01 / THE OVERVIEW
One national election.
Different chamber dynamics.
The source 2026 U.S. Senate and House Full Analysis presents a stronger Democratic position in the House than in the Senate. It reproduces DDHQ historical forecast benchmarks, with evidence cut off at 8:00 a.m. EST on November 3, 2026.
The distinction is important: this is an editorial presentation of the source research, not a new Bellelytics forecast, a live election feed, or an independently re-verified candidate or polling database.
An 84% probability still leaves room for the other outcome. Forecasts can miss, and closely related races can move together.
| Chamber | D control | R control | Projected seats |
|---|---|---|---|
| House | 84% | 16% | 241 D / 194 R |
| Senate | 65% | 35% | 52 D-caucus / 48 R |
Source: source report, p. 1; DDHQ House and DDHQ Senate references. External pages may now display a different snapshot.
02 / THE HOUSE
Many roads to 218.
The House portion covers all 435 voting districts. The report describes several possible routes to a majority: contested redrawn Southern seats, Midwestern swing districts, and suburban contests around major metropolitan areas.
Its chamber-level projection is not a declaration that each individual seat has been decided. Summing race-level probabilities produces an expected seat count; the provider's separately published seat projection is a different summary. The two should not be presented as interchangeable.
The public race directory lets you inspect the nominee roster and source date without turning a historical forecast into a live call.
Explore the House directory →Source: source report, pp. 1 and 4; workbook, “National Outlook” and “Evidence and Method.”
03 / THE SENATE
The arithmetic behind the path.
The report's Senate baseline contains 34 Democratic-caucus holdovers and 31 Republican holdovers. Of the 35 seats on the ballot, 13 are Democratic-held and 22 are Republican-held.
That leaves four Republican-held pickups to reach 51, assuming every Democratic-held seat is retained. An additional loss would require an additional pickup. This is illustrative seat arithmetic, not the probability that an entire sequence will happen.
New independent winners remain a separate category in the source analysis. A modeled independent win is not automatically a Democratic-caucus seat, and no future caucus commitment is inferred.
Explore the Senate directory →Source: source report, pp. 2–3; workbook, “Evidence and Method.”
04 / WHAT COULD CHANGE THE PICTURE?
The uncertainty is part of the story.
A shared polling miss can move many seats.
Race outcomes are not isolated coin flips. Common polling errors and different turnout patterns can affect multiple contests at the same time. Multiplying individual win probabilities will not reconstruct the provider's chamber-control forecast.
A flip is not automatically an upset.
A seat changing parties may already be the forecast's most likely outcome. The report's upset screen instead highlights modeled outsiders with meaningful chances against a more clearly favored opponent. It does not establish an independent forecasting advantage.
Ratings, vote shares, and probabilities answer different questions.
A qualitative race rating is not a numerical chance of victory. A poll's candidate share is not a win probability. A projected margin is not a confidence interval. Keeping these separate is essential to reading the research fairly.
Missing coverage is not a zero chance.
The workbook retains all nominees in its source snapshot, while the provider prices selected leading party groups. Unpriced candidates are not assigned invented probabilities. Ranked-choice and same-party ballots also carry their original coverage notes.
Source: source report, pp. 5–7; workbook, “Evidence and Method.”
05 / THE RECORD BEHIND THE BRIEF
Transparent about the limits.
The research covers 470 unique election records, a 50-state House summary, and 1,282 nominee entries from the October 8 Clerk snapshot. Later withdrawals were not exhaustively audited. Historical provider entries were retrieved later, so the research is a reconstruction rather than an immutable archive captured at the original time.
Coverage checks do not establish forecast calibration or predictive accuracy. The earlier Pennsylvania experiment is separate and is not blended into the national benchmarks.
This public edition contains an attributed overview and a nominee directory. The original research files and complete race-level provider forecast database are not distributed here. Republishing those materials requires a separate rights review.
Review sources & methodology ↗