Call
the Map
FORECAST10 DAYS OUT

Changelog

Changes to the site, day by day: new features, fixes and corrections. Routine data updates are not listed.

October 2026

1 change
  • Removed every Wedgewood Polls poll from the forecast and from every poll list on the site, after the firm said the numbers it had published were fabricated. That includes the Nebraska poll it still calls real: none of its numbers can be checked. The forecast had counted its polls in 18 races until the evening of October 20; in the update that removed them, Nebraska's governor race and Mississippi's Senate race moved the most.
2 changes
  • The early vote tracker now has counts for 39 states, up from 12. For the states it does not read itself, and on days a state's own file is behind, the counts come from civicAPI, which compiles them from state and county election offices; each such count is marked.
  • The early vote headline now counts the ballots in every state shown, not only the states compared by party.
17 changes
  • Published The Drift: fifty years of how every state and county moved against the nation, 1976 to 2024.
  • Pages open faster after news, market and early-vote updates, and the busiest pages render faster.
  • The front page and the Scenario Machine download less, so they open sooner on phones.
  • Every page shows its first content sooner on phones, and the national polls page responds to taps right away.
  • Small grey labels and links in fine print are easier to read and to spot, in both light and dark themes.
  • Fixed district lines on The Map when zoomed in: they now follow the official boundaries street by street.
  • The Map now says it is loading until the House districts are coloured, instead of showing a blank map.
  • Fixed map cards for races between two candidates of the same party, which showed a generic label instead of both names.
  • Louisiana House races now name the Democrats on the ballot where no party has a nominee, instead of "A Democrat".
  • The Scenario Machine fills the screen on computers, with the controls always beside the map. Any race can now be edited by hand, and setting a Senate or House winner shows how the other races move in the forecast's own simulations.
  • A Share button on the Scenario Machine makes a picture of your map or race calls.
  • The Scenario Machine can show tiles or a real map on every tab, and its side panels stay beside the map as the page scrolls.
  • Set any race's winner and margin in the Scenario Machine, and every other race moves with the national mood that result implies.
  • Pick the Winners now lives inside the Scenario Machine.
  • Share pictures from the Scenario Machine read like a results graphic: exact margins, county maps and the candidates' photos.
  • Early Vote opens with a plain summary, a board comparing every state with the same day in 2022, a daily reading of what changed and pictures to share. Iowa's new file is read again.
  • A heart in the menu opens the support note at any time, and the note stays open until you close it.
9 changes
correction

Fixed: Alaska's Senate race

Alaska's Senate forecast used an out-of-date ballot and, in some polls, counted two different candidates named Dan Sullivan as one. Both are fixed. Mary Peltola's chance of winning rose from about 45 to about 50 in 100, and the Democrats' chance of controlling the Senate rose by under one point.

  • The ballot: David Leslie (D), who placed fourth in the August primary, withdrew on August 31, and Gerald Heikes (R), who placed fifth, took his place on the November ballot. The forecast still listed Leslie.
  • The polls: Senator Dan S. Sullivan and Daniel J. Sullivan Jr., a separate candidate from Petersburg, are both on the ballot. In three polls we added their numbers together and credited the total to the senator (Alaska Survey Research, October 1 to 3: 50 where the senator had 45; Rasmussen Reports, September 13 to 14: 46 where he had 39; Alaska Survey Research, September 10 to 12: 48 where he had 44). Each poll now counts the senator's own first choices against Peltola's, and the other Sullivan and Heikes as their own small shares.
  • The ranked-choice count: because of the two names, the race's own ranked-choice polls had not been matched to the candidates, so the count of eliminated candidates' votes used only our general assumptions. It now also uses the transfers those polls report, as in every other ranked-choice race; Heikes's votes go mostly to Sullivan.
  • The same Minnesota 1st District poll by SurveyUSA was listed twice under two field dates; it is now listed once. That race's forecast did not change.
  • Added Florida, Maryland and Nevada to Early Vote.
  • Added a newsletter sign-up line to the front page and footer.
  • Added a Show all control to Biggest moves.
  • Added a press citations line to the About page.
  • Put the logo on three articles' share images.
  • Shortened page titles in search results.
  • Sped up page loads, including House pages.
  • Changelog now lists every day's changes.
7 changes
  • Published Nobody's Candidate.
  • Changed share images to show the logo instead of the site address.
  • Rewrote several page headlines and search titles to say what each page is.
  • Fixed old state-legislative addresses to open the matching state page.
  • Removed a stale-feed banner from the Wire and front page.
  • Rearranged the Wire so stories beside the lead differ and columns end together.
  • Early Vote now shows a state's last count and our last check as separate dates.
2 changes
  • Added a From the Data Desk row of our own articles to the Wire.
  • Fixed the menu colors over cover photographs.
2 changes
  • Added a note to Early Vote explaining what the summary adds up to.
  • Fixed the Other share in states where candidates appear on several party lines.
8 changes
product

