Polls vs Poly

About Polls vs Poly

The first platform to systematically compare poll-based election forecasts with blockchain prediction market odds across every competitive 2026 race.

--
Verified Polls
--
Poly Markets
50
States Covered
7/7
Primaries Correct

What This Is

Polls vs Poly is an open-source election forecasting platform for the 2026 U.S. midterm general elections — Senate, Governor, and all 435 House districts. It runs a poll-weighted Bayesian model alongside real-time Polymarket prediction market data, then shows you exactly where the two systems agree and where they don't.

Important: This model forecasts general elections (November 3, 2026), not primaries. Where primaries haven't occurred yet, projections assume the current frontrunner or incumbent. Primary dates and results are tracked and displayed, but win probabilities refer to the general election matchup unless explicitly labeled as a primary race (e.g., KY-04 Massie vs Gallrein).

How the Poll Model Works

The poll-weighting algorithm is modeled on the methodology pioneered by Nate Silver at FiveThirtyEight:

Recency Weighting

Polls decay on a 30-day half-life. A poll from two months ago carries one-quarter the weight of one released today. This ensures the model adapts to shifting voter sentiment rather than anchoring on stale data.

Pollster Quality Grading

Each polling firm is graded on a scale from A+ to F based on historical accuracy and methodology (mirroring the Silver Bulletin/FiveThirtyEight rating system). Lower-rated firms are automatically discounted. Partisan-sponsored polls receive a 40% penalty, and their margins are regressed 1.5 points toward center.

Sample Composition

Likely voter (LV) screens receive full weight. Registered voter (RV) polls are weighted at 75%, and adult population polls at 50%. Sample size is factored using square-root scaling, capped at 2.5x for very large samples.

State Fundamentals

A partisan lean prior (similar to Cook PVI) prevents the model from being overconfident when polls show a party winning in hostile territory. For example, a Democrat leading by 6 points in Alaska (Trump +14) is treated with significantly more skepticism than the same lead in Pennsylvania. The fundamentals share is small (6-15%) — a sanity check, not the dominant signal.

Uncertainty & Win Probability

Win probabilities are computed using a normal CDF over the weighted margin divided by total uncertainty. Uncertainty accounts for:

How Polymarket Data Works

The platform pulls real-time odds from Polymarket's Gamma API across all active 2026 election markets. Data is fetched in parallel using 15 threads and cached for 15 minutes with automatic warmup on server start.

Polymarket odds represent the implied probability derived from actual money wagered by traders. Unlike polls, which sample voter intent, prediction markets aggregate the collective judgment of participants with financial skin in the game.

Where They Disagree

The Disagreements view on the map highlights races where polls and prediction markets reach different conclusions. Key current divergences include:

When polls and markets agree, confidence in the projection is higher. When they disagree, it signals genuine uncertainty that neither source alone captures.

Track Record

Seven states held their 2026 primaries between March and May. In races where the model had polling data with specific candidates, it correctly identified every nominee. In races without polls, the model relied on analyst ratings only.

Polled Races — Specific Nominee Predictions

These races had polling data naming specific candidates. The model predicted who would win:

StateRaceDateOur PredictionActual Result
TXSenate DMar 3Talarico wins primaryTalarico 53.0%, Crockett 45.7%Correct
TXSenate RMar 3Cornyn vs Paxton top 2Cornyn 42.0%, Paxton 40.5% → runoffCorrect
TXGovernorMar 3Abbott (R) vs Hinojosa (D)Abbott 82.6%, Hinojosa 59.9%Correct
NCSenateMar 3Cooper (D) vs Whatley (R)Cooper 92%, Whatley 64.6%Correct
OHSenateMay 5Brown (D) vs Husted (R)Brown 89.5%, Husted unopposedCorrect
NESenateMay 12Ricketts (R) vs Osborn (I)Ricketts 81.8%, Osborn running independentCorrect
WVSenateMay 12Capito (R) winsCapito 66.5%Correct

Rating-Only — No Specific Nominee Prediction

These races had no polling data. The model assigned a general election rating based on state partisanship, not a prediction of which candidate would win the primary:

StateRaceDateOur RatingActual Result
ILSenateMar 17Safe D (no polls)Stratton (D) 40% over Krishnamoorthi 33%Rating only
ILGovernorMar 17Safe D (1 poll)Pritzker (D) unopposed, Bailey (R) 54%Correct
INMay 5No Senate raceNo Senate raceN/A

Summary: 7 specific nominee predictions, 7 correct. 1 rating-only call (IL Senate) where we correctly identified the party but did not predict the primary winner.

The next major test is the Kentucky 4th District Republican primary on May 19, where the model projects Trump-backed challenger Ed Gallrein to defeat incumbent Thomas Massie with 77% probability — based on two May polls showing Gallrein leading by 5-8 points after trailing by 5-10 points in April.

What This Is Not

Data Sources

Every poll in the database is verified with a direct source URL linking to the original pollster publication or news report. Sources include:

Built By

Polls vs Poly was built by Mike Glenn of Devion Software Developers. The platform is fully open-source.

Disclaimer: Polls vs Poly is an independent, non-partisan election forecasting project. All projections are probabilistic estimates, not guarantees. Past accuracy does not guarantee future performance. Election outcomes depend on factors that no model can fully capture. Always consult multiple sources when forming your views.