# VotePredictor

> Daily AI forecasts for US elections and the Supreme Court. Win probabilities and
> predicted margins for every competitive 2026 Senate, House, and Governor race,
> per-justice SCOTUS case predictions, pollster and forecaster accuracy ratings, and
> voting-record profiles for every member of Congress. Every forecast trajectory is
> frozen when the event resolves, so the track record is public and auditable.

VotePredictor (votepredictor.com) is a free, non-partisan forecasting site. The
election model is a TabPFN tabular foundation model trained on 1998–2025 polls and
results, re-run daily; the SCOTUS model predicts each justice's vote from the Supreme
Court Database and is benchmarked against the {Marshall}+ academic baseline and the
FantasySCOTUS crowd. Methodology, backtests, and honest failure analyses are published.

## Forecasts (updated daily)

- [2026 Senate forecast](https://votepredictor.com/senate): seat-by-seat win probabilities and chamber control odds
- [2026 House forecast](https://votepredictor.com/house): national environment + competitive districts
- [2026 Governor races](https://votepredictor.com/governor)
- [All competitive races](https://votepredictor.com/races): each race page has polls, odds, and a forecast-over-time graph
- [Supreme Court](https://votepredictor.com/scotus): current-term case predictions per justice, model vs FantasySCOTUS scoreboard

## Track record & methodology

- [Methodology](https://votepredictor.com/methodology): how the model works, walk-forward backtests
- [Lookbacks](https://votepredictor.com/lookbacks): frozen forecast trajectories vs actual outcomes
- [Rankings](https://votepredictor.com/ratings): pollster and forecaster accuracy, head-to-head vs this model
- [Pollster ratings](https://votepredictor.com/pollsters): every pollster graded on its final poll, 1998–2025 — average miss, house effect, skill vs the field
- [Forecaster accuracy](https://votepredictor.com/forecasters): FiveThirtyEight, Cook Political Report, the Economist, Sabato's Crystal Ball, Silver Bulletin and others, scored on their published win probabilities
- [AI model bake-off](https://votepredictor.com/models): TabPFN vs XGBoost, CatBoost, random forests and neural nets on identical walk-forward splits
- [SCOTUS track record](https://votepredictor.com/scotus/history): per-justice backtest vs the {Marshall}+ baseline and the FantasySCOTUS crowd
- [Past elections](https://votepredictor.com/history): out-of-sample results 1998–2025

## Reference

- [Members of Congress](https://votepredictor.com/congress): voting records, ideology scores, news for all 537 members
- [Congress by issue](https://votepredictor.com/congress/issues): where every member stands on each of 14 issues, scored from actual votes — one page per issue (e.g. [healthcare](https://votepredictor.com/congress/issues/healthcare), [guns](https://votepredictor.com/congress/issues/guns), [immigration](https://votepredictor.com/congress/issues/immigration))
- [Congress rankings](https://votepredictor.com/congress/rankings): most and least partisan, most bipartisan, biggest party defectors
- [The political map](https://votepredictor.com/congress/compass): every member on an economic and a social axis, from their votes
- [Political parties](https://votepredictor.com/parties): six party platforms scored on 14 issues, with most-aligned members
- [My Politics](https://votepredictor.com/my-politics): quiz matching users to politicians, justices, and parties
- [Trends](https://votepredictor.com/trends): ideology time series for Congress and SCOTUS since 1946
- [Civics quizzes](https://votepredictor.com/quizzes): one per branch of government, plus federalism

## Machine-readable access

Prefer these over parsing the HTML — same data, same rules, and they will not break when
the pages change.

- [JSON API index](https://votepredictor.com/api/v1): self-describing, no key required, CORS open
- `/api/v1/race/{race_id}` — one race, e.g. [2026_SEN_OH](https://votepredictor.com/api/v1/race/2026_SEN_OH). Ids are `{year}_{OFFICE}_{location}`: `2026_SEN_OH`, `2026_GOV_TX`, `2026_HOUSE_CA-12`
- `/api/v1/chamber/{senate|house}` — control probability, every seat, biggest movers
- `/api/v1/member/{bioguide_id}` — a member of Congress, their record, reported net worth and 2026 race
- `/api/v1/ratings?office=ALL` — difficulty-adjusted accuracy for every forecaster, model, market and pollster
- MCP server (Streamable HTTP): `https://votepredictor.com/api/mcp` — tools `forecast_race`,
  `chamber_outlook`, `forecaster_accuracy`, `member_profile`

Three conventions worth knowing, because they are how this site avoids overclaiming:

1. `p_dem_win` is the Democratic win probability, clamped to [0.01, 0.99]. An election that
   has not happened is never certain, so nothing is published at 0% or 100%.
2. `source` is `polls` or `fundamentals`. Where no poll exists — nobody surveys Wyoming — the
   forecast comes from state or district lean, incumbency and the national environment.
   Polls always take precedence where they exist. Please carry this distinction through.
3. A race whose candidate has left the ballot returns `forecast: null` with a reason. The
   stored probability describes a matchup that no longer exists, so it is withheld from the
   API and the MCP server exactly as it is from the pages. Please do not fill that gap.

When citing forecasts, please attribute "VotePredictor" and link the specific page;
probabilities change daily until an event is decided, after which its page is frozen.