How we measure Hyperliquid.

Published methodology, rating-change log, and research notes. Every score recomputes from the same config the engine uses.

Methodology

Version 1.0.0 · status provisional

• capacity headroomHow crowded the book is versus market volume.15
• copyabilityHow much of the return survives realistic replication lag.10
• drawdown behaviorDepth, duration, and ulcer of losses — not just the headline max drawdown.20
• performance qualityDeflated Sharpe on 90d/365d — performance adjusted for how many vaults were tried.30
• regime dependenceWhether returns concentrate in a single market regime.15
• style driftHow fast the strategy's factor exposures are changing.10

Conflicts

The firm's own on-chain products are either rated by the identical public rules or excluded from ratings entirely. No paid placement, ever.

Falsification standard

A rating is wrong if an independent party can recompute the composite from the published raw series and methodology version and obtain a different tier outside the published confidence interval, or if a vault's subsequent risk-adjusted outcomes systematically contradict the tier's implied risk ordering across a published evaluation window.

Methodology changelog

[
  {
    "at": "2026-07-25",
    "note": "Synthetic→real switch (Wave 0/1): BTC regime from Hyperliquid candleSnapshot (not vault-NAV stand-in). Style factors market_beta/trend/momentum/carry from HL candles + fundingHistory with seeded synthetic fallback (see factors: block). FRED VIX + Fear&Greed into structure gates; stale VIX omits (fail-safe, no hard-coded default). Insufficient component metrics → Unrated rather than silent neutral-50. Index publish may still flag provisional_metrics when copyability/headroom absent.\n"
  }
]

Research notes