Due diligence for Polymarket copy-trading, in plain English. How to vet a wallet before you copy it, how farming turns copiers into exit liquidity, and what copying actually returns once you subtract fees, slippage, and latency.
Fourth edition, and the first honest trend line: the blended flag rate fell from 47.8% to 35.0%, but the same 2,019 wallets from July are flagged at exactly 47.8% again. The drop is who we added — and half of them are too thin to read.
Eleven days before the Wild Card the graded MLB board is 1,663 wallets, three in five of them farming-flagged, with a clean cohort that is net negative and fifteen month-old positive verdicts. New this time: a ledger of every pre-game moneyline call these wallets held at first pitch, which hit 727 games against 747.7 expected at the prices paid.
Five weeks before kickoff we promised to re-grade the NFL board once real games had been played and publish what changed, either way. The board's realized profit went from $524k to $2.80M in nineteen days, the clean cohort that existed before Week 1 made nothing net, and nearly two-thirds of the board's profit is one flagged wallet whose Week 1 was a pre-kickoff bet on the Chiefs.
College football runs about 120 Polymarket game events a week against the NFL's 16 — 123 events and $12.5M in one week — and the graded wallets working it are 82% flagged, mostly trading in play. The dated baseline for the season, and what to check before copying a Saturday record.
Twelve days before kickoff, the typical clean NFL wallet is losing money after fees and latency. Clean wallets lose more often and far less than flagged ones — quote only half of that and the finding inverts. Fifteen clean wallets on the board carry a real edge above 1%.
The Wild Card round opens September 29. Our graded MLB board says the wallets worth diligence are fifteen tiny, brand-new accounts that are collectively down — while the board's entire $7.68M of realized profit sits with the farming-flagged cohorts.
A hot week can clear every scoring gate; only time can prove it wasn't luck. What we found on the NFL and MLB boards, why the youngest records look the best, and the observation-span column we now print beside the score.
The most common question in copy-trading has a prior, and the prior is brutal: at launch (2026-08-15), 1.5% of graded wallets cleared the full bar. The definition, the launch reading, and why we will never quote it as a trend.
Five weeks before kickoff the board is quiet enough to see who was actually there beforehand. Of 319 wallets with NFL exposure exactly one clears a positive verdict — and the clean cohort's entire profit is carried by 29 wallets.
The category just had its biggest month ever. The portion of it on a venue where a trader is an identifiable, auditable account shrank — because regulated access means KYC, intermediation, and no public ledger.
Tennis reads 5.75% profitable until you split it — the wallets that pass our farming check kept 0.12%. Esports wallets that pass traded $24.3M and finished up $446. The categories that look profitable look that way because of the cohort you cannot copy.
1,237 graded wallets, $1.4B of covered World Cup volume, and one champion. The grade bands didn't just sort who was right — they sorted who was early, who chased, and who sold the winner at 42 cents. Plus the final's cruelest lesson: both 90-minute match markets resolved No.
The July decision market has real two-way risk — no-change at 74¢, a 25bps hike at 25¢. Our Fed board says what usually happens to the money: six hundred graded wallets pushed $122M through rate markets and collectively kept five basis points of it.
Copy-bots are takers on entry and exit, so the new fees land on them twice per round trip — against a field whose median fee-adjusted edge is already −1.4%. The stack, the arithmetic, and why our scores didn't move.
Prediction-market volume is up ~5× in seven months and sports is Polymarket's biggest category. The H2 calendar, the market sizes as of mid-July, and the lesson from the World Cup board: event windows concentrate liquidity — and farmers.
The World Cup profitboard is the most tempting shortlist Polymarket has ever minted — and the first wallet our model grades a copy candidate sits at #55, with $49 of World Cup profit. The top 50: 94% flagged, zero candidates, median score 30.
Third edition of the recurring report: just under half the actively-traded universe carries a farming-risk flag, the median wallet keeps −4.8% after costs, and 23 wallets out of 2,017 clear every gate. The bar didn't move — the population grew.
2026's Polymarket insider cases share a profile that looks irresistible from the outside — a near-perfect record on a few correlated markets, built fast. It's the record copiers chase, and the least copyable kind there is.
The World Cup pushed Polymarket to a record month and is minting sports 'winners' by the thousand — but sports is structurally the worst place to copy a wallet. Why speed, entry prices, and one-tournament records break copying.
Search 'polymarket copy trading bot github' and the top result might be a credential stealer. Researchers documented fake copy-bots built to lift wallet keys — a provenance checklist before you run any bot.
The wallet you're tracking may be one mask of a larger operation — a brand while fresh addresses do the quiet work. Polymarket's own newsletter says sharps run secondary and tertiary accounts. Why funding trails are how you see the whole book.
Every scoring product claims its number works. We graded ours against 30 days of realized outcomes and it doesn't — yet (ρ = −0.001). Here's the full read, why an honest model can land here, and what we're doing about it.
The same bot that copies the trade can lose the funds. Polycule's ~$230K hack was a custody failure, not a strategy one — two risks every copier manages: who you copy, and what the bot can touch.
Most Polymarket profit is concentration and liquidity provision, not copyable skill — a $67B study finds winners make markets and losers take them, and a copier is always a late taker.
An airdrop pays people to manufacture volume — and manufactured volume is what makes a wallet dangerous to copy. 35% of active wallets already carry a farming flag; the POLY airdrop pours fuel on the fire.
The surest way to lose money copying a winner is to copy one whose edge was information you don't have. 2026's prosecutions show why an information edge is episodic, legally fragile, and invisible until the arrest.
A 95% win rate can lose money — the break-even win rate of any bet is its price. And the stat is for sale: farming-flagged wallets average a 42% win rate vs 36% for clean ones. What to read instead.
A Columbia study estimates ~25% of Polymarket's all-time volume is consistent with wash trading — and among actively-traded wallets, our farming-flag rate is 35% — 55% once thin records are set aside. Different denominators, same lesson.
Copy-bots watch for visible trades, so sharps trade invisibly: sub-$200 iceberg slices, YES+NO merges with no sale on the book, fresh wallets. What your bot can't see, and why it matters.
Not on average — the median wallet has negative edge before copying costs are added, and the records that attract copiers are the least trustworthy. The base rates, and the four gates that beat them.
The leaderboard is the obvious place to find a wallet to copy — and close to the worst. We scored the 193 names on it: 72% are farming-flagged, only 3 worth copying.
About a third as of the October 2026 re-score, and more than half of the wallets with enough history to judge: of 11,721 active Polymarket wallets, 35.0% carry a farming flag and 24.2% a high-confidence one — and on the leaderboard itself, 61.3%. What farming is, and how to spot it.
Polymarket's leaderboards rank by raw profit, so we scored the 193 names at the top. 72% are farming-flagged, 57% severely, and only 3 are worth copying — the dirtiest slice of the platform we've measured.
We scored 385 of Polymarket's most active wallets. 37% are farming-flagged, the median has no edge before you even copy it, and only 13 cleared the bar. The full distribution.
The leaderboard ranks by raw profit — the one number that tells you nothing about whether copying makes you money. Here's what to check instead.
A farmer builds quietly, advertises a clean record, attracts copiers, then exits into them. These are the specific behaviours that give it away.
A wallet's headline return assumes zero fees, zero slippage, and zero delay. You get none of those. Here's the gap, and how to size it.