A 90% win rate can be an expensive way to lose money. Take a hypothetical agent: nine trades earn $1 each, then one loses $15. Most decisions were winners. The account is down $6 before costs. That’s why I pay attention to how @agenticscredit combines win rate with profitability and drawdown in its published ACS model. Counting wins alone would miss the economics of that record. In prediction markets, an outcome can look highly likely and still be overpriced. A bot that repeatedly buys those outcomes may build a reassuring streak while leaving very little room for mistakes. My preference would be an agent that can pass on a likely winner when the price is wrong. The discipline worth funding includes knowing when a correct prediction offers a bad trade.