Loading…
Below are the full 2026 chamber projections and race-by-race tables aggregated from the county-level voter swarm. Each county's voting-age population is partitioned into demographic cells (Census CVAP × Cooperative Election Study), simulated across 5,000 correlated runs, and rolled up into states and redrawn 2026 congressional districts.
Senate · 2026 Class 2 + Special Elections
—
Faded squares represent seats not up for election in 2026; solid squares represent the 35 seats on the ballot, colored by the favored party in the simulation.
| State | Incumbent | Party | 2024 Pres | P(Dem) | Rating | Polls |
|---|
House · All 435 seats · 120th Congress
—
—
| District | Projected | P(Dem) | Rating | Polls | Counties |
|---|
Presidential · Out-of-sample validation
—
| State | Projected | Actual | P(Dem) | Rating | Polls | EV |
|---|
Out-of-sample evaluation
Each past election was evaluated strictly out of sample: the county prior and CES survey parameters were fit using only elections prior to the target year. For example, the 2016 backtest used only data from 2012 and earlier.
—
Model accuracy & uncertainty
While a voter-level simulation can output arbitrarily fine decimal numbers, its actual precision is bounded by its historical out-of-sample error. Across the four backtested presidential cycles (2012–2024), the model's mean absolute error (MAE) at the county level serves as the baseline uncertainty floor used to calibrate simulation intervals.
—
Explore the mathematical roll-up sandbox, out-of-sample accuracy benchmarks, and 4-turn synthetic agent transcripts on our companion methodology page.
pipelines/swarm.py. Switch to the Monte Carlo Simulator, return to the 2026 Overview, read the Methodology & Past Elections Archive, read the Blog Post, or ☕ Buy Me a Coffee.