About

Why this exists

The 2026 US midterms were held on Tuesday, November 3. This site launched the next morning, while plenty of races were still being counted and "why haven't they called it?" was everywhere.

The honest answer is usually maths, and maths is more fun when you get to press the button yourself. Too Early to Call puts you on a decision desk: you watch votes arrive in the order they're actually counted and decide when a race is over.

Nonpartisan by design

Every state, county, candidate and party in the game is invented. The Lantern Party and Harbor Party aren't stand-ins for anyone. Which party leans toward mail ballots changes from race to race, so you can't win by assuming one side is always the "mirage". Nothing here describes, predicts or comments on any real 2026 result.

How the simulation works

• States and counties. Each race gets 7 to 12 counties: one or two big urban counties, a couple of swingy suburban ones, and a rural remainder that leans the other way. Each county has a "last time" baseline.
• Ballot types. Votes are split into early in-person, mail and Election Day ballots, and the two parties' voters favour different methods by different amounts in each race.
• Counting profiles. There are four state types: Early votes first, Mail counted last, Counts for days and Fast count. They decide when each batch of ballots gets reported. Big counties report more slowly. Late-arriving mail can break differently from early mail.
• The desk model. Like a real desk, the model doesn't know the final result. It knows the votes counted so far, an estimate of the expected total vote (which can be off by a few percent), and approximate pre-election data on how each ballot type leans. It estimates how the outstanding votes will break, with correlated uncertainty, and turns that into a win probability. That's what the Rookie desk sees.
• Rivals. The Wire calls only at 99.95% model certainty and declares sub-0.5-point races too close to call. Channel 7 calls whoever leads by a big enough margin with enough vote in, which makes it easy prey for the mirage.
• Shared nights. Every Election Night is generated from a seed number. Share the link and your friend gets exactly the same eight races and the same votes.

It's a simplification. Real desks also use voter surveys, precinct-level history and phone calls to county clerks. But the core logic of expected vote, outstanding vote and path to victory is the same.

Sources

The real-world facts on this site (history, quiz, explainer) come from:

• WBUR / AP: How the Associated Press calls election winners (Nov 2, 2026): wbur.org
• PBS NewsHour: How does the AP call 2026 midterm winners? 4 things to know: pbs.org
• NPR: How NPR makes race calls in the 2026 election: npr.org
• CBS News: Why some 2026 races won't be decided on election night: cbsnews.com
• FairVote: An analysis of statewide election recounts, 2000–2023: fairvote.org
• Wikipedia: Dewey Defeats Truman: wikipedia.org
• Wikipedia: 2008 United States Senate election in Minnesota: wikipedia.org
• US Election Atlas: 2000 election timeline: uselectionatlas.org
• NPR / Alaska Public Media: Luck of the draw: games of chance not uncommon in deciding tied elections: apr.org
• NPR: AP explains calling Arizona for Biden early (2020): npr.org
• It's About TV: Election night 1980 coverage: itsabouttv.com
• Ballotpedia: Recount laws in Pennsylvania: ballotpedia.org
• Protect Democracy: The "blue shift" and "red mirage" explained: protectdemocracy.org

Credits & privacy

Built as a one-day project for dailyma.de. Typefaces: Playfair Display, IBM Plex Sans and IBM Plex Mono via Google Fonts. No trackers, no accounts, no ads. Your best scores and theme are saved only in your own browser's localStorage.

Not affiliated with the Associated Press or any news organisation. "The Wire", "Channel 7" and "The Morning Ledger" are fictional.