Live Calibration

Live scoring begins as 2026 races resolve. No 2026 race outcomes are recorded yet. The model is freezing its predictions weekly into forecast_snapshots so each race can be scored the moment its winner is recorded in race_outcomes.

Snapshots

The forecast snapshotter runs weekly (Monday 13:00 ET) and writes today's projected P(D) for every forward 2026 race into forecast_snapshots. So far: 8507 snapshot rows total across all races; 0 snapshot rows are for races already resolved and contribute to the live Brier above.

Model vs market, in progress

Maine Senate. On the June 12, 2026 snapshot the model priced Maine at 6% D while the betting market priced it at 62% D. That is one of the largest model-vs-market gaps on the board this cycle. The Democratic nominee has since changed twice: Graham Platner won the June 9, 2026 primary, withdrew on July 10, 2026, and the Maine Democratic Party nominated Troy Jackson at its July 25, 2026 convention. The June 12 snapshot was therefore priced on a candidate who is no longer in the race, and the model’s Maine inputs were refreshed for the new nominee on August 13, 2026. This race is not yet resolved and is not in the live Brier above.

Framing note: this is model vs. market in progress, not a claim of a resolved race. The June 12 reading is preserved as a locked call and will be scored as-is when Maine votes on November 3, 2026, but it should be read as a call made under the pre-withdrawal candidate field, not as a live forecast of the Collins versus Jackson matchup. The model leans on Maine’s personal-vote spread for Susan Collins (R); whether that feature is over-weighted will be scored on this page in the live-Brier table below.

Backtest Brier (2024)

The published forecast reports an out-of-sample Brier of 0.0978. This is a backtest number, not live. The forecast is fit on the two most recent completed cycles, so it cannot hold one of those out to score itself. The reported Brier is instead the genuine out-of-sample score of an identically specified forecast measured on a cycle it did not train on. The window it serves from and the window its accuracy is measured on are reported separately and never conflated. It is shown here for context only and is never combined with the live Brier above.

Reliability curve

Accuracy by cycle

Out of sample Brier for every holdout cycle, Senate and Governor against House. Lower is better, and every bar has to clear the coin flip line to mean anything at all.

Backtest explorer

351 settled backtest races, each shown at the earliest cutoff the backtest scores (90 days out, the hardest of its four cutoffs). The pooled and per cycle Brier figures above average all four cutoffs, so a single row's miss will not reproduce them exactly. The model called 297 of 351 correctly on the side it favored.