Polymarket 'No' bot keeps losing
A bot that systematically placed only 'No' bets on Polymarket continued to lose money even though 73% of the platform’s contracts resolved 'No', highlighting that high base rates don't automatically translate to profitable trading. The report called the result a microstructure example of pricing and expectancy dynamics on prediction platforms. (protos.com)
A Polymarket bot built to buy only “No” shares is still losing money, even though about 73% of resolved contracts on the platform end that way. (protos.com) Protos reported that artist and former Apple researcher Sterling Crispin open-sourced the bot, called “Nothing Ever Happens,” on April 12, 2026. A screenshot attached to the post showed a portfolio of about $2,859, and the article said the code repository had drawn more than 400 stars by April 14. (protos.com) Polymarket’s own accuracy page says 73.3% of resolved markets finished “No” and 26.7% finished “Yes.” The same page says the platform’s 4-hour pre-resolution accuracy was 96.7% and its 1-month accuracy was 90.4%. (polymarket.com) Prediction-market shares are priced like probabilities: a share bought at $0.65 pays $1 if it wins and $0 if it loses. Polymarket’s fee schedule says taker fees are calculated as Θ × C × p × (1 − p), which means costs change with price and trade size instead of staying flat. (docs.polymarket.us) That math is why a high “No” hit rate does not automatically produce profit. Protos gave the example of a “No” priced at $0.75, which returns only $0.25 on a win while still risking a full loss when the 26.7% minority outcome happens. (protos.com) Crispin told readers the strategy needs lower entry prices than the platform-wide 73% figure suggests. Protos said he wrote that the bot “has to buy below $0.73 long term,” uses a configurable ceiling of $0.65, and checks for markets closer to $0.50. (protos.com) Polymarket’s own data points in the same direction. Its accuracy page says expected and actual outcomes “closely align” across probability ranges, which means obvious base-rate trades are often already reflected in market prices before a bot arrives. (polymarket.com) The fee schedule adds another drag. On Polymarket US, the current exchange-wide schedule took effect at 3 p.m. Eastern Time on April 3, 2026, and says makers receive rebates while takers pay fees, so a bot that crosses the spread to buy available “No” shares can start each trade at a disadvantage. (docs.polymarket.us) Protos framed the result as a lesson in market structure rather than a verdict on whether most events happen. The bot keeps running, but the early returns show that being right more often is not the same as being paid enough when you are right. (protos.com)
