U.S. House

10 states redrew their congressional maps for 2026 — TX, CA, MO, OH, NC, UT, FL, TN, LA, AL — so this forecast applies each redrawn district's new 2024-presidential lean (181 districts) instead of the old lines, with the national environment added as a uniform swing from the current generic ballot (D+6.5, swing D+2.7).

Model: each district's 2024 presidential margin on its 2026 lines, measured relative to the national vote, plus one intercept fitted on the 2016–2024 cycles, plus the national environment as a uniform swing. That is the whole model. It replaced a tree on lean, incumbency, money and expert ratings because it beat that model on every metric in every walk-forward cycle; the incumbency premium the tree learned kept shrinking. New-map leans are sourced per district for the ten redrawn states. The control and seat-total estimates are more robust than individual toss-up calls. The margin knows nothing about who is running, so an open seat and a long-serving incumbent with the same lean get the same margin; the win probability adds a small incumbency premium (about two points, re-estimated each cycle with a two-cycle half-life and preregistered separately), so a district can show a narrow Democratic margin with a below-even Democratic probability when a Republican incumbent is running. Where a district has polls, the poll average is blended in by inverse variance (preregistered separately), with about half the weight on a single poll and more as polls accumulate, and the same blended number is what every page, the API and the seat simulation use. Seats with a large presidential-vs-House split (e.g. South Florida / South Texas Hispanic districts) read too Democratic here.