Turnout and vote-count estimates are no longer shown

Race pages, race cards and county tables no longer show a forecast turnout or a forecast number of votes. No margin, vote share or chance of winning changed.

  • The turnout estimate started from each state's 2022 ballots and used no information from the 2026 cycle, so it would likely have understated a high-turnout year. It was never an input to a margin or a chance of winning.
  • Vote shares and margins do not depend on how many people vote and are unchanged. Senate and governor race pages now show the state's actual turnout in the 2018 and 2022 midterms, and House pages the 2024 presidential vote on the district's ground, each labelled with its year.
  • Published The House Always Wins.
  • Added Pick the Winners, where you choose race winners.
  • The Map now shows vote circles in place of district fills.
  • Early Vote now shows every district by party, with counties on request.
  • Early Vote now compares each state with 2022 only on like-for-like terms.
  • Added a June primary map to same-party race pages.
  • Race pages now show one table of polls with the polling average first.
7 changes
  • Launched Early Vote, tracking early ballots in North Carolina, Pennsylvania, Maine, Colorado and Iowa.
  • Added Oklahoma to Early Vote.
  • Added Where the votes are, a county map, to every race page.
  • Added a Bubbles view with a circle for each county.
  • Redrew history charts as two chance lines with the days left out named.
  • Added a smooth trend line to poll charts.
  • Fixed poll chart tooltips that covered the nearest dot.
6 changes
  • Added a small support request on every page.
  • Polls pages now show the polls' own average.
  • Rewrote The Ohio Split as reported analysis.
  • Made Biggest moves windows consistent across time spans.
  • Redrew national charts to cover every day of the line.
  • Fixed history charts to show every day under the current method.
7 changes
correction

Fixed: races with a third candidate, an independent or a ranked-choice ballot

We found a flaw in how races with a third candidate, an independent or a ranked-choice ballot were handled, and fixed it. Here is what was wrong and what is different now.

  • The flaw: a third candidate entered a race's forecast as soon as a secondary source listed one, whether or not the candidate was on the certified ballot and however little polling there was. The race's polls were then re-read, and polls that had named the new candidate could drop out of its average. Kansas's Senate race and Arkansas's governor race changed overnight this way on October 4 and 5. A race with a third candidate was also centred on a plain average of the polls naming all its candidates: an April poll counted as much as an October one, party-sponsored polls were not adjusted, the correction for past polling misses was not applied, and the result was not blended with past results or the national mood.
  • A third candidate is now recognised only from the certified general-election ballot, by an official election-authority record. A recognised candidate is a minor candidate within the 'other' vote, and the race stays an ordinary two-candidate race that keeps all of its polls.
  • A third candidate is forecast separately only when on the certified ballot and averaging at least 5% across two polls from two pollsters. An independent standing in place of a major party's nominee, and candidates on a ranked-choice ballot, are forecast separately as before. The same rule applies to every race.
  • Polls naming every candidate are weighted like the polls of every other race (newer and larger polls count for more; party-sponsored polls are handled as elsewhere), and one poll release counts once.
  • Where a Democrat and a Republican lead a field with a third candidate, the margin between them is the midpoint of two readings that count equally: those polls as weighted, and the forecast's usual method with its polling-error correction and its blend with past results and the national environment.
  • Where a third candidate is in the top two, or a nominee is a token candidate, every candidate is centred on the weighted polls of the full field.
  • A third candidate's share combines their polls with how candidates of the same kind have finished in past elections. With no polls the share is that historical figure; with few polls it sits between the two; a high share requires real polling. The range drawn around the share narrows as more polls measure it.
  • An independent standing in a major party's place starts from that state's Democratic baseline moved 6 points toward the independent, before the race's own polls are blended in.
  • On a ranked-choice ballot, where polls published the final round between two candidates, that result counts for half in how transferred votes are simulated.
  • Races affected: the Senate races in Kansas, Mississippi, Montana, Nebraska, South Dakota and Idaho; the governor races in Alaska, Arkansas, Idaho, Maine, Nebraska and Rhode Island; and Alaska's House seat. Kansas's Senate race and the Arkansas and Idaho governor races return to ordinary two-candidate forecasts. No other race's estimate changes. Earlier published forecasts stay in the archive as they were published. The history charts show every earlier day with the corrected method, so the correction does not appear as a step in a race or a chamber; those days are recomputed, not forecasts issued then.
  • Added a polls ledger linking each race name to its polls.
  • Race pages now show the polls behind a candidate-field headline.
  • Rewrote The Ohio Split with new photographs and Ohio county results.
  • Changed Lines: Circles became Centered, with close seats in purple.
  • Poll average lines now bend at each poll's field date.
  • Rewrote pollster page titles and descriptions to answer searches.
6 changes
  • Search now finds a race by postal code.
  • Every race now shows one leader on all pages.
  • Polls pages now show one poll average with its weights and a daily line.
  • Fixed race notes that could contradict the forecast.
  • Lines districts are now drawn as single pieces.
  • Added three forecast updates a day.

