What the market priced and what the polls measured about the same election, day by day, before it happened.
Dataset Card
By AFOS Analytics, published under cc-by-4.0, revision f1fcd7f72f09.
AFOS — US 2026 Midterms Divergence (v1.0.0, pre-electoral)
What the market priced and what the polls measured about the same election, day by day, before it happened.
Pre-electoral bundle. The election is on November 3, 2026. There is no certified result to score these forecasts against, so this is not the gold-standard version. See
DATASHEET.md. The v2, after certification, closes the cycle.
What is inside
- 419 national generic-ballot polls from 71 pollsters, fieldwork 2025-04-19 → 2026-09-17, 417 of them with the institute's own source URL
- 23784 market price rows across 9 Polymarket contracts: chamber control, seat distributions, turnout, popular-vote margin, and whether the election happens on schedule
- 517 headlines from 52 daily collections over a fixed list of 23 outlets
- One already decided race with its verified outcome: the Texas Republican Senate primary runoff
The one case that is already scored
The Texas Senate Republican runoff resolved on May 26, 2026. Ken Paxton beat John Cornyn with 63.8% of more than 1.38 million votes.
The market had priced Paxton between 57% and 62% for five straight weeks, then moved to 94.5% on May 21, five days before the vote. The price was a probability and the result is a vote share, so the two are not the same quantity and do not subtract. What the series does let you ask is a real question: was the pre-jump price well calibrated, and what moved on May 21.
How to read it without making the classic mistake
Do not subtract a market probability from a poll margin. One is the chance of controlling a chamber; the other is a lead in vote points. In 2012 Democrats won more votes and fewer seats, and the gap can be entirely geography.
Filter by market_slug. House and Senate name their outcomes identically.
Do not compare LV, RV and A samples as if they were one series.
Reproducing
node scripts/build-us-2026-dataset.mjs # data
node scripts/build-us-2026-metadata.mjs # datasheet, dictionary, package, checksums
Both read only files versioned in the AFOS repository. CHECKSUMS.txt carries SHA-256 for every file.
Citation
See CITATION.cff. Harvard Dataverse DOI: 10.7910/DVN/XRUT8U. Cite: Felipe, Andre, 2026, "AFOS · USA 2026 Midterms Electoral Divergence Dataset (v1, pre-electoral)", https://doi.org/10.7910/DVN/XRUT8U, Harvard Dataverse.
The citation above names NO version number on purpose: the DOI resolves to the latest release, and a hard-coded version goes stale the next time any file changes.
Data CC BY 4.0, code Apache 2.0.
Structure
distribution_coherence 2,183 rows
| Split | Rows | Size |
|---|---|---|
| train | 2,183 | 197.3 KB |
election_as_scheduled 184 rows
| Split | Rows | Size |
|---|---|---|
| train | 184 | 28.0 KB |
generic_ballot 419 rows
| Split | Rows | Size |
|---|---|---|
| train | 419 | 85.9 KB |
governors_distribution 4,207 rows
| Split | Rows | Size |
|---|---|---|
| train | 4,207 | 635.3 KB |
house_control 462 rows
| Split | Rows | Size |
|---|---|---|
| train | 462 | 58.7 KB |
house_seats_distribution 4,790 rows
| Split | Rows | Size |
|---|---|---|
| train | 4,790 | 694.5 KB |
popular_vote_margin_distribution 4,144 rows
| Split | Rows | Size |
|---|---|---|
| train | 4,144 | 558.8 KB |
press_timeline 517 rows
| Split | Rows | Size |
|---|---|---|
| train | 517 | 162.9 KB |
senate_control 747 rows
| Split | Rows | Size |
|---|---|---|
| train | 747 | 95.6 KB |
senate_seats_distribution 2,750 rows
| Split | Rows | Size |
|---|---|---|
| train | 2,750 | 385.2 KB |
texas_senate_primary 385 rows
| Split | Rows | Size |
|---|---|---|
| train | 385 | 48.5 KB |
turnout_distribution 6,115 rows
| Split | Rows | Size |
|---|---|---|
| train | 6,115 | 733.8 KB |
Details
- Repository
- AFOS-Analytics1/usa-2026-midterms-divergence
- Publisher
- AFOS Analytics
- Task category
- Not stated by the source
- Tags
- elections, prediction-markets, polling
- Size category
- Not stated by the source
- Languages
- en, pt, es
- Revision
- f1fcd7f72f09061855a04bd6027b56c211d84063
- Last updated
- 2026-09-20
Files
27 files, 3.8 MB in total.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| CHECKSUMS.txt | Data | 2.3 KB | — |
| croissant.json | Data | 3.4 KB | — |
| data/distribution-coherence.csv | Data | 192.6 KB | — |
| data/election-as-scheduled.csv | Data | 24.5 KB | — |
| data/governors-distribution.csv | Data | 546.5 KB | — |
| data/house-control.csv | Data | 50.4 KB | — |
| data/house-seats-distribution.csv | Data | 594.8 KB | — |
| data/popular-vote-margin-distribution.csv | Data | 468.5 KB | — |
| data/senate-control.csv | Data | 82.2 KB | — |
| data/senate-seats-distribution.csv | Data | 330.2 KB | — |
| data/texas-senate-primary.csv | Data | 41.5 KB | — |
| data/turnout-distribution.csv | Data | 596.3 KB | — |
| data/us-2026-case-summary.json | Data | 1.9 KB | — |
| datapackage.json | Data | 17.4 KB | — |
| polls/us-generic-ballot.csv | Data | 67.2 KB | — |
| press/us-press-timeline.csv | Data | 152.2 KB | — |
| raw/us-2026-texas-senate-primary-result.json | Data | 918 B | — |
| raw/us-polls-data.json | Data | 202.6 KB | — |
| raw/us-press-archive.json | Data | 409.7 KB | — |
| DATASHEET.md | Documentation | 7.3 KB | — |
| DATA_DICTIONARY.md | Documentation | 3.1 KB | — |
| LICENSE-APACHE-2.0 | Documentation | 119 B | — |
| LICENSE-CC-BY-4.0 | Documentation | 144 B | — |
| README.md | Documentation | 4.6 KB | — |
| CITATION.cff | Other | 725 B | — |
| banner.png | Other | 43.0 KB | 43640bef3e50 |
| .gitattributes | Repository | 2.5 KB | — |
License and Download
- License
- cc-by-4.0
- Access
- No access gate
Released by AFOS Analytics through its official repository on Hugging Face. Read the license.