We mapped 25 Polymarket categories. Clean wallets keep 0.45% of what they trade.
Every copy-trading pitch starts at the same place: a category where somebody is visibly winning. Tennis wallets up seven figures. An esports specialist with a chart that only goes up. The pitch is never wrong about the money — the money is real and it is on-chain. It is wrong about whose money it is.
We mapped all 25 category boards we grade — $2.79B of board-attributed volume and $71.4M of realized profit, drawn from a graded universe of 2,646 wallets that have traded $3.24B in total — and split every board by whether the wallet passes our farming-risk check. (The board figures are lower than the universe totals because only trading we can attribute to one of the 25 categories lands on a board.) The result is the same on board after board: wallets that pass keep 0.45% of what they trade. Wallets that fail keep 3.20%. The categories that look profitable look that way because of the cohort you cannot copy.
All figures below are a point-in-time pull from CopyGrade's live scored set via the read-only database, dated 2026-08-05; the method is documented here and every published aggregate is here.
The headline split
Take the graded universe whole, before any category cut:
| Graded universe (2026-08-05) | Wallets | Volume | Realized profit |
|---|---|---|---|
| All graded wallets | 2,646 | $3.24B | $76.6M |
| Carrying a severe farming-risk flag | 987 (37.3%) | 50.4% of volume | 60.7% of profit |
Thirty-seven percent of the wallets hold sixty-one percent of the money. That is the whole problem stated once. A P&L leaderboard sorts on the last column, which means a leaderboard is, structurally, a list of the wallets most likely to fail a farming screen — which is what we found the first time we scored one.
Now cut it by category. Across the 25 boards, restricted to wallets that pass the farming check:
The map
Every board we grade, sorted by volume. Take is realized profit as a percentage of volume traded — what the cohort kept per dollar it pushed.
| Board | Volume | All wallets | Passing cohort | Flagged cohort |
|---|---|---|---|---|
| Soccer | $1.20B | 2.18% | 0.69% | 2.79% |
| Esports | $349M | 4.25% | 0.00% | 4.33% |
| MLB | $277M | 2.33% | −0.06% | 2.39% |
| Tennis | $248M | 5.75% | 0.12% | 6.91% |
| NBA | $205M | 2.34% | 0.61% | 3.57% |
| Fed & rates | $70.9M | 0.06% | 0.06% | −0.01% |
| EPL | $55.0M | 1.03% | −0.03% | 4.05% |
| Crypto prices | $51.8M | 1.26% | −0.15% | 1.72% |
| Commodities | $45.3M | 0.07% | 1.34% | −0.17% |
| NHL | $43.1M | 0.60% | 0.00% | 2.06% |
| Bitcoin | $40.6M | 1.63% | 0.32% | 2.11% |
| UFC | $40.5M | 3.66% | −0.01% | 4.42% |
| Pop culture | $32.8M | −0.05% | 0.09% | −0.01% |
| UCL | $24.8M | 1.06% | 0.06% | 1.36% |
| NFL | $23.2M | 0.93% | 0.82% | 1.37% |
| Markets & finance | $22.0M | 1.17% | −0.43% | 1.89% |
| WNBA | $15.7M | 3.31% | 0.00% | 3.59% |
| Big tech | $10.5M | 0.09% | −0.17% | −0.59% |
| Cricket | $8.08M | 1.05% | 0.09% | 1.28% |
| Ethereum | $6.18M | 0.19% | −0.89% | 0.85% |
| Boxing | $5.07M | 0.08% | 0.04% | 0.92% |
| IPOs | $4.32M | −0.77% | −0.04% | −0.90% |
| Weather | $3.93M | 0.16% | 0.30% | −0.16% |
| Golf | $2.48M | 1.40% | −1.09% | 1.73% |
| F1 | $2.30M | −3.84% | −0.48% | −5.59% |
Four things fall out of that table, and none of them is the thing the category pitches say.
The flagged cohort out-earns the passing cohort on 18 of 25 boards. Six of the seven exceptions — Fed & rates, pop culture, big tech, IPOs, weather, F1 — are not really wins: both cohorts sit within a rounding error of zero and the flagged wallets simply lose more. Commodities is the one genuine exception in the table, and it's worth naming: the passing cohort took 1.34% on $7.6M while the flagged cohort lost money on $33.6M. One board out of twenty-five, carrying 1.6% of the mapped volume.
