Monte Carlo · 100,000 runs · Polymarket + Kalshi probabilities
Running simulations…
Fetching Polymarket + Kalshi prices, then running 100,000 scenarios
First load can take a few minutes — cached for 6h after that
We run the election 100,000 times using betting-market prices as each race's baseline odds. Races don't miss independently: every simulated election draws a shared national swing, a swing per region, and one per state (fat-tailed, so 2016-style systemic misses happen occasionally), then decides each race around its shifted margin. The left chart shows how often Democrats finished with exactly that many seats; bars past the majority line mean Democratic control. The right chart is the same question answered by the markets directly — traders buy seat-count brackets ("Democrats win exactly 51 seats"), so those bars are the market's own distribution, with our simulation overlaid as the dashed line for comparison.
| State | Current | Win % | Poly spread | Mkt margin | Poll avg | Model avg | '24 Pres | Cook PVI | Handicappers | Fundraising (CoH) | Source |
|---|
Same idea as the Senate section: 100,000 simulated elections on the left, each bar the share of simulations landing on that exact seat count. All 435 districts are simulated — districts without a betting market carry their 2024 margin as a cushion and can still flip when a simulated wave breaches it. On the right, the House seat brackets the markets themselves trade, against the same simulation curve — where the market bars run wider than the dashed line, traders are pricing in more uncertainty than the calibrated model carries.
All 435 districts simulate: — at market prices; the other — carry their 2024 margin as a cushion and can flip when a simulated wave breaches it.
| District | Win % | Poly spread | Mkt margin | Poll avg | Model avg | '24 Pres | Cook PVI | Handicappers | Fundraising (CoH) | Source |
|---|
A gerrymander wins extra seats by spreading a party's voters efficiently — many districts won by deliberately thin margins. The catch: if the political wind shifts, all those thin seats flip together. That's a "dummymander." The chart counts, across 100,000 simulations, how many engineered seats the drawing party loses; the table lists each engineered seat and its market-implied chance of backfiring.
| Seat | Drawn by | Map | Mkt margin | Poll avg | '24 Pres | P(backfire) |
|---|
Pick any two metrics and see how the races line up. Dots are colored by market win odds and labeled by race; click any dot to open that race's details, and selecting a race in the tables highlights its dot here. Money axes use log scales. When the Y axis is a probability, a fitted logit curve is drawn, and when an axis is the market win probability a yellow line marks each chamber's marginal seat under that ordering — the one on the margin of majority (the time machine re-ranks each day, so the line jumps to whichever seat is marginal on the day shown). '24 presidential margins use The Downballot's 2026-lines calculations; redrawn districts are omitted only from PVI and last-result axes (still old-lines data).
Everything above is a snapshot of right now. This section records one snapshot per day so you can see the story move: control odds drifting, projected seats creeping, and which races are tightening. Gaps in a line mean no snapshot was recorded that day.
Each ridge below is one week's seat distribution (the newest ridge is today's) — like stacking each week's chart behind the next. Ridges getting taller and narrower over time = growing confidence about the outcome. The first pair comes from our simulations, the second directly from the market's seat-count brackets.
Every fresh poll is news — this table asks whether the news mattered. For each recent poll we compare the race's market-priced D-win probability just before the poll landed to where it stood 24 and 48 hours later. Impact is measured from daily snapshots, so a dash means the window isn't covered yet (snapshots build forward from the first recorded day). A move isn't proof the poll caused it — other news moves prices too — but big jumps right after a surprising poll are usually no coincidence.
| Race | Pollster | Asked | Released | Sample | D | R | Margin | Rating | Weight | Moved avg | Mkt D win | Market impact (24h / 48h) |
|---|
🎞 animated history-replay chart (GIF, 1080×1080) — rendered in your browser, made for social posts
Every number on this page is someone else’s work. This site only runs a simulation on top of it — the polling, the race ratings, the statistical forecasts, the campaign-finance filings and the live prediction-market prices are all produced by the organisations below, and we believe they are the best public sources of information on the 2026 US midterm elections. If this dashboard is useful to you, they are the reason. Please read them directly, and subscribe to or support them where you can.
Ratings and forecasts are reproduced here for comparison and remain the property of their publishers. Where a source publishes both a rating and a model, the two are shown in separate buckets because they are different kinds of claim. Every link above is a normal followable link — nothing on this page is sponsored, affiliated or paid for.