The Geodesic projection, open for you to drive. Every dial below changes one piece of the model's structure, and 20,000 simulated elections rerun in your browser each time. The locked call, published Oct 27 before any votes were counted, is the starting position.
| Scenario | Senate control | Exp. Senate seats | House control | Governor majority |
|---|
43 Democratic-caucus seats are safe or not up. The 13 races below decide it. Nudge any race to see what it does to control. The highlighted row is the tipping point: the race that would deliver the 51st seat.
| Race | Matchup | Polls | Poll avg | Trend /wk | Turnout geo. | Your nudge | Margin | P(Dem) |
|---|
Democrats hold 24 governorships now. 10 Democratic seats are safe or not up; Alaska's ranked-choice race is treated as a coin flip. Blank matchups are races whose nominees are still being verified in our source files.
| Race | Matchup | Polls | Poll avg | Trend /wk | Turnout geo. | Your nudge | Margin | P(Dem) |
|---|
The House is coarser than the Senate. Each seat sits at the position of its Cook Political Report rating (Oct 16), with positions measured from this year's statewide races. There is no district-level geometry yet, so the turnout dials don't touch the House.
| Cook rating | Seats | Position (D margin) | Expected D wins |
|---|
Each race starts at its polling position: every public poll since Sep 12, with any trend damped to a quarter of its slope (κ = 0.25). That damping beat the raw average on 185 past races, 3.93 vs 4.09 pts of average miss.
Where a poll published party crosstabs (TX, NC, OH, ME, MI and IA Senate; MI, OH, ME and NY Governor), each party's voters are placed in a Galileo space built from candidate favorability. A voter's stake is how lopsided their distances to the two candidates are, measured in elliptic coordinates. People with more at stake turn out more. The dial picks how mass enters:
Mass comes from seven synthetic partisan "brains" (Woelfel's Huey network) trained on 37,554 news articles and measured by how far each one moves when fed the other side's coverage.
From 329 Senate and governor races, 2010–2022, every race's miss splits into a shift shared by every race that year (spread 2.95 pts) and a race-specific remainder (4.45 pts). Because the shift is shared, races move together. That is why control odds are not a product of independent coin flips.
Every finding that held up in AAPOR's reviews of 2016, 2020 and 2024 is about who is in the electorate, not how people answer: late deciders, nonresponse, new voters, Republicans in Republican-leaning areas, Hispanic voters, and uneven turnout. None of the reports found "shy" voters or herding. Three additions follow from that, each a separate variant beside the locked call:
Probabilities are model outputs from 20,000 simulations with a fixed seed, so small differences (±1 pt) from the published tables are simulation noise.