The Polymarket leaderboard is full of genuinely profitable traders. On a per-position basis the typical top-100 trader earns between 1.4% and 2.1%, and at the volume these accounts run, that compounds into millions of dollars.
That thin margin is also what makes many of them poor traders to copy trade. Copying a position costs somewhere between 3.5% and 4.1% once you account for entering at a worse price than they did and paying trading fees on the way in. So following a top trader costs roughly two to three times the edge you'd be following them for. Successful copy trading on Polymarket has much less to do with picking a top trader than with picking the right one.
We looked at the Polymarket leaderboard filtered two ways, all-time profit and the last 30 days, and ran all 200 top-100 slots through Copy Score, our signal for whether a trader's record still holds up once you subtract what it costs to follow them. Here's what came back.
What We Measured
We pulled the top 100 from both Polymarket profit rankings on September 8, 2026, and asked the same question of every trader. On a trailing window of their resolved positions, after assumed slippage and fees, does anything survive? Copy Score is the filter we use to answer that question. A leaderboard only tells you how much money a trader made. This asks whether that record still looks worth following once you've paid to follow it, and a big earner can easily fail.
Everything below comes from one measurement on one date, held in a frozen snapshot so each figure can be re-derived. These are modeled results on a historical record, not realized returns from copying a leaderboard.
How we read a trader
Copy Score reads a trader's resolved positions over a trailing window and subtracts an assumed cost of copying them. Zero is the line. Above it, the record survives those costs. Below it, it doesn't. This report quotes the underlying percentage so the arithmetic is visible. Each score also carries one state.
Two of those labels describe our evidence rather than a trader's performance. We call a record thin when it holds few resolved positions, because a spectacular number over a handful of bets can just be one good week and we can't tell from the outside. We call it not comparable when the trader closes out early, since Copy Score assumes a copier mirrors the position and holds it to the market's result.
When this report says edge, it means the shrunk per-position figure the score is built from, before copy costs come off. Shrinking a short record toward the population is what stops one lucky market reading as a durable edge, and it matters a great deal here: the all-time median is 2.07% shrunk against 10.52% raw.
The trader types we flag
Separately from the score, we flag trading styles that are hard to copy trade, no matter how good the trader is. A flag is a heuristic read off the trading pattern. It tells you where a copier's money would tend to go instead.
An unflagged wallet isn't an all-clear. It only means we found nothing in the trading pattern that would stop a copy from working, which is weaker than establishing that the wallet is an ordinary directional trader. The full list is in the trader types we flag.
Two last notes. Every modeled score here rests on a trailing window of resolved positions, so it's not a forecast and not a trade-by-trade replay. And ranks are positions on the morning of September 8, on boards that reorder daily.
Copying Costs More Than the Edge Is Worth
When a leaderboard fails a copy screen, the instinct is to assume the traders aren't as good as they look. On this evidence, that's the wrong read. The real problem is that a copier needs margin left over after slippage and fees, and a leaderboard trader's margin is thin to begin with. A per-position edge of one or two percent runs into a per-position copy cost of three or four, and there's nothing left on the other side.
What the Typical Leaderboard Trader Earns, Against What Copying Them Costs
Copy costs exceed the median edge on both boards
Source: Polycopy data warehouse, frozen snapshot of September 8, 2026. Medians across the 82 measurable wallets on the 30-day board and the 60 on the all-time board. Edge is the shrunk per-position figure the score is built from, before copy costs. Cost combines trading fees with a slippage term averaged from our own bot fills. These are medians of separate distributions and do not subtract to the median net score.
On the 30-day leaderboard the median cost of copying is about three times the median edge. On the all-time one it's about 1.7 times. Either way the subtraction runs the wrong way for most of the board, which is why the median score is negative on both. That doesn't say these traders lose money, because most of them make it. It says the part of their return a copier could realistically capture is smaller than the cost of capturing it.
77% of Leaderboard Traders Come Out Negative After Copy Costs
One leaderboard failing a screen is a fact about that leaderboard. Two failing at the same rate is a fact about copy trading.
Of the 82 wallets we could measure on the 30-day leaderboard, 63 post a negative modeled score. Of the 60 on the all-time one, 46 do. That's 76.8% and 76.7%. The two boards share six wallets out of a hundred, so this is the same result arriving twice from almost entirely different people, selected on different criteria over different periods.
