Methodology v1.2

Ratings are computed automatically from the public Hyperliquid info API (portfolio history, positions, recent fills). No private data is used. Universe: open vaults by TVL plus top leaderboard wallets (≥ $10K equity), plus any claimed wallet.

Eligibility

Rated only with ≥ 30 days of history, ≥ 20 closed trades and ≥ $1,000 equity. Otherwise "NR" (Not Rated).

Four pillars

• Returns 30% — annualized time-weighted return. Each period's return = PnL change ÷ prior equity, so deposits and withdrawals do not count as performance.
• Risk 30% — max drawdown 35%, Sortino 25%, Sharpe 15%, current leverage 15%, liquidations in recent fills 10%.
• Consistency 20% — monthly win rate 60%, PnL stability (1 − best month's share of all positive PnL) 40%.
• Track record 20% — history length 40%, closed trades 30%, equity size (log scale) 30%.

Each metric is mapped linearly onto 0–100 between a 'bad' and a 'good' bound (see src/lib/scoring.config.ts).

Grades

AAA ≥ 85 · AA ≥ 75 · A ≥ 65 · BBB ≥ 55 · BB ≥ 45 · B ≥ 35 · CCC ≥ 25 · D ≥ 0

Anti-gaming flags (score penalties)

• Deposit jump: equity jump not explained by PnL > 100% of prior equity in the last 45 days (−5)
• Low liquidity: > 30% of traded notional in coins with < $5M daily volume (−10)
• Wash-like: > 40% of closes are ~zero-PnL round trips within 60s (−15)
• Short history: < 90 days (−3)
• Extreme leverage: > 20x effective (−10)

Copy-risk check

Each wallet page estimates what copying it would have cost: the worst drop of its return index over the last 90 days (in % and $ per $1,000), a sizing heuristic (keep 1.5× its worst historical drawdown under ~10% of your portfolio, max 25%), and stop signals: leverage far above its own usual level, a drawdown deeper than its historical norm, a shift into low-liquidity coins, size inflated by a large deposit, or a score drop. These are heuristics from public data, not advice.

Track-record verification

Since 2026-10-30 every rating is stored as a daily snapshot (grade, score, flags, full metrics, scoring version). This lets anyone later backtest whether the ratings work: did wallets flagged high-risk actually lose more over the following 30/90 days than unflagged ones, and do AAA wallets keep outperforming? The scoring version is recorded so method changes stay visible. We will publish these results once there is enough history.

Limitations

Only the most recent ~2,000 fills are analysed per wallet. Portfolio history is sampled sparsely by Hyperliquid, so drawdowns between samples can be missed. Vaults that trade via child vaults (e.g. HLP) are counted by PnL periods instead of fills. AI-agent sub-wallets (API wallets) hold no funds and appear empty — the master account is what gets rated. Ratings refresh daily at most.

Not investment advice.