Your most frequent questions, covered.

Four institutional sources, all live: NOAA's GFS with a 16-day horizon and the European Centre's ECMWF at 10 days, both through Open-Meteo; the UK Met Office's UKMO at 7 days, again through Open-Meteo; and hourly NWS data pulled directly from api.weather.gov for US locations. On top of that, NOAA NCEI's GHCND station archive provides a 10-year historical prior that the Bayesian step blends into the live consensus.

At the heart of the probability core is an Abramowitz-Stegun approximation of the Normal CDF. The four model temperatures are first weighted into one consensus (ECMWF 35%, GFS 25%, UKMO 20%, NWS 20%), and sigma is then set according to how far off resolution is (0.8° inside six hours, rising to 5.5° past ten days). Every outcome type — exact value, range, "or higher", "or lower" — gets its own CDF integration. The edge is just our probability minus the market price, and the Bayesian fold-in narrows sigma whenever the models agree with the historical record.

Sizing starts from the Kelly criterion (f* = edge ÷ implied odds) and is then cut to a quarter of that value for safety, with a hard cap of 15% of bankroll per trade. The order that actually goes out is the lower of (bankroll × kellyFraction) and the maxPerTrade value in your config. Out-of-the-box defaults: $500 bankroll, $50 maxPerTrade.

All settlement happens on Polygon (chain ID 137) via Polymarket's CLOB at clob.polymarket.com. pUSD is the trading asset post-CLOB-V2 (Polymarket auto-wraps USDC.e at deposit), and on the first wallet pair-up the engine signs approvals for the Exchange (0x4bFb41d5), NegRiskExchange (0xC5d563A3) and NegRiskAdapter (0xd91E80cF) contracts using ethers.js together with viem.

Five exit checks run side by side, and whichever triggers first closes the position: (1) Profit target — at least 60% of the original edge captured, (2) Convergence — the market has moved within 2% of our probability, (3) Trailing stop — 40% or more surrendered from peak unrealized P&L, (4) Drawdown — the position is 15% below entry, (5) Time decay — less than two hours to resolution while in any profit. The exit is routed directly through CLOB.

It is nearly always one of these five things — go through the list before deciding the engine is broken:

1. Pair the wallet first — without a paired Polygon wallet no order can be routed. Open Settings and check that the connection went through.

2. Give the book some time — weather books don't behave like crypto books. Twenty quiet minutes between trades is normal; when conditions line up you might see ten fire within two minutes.

3. Verify the proxy (funder) address — a proxy with a typo fails silently. If even one character is wrong, on-chain execution will never settle.

4. Sanity-check the risk config — open Settings and check that bankroll, maxPerTrade and minEdge are the values you intended. While there, confirm the proxy holds enough pUSD to cover the order sizes you've set.

5. Top up POL for gas — the step people skip most often. Gas is paid in Polygon's native token; a single dollar of POL in the wallet covers plenty of fills.

More or less free. There's no subscription and nothing to pay up front — just a small per-trade fee, and you choose how it's charged: $0.015 on every executed order whatever the result, or $0.03 only on winners. Choose the one that suits you.

No — and any tool claiming otherwise is selling you something risky. The engine is under active development, with new releases landing nearly every day, and our internal benchmarks usually come in between 18% and 48% over a 24-hour window. Still, don't plan around instant results: weather markets settle on weather time, not crypto time, and luck pushes outcomes around the average. Only stake money you can afford to lose.

Yes — close anything you opened manually before you start. The exit engine manages only the positions it opened itself, since it records every fill the moment it happens: exact entry price, size, side (BUY_YES or BUY_NO), market ID and timestamp. That record is what makes precise P&L and clean exits possible.

Positions opened by hand (or by another tool) have no such record. The engine would need to reconstruct an entry price from historical data, which gives inaccurate P&L and can trigger exits at the wrong time.

Since nothing carries over between sessions, this is what happens in practice:

Engine-opened positions — fully managed by the exit monitor, with the true entry recorded and proper automatic exits.

Hand-opened positions — listed in the Open Positions panel so you can see them, but the engine leaves them alone.

To keep things clear, close any existing weather positions before you press Start.

If you're hoping for overnight riches, this isn't the tool for you — it's worth being honest about that. Weather books just don't move like crypto books. Most positions have to wait 24 to 48 hours for the actual weather to play out and the market to resolve.

The gains are genuine, but they build up gradually. The engine is designed for repeatable, edge-driven returns — far more like a steadily compounding book than a lottery ticket.

Whether it makes you rich depends on your bankroll, your settings, how long you run it and what the market offers. What we can say is that with sensible expectations and the patience to let it work, the results tend to speak for themselves over time.

New builds go out almost daily — small fixes, fresh strategies, better instrumentation. None of it is "set and forget."

Right now the focus is expanding the strategy library and benchmarking the next generation of frontier reasoning models — GPT-5, Gemini 3.1 and DeepSeek R2 — so traders can pick the engine that fits their priorities: more analytical depth, quicker turnaround, or a more conservative stance.

There's a lot in the pipeline and it's moving quickly.

If you've already discovered MeteonBot, you're early. We haven't done any marketing yet — you're ahead of everyone else.