How to read a Polymarket trader’s track record
A win rate means little until you set it against the prices the trader paid, the number of settled bets behind it and the category it comes from. For example, shares bought at 70 cents win about 70% of the time on the odds alone, so a 73% win rate on them is only 3 points better than the odds.
Why are trader records so easy to misread?
Prediction-market records are easy to see and easy to misread. Polymarket trades settle on a public blockchain, so leaderboards, screenshots of big wins and accounts billed as “smart money” travel fast. This guide is for anyone who looks at one of those records, including journalists who want to quote one, and wants to know what it really shows. It uses figures from our own measurement of Polymarket esports bets, and arithmetic you can check with a calculator.
Why is a win rate not enough?
Every share pays $1 if its outcome happens and nothing if it does not. So the price paid is the chance the market gave at the time: a share bought at 40 cents was a 40% chance.
That makes a win rate a test the price has already taken. Buy shares at 90 cents and the odds alone say you will win about 90% of the time. A 90% win rate there shows the prices were right, not that the trader saw something the market missed.
The useful number is the gap. Average the prices a trader paid across their settled bets, then take that average away from the share of bets they won. We call this beating the odds, and we measure it in percentage points. A trader who won 55% of bets bought at an average of 40 cents beat the odds by 15 points.
This matters in practice. In our research note on esports records, measured on 26 September 2026, the tenth of Polymarket wallets with the highest win rates after their first 20 esports bets had paid 82 cents a share on average. At 82 cents, the odds alone say a bet comes in about 82% of the time. A high win rate in that group is mostly the price talking.
How many settled bets does a record need?
More than most people think, and the number depends on the category. Our research note of 26 September 2026 sets out how many settled bets a Polymarket record needs before it says more about the trader than about luck: weather 100, stocks 98, esports 88, macro 50, and 20 for sports, politics, world, crypto and other.
Twenty is also the floor: no category goes below it. These are the figures as of that date. Esports became a category of its own that same day, and we will measure its number again as more esports bets settle.
The idea behind these numbers is called shrinkage, and it is simpler than it sounds. Treat a record as a blend of two things: the trader’s own results, and the average of everyone betting in that category. After n settled bets, the trader’s own results get n / (n + k) of the weight and the category average gets the rest.
k is measured, not chosen: it compares how much one trader’s results swing from bet to bet with how much traders really differ from each other. For esports it came out at 87.04. So after 20 esports bets, a record carries less than a fifth of the weight (20 / 107.04, about 19%). After 88, it carries about half. The noisier the category, the more bets a record needs before it speaks for itself.
Is a perfect start a sign of skill?
Not on its own. In the same esports measurement (26 September 2026), 27 of 914 wallets won every one of their first 20 esports bets on Polymarket, about 3 in every 100. A perfect start is common enough that it cannot be the test.
Arithmetic shows why. If every price is fair and a trader has no edge at all, the chance of winning 20 bets out of 20 depends almost entirely on the price paid: about 1 in a million at 50 cents a share, about 2 in 100 at 82 cents, and about 12 in 100 at 90 cents.
A streak bought at favorite prices is something luck produces all the time.
Why do bets at 97 cents or 2 cents say so little?
Extreme prices squeeze the information out of a record. The examples below are pure arithmetic on made-up records, and say nothing about any real trader.
At 97 cents, a share wins about 97% of the time by construction. Take a record of 100 such bets with 98 wins. It beats the odds by 1 point, and one more loss would erase that. The most it could ever beat the odds by is 3 points, because nobody wins more than 100% of the time. If the prices are fair, a trader with no edge at all wins 98 or more of 100 such bets about 4 times in 10. The payoff is lopsided too: each win earns 3 cents a share and each loss costs 97 cents, so one loss cancels about 32 wins.
At 2 cents, it runs the other way. The odds say about 2 wins in 100. A record with 4 wins has a 4% win rate, which looks terrible, yet it beat the odds by 2 points and won twice as often as the prices implied. The whole difference is two bets. A trader with no edge reaches 4 or more wins in 100 such bets about 1 time in 7, and wins none at all about 1 time in 8.
In both cases the win rate barely moves with skill and swings with luck. A record built mostly on prices near 0 or near 100 needs far more bets before it tells you anything.
How do trading costs change a record?
The price on the screen is not what a trade costs. Three things sit on top of it: the spread, slippage and fees.
The spread: the gap between the best price to buy (the ask) and the best price to sell (the bid).
Slippage: a large order uses up the best price and moves on to worse ones. If 100 shares are offered at 62 cents and 200 more at 64 cents, buying all 300 costs about 63.3 cents a share on average.
Fees: each venue sets its own, and some markets charge none.
Your all-in price per share is the win rate you need to break even. A winning share pays $1, so paying 63.3 cents means winning more than 63.3% of the time over many bets just to come out level.
That changes how a record reads. A hypothetical example: a record shows 60 wins in 100 bets at an average screen price of 58 cents. On screen prices it beats the odds by 2 points. If the spread, slippage and fees added 3 cents a share, the all-in price was 61 cents, and the same record sits 1 point below break-even. So ask whether a record uses the prices actually paid, or quoted ones.
Why read a record one category at a time?
A record pooled across sports, politics and crypto can hide that a trader is average in the category you care about. A made-up example: 300 sports bets that beat the odds by 4 points and 100 politics bets that beat them by 0 pool into 400 bets that beat the odds by 3 points. Someone reading about politics sees “3 points over 400 bets”. The politics record alone is 0 points, from 100 bets.
Categories also differ in how many bets they need, as the figures above show, and in the prices traders typically pay. Read the per-category record, check its own count of settled bets, and judge it against that category’s bar.
What should you ask before you trust a trader’s record?
1. How many settled bets does it hold in the category you care about? Open positions are not results.
2. What was the average price paid, and how far did the win rate beat it?
3. Are those the prices actually paid, after spread, slippage and fees?
4. Is it the full history, or a chosen slice: a screenshot, a good month, the one account out of many that did well?
5. Does it still look good without its single best bet?
6. Is the trader taking sides, or making markets on both sides of the same question? A market maker’s win rate says nothing about judgment.
7. What happened after the record was picked out? Results counted only after a trader was chosen are the fairest test of whether a record holds up.
How does Kahyna apply this?
We hold ourselves to the same questions. Kahyna publishes its own track record, losses included, with every win rate shown beside the number of settled calls behind it. When we look at Polymarket wallet records, we judge them against the odds paid, require a measured number of settled bets in each category, and publish every threshold on our methodology page.
The esports figures in this guide come from our research note on esports records, measured on 26 September 2026.