The passing cohort clears 1% on exactly one board out of 25. Commodities, at 1.34%, on $7.6M of clean volume. Second place is the NFL board at 0.82% — a board that hasn't played a regular-season snap yet. Everything else is under 0.7%, and ten of the 25 boards put the passing cohort underwater.
The biggest, most-pitched categories are the worst offenders. Tennis shows a 5.75% take overall — the highest number in the table, and the sort of figure a signal service builds a landing page on. Split it and the passing cohort's share of that is 0.12%. Esports is starker still: the wallets that pass the farming check traded $24.3M and finished up $446, a take of two thousandths of one percent. On the NHL board, the passing cohort pushed $28.8M for $671. These are not small samples of tiny wallets — they are tens of millions of dollars of real trading that netted lunch.
The macro boards are near-perfectly zero-sum in aggregate. Fed & rates: 0.06% across $70.9M, which is the finding we published before the July FOMC and it has not moved. Big tech, 0.09%. Weather, 0.16%. Commodities, 0.07% — though that one is the exception noted above, where an aggregate near zero hides a passing cohort at +1.34% and a flagged cohort in the red. When every input is public and simultaneous, nobody's aggregate edge survives the spread — and the new taker fees come out of a pot that was already empty.
Why the flagged cohort's profit is not an opportunity
The obvious bad reading of this post is "so copy the flagged wallets — that's where the money is." Three reasons that inverts the finding.
Some of that profit is an artifact of the behaviour that got them flagged. Self-trade wash and manufactured-volume churn move size between related accounts. Realized P&L computed from a public tape reads those transfers as trades, and one leg books a gain. A number that exists because two accounts traded with each other is not a number a third party can join.
Some of it is real but structurally uncopyable. Iceberg accumulation — our most common signature by far — is a wallet slicing one large intention into hundreds of small fills so the tape doesn't show it. That wallet is deliberately unfollowable, and by the time your bot has confirmed a pattern the position is built and the price has moved. Same for insider-shaped edge: the profit is real, and it is the profit of knowing something before the market, which is the one thing a mirror can never copy.
And some of it is bait. The copy-bait signature is a wallet building a legible, attractive record precisely so that followers arrive and become the exit. High take, clean-looking chart, and you are the counterparty.
So the flagged cohort's 3.20% is not an edge you were locked out of. It is the measurable signature of profit that does not survive being copied — which is exactly why our score vetoes on it.
What a copy trader should actually do with this
- Stop shopping by category. There is no category where the honest cohort is reliably paid. "Tennis is profitable" and "esports specialists print" are true statements about a population you cannot join. Pick wallets, not sectors.
- Split any performance claim by farming risk before you believe it. Any board-level, sector-level, or "top traders in X" number that hasn't been split this way is dominated by the flagged cohort — that's not an accusation about any product, it's arithmetic about who holds the profit.
- Set your expectations at the clean number, not the headline. If the passing cohort keeps 0.45% of volume traded across a $2.79B sample, then a copied wallet's realistic expectation before your own fees and latency is somewhere in that neighbourhood. Run it through the Copy Simulator with fees on and see what survives.
- Treat a near-zero-take board as a signal about the market, not the trader. Nobody is beating the Fed board. A wallet with a great FOMC record on a board that keeps six basis points is showing you variance with a story attached.
Caveats
These figures cover CopyGrade's covered universe — 2,646 graded wallets — not all of Polymarket; they are attributed per board, so a wallet active in three categories appears on three boards and the per-board wallet counts sum to more than 2,646. Realized profit is computed from public trade history against a published methodology, and it necessarily understates wallets holding good unresolved positions. Board totals are lifetime-to-date under our coverage, not a single window, so a board's figure blends different market regimes. A farming-risk flag is an algorithmic risk assessment, not an accusation of conduct — it says a wallet's on-chain pattern resembles patterns that historically precede uncopyable outcomes. And our own score's predictive validation is still null: this post measures who holds the money today, which is a description, not a forecast.
CopyGrade is analysis-only — it never executes trades, holds funds, or custodies keys, and a CopyGrade Score is a documented research opinion, not financial advice.