Behind the Calls
The networks will tell you who won. The Race Caller is built to show you why, and to put my judgment on the record before a single vote is counted. Here's exactly how it works, where the numbers come from and how I'm keeping myself honest.
What's different
Built to sit alongside the networks, not replace them.
| Typical election-night coverage | The Race Caller | |
|---|---|---|
| On the record | Analysts talk through scenarios; projections come once a race is decided. | A winner picked in all 125 races, locked before each state's polls close, with a full post-game analysis afterward. |
| Context | The vote count, with commentary. | Three forecasters' ratings, polling averages, prediction-market odds, early vote and the county numbers each candidate needs, side by side for every race. |
| Polling misses | Polls are treated as the baseline. | A poll-error dial that redraws the map if this year's polls miss the way they did in 2024, 2022 or 2025. |
| What's left to count | Percent reporting. | Votes left, the share the trailing candidate needs, and which way each state's late count usually breaks. |
| Accountability | Last night's analysis moves on to the next story. | A full post-game analysis once the count is done: how the picks held up against the polls, the markets and the forecasters. |
How I pick
Every race gets a winner. No hedging, no "too close to call."
Start with the polls
Each race starts at its polling average and the consensus of Cook Political Report, Inside Elections and Sabato's Crystal Ball.
Go against them only for a reason
I break from the polls when something specific points the other way: how a state's late votes usually break, county benchmarks, ratings moves, prediction markets or how polls missed there before. Those picks are flagged vs polls.
Lock it before the polls close
Each pick locks automatically when that state's polls close and can't be changed. Every version of the site is archived with a timestamp in a public GitHub record, so the picks are provably on the record before results come in.
Election night
Results, my calls and the math behind them.
Results
Vote counts and calls come from Associated Press–sourced reporting by major news outlets and from official state results sites. The desk checks for new results about every 15 minutes, and the website refreshes every 7 to 8 minutes through election night, with a countdown at the top of every page.
A race shows as called only when the AP or a major decision desk projects it. Never on my say-so.
My calls
Separately from the AP, I make my own calls as the votes come in. Each one is timestamped and compared with the AP's call, so you can see whether I was early, late, right or wrong.
The Call Watch flags races where the remaining vote can no longer realistically close the gap, adjusted for which way each state's late count usually breaks. It also flags races headed for an automatic recount or runoff.
Why count order matters
Where the numbers come from
After the count
Once races are called, I'll publish a full post-game: how my picks held up against the polls, the markets and the forecasters, where the polls missed and why, and what actually moved the vote.
The Race Caller is an independent project by Jon Friedman. It is not affiliated with or endorsed by the Associated Press, any news organization, ratings service or prediction market. Official results come only from state and local election officials.