Our prediction comes from the polls, with Undertow AI making the final call. Betting-market odds are Polymarket prices, shown only for comparison; they are never used in the forecast. Qloo-only is what each state's tastes alone predict, plus the national shift in the polls. Click a race to open it on the map; click a column heading to sort.
The map shows where. These charts show how much: where Qloo's tastes disagree with the polls this year, how well tastes alone reproduce the real 2024 result, and how the signal did when we tested it on 2020→2024.
Each row is a race. Amber is our call, made from the polls. Teal is the Qloo-only forecast: the vote each state's tastes predict, plus the national shift the polls show. Sorted by the gap; bold rows have a different winner. Click a row to open the race.
We fit the taste model to 2020, then sorted counties by how far their tastes sat from their 2020 vote. Counties whose tastes were bluer than their vote really did swing toward Democrats in 2024, and redder ones toward Republicans: steadily, but by fractions of a point. Bars show the swing beyond what the 2020 margin and county size explain; the whiskers are 95% intervals.
The same test for whole states: a state's taste gap did not predict its 2020→2024 swing (the line is flat to slightly backwards). That is why Qloo never moves a statewide call here; the polls do.
Caveat for every backtest: Qloo's data describes tastes today, not in 2020, so these are not clean out-of-time forecasts. All models are scored on states they never saw.
The regression predicts each county's 2024 presidential margin from its 24 strongest tastes and its size, across 3,104 counties. "Held-out state" means the model never saw that county's state.
| Measure | Training fit | Held-out state | County size only, held-out |
|---|---|---|---|
| Average miss | 15.2 pts | 15.5 pts | 19.7 pts |
| Average miss, weighted by votes | 11.3 pts | 11.8 pts | 19.6 pts |
| Root-mean-square error | 20.6 pts | 20.9 pts | 26.4 pts |
| R² | 0.57 | 0.55 | 0.29 |
| R², weighted by votes | 0.78 | 0.76 | 0.42 |
| Correlation | 0.753 | 0.743 | 0.546 |
| Wrong winner, counties | 9.1% | 9.4% | 12.2% |
| Wrong winner, weighted by votes | 13.8% | 14.5% | 24.8% |
This shows how well today's tastes match the 2024 vote, not how well they predict a future one. For comparison, always guessing Republican gets 14.4% of counties wrong but 48.7% of votes wrong, which is why the vote-weighted rows are the fairer test.