A 2026 midterm forecast, locked before Election Day and scored against every result as it comes in. No edits after the fact: the fingerprint at the bottom proves it.
Every forecaster's numbers as published on the day Claude's forecast was locked. Scoring fills in as races are called.
| Forecaster | Senate | House | Races | Called right | Brier ? | Head to head with Claude ? |
|---|
Bars are the share of 40,000 simulations ending at each seat count. The marker is today's tally: seats already called plus the expected value of races still out.
Same idea for the House. 218 seats is a majority.
Both numbers are the win chance for the Democrat (or independent, where marked) as published before the election. A stamp appears once a race is called.
| Race | Matchup | Claude | Experts | Forecast margin | Result | Verdict |
|---|
Races where Claude's number is furthest from the average of the other forecasters. Once a race is called, the stamp shows whose number was closer to what happened.
Called races where the model gave the eventual winner the least chance.
Called races grouped by the model's win chance. An honest forecast sits on the diagonal: races given 70% should be won about 70% of the time.