For every candidate we publish one number: if they win their party’s nomination, how likely are they to win the general election? It is read directly from live prediction markets — not a poll, not an opinion.
Electability is a conditional probability, P(win the general | nominated). No market prices it directly, but two markets that already exist pin it down. Winning the presidency requires winning the nomination first, so a market’s “wins the presidency” price is the probability of winning both. Divide by the nomination price and the nomination cancels:
Both prices come from two independent, real-money venues — Polymarket and Kalshi — each listing per-candidate nomination and presidency markets. We read the live order books, take a fair mid-price, and blend the venues by liquidity (near-the-money depth and 24-hour volume): the market where the real money on a candidate trades counts more. Nothing is hand-entered.
A poll taken years out mostly measures name recognition, and its respondents have nothing at stake. A market is thousands of forecasters with money on the line, repricing on every debate, endorsement, and filing in real time. Being wrong costs them — which is why, across decades of research, prediction markets rank among the best-calibrated forecasters available, regularly matching or beating expert panels and poll aggregators. Every source of signal, from polls to private information, ends up priced in and weighted by conviction.
A poll asks people what they think. A market asks them to prove it.
Each estimate carries a range. It widens when the two venues disagree or the books are thin, and tightens when they agree and run deep. The dot is the estimate; the range is how much to trust it. For a longshot — priced on a rare event by few traders — we shrink toward the field average and widen the range rather than manufacture false precision.
We log every estimate daily and score it against outcomes as races resolve, using the standard Brier score. A forecast you cannot check is only an opinion; ours is built to be graded.
Conditional, not causal. “If nominated” selects the futures in which a candidate already broke through — usually their favorable ones — so a longshot’s electability reads high. We correct for it but cannot fully remove it. And markets, while hard to beat on average, are not oracles: every price is a moving snapshot.
A small, independent team of researchers and data scientists. (More about us soon.) We build the model; the markets set the prices.