MethodologyUpdated Sep 5, 2026

One pipeline: blend the record, simulate the House, calibrate against history.

Model read

83.1%Dem win chance

244.4Expected seats

244.4 seats expected; 80% of simulations land between 207 and 281 seats.Open forecast

1 · Blend

Prior House results, district fundamentals, the national environment, and direct polling where it exists are blended into one margin per district.

layered margin blend

2 · Simulate

The House is simulated thousands of times. Each simulated election is a full 435-race map with national, state, and district errors moving together; Democratic seats are counted against the 218-seat majority line.

Monte Carlo simulationmajority = 218 seats

3 · Calibrate

Final probabilities are recalibrated against past cycles and padded so the published ranges stay honest about real-world error. Where a race's data stopped, the last winner stands in.

noneuncertainty ×1.35freeze: prior cycle winner proxy

Backtest

0.073Brier score · 0 is perfect

0.257Log loss · lower is better

75%Correct chamber winner

61.2%Results inside 80% range

87.5%Seat totals inside range

14.7Average seat miss · seats

Scored on 8 past election years (2010–2024) — each one predicted before seeing its outcome.

Built from 5 data feeds · 12,062 records passed checks · State-certified election results · FEC campaign finance disclosure · Congressional district boundaries · Sources

Forecasts are probabilistic estimates, not predictions.