September 2026

1 change
  • Updated Lines with close-up outlines, town-preserving maps and Voting Rights Act floors.
3 changes
  • Added a dated updates log and explanations of every measure to Lines.
  • Fixed The Map breadcrumb, Back and Forward behavior and state labels.
  • Fixed spellings of Bartlett, Fort Worth and St. Paul on Lines.
7 changes
  • Launched Lines, an interactive on redistricting with 440 maps.
  • Made The Map faster and opened phones on the whole country.
  • Added a county URL to The Map, so a selected county can be shared.
  • Added a 44px Full race link and one-line breadcrumb on phones.
  • Added a support ledger showing donations.
  • Changed Lines to show each map at its best, with district descriptions on hover.
  • Added county ranking by votes cast to Map search.
3 changes
  • Added share images for Lines.
  • Added extreme gerrymander maps and town-and-country maps to Lines.
  • Fixed nominees showing as A Democrat or A Republican once settled.
7 changes
  • Added an Interactives menu.
  • Launched Your Call, a build-your-own-forecast interactive.
  • Published The Testimony.
  • Added vote circles to The Map.
  • Updated Scenario to match Lines.
  • Changed rating chips to read the chance on the same scale.
  • Fixed duplicate rows for one poll spelled two ways.
5 changes
  • Rebuilt The Map with Alaska and Hawaii insets, labels and legend keys.
  • Fixed a district painted in the wrong shade on The Map.
  • Fixed a county with a small swing showing as no data.
  • Added a county-level swing direction instead of flat gray.
  • Added a House county view that blends up to 18 races, and says so.
3 changes
  • Added hatching for flipped seats, a lead headline and a House cartogram to The Map.
  • Fixed the mobile map camera in Hudson Bay and the dark no-data swatch.
  • Fixed Senate lens and URL bugs on The Map.
3 changes
correction

Movement tools now compare on a current-model history, not the day-by-day published numbers

Biggest Movers and the momentum desk used to subtract two days of the published forecast directly. When a specification changes mid-window, that reads as a race moving even when the race did not: on September 9 a legitimate model change (below) moved several governor races by several points overnight with no election news behind it. These tools now compare on a separate history recomputed under today's specification for every day, so a model change is never read as a race moving.

