The snapshot at a glance
35 original picks
Published before Election Day, including additional race guides beyond the 28-call final article.
22 matched · 2 missed
Only 24 scored in the November 4 snapshot. Eleven were not scored for reasons including unresolved or not-yet-tracked results.
Using 22/24 would produce 91.7% among scored forecasts only. It would be wrong to present that number as our final accuracy across 35 picks. The unscored group is real and visible, not counted as wins.
The two misses we have identified
- South Carolina Senate: We picked Republican Darline Graham. The recorded call in our November 4 file was Democrat Annie Andrews.
- Texas governor: We picked Republican Greg Abbott. The November 4 file records a call for Democrat Gina Hinojosa.
We should not hide those picks under a headline celebrating the party-control forecast. The probability assigned and the reasons for each call matter more than selecting a plausible winner after the fact.
Which calls mattered beyond safe seats?
Among the Democratic pickups in our archived picks were Senate choices in Kansas and Iowa. The October 30 final-calls article labels both as editorial overrides of a formula that leaned the other way. Our post-election record should preserve that distinction. A correct override does not retroactively make the original formula correct.
What counts as a match?
- The forecast is timestamped before the election, with its original candidate and party.
- The result is a sourced projected winner call, not an early vote lead.
- If the race is not yet tracked or called locally, the prediction remains unscored.
- When later source-attributed calls become available, scores can be updated without changing the original forecasts.
For complete clarity, our public audit file identifies tracked and outside-the-feed races separately. That is why “11 unscored” should never be paraphrased as “11 races still uncalled nationwide.”
From election forecasts to accountability forecasts
ElectionTracker's impeachments and firings prediction page is the natural next test of this methodology. Its predictions should retain a date, a resolution criterion and a change log. A cabinet member being investigated is not the same as being fired, and impeachment is not removal.
Original evidence
- Archived 35-race prediction audit and dated results snapshot
- Our October 30 final calls and methodology
- Live local results board (distinct from the dated audit snapshot)
- Roll Call, national election results tracking, November 10
This audit describes a historical November 4 data snapshot and intentionally does not silently replace it with an unsourced November 10 tally.