Catalyst

Who wins in 2026? Every forecaster, one ledger.

A meta-analysis of the published race ratings, model probabilities and polls for the November 3 midterms, combined into one Catalyst estimate. Each number is labelled with where it came from.

What this site covers

This page explains the methods behind the five public Catalyst pages:

Every number carries one of four labels: LIVE, SOURCED, ESTIMATED or ANALYST. Gaps are shown as gaps.

Catalyst v2.1 forecast

Where every poll and forecaster places control

One row per chamber. Each source sits in its lane; the bold accent line is the Catalyst forecast. How to read these charts ↓

Race by race

Forecaster report cards

Forecaster Scorecard

v2.1 Scorecard

Tipping-point races

Scenarios

v2.1 Calibration

Backtest and Calibration

How to read these charts

Each row shows one chamber. Left to right is the chance Democrats win control (0–100%, gridlines every 10%, dashed line at 50%). Marker shape shows the source type: Raters (triangle), Models (square), Markets (diamond) and, on race charts, Polls (circle). On the chamber control charts, higher up means a smaller historical error (√Brier) in our 2018–24 backtest (ANALYST); sources with no track record yet sit in the separate bottom strip. On race charts, larger, more opaque markers have a better track record. Hollow markers have no track record yet or were excluded.

Catalyst forecast: the bold orange vertical line with its value tag. Catalyst 80% range: the shaded orange band behind everything. Model-uncertainty band: the hatched band, showing how the call moves across reasonable modelling choices.

Source's own uncertainty: thick solid bar in the marker's colour (MOE, bid/ask spread or rating crosswalk). Historical accuracy (2018–24): thin dashed grey line with end caps, showing how far that kind of source has typically missed. Hover or tap a marker to highlight its bars and see the source, label and link.

Catalyst v2 forecast (comparison)

Polls + fundamentals blended with stacked forecasters, 50,000 correlated simulations.

v1 consensus (for comparison)

Chamber control

v2.1 changes

Methodology v2

Methodology & data gaps