Atlas Ahead

Fifty-One

Thirty-five Senate seats are on the ballot in 2026, and 51 of the hundred decide who runs the chamber. Click any state with a seat up to change who wins it; the count moves with you. It starts where the polling starts — and where there is no polling, it starts where the seat already sits.

—Democratic

—Republican
51 SEATS FOR CONTROL · A 50–50 TIE IS BROKEN BY THE VICE PRESIDENT
Democratic hold or gain
Republican hold or gain
Hatched: no 2026 polling. Shown as the party that holds the seat today — a fact, not a forecast.
Purple ring: the models put this one in reach — P(flip) of 25% or more. That is fragility, NOT a predicted flip. Nothing here knows which way an error runs.
Dimmed: no seat on the 2026 ballot. 34 of these are Democratic and 31 Republican, and they are banked before any vote.

The 6 races the models put in reach

These carry a purple ring above. The ring means the polled margin is small relative to the error a race shaped like this one has historically carried — thin polling, or an average sitting far from how the state actually votes, or simply a lead inside the noise. It does not mean the race will flip, and it does not say which way. That distinction is the main finding of this whole project: magnitude is partly predictable, direction is not.

Iowa 50% flip
polls +0.0 · state lean -8.6 · 14 firms
The polling average is inside a point and a half of level, and it sits 9 points away from how the state has actually voted, on only 14 distinct firms.
Michigan 50% flip
polls +0.1 · state lean +2.8 · 30 firms
The polling average is inside a point and a half of level.
Texas 45% flip
polls -0.3 · state lean -5.7 · 37 firms
The polling average is inside a point and a half of level, and it sits 5 points away from how the state has actually voted.
Maine 34% flip
polls +2.2 · state lean +8.9 · 33 firms
The lead is small enough for a normal error to cross it, and it sits 7 points away from how the state has actually voted.
Alaska 32% flip
polls +3.0 · state lean -10.5 · 15 firms
The lead is small enough for a normal error to cross it, and it sits 14 points away from how the state has actually voted, on only 15 distinct firms.
New Hampshire 29% flip
polls +3.3 · state lean +7.5 · 18 firms
The lead is small enough for a normal error to cross it.

What this map is, and what it is not

Of the 35 seats on the ballot, 10 have general-election polling and the other 25 do not. The polled ones open on whoever leads the average of the final polls; the rest open on the party that holds the seat. Both are starting points for you to argue with, which is what the map is for.

No candidate is named anywhere here. Matchups are still settling, and this page identifies races by state — the same rule the forward panel follows in the Cloud Room. The seat counts reconcile to 100 at build time or the page does not build.

This is not the registered call. The pre-registered 2026 forecast freezes its spec on 2026-10-20 and its call on 2026-11-02. This is a toy for thinking with, built on polling that is still arriving.

Where the seat data comes from

Seat composition verified 2026-09-10. Built 2026-09-10 from data/senate2026/seats.json and forecast2026.json. Polling averages are our own unweighted mean of each race's final polls, not a published model's average.