Money vs. market — separate signals, not a single win percentage
Each race shows two independent model signals: a fundamentals lean (prior presidential or lagged House share, incumbency, and a national midterm environment) and a money signal (Democratic share of two-party FEC weball receipts). We do not publish a fused WhoDonated win percentage yet — stacking money on fundamentals did not beat either signal alone on competitive races in our 2018–2024 leave-one-cycle-out backtest.
What each signal means
- Fundamentals — district or state lean, incumbency, and a national midterm environment. For Senate we use the prior presidential two-party Democratic share; for House we use lagged House Democratic share.
- Money signal — Democratic share of two-party FEC cycle receipts for the seat (weball-style totals). Itemized donor pages on WhoDonated use the same underlying public filings.
- Markets — when permission allows, prediction-market prices appear as a third column. They are not polls and are not advice. Kalshi is not displayed without written permission.
Backtest (leave-one-cycle-out)
Trained and tested on 2018, 2020, 2022, 2024 (1863 races). Competitive races are those with a two-party margin under 10 points.
- Fundamentals: 85% accuracy overall, 64% on competitive races.
- Money only: 90% overall, 66% competitive.
- Naively combining the two did not beat either alone on competitive races in this first pass — so we show them separately.
Caveats
- Not a poll and not advice. Probabilities are model output under clear assumptions.
- FEC weball cycle totals can include a little post-election activity on historical cycles; 2026 uses live cycle totals.
- House lean uses lagged same-district Democratic share, which is weaker after redistricting.
- Senate lean uses the prior presidential two-party Democratic share by state.
- The 2026 national environment uses a placeholder +3 Democratic generic-ballot midterm setting.
- Nominees are inferred as the top-receipts Democratic and Republican candidates per seat; some primaries are still unsettled.
- Uncontested or one-sided fundraising races inflate money-only accuracy.