A weather bot with a high win rate still needs a loss column

A bot buys cheap weather contracts, exits after a forecast update, and shows a string of green results. The obvious question is how often it wins. The better question is what happens when it does not.

Count dollars, not celebrations

In a hypothetical run, nine gains of $2 produce $18. One loss of $25 leaves the run down $7. A 90% win rate can coexist with a losing strategy.

Ask for position size, average gain, average loss, fees, and open positions. A screenshot ending before a losing contract settles can make the strategy look cleaner than it is.

Check the forecast history

A fair replay uses the forecast available at the decision time. Final observations and later revisions must not leak into earlier decisions. Otherwise the bot is being graded with tomorrow's answer.

Keep the raw forecast file, retrieval time, bucket boundaries, and executable quote together. That bundle lets another person reconstruct the decision.

Split the sample by city and season. An apparent edge in one station's summer data may not transfer to a different climate.

The point is not to dismiss automation. It is to ask for a scorecard that includes the unpleasant days. A bot earns confidence through a complete record, not a beautiful winning streak.

Further reading: Prices and order books · Resolution rules · Current fees.

Explore the markets: Open Polymarket (referral link).