Jorge Garcelán

THE PROJECT

Why this projectWhy this project

An open forecast of the 2026 US midterms, made for WATIF TV and for Spanish-speaking readers and journalists: all 470 races, each with four readings side by side (the polls, the Polymarket price, and two models of its own) and the expert ratings next to them, so you can see where they agree and where they do not. Every number can be traced to its source and date.

It is not affiliated with any party, campaign or media outlet. The model is experimental: as it runs now it is replayed on 2018 and 2022 (see Validation), but treat it as one more signal, not a prediction of certainty.

The forecasts, predictions and models are independent: they are not tied to any party or campaign. The models are experimental: as they run now they are replayed on 2018 and 2022 (see Validation), but treat them as one more signal, not a prediction of certainty.

HOW IT IS BUILT

Data, model and code

• ModelMP-26: 50,000 simulations of all 470 races, built on its own baseline (2024 presidential lean, incumbency and the national environment) and the polls of VoteHub (CC BY 4.0) and Wikipedia (CC BY-SA 4.0).
• Census-26A second, research model for the House that reads only each district's census profile and region, never polls or markets: what the district's demographics alone would predict.
• DataPolls from VoteHub and Wikipedia (every Senate race and House district), Polymarket prices for each race and each chamber, Cook / Inside Elections / Sabato ratings, FEC filings, Census ACS, MIT Election Lab and county results.
• Election nightThe WATIF projections desk, following the networks and AP; the official count from civicAPI, county by county; the models re-read after each projection.
• CodeNext.js and TypeScript; the forecast is archived daily so every change can be audited.