We re-scored 2,017 Polymarket wallets. 23 are worth copying.
This is the third edition of our recurring report on the state of Polymarket copy trading. We re-scored the full universe under CopyGrade coverage — 2,017 actively-traded wallets, up from 1,297 in June and 385 in the first edition. Just under half carry a farming-risk flag. The median wallet keeps −4.8% after modelled trading costs. And 23 wallets — 1.1% — pass every test a copier should apply.
A note on the sample: 2,017 wallets that trade actively enough to be scored, snapshotted in late July 2026, each graded from its public Polymarket trade history by the same CopyGrade Score model as every prior edition. The model's weights and thresholds are unchanged since June — we published a 30-day validation of the score against realized outcomes on July 1 and deliberately changed nothing.
The headline numbers
| July 2026 (2,017 wallets) | |
|---|---|
| Median CopyGrade Score | 42 |
| Farming-flagged | 47.8% |
| — high-confidence (severe) | 37.3% |
| Negative fee-adjusted edge | 74.6% |
| Median fee-adjusted edge | −4.8% |
| Worth copying (all tests at once) | 23 (1.1%) |
Sixty-three percent of the scored set sits below a CopyGrade Score of 50. Three in four wallets do not clear their own trading costs.
Why there's no June-versus-July column this edition. Two things changed in how we build these numbers between editions, and both move the figures independently of anything traders did: we now charge Polymarket's real published fee schedule rather than a simplified assumption, which pushes fee-adjusted edge down; and we corrected an ingestion defect that had been storing duplicate copies of the same fill, which had inflated the repetition that farming detectors read. Printing a delta across that boundary would attribute our own pipeline changes to the market. So this edition reports levels. The next one, computed end to end on the corrected pipeline, restores the trend line — and it will be an honest one.
One number makes the copier's problem concrete without needing a trend: 354 wallets currently show a positive headline edge — and 214 of them, three in five, carry a farming-risk flag. The records that look most worth copying are, at this snapshot, mostly records our model assesses as manufactured-risk. A good-looking history is not evidence of a good target; at these base rates it is closer to the opposite.
The vetting funnel: 2,017 in, 23 out
Each row below is how many wallets pass that single test; the last row is how many pass all four at once.
| Test | Wallets passing | Share |
|---|---|---|
| Scored | 2,017 | 100% |
| Enough recent history to judge (≥20 trades / 90d) | 1,463 | 73% |
| Farming-clean | 1,053 | 52% |
| Fee-adjusted edge above 1% | 213 | 11% |
| CopyGrade Score ≥ 75 | 23 | 1.1% |
| All four simultaneously — worth copying | 23 | 1.1% |
Those last two rows are the same 23 wallets: at this snapshot every wallet scoring 75 or better also clears the history, farming and edge gates on its own. That is a property of this population, not a shortcut in the method — the four tests are still applied separately, and in earlier editions they did not collapse.
53 wallets (2.6%) are simultaneously clean, active, and genuinely profitable after costs — before the score bar is applied at all. The score bar then takes it to 23. The funnel is not narrow because we set it narrow; it is narrow because the same four unchanged gates keep catching most of the field.
Is that half-the-universe flag rate a composition effect?
Coverage grew by 410 wallets since the last snapshot, so the fair question is whether the flag rate is being set by who we added rather than by what wallets do. We split the population and measured.
| Cohort | Wallets | Farming-flagged | — severe | Median fee-adj. edge | Median score |
|---|---|---|---|---|---|
| Already under coverage | 1,607 | 49.2% | 38.8% | −3.3% | 41 |
| Newly covered since | 410 | 42.4% | 31.5% | −13.8% | 44.5 |
The newly-covered wallets are cleaner on flags than the incumbents, so coverage growth is pulling the blended flag rate slightly down, not up — the 47.8% is not an artifact of who we added. What the new cohort does bring is a far worse median edge (−13.8%): these are shallower, newer records, and a short history flatters a wallet until it doesn't.
The two cohorts land within seven points of each other on flags. Whatever is driving a flag rate near one in two, it is not sampling.
The leaderboard cut: one copyable wallet
The wallets currently on Polymarket's public profit and volume boards — 189 under coverage at this snapshot; the roster churns as windows roll — remain worse than the field, extending last edition's finding:
| Full set (2,017) | Current leaderboard (189) | |
|---|---|---|
| Median CopyGrade Score | 42 | 37 |
| Farming-flagged | 47.8% | 59.8% |
| — severe | 37.3% | 46.0% |
| Median fee-adjusted edge | −4.8% | −9.9% |
| Worth copying | 23 (1.1%) | 1 (0.5%) |
One wallet. The public boards rank by raw profit, raw profit rewards exactly the behaviours our model penalises, and at this snapshot precisely one board name survives every check. If your shortlist is the leaderboard, your base rate is one in 189 — see the graded leaderboard for the rank-versus-reality gap on today's board.
What this means if you copy
The base rates are the strategy. At 1.1%, picking a wallet by eye is a 98.9%-failure lottery, and the failure modes are asymmetric — a farmed record doesn't underperform, it turns its copiers into exit liquidity.
- Vet before you automate. The pre-automation checklist exists because most candidates fail it.
- Read edge net of costs. The headline return is a ceiling you'll never reach; three in four wallets are negative before you add slippage, and we now charge the real fee schedule to say so.
- Treat a clean-looking record as unproven, not clean. Three in five positive-edge records carry a flag this month. Check the behavioural signatures, not the PnL.
- Re-check what you already copy. Both the population and the way we cost it moved this month. Alerts exist so a changing target reaches you before your bot keeps mirroring it.
How we got these numbers
Point-in-time snapshot, late July 2026: 2,017 actively-traded wallets under CopyGrade coverage, re-scored within a single sync cycle by the current CopyGrade Score model from public Polymarket trade history. Edge figures are net of Polymarket's published trading costs; we quote medians throughout, which the long negative tail doesn't distort. Farming flags are algorithmic risk assessments — our documented opinion from disclosed public data, not verified conduct or a statement of fact about any trader, and each wallet's page carries a dispute path. The model and the pipeline behind it change in public, with dated notes — including the two changes noted above, which is why this edition reports levels rather than deltas. This edition reports the same weights validated — and found wanting, and kept anyway pending more data — in our published null result. Prior editions: June 8, June 2. Not financial advice.
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.