Methodology
How every number on this site is computed — and how to read it without fooling yourself.
Why PolyHunter exists
Every wallet tracker ranks Polymarket whales by raw PnL. That number lies to copiers twice. First: by the time you see the whale's trade and place yours (~3 seconds), the price has moved and the orderbook has thinned — you do not get their fill. Second: some of the biggest "winners" manufactured their record by wash trading. PolyHunter fixes both. We rank by copyable PnL and we vet every wallet before it reaches the leaderboard.
Copyability ratio — the exact math
copy_ratio_250 = (copier ROI at $250 size, 3s delay) ÷ (whale ROI). Both sides are per-dollar-invested returns, so the ratio is dimensionless and comparable across trade sizes. When a whale trades, our recorder snapshots the live orderbook at that exact moment. The simulator then walks that stored book as if you had placed a $250 order 3 seconds later, paying real slippage against real depth. How to read it: 1.00× means you keep 100% of the whale's edge. 0.7–0.9× is typical — slippage eats some. Above 1.00× happens when the whale moved the market and the book refilled at a better price for you. Below 0 means the edge vanishes entirely after slippage.
What copyability is NOT
It is not a money multiplier. A 33× ratio does not mean 33× your money — it means the simulated copier return was 33 times the whale's own return, which almost always happens because the whale's ROI was close to zero (a tiny denominator inflates the ratio). We flag these with a LOW N badge when the wallet has fewer than 30 recorded trades. Treat any ratio above ~5× as a statistical artifact until the sample grows.
Wash-trade detection — four signals
1) Self-opposite: buying both outcomes of the same market at prices summing to ≥ 1.00 (guaranteed loss), in mirrored sizes (within 15%), within 60 seconds. That trade only makes sense if you are manufacturing a win on one side while hiding the loss on the other. Genuine arbitrage sums below 1.00 and passes clean. 2) Round-trip: buying then selling the same token at a meaningfully different price (≥ 2¢ delta) in a tight window — filters out legitimate high-frequency scalping, which moves fractions of a cent. 3) Padding: loading up above 95¢ right before resolution to inflate win counts. 4) Statistical anomaly: win rates and patterns that are implausible for the sample size. Wallets are tiered clean / suspect / grey / blacklisted, rescanned every 30 minutes.
The data moat
Orderbook depth at the moment of a trade cannot be reconstructed later — Polymarket does not archive it. Our recorder has been capturing it live, 24/7, since 2026-07-08. Every day it runs is data nobody else can backfill. The simulator runs at 4 latency tiers (0s / 3s / 10s / 30s) and 3 size tiers ($50 / $250 / $1000); the leaderboard shows the $250 / 3s tier because it best matches a realistic manual copier.
Honest limitations
Our window starts 2026-07-08 — stats reflect what we have seen, not lifetime performance. Win rate and PnL only count markets that resolved inside our window. Wallets under 30 trades carry a LOW N flag because every metric is noisy at that sample size. Wash detection is probabilistic: suspect and grey tiers exist precisely because we would rather show uncertainty than fake confidence.
Questions or a wallet you think is misclassified? This will be a support channel soon.