Akashic
1876–2024
Forecast / scorecard

The forecast scorecard

The part nobody else does: the locked forecast, scored in public and kept permanently — Brier, calibration, the named misses, and the seat-forecast error. More transparent than the raters, more accountable than the quants.

ScorecardAwaiting the 2026 results. Below is the methodology proven against the 2018 and 2022 results — exactly what the 2026 scorecard will show when the outcomes land.

What will be scored

Two editions are locked and timestamped. When the certified 2026 outcomes arrive on the shared race spine, every call in each edition is scored automatically — and the original calls are preserved exactly as published.

2026 Midterm Forecast — edition 1
506 contests ratedlocked June 22, 2026
Results pending
2026 Midterm Forecast — edition 2
506 contests ratedlocked June 23, 2026
Results pending

Four things get scored: the Brier score on the win probabilities (overall and by office), the calibration curve (did our 70%-favorites win about 70% of the time?), the directional record with every miss named, and the seat-forecast error — was the actual composition inside our interval? The edition-1 → edition-2 comparison will show whether the revisions improved the call.

Proof the method works — the historical backtest

We can’t score 2026 yet, so here is the exact same scorecard run against cycles we already know the answer to. Each seat’s prior-cycle margin stands in for the call (the documented proxy), then it is scored against what actually happened. A stable cycle (2022) and a wave (2018) bracket what the real forecast will face.

2022 Senate + Governor — historical backtest
Called right
61 / 68
90%
Brier — overall
0.111
lower is better · 0 = perfect
Brier — Senate
0.085
Brier — Governor
0.136

Calibration

Did the favorites we put near a given probability win at that rate? On the line is perfectly calibrated.

perfect129173050607080901005060708090100Predicted win probability (%)Actual win rate (%)
2022 Senate + Governor, scored against the certified results. The “forecast” here is each seat’s prior-cycle margin — a deliberately weak documented proxy that sees no candidate quality or national mood, so it is a floor on a hand-rated forecast’s accuracy.

The misses, named

Every call for the wrong party, most-confident first.

Seat-forecast error

Predicted versus actual chamber composition — and whether the result landed inside our interval.

U.S. Senate
Predicted 49 DActual 51 D(+2 vs predicted)
inside 80%inside 95%
Governors
Predicted 23 DActual 24 D(+1 vs predicted)
inside 80%inside 95%
2018 Senate + Governor — historical backtest
Called right
51 / 66
77%
Brier — overall
0.147
lower is better · 0 = perfect
Brier — Senate
0.117
Brier — Governor
0.174

Calibration

Did the favorites we put near a given probability win at that rate? On the line is perfectly calibrated.

perfect1214152550607080901005060708090100Predicted win probability (%)Actual win rate (%)
2018 Senate + Governor, scored against the certified results. The “forecast” here is each seat’s prior-cycle margin — a deliberately weak documented proxy that sees no candidate quality or national mood, so it is a floor on a hand-rated forecast’s accuracy.

The misses, named

Every call for the wrong party, most-confident first.

+7 more.

Seat-forecast error

Predicted versus actual chamber composition — and whether the result landed inside our interval.

U.S. Senate
Predicted 47 DActual 47 D(on the nose)
inside 80%inside 95%
Governors
Predicted 19 DActual 23 D(+4 vs predicted)
inside 80%inside 95%

How the calls are made — and how this is scored — is on the methodology page. The scoring data is openly licensed, like the rest of Akashic. · Back to the forecast →