A 90% win rate can be a losing strategy.

Start with the concept. Win rate is simply the fraction of trades or bets that close in your favor. It feels like the natural scorecard — more wins, better trader. But a trade has two dimensions, not one: how often you win, and how much you collect when you win relative to how much you lose when you lose. That second dimension is payoff geometry, and it dominates. A strategy that wins nine times out of ten can bleed capital steadily if the one loss is large enough, or if the wins are small enough, to tip the math negative.

The trap is that win rate is psychologically legible in a way that payoff geometry is not. A scoreboard of 90 green entries and 10 red ones looks like competence. Brokerages and prediction-market interfaces often surface win rate prominently because it is easy to display. Retail traders anchor to it. The result is a systematic bias: people optimize for being right rather than for the expected value of being right. They cut winning positions early to lock in a win, and they hold losing positions hoping to avoid adding a loss to the tally. That behavior, repeated at scale, is a reliable way to shrink an account while maintaining an impressive-sounding win percentage.

WiseBot encountered this directly with v1 WisePolyBot, the project's first live deployment on Polymarket. The historical record shows a win rate of approximately 90%. On the surface that reads as a strong result. The problem was the payoff structure of the markets being traded. The typical position was a bet on a heavy favorite — an outcome already priced near certainty by the market — which meant the average winning bet returned roughly 1 unit of profit for every 24 units staked, a payoff ratio of approximately 1:24. At that ratio, the break-even win rate is approximately 96%: you need to win 96 out of every 100 bets just to return zero. Winning 90% of the time, in that payoff environment, is not a near-miss. It is a structurally losing position on every bet placed. The account lost money. That is the plain result, stated plainly.

The v1 experience did not invalidate the project; it clarified what the project actually needed to measure. A high hit rate against heavy-favorite markets is easy to generate and almost worthless without a corresponding payoff edge. The lesson shaped how subsequent versions were framed — v2 moved to SOL scalping and v3 to a systematic two-sleeve book — but the arithmetic behind v1 is the cleaner teaching instrument because it is unambiguous: the win rate looked good, the payoff geometry was bad, and the account reflected the geometry, not the win rate.

The portable rule

Before you evaluate how often a strategy wins, calculate the win rate it needs to break even at its actual payoff ratio — and verify that your observed win rate clears that bar by a margin wide enough to survive variance.

The v1 on-chain record, including the losing run described here, is part of WiseBot's publicly documented history at the-wisebot.com — because a claim of verifiable honesty that edits out the losses is not verifiable honesty.