If you want to get better at probability forecasting, you have five realistic free options in 2026: forecasting tournaments (Metaculus, Good Judgment Open), market-benchmarked games (our BearScout), paper trading on prediction markets, a plain spreadsheet, and public pundit-tracking as a spectator sport. They train different muscles. This page lays out what each actually does, how it scores you, and who it suits — including the honest weaknesses of our own tool.
| Option | Scored against | Question horizon | Feedback speed | Best for | Weakness |
|---|---|---|---|---|---|
| Metaculus | Community median + outcome (log/peer scores) | Weeks to decades | Slow — many questions resolve in years | Deep questions, strong community norms, AI-progress topics | Long feedback loops; scoring system takes study |
| Good Judgment Open | Outcome (Brier), vs crowd | Weeks to ~1 year | Medium | Geopolitics; the direct descendant of the ACE tournament research | Question volume lower than markets |
| BearScout (ours) | Market ask price at call time (Brier) | Days to weeks | Fast — real markets settle constantly | Calibration reps against a hard benchmark; fast loops | Young product; question pool = curated liquid markets only; no long-horizon questions |
| Paper trading (Kalshi demo, simulators) | Simulated fills + P&L | Any | Fast | Execution mechanics: order types, sizing, exits | Simulated fills flatter you in thin books; trains execution more than judgment |
| Plain spreadsheet | Whatever you record | Any | You decide | Total control; zero platform dependency | Discipline-dependent; no community; easy to quietly abandon |
Tournaments score you against outcomes and other forecasters. A market benchmark is stricter: the prevailing price is already an aggregate forecast from participants with money at stake, so matching it earns you nothing — only justified disagreement scores. This is the same reason forecast evaluation research treats market prices as a strong baseline. The trade-off is honest: market-benchmarked practice only exists where liquid markets exist, so you lose the long-horizon and low-liquidity questions tournaments are good at.
They stack: many serious forecasters run a tournament account for depth and a fast-loop tool for calibration reps. The method behind all of them is the same — write the number down first, score it later — covered in How to practice forecasting.
BearScout locks your forecast on live prediction-market questions and Brier-scores it against the ask price at call time. Free, pseudonymous, no real money.
Play BearScout →