WeatherBot review
Our verdict
The strongest disclosure record in the category, and the reason it tops our index. WeatherBot publishes the numbers most competitors decline to, prices fees in the user's favour, and documents its probability and exit models in enough detail to actually be evaluated. The headline 24-hour margin claim is the one place we would push back — it is self-reported and unaudited. Everything else here is well-evidenced by the standards of this market.
How it works
WeatherBot discovers weather markets through Polymarket's Gamma API, pulls forecasts from GFS, ECMWF, UKMO, NWS and HKO, and blends them against 10-year NOAA NCEI climatology to form a Bayesian prior. Edge is computed with a Normal CDF approximation using a dynamic sigma that widens from 0.8° at six hours out to 5.5° beyond ten days. Position sizing is quarter-Kelly, hard-capped at 15% of bankroll. Claude and Gemini review each candidate trade with the full forecast set, model-agreement score, base rates and live prices before anything fires. Exits run on five layers: 60% of edge captured, edge convergence under 2%, a 40% trailing stop, a 15% stop loss, and a forced exit within two hours of resolution.
Strengths and weaknesses
What works
- The most complete public stat block of any bot we reviewed — win rate, P&L, volume, user count and per-account average are all published.
- Fee model is unusually fair: the $0.03 option charges only on winning trades, so a losing streak costs you nothing in platform fees.
- Non-custodial with keys held client-side; the operator cannot move funds.
- Exit logic is fully specified rather than hand-waved, which is rare in this category.
- The Hong Kong resolution-source lock (reading the live HKO gauge on settlement day) is a genuine structural edge, not a marketing line.
What doesn't
- The claim of 18–48% profit margins within 24 hours comes from internal testing and is not independently audited — treat it as the aggressive end of the range, not an expectation.
- Weather markets only. If you want crypto, politics or sports coverage this is not your tool.
- A 71–80% win-rate band is wide; the site does not publish the sample size behind it or a drawdown curve.
- Depends on third-party forecast APIs, so an upstream outage is an execution risk that is not discussed.