Before last year, when someone told me “I’m pretty sure this will work,” I’d take it at face value. Now I don’t. Not because I’m cynical — because I spent a year watching thousands of people put real money behind “pretty sure” and most of them were wrong.

For the past year I’ve been tracking prediction market traders — grading them on ~40 metrics like Sharpe ratio, drawdown, and profit factor. The project is called 0xinsider. Prediction markets let you bet on real-world outcomes: elections, earnings, sports, crypto. If you think something has a 70% chance of happening and the market says 50%, you buy. If you’re right more often than you’re wrong, you make money. Simple in theory.

In practice, most people lose. And the way they lose taught me more about confidence and decision-making than anything I’ve read.

Here’s a sample of 47,342 traders, graded S through F:

Almost half land at D. Another 28% sit at C. S and A combined are under 10%. The distribution looks like most grading curves — except these grades are backed by real money, not assignments.

Here’s what each grade actually means in dollars. The left column is median P&L — the midpoint trader in that grade. The right bar shows what percentage of traders in that grade have lost money overall (negative lifetime profit):

The median S-grade trader has made $1.17M, and literally zero of them are in the red. The median D-grade trader — the biggest group — is down $23. That sounds harmless until you see that 78.6% of them have negative lifetime profit. They’re losing, just slowly. F-grade is worse: median loss of $42K, and 97.3% of them have lost money.

Here’s what surprised me. F-grade traders aren’t small. Their median trading volume is $2.2M. Seventy percent of them have pushed over $1M through prediction markets. These are people betting millions and still losing. It’s not a bankroll problem. It’s a judgment problem.

S-grade traders trade an average of 6,736 markets. F-grade traders trade 823. The best traders don’t just bet better — they bet more often, across more markets, with smaller individual positions. They’re diversified. The worst traders concentrate into fewer, bigger bets and get crushed.

The overall median P&L across all traders is −$0.90. Basically zero. A few S-grade outliers pull the average up to +$32K, but that number is meaningless for the typical trader. Most people are either treading water or slowly losing.

The data doesn’t show that people are bad at predicting things. It shows that people are bad at knowing how good they are at predicting things. The F-grade traders aren’t making small, cautious bets. They’re going big. They have conviction. They’re just wrong.

The traders who actually win don’t look confident. They look boring. Lots of small positions across thousands of markets. No dramatic all-in bets. When they’re wrong — and they are, often — it doesn’t matter because no single position can hurt them.

I catch myself thinking this way about everything now. Someone says “this project will take two weeks” and I wonder — is that a high-conviction bet on a few data points, or a calibrated estimate from deep experience? The prediction market data didn’t teach me to think in probabilities. It taught me that confidence is almost completely uncorrelated with accuracy, and that the people who do the best are the ones who already know that.