Accountability
How we will be scored
Anyone can claim to be accurate. We record every forecast before the votes are counted and grade it in public afterward, including the misses.
The rule
- Every day at 5 PM Eastern we save the Precision Combined forecast for every race in the Race Center, along with the chance each source gave on its own.
- The forecast we are graded on is the last one saved before polls open: Monday, November 2, 2026.
- Once a race is called, we score that forecast and publish it here beside the same score for Kalshi, Polymarket, PredictIt and the poll average alone. If a single source beat our blend, this page will say so.
- Saved forecasts are never edited. Each one records the method version it was made with (current: 2026-10-12), so a later change to the method can’t rewrite the past.
The score: Brier
The Brier score is the standard grade for probability forecasts. For each race it is the square of the gap between the chance we gave the favorite and what happened (1 if the favorite won, 0 if not), averaged across all races. Lower is better: 0 is perfect, and always saying 50/50 scores 0.25.
80% and they win
(0.8 − 1)² = 0.04
80% and they lose
(0.8 − 0)² = 0.64
50% either way
(0.5 − 1)² = 0.25
Confident and right scores well. Confident and wrong is punished hard. That is why a blend that hedges sensibly can beat a single market that is often right but sometimes badly wrong.
2026 scorecard
First results after November 3
| Forecast | Races scored | Brier score |
|---|---|---|
| Precision Combined | — | Not yet scored |
| Kalshi alone | — | Not yet scored |
| Polymarket alone | — | Not yet scored |
| PredictIt alone | — | Not yet scored |
| Public polls alone | — | Not yet scored |
| Precision member poll alone | — | Not yet scored |
We will publish the first scores within two weeks of Election Day and add each remaining race, including runoffs, as it is called.
Recording now: 4 daily snapshots of 95 forecasts saved since October 9, 2026.
Why this matters
The research says blending independent forecasts usually beats any one of them. We think that holds for our blend, but a belief is not a track record. This page is where we earn one, race by race.