  • The recomputed series (scripts/model/ctmm2/hindcast21.py) uses only the evidence each day actually had — a strict as-of hindcast — and is regenerated whenever the model changes.
  • The day-by-day published record is unchanged and stays what it always was: what the site actually said on each day. It is what any accuracy or track-record claim on this site reads, and it always will be — this addition is for movement tools only.
  • See data/model/model_revisions.json in the repository for the full, dated list of specification changes this covers.
  • Added an embeddable race card with a trend chart.
  • Added a State of the Race page.
6 changes
  • Added the incumbent mark to race pages and cards.
  • Made charts work by touch.
  • Changed the forecast history line to wear the leader's color.
  • Updated the Wire to drop right-wing advocacy outlets and add 52 newsrooms and 40 accounts.
  • Fixed Louisiana's departed incumbents showing as presumptive.
  • Fixed a House district page missing its incumbent's name.
5 changes
  • Published The Fetterman Inversion.
  • Updated Biggest moves to show races that moved, with a quiet-window fallback.
  • Fixed Republican figures on the ticker showing the wrong color.
  • Redrew the race poll chart as a smoothed trend over the polls.
  • Redrew the chamber history figure as a line again.
1 change
  • Added two poll averages to race pages, each named for what it is.
2 changes
  • Fixed poll age to use the site's clock.
  • Showed one row per poll release in the Wire.
4 changes
  • Launched Biggest moves with six time windows ranked by news.
  • Fixed the polls table overflowing on phones.
  • Led race cards with the race instead of the number.
  • Moved every clock to Eastern time.
4 changes
  • Fixed eight House districts that had an invented opponent.
  • Fixed a governors share card that showed the wrong seat count.
  • Rewrote the imprint page.
  • Fixed a one-party ballot showing as a Lean.
6 changes
model

Retired demographic composition features from the structural priors

White, Black and Hispanic 2010/2017 Census-share variables were dropped from the House, Senate and Governor structural priors. Measured at the cycle level (the unit the backtest holds out): leave-one-cycle-out accuracy moved 6.8355 -> 6.8393 in the House and 9.2278 -> 9.2132 in the Senate (an improvement), and the paired difference straddles zero in all three offices — a legitimate, evidence-neutral change, not new information about any race.

  • The regional shock and the Senate/governor poll-bias deviation are now drawn once per simulation batch and shared across offices instead of independently per office, matching how AAPOR's 2020 task force found this error actually behaves.
  • Because this changed the fitted specification, comparing two published days spanning September 9 mixes a model change into the number. The current-model history above (added 2026-09-23) is how this site now shows that comparison without the mix-in.
  • Education, income and median age remain in the model.
  • Added Alaska's ranked-choice count.
  • Added at-large House seats and top-four nominees.
  • Recorded candidate withdrawals, and noted Begich's on the page.
  • Fixed same-party race ratings and removed empty ribbons.
  • Rewrote the launch note.
7 changes
  • Fixed Rhode Island showing a third-place share on a second-place chance.
  • Fixed three Senate races showing an independent's chance as the Democrat's.
  • Fixed ten races that said polls were withheld but published them.
  • Fixed the governors board counting a seat Rhode Island's page gave away.
  • Fixed British spellings across the site.
  • Merged nine prose pages into three, now centered.
  • Added an icon to site search and more ways to write a district.
8 changes
model

The model version changed to CTMM 2.1

The published forecast moved from CTMM v1 to CTMM 2.1. Every race centre, chamber probability and interval on the site now comes from the new model.

  • CTMM 2.1 rebuilds the centres on validated v2 fundamentals, adds G1 governor calibration and core-plus-tail uncertainty for the House and Governor boards, and restores the v2 architecture for the Senate. Two House data defects were repaired with it: twelve seats carried a holding party taken from the sign of a reconstructed prior rather than from the member defending the seat, and fourteen were contaminated by 2024 districts that had no candidate from one major party. No calibration layer was adopted; nested leave-one-cycle-out testing rejected every candidate, so the identity stands.
  • Because the model version changed, the public history series restarts here. Readings from before this date remain in the archive and are not joined to readings after it: a line drawn across a model change measures the model, not the race.
  • Added a welcome article and state-shaped confetti that opens each race.
  • Added compact polls pages for each race.
  • Split the About page into About, Contact and Support.
  • Rebuilt the methodology page.
  • Showed one row per poll instead of one per crosstab.
  • Fixed the polls list marking a counted poll as not counted.
  • Changed all times to Eastern.
5 changes
publication

Launch

Call the Map opens with forecasts for all 435 U.S. House districts, 35 Senate seats and 36 governorships, from the sealed edition ctm-1.0-20260906-25e5a4bb5f67.

  • Every race page shows a chance of winning, a projected margin where the candidate field supports one, the middle-80% range, and the evidence behind the estimate.
  • The full methodology is published in two forms: a plain-language guide and the technical document with every equation and fitted coefficient.
  • Corrections that change a published number will be recorded on this page and in the edition record on the About page, with the edition they landed in.
  • Added automatic page refresh and two color palettes.
  • Made Bands the default map view.
  • Sped up the map and the site.
  • Rewrote the About and methodology pages.