Abstract
<p>Election forecasts report precise probabilities. For example, as of 15 July 2026 we forecast a 41.4% chance that Democrats regain Senate control, with 49.8 expected Democratic seats (90% interval 45–55). Those predictions depend on choices we make in building our model. How much does incumbency matter? How strongly do national conditions move individual races? This article asks whether Senate electoral history (2010–2024) can tell us which choices are right. For the U.S. Senate, we show it mostly cannot. Races in the same year all move with the same national tide, so even eight cycles of polling-era elections behave more like eight observations than hundreds. We build 120 reasonable versions of a Senate forecasting model, and 117 of them predict past elections similarly well. For 2026 they put the chance of Democratic control anywhere from 33.9% to 47.1%. This is an example of the “Rashomon” effect: models starting from different assumptions arrive at similar performance but divergent substantive conclusions. We explore the consequences of this phenomena for our forecast and Senate forecasts generally. We report both our own forecast of choice alongside the “Rashomon set” of models with similar performance. In overview, the range of predicted outcomes in the Rashomon set honestly reflects forecast uncertainty in this case.</p>