Modeled Copy Score After Assumed Copy Costs, Both Leaderboards
Two nearly separate populations, and the same shape of failure
Source: Polycopy data warehouse, frozen snapshot of September 8, 2026. The 82 wallets with a comparable Copy Score on the 30-day board and the 60 on the all-time board. Modeled scores on a trailing window of resolved positions, not a forecast and not a trade-by-trade replay. Wallets whose score is Not comparable are excluded, because their figure divides a real profit by a residual cost basis.
The shape differs in a way worth noting. The all-time leaderboard clusters just below the line, with 25 of its 60 wallets between -2% and 0%. The 30-day leaderboard fails harder, with 26 wallets between -5% and -2% and 15 more between -10% and -5%.
The 30-Day Leaderboard Has Fresher Traders and Worse Scores
Here's a result we didn't expect.
The obvious criticism of screening the all-time leaderboard is that nobody picks a trader to copy from a hall of fame. Two thirds of that board has gone quiet: 66 of the 97 wallets we hold trade data for placed no trades in the 30 days before we measured, and a dormant trader can't be copied because there's nothing to follow. Switch to the 30-day leaderboard and you remove every one of them by construction, so the screen ought to look considerably better.
But it actually looks worse. The median modeled score falls from -1.31% to -2.31%.
Two things move against a copier at once on the 30-day board. The median edge is lower, 1.39% against 2.07%, and the median cost of copying is higher, 4.12% against 3.51%. Composition explains the direction. The 30-day leaderboard carries 14 high-frequency wallets against the all-time board's 8, and frequency multiplies cost. A trader placing hundreds of orders a day pays the copy cost hundreds of times, and their per-position edge has to clear it every single time.
What the 30-day leaderboard does improve is how much of it we can read at all. It gives us 88 of 100 wallets we can score against 65 on the all-time board, only 12 of 88 carry a thin-record flag against 34 of the 65, and 32 of the all-time board's wallets carry no score whatsoever. So the 30-day leaderboard is the better list to screen. It just doesn't hold better traders to copy.
What Happens to Each Board's Top 100 at Every Test
The 30-day board survives the coverage tests far better and ends up in the same place
Source: Polycopy data warehouse, frozen snapshot of September 8, 2026. The steps in order are: on the board, we hold a record, enough recent trading to score, a number comparable to other traders, clears the modeled cost of copying, record not flagged thin, and carries none of the three exclusion flags we assign for trading a copier cannot mirror.
Leaderboard Rank Tells You Nothing About Copyability
If a leaderboard were useful for picking someone to copy, the traders worth copying would cluster near the top. But they don't.
On the 30-day leaderboard, the first wallet to clear the bar on a clean, deep record sits at #9. On the all-time leaderboard the first sits at #30, and nothing in its top 28 clears at all. Below those points the passing wallets scatter across the whole hundred, three of them inside the last ten ranks of the 30-day board.
Wallets That Clear the Modeled Cost of Copying, by Position on the Board
No concentration at the top of either board
Source: Polycopy data warehouse, frozen snapshot of September 8, 2026. A wallet counts here when its Copy Score state is Proven or Ahead, its record is not flagged thin, and it carries none of the three exclusion flags. The bars sum to the 14 wallets on the 30-day board and the 7 on the all-time board that meet all three conditions.
Rank is not just uninformative, it's unstable. Of the 14 wallets that cleared on the 30-day board, four had dropped out of both top 100s when we went back a day later to identify them.
Most Failures Aren't Market Makers, They're Ordinary Traders
The instinctive explanation for a leaderboard full of hard-to-copy traders is that it's full of market makers and other advanced trading styles a copier can't mirror. That explanation is true, but not to the degree you'd expect.
Eighteen of the 100 wallets on each board carry at least one flag. Set that against 63 and 46 negative modeled scores. Even if every flagged wallet were also a negative one, at least 45 of the 30-day board's failures and at least 28 of the all-time board's would carry no flag at all. The bulk of what this screen catches isn't machines. It's ordinary traders whose margin is thinner than the cost of following them.
Trading Styles We Flag Across Both Leaderboards
The categories overlap, and the mix shifts between the two boards
Source: Polycopy data warehouse, frozen snapshot of September 8, 2026. Flags are assigned from trading patterns. A wallet can carry more than one, so the bars sum to more than the 18 wallets on each board that carry at least one.
