Paste a Hyperliquid address. See how much of its record counts.

How it reads a wallet

Returns come from PnL, not account value. Account value moves when the owner deposits or withdraws; PnL does not. Each week’s return is the change in cumulative PnL divided by the account value at the start of the week. Weeks that start below $1,000 are skipped.

Weekly, because the history is. Hyperliquid keeps older history about a week apart and recent history minutes apart. Everything is put on one weekly grid; a week with no new data point is merged into the next one, not counted as a zero.

Short records are not over-read. A t computed from a handful of weeks has fat tails: with eight weeks and no skill at all, t ≥ 2 turns up about twice as often as the bar assumes. So t is taken from Student’s t with the record’s effective degrees of freedom (fewer when a few long gaps carry the record) and put back on the usual scale. For a long record it is close to Sharpe × √years; for a short one it is lower. Tested: with no skill, the share of records reaching 2 stays at the nominal 2.3% however sparsely the history was sampled.

Two bars. If you came to this wallet with no prior selection, t ≥ 2 is the usual bar. If you found it by sorting a leaderboard, you picked the best of many, and the bar rises with the number you picked from (Bonferroni on 48,407 accounts: 4.90). Most people find wallets on leaderboards. That bar counts every account as independent; accounts that trade alike are fewer independent bets, which would lower it to about 2.6 by our rough estimate (an order of magnitude, not a measurement).

What t cannot see. Tail risk that has not happened yet. A strategy that sells insurance can show no losing week for a long time. Where the record has no losing week, or one week made most of the gains, the page says so. A seller that scales its positions to recent volatility can keep its weeks looking ordinary before a large loss and, scaled down, around it, so these notes can miss it.

Other wallets. The check scores the one address you paste. Whatever its owner makes in other wallets, a copier of this one does not get. And someone who runs ten wallets and shows only the best is showing the best of ten, which should face a higher bar; we cannot tell how many wallets they run.

The drawdown reading. Most money arrives after a strong run and leaves in the drawdown that follows, which is often an ordinary one; the gap between a strategy’s return and its investors’ return is, in one manager’s words, “usually the biggest cost” (Patrick Welton on Top Traders Unplugged). So the page says how far the record is below its own peak, in weekly returns added up since the peak (the same returns t uses; compounding would describe a portfolio nobody held, since accounts add and withdraw money), and how much of its time a record with the same measured edge and volatility spends at least that far down. Under 2% is rare enough to doubt the edge; 10% or more is ordinary. The share comes from the drawdown distribution of a random walk with that drift, corrected for weekly sampling, and is checked against simulation.

Evidence is not what a copier gets. t measures the record the trader made. A copier enters after the trader, at another size and with other fees; the shorter the holding period, the more of the edge the delay costs, and a size that works for $50k may move the market at $5M. A record that clears the bar is evidence about the trader, not a forecast of your return.

What it counts. The PnL Hyperliquid records for the account: perps and the spot account, which holds HIP-4 outcome contracts. Trading on HyperEVM, such as Papertrade or HyperEVM options protocols, is not in it. A wallet that sells options on HyperEVM and hedges with perps shows only the hedge.

Not investment advice. Public data from api.hyperliquid.xyz, fetched by your browser. The address is never sent to us. We count page views without cookies (Cloudflare Web Analytics), and each check as one tally with the page language and nothing else. Method: Method, Lo (2002), Bailey & López de Prado (2014).