Two models reading the same polls are 20 points apart on control
UnclassifiedDDHQ says 54%, FiftyPlusOne 72.4%. Each matches what its own seat numbers imply, so the gap is in the seats: FiftyPlusOne is higher in ten of fourteen by 2 points or more, most of all in Kansas (25.7 points apart across sources), Ohio (16.5) and Maine (15.8). DDHQ's number is dated October 2, the last its page gave the reader; Silver Bulletin, read once on October 4, said 72.8%.
Why these should match, and what could explain the gap
Both models take the same public polling and the same seat map, and each one's control number sits inside the range its own seat probabilities imply. So the gap is not about how seats combine; it is the sum of seat-level differences. Kansas is widest: FiftyPlusOne's number fell from 51.2% to 48.7% in its October 26 reading, against DDHQ's 23% and Kalshi's 39.5%, both unchanged since October 24. Against the five-aggregator poll average of Hamilton +0.8, FiftyPlusOne takes 1.0 point off his lead, Kalshi 2.5 and DDHQ 5.6. In Ohio the poll average is Brown +3.8; DDHQ's model takes 3.6 points off his lead, FiftyPlusOne, down from 64.5% to 62.7%, takes off 1.7, and Kalshi, unchanged at 67.5%, takes off 0.9. In Maine, where the poll average is Jackson +2.0, DDHQ takes 1.5 points off his lead and FiftyPlusOne adds 1.2; Kalshi fell from 60.5% to 59.5% and takes 0.4 off. Expected Democratic seats are 50.5 for DDHQ and 51.5 for FiftyPlusOne, either side of the majority line. Kalshi (63.5%, up 1 point since October 25) and Polymarket (64.5%, down 2) are still between the two models. FiftyPlusOne's control number fell from 73.5% to 72.4% in the same reading, which narrowed the spread from 19.5 points to 18.4.
- Maine is where structure and polling point opposite ways: incumbents in Collins's position have lost 8 of 9 midterm races since 2006, but her 2020 result beat the final polling average by 12 points.
- Ohio and Kansas turn on how far to mark down a Democratic poll lead in a state Trump carried. The record runs from 2 to 3 points in the contested midterm races of 2018 and 2022 to 5.3 across all Republican-state midterm races since 2014. Brown adds a wrinkle: in 2024 he ran 7.6 points ahead of the top of the ticket and still lost by 3.6.
- Timing: DDHQ's reading is dated October 2. Its page showed no newer one when read on October 3 and 4, and the daily reader has recorded no newer reading on any day from October 5 to 26, so a newer DDHQ number, if one exists, is not on the board. FiftyPlusOne's is dated October 26. Every poll published in between, including New York Times/Siena surveys of Alaska and Ohio fielded through October 1 and YouGov's Kansas survey fielded through October 4, may be in one and not the other.
- A third model sits with FiftyPlusOne. Silver Bulletin's Deluxe forecast, read once on October 4 from its subscriber page and not a board source, has control at 72.8% and publishes projected margins, so its markdown can be read directly: it adds 0.8 points to its own poll average in Maine and takes 2.1 off in Kansas and 0.4 off in Ohio, nearer FiftyPlusOne's inferred markdowns than DDHQ's in all three. Its Classic version, without expert ratings, has control at 70.0% and differs by 0.2 to 0.7 points in those seats, so expert ratings, which FiftyPlusOne also uses, are worth 2.8 points of control inside Silver Bulletin's model, against a gap of 18. Agreement between two models identifies DDHQ as the one that differs and leaves open which is closer to the result.