RocketHow RocketScience forecasts prediction markets
RocketScience watches Polymarket and Kalshi markets in finance, politics and crypto, researches each one on the open web, and publishes a live probability with a plain YES / NO / Not sure call. When the world changes suddenly, the forecast changes with it.
1. Live market data, not snapshots
We stream prices from the exchanges over live connections, so a move is seen within a second. Markets are routed to dedicated finance, politics and crypto analysts, which study the closest deadlines first.
2. Separating noise from news
Prediction-market prices jitter all the time. We smooth ordinary noise, so a one-tick blip does not flip a call. But a real shock — a move of 10 points within five minutes, or a statistically extreme jump — is treated as a regime change: the forecast follows the market immediately, older evidence is down-weighted, and a fresh news search starts at once. Think of a cricket match where one team is 89% to win, then a miracle over turns the game: the forecast should drop in seconds, not hours.
3. Reading the news
For every market we search current news coverage, read the articles, and judge whether each one makes YES more or less likely for that exact question. Recent articles count more than old ones.
4. Domain models
For crypto price markets we also run a spot-price model: the live Coinbase price, the strike, the time left and the asset's volatility give an independent probability that the price will finish above (or touch) the target.
5. One number, one call
These inputs are combined in log-odds into a single forecast. We call YES at 60% or more, NO at 40% or less, and Not sure in between — with a small buffer so calls don't flicker. When our forecast is at least 5 points away from the market, we flag an edge signal.
6. Graded in public
Every resolved market is graded against the call we made up to 24 hours before resolution. The results — good and bad — are on the track record page.