For prediction markets

Independent, uses no polls, no surveys, no market prices as inputs.

471 races covered
2026 cycle

35 House districts in 7 states that redrew their maps for 2026 are held for re measurement on their new lines. Market prices helped us catch districts being forecast on old lines. We held them the same day. They return as each is re measured.

DatumSignals™ covers 471 races for the 2026 cycle -- every Senate seat, every governorship, and 400 House districts. Of 435 House districts, 424 have a price on Kalshi or Polymarket today; we forecast all 400. Every race is re-verified against public nominee records weekly.

See the full methodology →

What makes this independent

The model runs on structural conditions -- partisan lean, incumbency, fundraising, economic environment, public mood -- not on polling averages and not on prediction-market prices. Both are deliberately excluded as model inputs. A forecast built from your own order book can't tell you when your order book is wrong; ours can, because it was never downstream of it.

Prediction markets have done well pricing the races with national attention: competitive Senate seats, governors, the generic-ballot fight for control of Congress. Below that tier, coverage thins out fast -- of 435 House districts, 424 have a price on either exchange today. Not because those districts don't matter, but because building an independent forecast for hundreds of districts, refreshed daily, has historically been more infrastructure than any single market wanted to build. That's the gap we built for.

What we're proposing

A data license: DatumSignals forecasts as a pricing input for races you don't currently cover -- starting with House districts and down-ballot governors, particularly early in a cycle before organic trading volume exists to do price discovery on its own.

Concretely, that could mean:

This isn't a pitch to replace anything that already works. It's an offer to fill the part of the map neither of us can currently price well.

Why this model, and why now

Every product on this page depends on one thing: a forecast granular enough, current enough, and sharp enough to price correctly. That's a shorter list than it sounds.

Coverage sets us apart first. Most forecasters stop at Senate and Governor. Ours runs 471 races, including every House district -- exactly where markets have the least price discovery today and the most room for a new product to matter.

Reliability is what makes that coverage usable. We run a weekly verification layer built specifically to catch the errors -- a stale candidate, a mis-coded incumbency -- that make a forecast untrustworthy for anyone pricing real money.

DatumSignals
Uses public polling
Uses market prices as input
Covers House districts
Daily data verification
State-level campaign finance
Built by former campaign operators
Typical forecasters
Uses public polling
Uses market prices as input
Covers House districts
Daily data verification
State-level campaign finance
Built by former campaign operators

The real differentiator is what feeds the model, and what deliberately doesn't. We don't use public polling as a model input. Polling is sparse below Senate/Governor, expensive to commission at scale, and it measures stated intent rather than the structural conditions -- partisan lean, incumbency, fundraising, economic environment -- that actually determine outcomes. Building on fundamentals instead of polls is why we can forecast 400 districts nobody polls at all, and it means our signal is genuinely independent, not a repackaged poll average.

We also pull state-level campaign finance directly for governor races -- a signal most national models skip entirely, because governors don't file with the FEC.

Broad coverage, a live accuracy system, and a fundamentals-only design nobody else in this space has assembled the same way. Our Brier score isn't a number we quote -- it's a live computation against a public methodology and dated outcomes (currently 0.0978 test Brier (sengov-2026-govmoney-v1)). Anyone can rerun it and get the same answer. Full model card →

DatumSignals is built by David B. Wheeler and a team with decades of hands-on political campaign experience -- not a data-science exercise applied to elections from the outside. We've seen firsthand how campaigns actually spend, target, and win, and that operational reality shaped the model's design from the start.

Grow your market, not just your prices

You already have deep liquidity where national attention is. The bigger opportunity is everywhere that attention isn't yet.

403 races we forecast with no market today

435 House districts with no meaningful liquidity today. Every one is a possible new contract. We forecast all of them, daily, with a verification system built to catch errors before they ever reach a listing.

A product roadmap, not just a feed. Factor baskets, volatility indices, correlated-race parlays, divergence signals, fundraising props -- see below. Each is a new reason for a trader to open the app, built on infrastructure you'd otherwise have to build yourselves.

A partner built to expand your addressable map, not repackage what you already see. New contracts, new baskets, new reasons to trade.

Exclusivity or a no-cost beta -- your call. We're ready now. If being first to full House coverage matters, that's worth talking about before someone else gets there.

Data we collate

FEC filings 12,437,012 Candidate committees, receipts, disbursements, back to 2014
State campaign finance 64 Direct filings and state ethics/disclosure boards, covering governor races FEC doesn't touch
Certified election results 2,048 Historical outcomes across multiple past cycles, used to validate the model against reality
Officeholder & candidate status Daily Verified against public nominee records
Economic indicators 7,650 State-level labor market and consumer data, refreshed on a rolling basis
Prediction-market pricing 68 Kalshi and Polymarket, used as a comparison layer only, never a model input

By design, deliberately excluded: public polling. Our model measures structural conditions, not repackaged poll averages.

What comes out -- model architecture, weighting, how signals combine -- stays ours.

Product roadmap

We're building beyond the core forecast feed:

Correlated-race data Race-to-race covariance from our underlying model factors, so correlation risk on parlays and baskets is priced correctly instead of assumed independent
Daily divergence feed A live, date-stamped feed of the largest gaps between our forecast and current market pricing, refreshed daily
Factor baskets Races grouped by shared exposure (economic anxiety, national-environment sensitivity), priced from our underlying model factors, not replicable from price data alone
Volatility index A running index of the most-contested races, updated daily
Candidate fundraising markets Head-to-head and league-table props on FEC filings, resolving each quarterly filing period

Talk to us about which to build first for your market.

Turn winning positions into more coverage

Some traders on your platform are already tracking the races that matter to them personally -- donors, activists, people with a real stake in the outcome, not just a market view.

DatumSignals gives that trader base something to trade on: verified weekly, every competitive race, a forecast built to be right rather than to move with sentiment.

1Win
2Redeploy
3New position

Consider a feature built on a simple idea: winning positions can be redeployed, not just cashed out. A trader who calls a race correctly can roll proceeds directly into their next position -- earlier, before a race becomes a national story and the market's already efficient.

We supply the data layer. Product design and compliance are yours to build with your own legal team.

Beyond 2026

The method underneath this -- district-level partisan history, incumbency, national environment, campaign finance, economic indicators -- isn't specific to US elections. It generalizes to any democracy with single-member districts and comparable public data. Not every market is a fit, and the accessible set is narrower than "any EU country": Canada is the clearest near-term candidate, and prediction markets are currently accessible in Germany, Spain, Denmark, and Greece.

That access is a gray-zone status, not a settled legal one -- worth saying plainly rather than implying otherwise. No dedicated EU framework for political prediction markets exists yet, and a MiCA compliance deadline in July 2026 could reclassify how these platforms operate in any of those four countries. "Accessible today" is not the same as "durably licensed," and the picture moves fast: several jurisdictions that are closed or hostile today only restricted access in the last several months. Gibraltar and Malta are both actively building real regulatory frameworks and are worth watching as the more durable path, even though neither has one in place yet.

That's a later phase, not something we're offering today. If international coverage is something you'd value, that's a reason to prioritize it sooner rather than later.

Built by people who've run campaigns

DatumSignals is built by David B. Wheeler and a team with decades of hands-on political campaign experience -- not a data-science exercise applied to elections from the outside. We've seen firsthand how campaigns actually spend, target, and win, and that operational reality shaped the model's design from the start.

Next step

We're ready to move on this now -- including exclusivity for the right partner, or a no-cost beta so you can see the feed against your own books before anything's signed.

Reach out directly: David Wheeler, david@parallaxadvisory.llc