How the copiability score and the liquidation cushion are computed

• The score is a weighted geometric product of six components: consistency 25%, activity 25%, quality 20%, expectancy 10%, risk 10%, returns 10%.
• A wallet is scored only if it clears 10 ranking gates: perps only, a net winner, enough trades and coins, recent activity, a holding time a person can follow, a funded perp account.
• The cushion before liquidation is measured at the account level, because in cross margin Hyperliquid liquidates on the equity of the whole account.
• A value that cannot be computed is never replaced by a guess: no component, no score; an unknown number shows as a dash, never as a zero.

Each component lies between 0 and 1, and the score is their weighted geometric product:

score = 100 × Expectancy^0.10 × Risk^0.10 × Consistency^0.25 × Quality^0.20 × Returns^0.10 × Activity^0.25

• Consistency · 25% — the share of TIME the all-time perp PnL curve spends rising, weighted by duration — Hyperliquid's samples are irregularly spaced, and counting points would weigh years of calm like days of storm —, then shrunk by the Wilson interval (lower bound, z = 1.96) with the number of curve segments as the sample size.
• Activity · 25% — the trade count on a log scale, min(log10(trades + 1) ÷ 3, 1), full at 1,000 trades: enough closed trades for the other components to mean something.
• Quality · 20% — the square root of the Wilson lower bound (z = 1.96) of the win rate: a win rate is believed only as far as the trade count proves it — 90% over 10 trades weighs far less than 90% over 1,000 —, and the square root keeps it from dominating the score.
• Expectancy · 10% — the all-time ROI published by Hyperliquid's leaderboard, squashed as ROI ÷ (ROI + 0.5): an ROI of 50% gives 0.5, a higher ROI gets closer to 1, and a negative ROI gives 0.
• Risk · 10% — final PnL ÷ (final PnL + maximum drawdown), on the all-time perp PnL curve: no drawdown gives 1, a drawdown as deep as the PnL gives 0.5, and a curve ending at or below zero gives 0. This is the component that catches small regular gains followed by a collapse.
• Returns · 10% — the profit factor (gross wins ÷ gross losses) on a log scale, ln(1 + min(PF, 50)) ÷ ln(51): a profit factor above 50, or no losing trade at all, counts as 50, so one extreme trade cannot dominate.

The product is geometric: a weak component pulls the whole score down and cannot be made up by the others. The weights add up to 1, which keeps the score between 0 and 100, and they are never re-weighted: if one component cannot be computed — unknown ROI, unusable PnL curve — the wallet gets no score and is not ranked. On screen, the score is rounded to the nearest whole number.

The weights were chosen on August 16, 2026 by replaying 13 candidate formulas over 8 weekly periods. Expectancy, risk and returns turned out to be strongly correlated — one signal counted three times —, so together they weigh 30%; activity was the only component uncorrelated with the others, hence its 25%.

• High: 65 and above.
• Medium: from 45 to under 65.
• Low: under 45.

The thresholds were set on August 16, 2026 from the real distribution of the 145 wallets ranked that day (highest score 79.6); the top band was then roughly the top 15%. Nobody reaches 100: the Wilson interval caps what any finite track record can prove, by design.

Before any score is computed, a wallet must clear every one of these gates, in this order:

1. It has traded: an account with zero all-time volume on Hyperliquid's leaderboard stops here.
2. Perps only: spot trades are left out, and an account with no perp PnL in its history is not a trader you can copy.
3. A net winner: positive all-time perp PnL, and positive realized PnL over its latest fills.
4. At least 5 trades over its latest fills.
5. At least 3 different coins traded.
6. An account worth between $1,000 and $10,000,000 on Hyperliquid's leaderboard — the total value of the account (perp, spot and vaults), not the perp side alone.
7. At least one trade in the last 30 days.
8. A median holding time between 30 minutes and 14 days. Shorter is scalping, too fast to follow; longer means positions opened and left, with no entry or exit to copy in time. A median that cannot be measured (no closed position) also excludes the wallet.
9. At least $100 of perp equity. This gate is checked again every time the ranking is read, so an account emptied after its analysis drops out before the next pass; an unknown equity does not exclude.
10. All six components of the score can be computed.

The win rate itself is not a gate: it only enters the score, through the quality component.

The page of a wallet that is not ranked says which gate it missed, when the analysis recorded it. The ranking is refreshed by an analysis pass that takes about a day, ranked wallets first.

In cross margin, Hyperliquid liquidates on the equity of the whole account, not on the margin of each position: one losing position eats into the collateral of all the others. That is why a wallet page shows the cushion before liquidation at the account level, not only a liquidation price per position.

cushion = (account value − maintenance margin) ÷ account value

Both values come from Hyperliquid's public account state for the main dex. The maintenance margin is estimated position by position as half the initial margin at that market's maximum leverage — the rule Hyperliquid applies to cross positions. 100% means no maintenance margin is used; the closer to 0%, the closer the account is to liquidation. The same panel shows equity, total notional, margin used, maintenance margin and withdrawable.

An isolated position has its own margin and its own liquidation price: the positions table shows that price and the distance to it, at the live price.

These limits are known and kept; each is written here rather than hidden behind a number.

• In the win rate and the trade count, a “trade” is a closing fill, not a full round trip: a winner closed in three pieces counts as three winning trades, which flatters traders who scale out. Counting round trips was replayed against the current rule and did not rank traders better, so the rule was kept (decision of October 13, 2026).
• Win rate, trade count, profit factor, realized PnL, coins traded, median holding time and last trade are computed on the latest 2,000 fills Hyperliquid returns for the account; older history is not read. The PnL curve, the drawdown and the ROI, on the other hand, cover the whole history.
• Fills from TWAP orders (the slices of a TWAP) are absent from the fill history Hyperliquid returns, so they are not counted.
• “Perp equity” (the $100 gate, the card tile and the size filter) is the last point of Hyperliquid's perp portfolio history. For a unified account it follows the margin committed to positions, not the capital — and most ranked accounts are unified (77% of 150 drawn at random on September 27, 2026).
• The expectancy reads the all-time ROI of Hyperliquid's leaderboard, which covers the whole account (perp, spot and vaults), not the perp side alone.
• For a unified or portfolio-margin account, the capital sits in the spot USDC balance, which the perp account value of the main dex does not measure: the cushion shown is computed on the perp account only and can differ from the real one.
• Positions on HIP-3 builder dexes (stocks, indices, commodities) are not read into the cushion of a wallet page.
• Ranking numbers date from the last analysis pass (about a day). A wallet page reads positions and the cushion when it opens, or shows its last read with its age.

Every card of the copy trading leaderboard shows the score, and “How it's calculated” opens its six components; the full list of ranked traders links to each ranked wallet's page, which shows the cushion. The docs cover the rest of the site.

Automated copy trading is not open to the public: the ranking and the wallet pages are analytics.

The score describes a public track record. It is a heuristic, not a prediction, and not investment advice.