50 states · upper, lower & Nebraska
Explore the prior, the information behind it, and the range of elections that could happen together.
Each model is a separate joint posterior. Conditions update its supported chambers together; binary, four-state, broad candidate and standalone Nebraska draws are never merged.
The broad candidate model is experimental and is not selected for daily forecasts. Converged sampling does not establish predictive accuracy. Read its held-out results separately from fit diagnostics in Historical research.
Loading official display geometry.
No outcomes selected.
Historical results, district plans and candidate records must pass coverage checks and retrospective validation before this chamber receives probabilities.
Use the synthetic example to explore the scenario picker. Its numbers are generated for interface testing and say nothing about the 2026 election.
Keep only joint elections satisfying every outcome you choose.
Choose a winner to condition the entire state. Select a district name to inspect its evidence.
Interpretable by construction
No prediction markets. News may enter only after a reproducible historical coding rule improves held-out validation.
Historical accountability
Hold out whole election cycles. Track probability calibration, vote-margin coverage and chamber seat distributions; publish the uncertainty and data gaps.
Research results · publication withheld
Loading the research artifact.