How it works, and why it should be more accurate

A forecast company that puts three ways of reading America side by side: what people bet on prediction markets, what pollsters measure, and, from December 2026, what our own verified, paid members tell us. We weigh them in the open, publish one number per race, and grade ourselves in public.

One win chance and one set of odds for each race. It blends Kalshi, Polymarket, PredictIt and the public poll average through the Precision Funnel™: no source has a fixed share. Each market is weighted by the dollars traded on it (thin markets are left out) and the polls by how many recent, solid polls exist, and each race page shows the weights it got and why. See it in the Race Center and the exact formula on the methodology page.

Market prices come straight from the public data Kalshi, Polymarket and PredictIt publish, and refresh about every 30 minutes. Polls come from VoteHub under the CC BY 4.0 license. PredictIt data is shown for non-commercial purposes only. Every source is credited and linked on each race page, with the time it was read.

No. The probability on every race is computed by our published formula from the stored market and poll data. The short written note on a race is drafted by an AI model and checked against our stored number before it appears. The model never sets or changes the probability.

Because each source fails differently. Markets react within minutes but can be pushed around when trading is thin. Polls ask voters directly but lag the news and can share the same blind spots. When independent forecasts make different mistakes, averaging them cancels part of each error. Decades of research back this up; we summarize it on the research page.

Traders have money on the line, so they have a reason to be right rather than to sound right, and prices move the moment news breaks. In studies going back to the 1980s, election markets have often beaten polls, though not always.

Polls are the only source that asks voters directly. They anchor the blend when markets drift on enthusiasm or a few large bets.

Public polls are slow, expensive and scarce in most races. A large verified panel lets us ask the questions that matter, in every state, quickly. It is the third view in the blend, and it joins only after its weight is published and it has a track record.

It can, which is exactly the problem we built around. Open online panels that pay for finishing attract bots and careless answers. In one opt-in poll checked by Pew Research Center, 12% of adults under 30 claimed to be licensed to operate a nuclear submarine. Paying people does not make them honest. Our answer is three layers together: every member is verified as one real adult, they are paid for their time, and a scoring rule rewards honest answers so invented ones lose out.

A bonus for an honest answer and a good read of the room. After you answer, you also guess how the rest of the panel will split. When the poll closes, two things are scored: whether your answer was more common than members expected, and how close your guess was. Answers that beat expectations rank higher; a guess that is far off pulls you down, but a perfect guess never adds on its own. The top of the poll earns 2× the usual tokens and the bottom keeps the usual award. Full details are on the methodology page.

Because it lets us reward honesty without knowing the “right” answer. People who answer honestly tend to underestimate how many others share their view. When the results come in, honest answers beat the panel’s prediction more often than invented ones do, and the bonus follows that gap.

No, and that is the point. A majority answer that everyone saw coming doesn’t score well. A minority answer can earn the bonus if more members held it than the panel predicted. Paying for the popular answer would push people to agree; paying for the surprisingly common answer does not.

Your own view is evidence about other people, so a view that is honestly held usually turns out more common than the group predicted. Someone inventing an answer has no such private signal, so made-up or socially safe answers tend not to beat the prediction.

No. The rule is Bayesian Truth Serum, published by Drazen Prelec of MIT in Science in 2004. The Candor Score is our implementation of it: how we apply it to our panel, calibrate it and turn the score into a token award.

In a lab study (Weaver and Prelec, 2013), recognition questions were salted with made-up items. People paid by the rule claimed to recognize 13% of the fakes, against 24% for controls. When also paid to exaggerate, they claimed 16%, against 55% for people paid to exaggerate without the rule. That was a lab result on quiz items, not an election poll. We will publish our own results once the panel is in the field.

It is built to be hard. The guess is scored with a proper scoring rule, so the best strategy is to report what you really expect. Your own response is left out of the shares used to score you. Only responses that pass attention checks are scored, polls too small to score fairly are skipped, and identity checks stop one person collecting twice.

No. It rewards an honest answer over an invented one. Quotas and weighting make the panel look like the country, and those weights are published with each result. Whether a forecast was right is a separate question, graded in public on the scorecard.

The cash fee for finishing is the same for everyone, whatever your score. Tokens range from the poll’s usual award (the floor) to twice that award (the cap). The cap is 2× the poll award, not a price; tokens have no cash value.

No. Your cash payment for a completed poll is the same either way and is never scored. The truth bonus only adds tokens on top, and no completed response earns less than the usual award.

After launch, verified members are paid for each completed poll. Founding members also earn Precision Tokens for verifying their account, taking the Founding Poll, inviting friends who join and verify, and through the truth bonus. Amounts are subject to final terms.

Nothing today, and we make no promise they ever will be. During pre-registration they are locked: they cannot be sold, transferred or redeemed. Any future conversion would happen only through an offering qualified under U.S. securities law, which may never happen. See the Terms.

Reserve a founding spot on the Join page now. Pre-registration opens in early November 2026 and the panel launches in December 2026.

Never. We do not sell, rent or trade your name, email or phone number, and mobile opt-in data is never shared for anyone else’s marketing. Poll answers are published only in aggregate. See the Privacy Policy.

We save every forecast every day and grade the last one before polls open with the Brier score, beside each source on its own. If a single market beat our blend, the scorecard will say so.

No. Our first member polls come with the panel launch in December 2026. Until then, no number on this site is presented as a Precision poll.

We fix it on the page and log it in our corrections log with the date.

An independent company with research operations in Cambridge, Massachusetts. We are not owned or funded by any campaign, party committee or media organization. Our founding team will be named at launch. More on the About page.