The mix does shift between boards, and it shifts in the direction that hurts a copier. The 30-day leaderboard carries nearly twice as many high-frequency wallets, 14 against 8, and half as many arbitrage wallets, 5 against 10. High frequency is the flag most closely tied to cost. And being hard to copy still isn't the same as being unskilled. A market maker running a genuinely profitable book can be a poor copy for reasons that have nothing to do with how good they are.
Every Leaderboard Trader That Passed Is Only Just Ahead
Copy Score has two positive states. Proven means the modeled score sits comfortably above assumed copy costs, and Ahead means it clears them but only just.
All 14 wallets that passed on the 30-day leaderboard are Ahead. Not one reaches Proven. Each had already passed the record-depth and flag tests, so thin data isn't what's holding them back. Margin is. Five of the 14 clear by less than 1.5%, and one clears by 0.1%, which a slightly worse fill would erase.
The Fourteen Wallets on the 30-Day Board That Cleared on a Clean, Deep Record
All fourteen are Ahead rather than Proven
| Trader | Board rank | Modeled score after copy costs | Resolved token positions in the measurement window | Trades in the last 30 days |
|---|---|---|---|---|
| Talvez10 | #9 | 5.2% | 553 | 2,648 |
| Left the top 100 | #76 | 4.7% | 331 | 1,443 |
| one8tyfive | #92 | 3.9% | 1,451 | 4,300 |
| GatheringData | #52 | 3.5% | 343 | 4,862 |
| Elenes | #53 | 3.1% | 96 | 352 |
| 0xd570...e4f8 | #10 | 2.2% | 53 | 2,882 |
| justaluckydude | #40 | 2.2% | 15,090 | 22,151 |
| Left the top 100 | #85 | 2.1% | 75 | 1,466 |
| Left the top 100 | #97 | 1.7% | 247 | 3,723 |
| TheOpportunist | #70 | 1.4% | 125 | 1,225 |
| GrizzliesSuck | #75 | 1.4% | 262 | 89 |
| Left the top 100 | #37 | 1.2% | 492 | 4,282 |
| ArturitoFilito | #91 | 0.6% | 561 | 1,999 |
| CyberScore.live | #69 | 0.1% | 538 | 5,227 |
Source: Polycopy data warehouse, frozen snapshot of September 8, 2026. Names are Polymarket's public leaderboard handles, resolved on September 9. Four wallets had left both top 100s by then and could not be identified. A snapshot of one day's board, not a list of traders to copy. Copy Score is an informational signal, not a recommendation. Ranks move daily. Modeled score is based on a trailing window of resolved positions, not a forecast, and is not a trade-by-trade replay.
Exactly one wallet out of the 200 slots carries a Proven rating. It wouldn't survive the other filters either. It has 13 resolved positions, and its own thin-history warning is attached to that Proven score. So across the 153 wallets we can score on the two boards, not one is both comfortably above assumed copy costs and free of a caveat.
Copying Well Takes More Than Picking a Winner
Every finding here points the same way. The median leaderboard trader earns less per position than it costs to follow them. Three in four come out negative once those costs come off, on two boards that share six traders out of a hundred. And where a trader sits on either board tells you nothing about which side of that line they land on.
What that adds up to is that reading names off a leaderboard isn't a copy trading strategy. A leaderboard ranks how much money a trader made, while copying is decided by how much margin survives after your own costs, and those are different questions with different answers. The costs themselves are specific and knowable. You enter after the trader does, so you pay a worse price than they got. You pay trading fees on the way in. If the trader runs at high frequency you pay both of those hundreds of times a month, which is exactly why the fresher board scored worse than the older one.
None of that is an argument against copy trading. Twenty-one of the 200 slots we screened did clear their costs, and this report names them. It's an argument that the selection has to happen on the variables a leaderboard doesn't show you, like how much of a trader's edge survives a realistic fill, how deep the record behind it is, whether the style can be mirrored at all, and whether they're still trading at all. Screen on those and the shortlist looks nothing like the top of the board.
How to Read the Polymarket Leaderboard for Copy Trading
- Compare the trader's edge to the cost of copying. A trader clearing 2% per position on their own book is doing well and is still a losing copy if following them costs 4%. This is the single thing a leaderboard cannot show you.
- Ignore leaderboard rank. The first wallet worth copying sat at #9 on one board and #30 on the other, and the passing wallets scattered to rank 97. Position on a profit leaderboard carries no information about copyability.
