METHODOLOGY

Data and model methodology

Sources, transformations, current coefficients and the validation gates for MP-26.

MODEL PIPELINE

From raw observation to published probability

Collect

Polling, official results, ratings, demographics and campaign-finance data.

Normalize

Preserve field dates, sample, population, mode, sponsor and source URL.

Estimate

Weight the national environment and run 50,000 seeded, correlated simulations.

Validate

The starting point's coefficients are fitted on the 2014–2022 midterms, and the model as it runs now is replayed on 2018 and 2022 (see Validation); the model stays experimental.

Explain

Use AI for retrieval and synthesis while keeping every numeric claim grounded.

CURRENT STATUS

Experimental v0.5

The whole model is built in-house: every race starts from its own baseline (the 2024 presidential lean, incumbency and, in the House, the incumbent's 2024 personal vote, moved by the national environment), is blended with its polls and runs through the 50,000-draw simulation. Polls come from VoteHub (CC BY 4.0) and, for the Senate, Wikipedia (CC BY-SA 4.0); recent polls weigh most and party-sponsored polls are corrected. Both chambers are simulated race by race, and the baseline is calibrated on the 2014–2022 midterms (see Validation).

Every run is seeded and reproducible.
Chamber odds equal the aggregate of the race odds.
Coefficients and intervals are public.
Candidate finance does not yet affect probabilities.
Jev only triages signals in the pilot.

See the pipeline, animated

SOURCE REGISTER

Primary inputs

VoteHubPolls, generic ballot included (CC BY 4.0); Senate polls also from Wikipedia (CC BY-SA 4.0)Cook Political ReportHouse race ratings FECCandidates + campaign finance Census Bureau119th Congress boundaries