28, 1,063 likely voters: Jolly 48, Donalds 47 for governor and Moody 48, Nixon 47 for Senate. The governor average now carries every public poll of the race, 21 in all, including Stetson University at Donalds 51, Jolly 39. Governor: Jolly +0.5, a toss up at 52 percent. Senate: Moody +5.1. Governor control odds 84 percent.

The site is now On Point Politics. On Point Polling still fields the polls, and every forecast number carries over unchanged.

The House was re-run with every statewide change from today, so the district simulations share the same national and demographic shocks as the Senate and governor runs.

Both Kansas races were re-run with new polls from Wedgewood, GBAO and Tavern Research.

Governor races now respond to about 60 percent of the national swing and of the state's partisan lean, beta 0.6 and lambda 0.6, with the rest moved to state level uncertainty, because governor races are more local. Every governor race was re-run.

Both races were re-run with new polls.

Re-run with InsiderAdvantage, 1,200 likely voters, 48 to 46, and YouGov, 3,299 likely voters, 46 to 44.

Re-run with the Braun Research poll: Scott 44, Janoo 34, among 817 likely voters.