- Screen for copyability, not profit. Discover shows Copy Score on each trader as an informational signal, asking whether anything survived on their resolved positions after copy costs. It usually surfaces a different set of names than the leaderboard does. It isn't a shortlist or a recommendation to copy anyone. We go through the full approach in what traders to copy on Polymarket.
- Read the depth of the record, not just the score. A large return over a few dozen resolved positions can mean someone got very lucky once. On the all-time leaderboard, 34 of the 65 scoreable wallets are flagged thin.
- Use the 30-day leaderboard rather than the all-time one, and screen it anyway. Far more of it can be read at all, with 88 of 100 scoreable against 65 and a third the thin-record rate. Its traders still failed this screen at the same rate.
- Check the trader is still trading. Two thirds of the all-time leaderboard placed no trades in the month before we measured, and you can't copy a trader who has stopped.
Data and Methodology
Source: Polymarket's all-time profit leaderboard and 30-day profit leaderboard, top 100 by profit on each, both retrieved on September 8, 2026 in a single run, joined to Copy Score records in the Polycopy data warehouse. Reading both in one run is what allows the two to be compared, because the difference between the boards is then the ranking window rather than the measurement date. The inputs behind every figure are held in a frozen internal snapshot, keyed by board rank, so each number can be re-derived without re-reading the warehouse. Classifier registry version clf_v14.
Copy Score is a model, not a replay of copied trades. It reads resolved positions in a trailing 90-day window and shrinks a thin record toward a prior. It then subtracts an assumed cost of copying. You buy after the trader does, so the model charges a worse price, and you pay trading fees on the way in. The slippage term is a historical average taken from our own bot fills. What's left is the score. It describes a modeled reading of a past record, and it is not a recommendation, not realized copier P&L, and not a prediction about any single trade.
Where this report says "edge", it means the shrunk per-position figure the score is built from, before copy costs. That's the quantity the score subtracts costs from, and it's materially smaller than a trader's raw per-position return. Across the all-time leaderboard the median raw figure is 10.52% against a median shrunk figure of 2.07%, so quoting the raw number alongside a copy cost would overstate every trader's margin.
An unweighted median of wallet scores is not the return of a portfolio that copied the board. Wallets whose state is Not comparable are excluded from every modeled-score figure here, because their number divides a real profit by a residual cost basis and would read as an accusation we cannot support.
Two limits on coverage, both as they stood on the measurement date. The trailing window closes when a wallet's statistics were last recomputed, and we don't recompute a dormant wallet, so a score on an inactive wallet describes an older period than the measurement date. On the all-time leaderboard the median such lag was 19 days and 36 of the 60 measurable wallets exceeded a week, while on the 30-day leaderboard the median lag was zero. Separately, we don't hold a record for every wallet on either board. Eight of the 30-day board and three of the all-time board had no classifier record, and a further ten and three had no trade statistics. Those were recent arrivals our warehouse hadn't picked up yet, and they're reported as unknown rather than assumed to be anything.
Both of those limits have since been reduced. In the week after this measurement we found that a set of Polymarket contracts handling position exits had never been registered in our pipeline, which had been leaving wallets un-recomputed for longer than they should have been. On a re-measurement taken on September 11, 2026 the all-time median window lag was zero with 7 of 60 windows over a week, and the 30-day board carried 97 scoreable wallets against 88 here. The figures in this report are the frozen September 8 measurement and have not been restated. That re-measurement moved the median edge, cost and net figures by less than a quarter of a percentage point each, because the returns here are computed from fills and market resolutions rather than from the position ledger that was repaired.
Two questions this snapshot cannot answer. It can't tell you whether the wallets clearing today keep clearing, because the 30-day leaderboard is a momentum list by construction and nothing here measures what happens next. And it isn't a measurement of realized copier profit and loss, because no trades were copied. It's a filter that provides a signal.
Trader names in the clearing table are the public handles Polymarket displays on its own leaderboard, resolved on September 9, 2026, one day after the measurement. Four of the 14 had left both top 100s by then and are shown as such. No wallet addresses are published in full.
This report is for informational purposes only. It is a dated snapshot and analysis of publicly available Polymarket trader data and is not a forecast of future performance. Past performance is not indicative of future results. Copy Score is an informational signal, not financial advice and not a recommendation to trade or to copy anyone. Polycopy is an independent third-party tool, a verified member of the Polymarket Builders Program, and is not affiliated with or endorsed by Polymarket. All data sourced from the Polycopy data warehouse.