Is the Polymarket leaderboard accurate?
The numbers are not fabricated. They are just built from a record where wins reliably appear and losses frequently do not, and the gap is far bigger than most people assume.
The short answer
The leaderboard is accurate about what it measures, and what it measures is not what most people read it as. It counts positions that produced a settlement event. On Polymarket a large share of losing positions never produce one, so they are missing from the total.
This is not Polymarket hiding anything. It is a consequence of how redemption works, and it affects every tool in the ecosystem that reads the same data, including every third-party tracker.
Why losses go missing
A winning position pays one dollar per share, so the holder redeems it, and that redemption is written to the blockchain.
A losing position pays nothing. Redeeming it still costs gas. So essentially nobody does it, the worthless tokens sit in the wallet indefinitely, and no settlement event is ever written.
Any figure assembled from settlements therefore captures wins far more reliably than losses.
What that looks like on real wallets
On 11 August 2026 we rebuilt the five wallets then ranked highest by weekly profit. Every address is public and anyone can repeat this.
| Wallet (rank on 11 Aug) | Profile shows | Actual record | Losses with no on-chain record |
|---|---|---|---|
| quavoo (#1) | 2182–96 95.8% | 2221–2812 44.1% | 2,723 |
| ferrariChampions2026 | 3583–1812 66.4% | 4022–4060 49.8% | 2,248 |
| Djdjdjekekek | 151–15 91.0% | 151–160 48.6% | 151 |
| HOG993 | 570–160 78.1% | 572–299 65.7% | 182 |
| pleaseplease123 | 200–1 99.5% | 200–222 47.4% | 221 |
Four of the five are within three points of a coin flip. Their profiles read anywhere from 66% to 99.5%. The most extreme case is the fifth: a profile showing two hundred wins against a single loss, and a real record of 200 and 222.
The fourth wallet is the one worth pausing on. HOG993 comes out at 65.7%, genuinely above a coin flip, and the correction only moves him twelve points instead of fifty. Whatever this method is doing, it is not simply deflating everybody.
None of these traders is a fraud and none of these numbers is fabricated. They are genuinely up a great deal of money. For most of them the edge is position sizing rather than accuracy, which is a completely different story about the same person, and a much harder one to copy.
The wider sample, and the part that actually matters
We have now rebuilt twelve wallets. The real win rates spread from 44% to 74%, so "they are all coin flips" is too neat a story and we are not going to tell it.
What has not varied once is the direction. All twelve profiles read higher than the reconstructed record. Not eleven, not mostly. Twelve out of twelve. The gap runs from 12 points at the narrow end to 52 at the widest, and one profile shows 33 wins against no losses at all for a trader who is really 33 and 23.
That is the claim worth taking away. Not that every top trader is secretly average, but that a published win rate is an upper bound on every wallet we have checked, and you have no way of knowing from the outside whether a given one is off by twelve points or fifty.
How to check it yourself
- Walk the full activity history for the wallet. Page backwards using a time cursor rather than an offset, because offset paging stops returning rows past roughly five thousand and looks exactly like reaching the end of history. Pin the start of the range too, or the lookback quietly caps at about three years.
- Handle splits and merges. They arrive with an outcome index of 999, meaning every outcome at once, because splitting collateral mints one unit of each side for a dollar. Skip them and later sales of one leg look like selling something that was never bought.
- Replay the history into a position per outcome, so you know what the wallet was holding when each market closed.
- Ask each market how it actually resolved and settle the leftovers against that. A position carried through resolution is worth a dollar a unit if it won and nothing if it lost, whether or not anybody ever claimed it.
The step most people get wrong
It is tempting to treat "still holding it, never redeemed it" as proof the position lost. It usually did. But that test cannot tell a worthless position apart from one whose market has not closed yet, and it books an unclaimed winning position as a loss.
Look the resolution up instead of inferring it. It is one extra request per market and it is the difference between a number you can defend and a number that is probably right.
What this means if you are copying a trader
A published win rate on any of these profiles is a lower bound on how often someone is wrong, not a measure of how often they are right. Before you follow anyone, rebuild them. You will sometimes find the opposite of what the board suggested, and you will often find that the edge is sizing rather than prediction, which is a much harder thing to copy.
Common questions
- Is the Polymarket leaderboard fake?
- No. The figures are computed from real on-chain data. The issue is completeness rather than honesty: it counts positions that settled, and losing positions on Polymarket frequently never settle because redeeming a worthless position costs gas and pays nothing.
- How wrong can a leaderboard number be?
- On the wallet ranked first by weekly profit on 11 August 2026, 2,723 losses had no settlement event. Read from settlements alone that trader appears to be 2182 and 96, a 95.8% win rate. The reconstructed record is 2221 and 2812, a 44.1% win rate. The two traders ranked below him came out at 49.8% and 48.6%.
- Does this affect profit figures as well as win rates?
- It affects any figure assembled from settled positions. A trader still holding worthless unredeemed tokens has capital tied up in positions that already resolved against them, so both the win rate and the realised result read better than reality.
- Do other Polymarket trackers have this problem?
- Any tool reading settlement events has it, which is most of them. The only way around it is to diff a rebuilt history against the wallet current holdings and book the difference at zero.