Methodology

This is a small, experimental election model. It is not a pure polling average and not a heavyweight simulation, but a simple set of estimates meant to answer one question: if the election were held under today’s conditions, what kind of result would be reasonable to expect?

What goes into the model

Public polling: When reliable polls exist, they are the starting point. The model pays more attention to fresher, higher-quality polling and less to older, sparse, or inconsistent data.
Basic fundamentals: State partisanship, incumbency, national environment (generic ballot, presidential approval), and recent election results help anchor races where polling is thin or noisy.
Current conditions: Major news events, candidate quality, scandals, retirements, and other context can gently nudge a race toward one side or the other when the environment clearly shifts.
AI-assisted synthesis: An AI system is used to read and combine these inputs, keep track of the overall picture, and update numbers as information changes. The structure is intentionally simple and judgmental rather than a black-box algorithm.

How to read the numbers

Two-party vote share: General elections are shown as projected two-party vote share, with Democrat and Republican percentages that add to 100.0 and are rounded to one decimal place.
Primaries: Primary charts show the estimated share of the vote for named candidates in that contest. Small “other” or undecided buckets are not displayed on the main cards, even though the model internally accounts for them.
Leaders and margins: Labels like “Cooper +2.0” represent the model’s estimate of the margin between the top two candidates, rounded to one decimal. A small lead does not mean a safe race; it just reflects where things sit right now.
House control: The House page focuses on the national generic ballot and a simple directional read on which party is more likely to control the chamber, not an exact seat-by-seat projection.

Limits and caveats

Directional, not certain: These are best read as directional estimates, not precise predictions of final results and not advice for trading or betting.
Data quality varies: Some races have rich polling and history; others rely more heavily on fundamentals and judgment. The thinner the data, the more cautious the interpretation should be.
No inside information: The model only uses public information and broad indicators. It does not incorporate private polling, proprietary data, or campaign internals.
Subject to change: As new polls, events, and fundamentals shift, the numbers will move. Snapshots on this site are just that—snapshots of a moving target.

Roadmap

Over time, this project may expand to include simple race ratings (for example, “Lean D” or “Toss-up”) and more detailed state-by-state primary views. For now, the priority is to keep the model small, transparent, and readable.