Sources and methodology

Reproduce a vault snapshot

Open the vault's evidence link or request /api/v1/vaults/0xADDRESS/evidence. It contains the original redacted Hyperliquid history, SHA-256 receipt, selected history window, materialized observations, daily closes, calculation inputs and output. Account values and PnL are separate series.

Source timestamps matter: a later API response need not equal an earlier snapshot. A hash proves internal consistency, not that Hyperliquid signed the response. Missing, stale or quarantined history does not produce a public risk score.

Upstream: Hyperliquid vault directory (TVL) and Hyperliquid Info API (vaultDetails).

Risk

weights

perf: 0.35
drawdown: 0.25
liquidity: 0.25
data_risk: 0.15

upper inclusive buckets

volatility: [[0.003, 10], [0.01, 25], [0.02, 45], [0.04, 65], [null, 85]]
worst_loss: [[0.005, 10], [0.02, 35], [0.05, 65], [null, 90]]
drawdown: [[0.01, 10], [0.05, 35], [0.12, 60], [0.25, 80], [null, 95]]
tvl_volatility: [[0.01, 15], [0.03, 35], [0.08, 60], [null, 85]]

tvl lower inclusive buckets

[[100000000, 10], [20000000, 20], [5000000, 35], [1000000, 55], [0, 75]]

components

perf: int(0.6 * volatility_bucket + 0.4 * worst_loss_bucket)
drawdown: drawdown_bucket
liquidity: int(0.7 * tvl_size_bucket + 0.3 * tvl_volatility_bucket)
data_risk: int(0.7 * 10 + 0.3 * (10 if daily_return_count >= 30 else 20))

total

clamp(round(sum(component * weight)), 0, 100); Python ties-to-even rounding

bands

low: [0, 33]
moderate: [34, 66]
high: [67, 100]

missing data

null, never APR-derived volatility, drawdown estimates, or a default safe score

eligibility

Recorded, hash-matching, trusted PnL history <=48h old; >=24 consecutive completed UTC daily returns in the trailing 30-day window; positive account values.

History

interval return

(pnl_usd[t] - pnl_usd[t-1]) / account_value[t-1]

share index

S[0]=1; S[t]=S[t-1]*(1+interval_return[t])

daily close

Last sample in each UTC date, excluding the UTC date of the latest sample (unfinished day).

window

Daily closes in [last_completed_close - 30 days, last_completed_close]. No forward fill; a missing UTC date withholds risk and drawdown.

drawdown

max(1 - S[t]/max(S[0:t+1])) over that window, including the initial close; positive fraction

volatility

Sample standard deviation (ddof=1) of daily share-index returns; not annualized

sharpe

mean(daily returns) / sample_stdev(daily returns) * sqrt(365); zero risk-free rate; null when stdev <= 1e-9

worst day

min(0, min(daily returns)); positive days are not losses

tvl volatility

Sample standard deviation (ddof=1) of log(account_value[t]/account_value[t-1]) at the same daily closes

risk size input

Latest recorded account_value from the same source receipt, not current directory TVL

limitations

Sampled PnL/prior-equity estimate, not exact investor TWR or fee-adjusted realized return; intraperiod flow timing unavailable. Daily-close drawdown can miss intraday losses. Window coverage and source timestamps are disclosed per vault.

Tvl

endpoint

https://stats-data.hyperliquid.xyz/Mainnet/vaults

selector

item.summary.vaultAddress == vault_id; item.summary.tvl

type

Observed directory equity, not return, available withdrawal liquidity, or proof of reserves

cross check

vaultDetails.portfolio[window].accountValueHistory[-1]; different sampling times/scopes can differ. No top-level tvl field required.

Ranking

risk adjusted

clamp(r30, -0.95, 1) / max(0.15, risk_score/100) * (0.8 + 0.2 * log_normalized_directory_tvl)

tvl normalization

Clamp TVL between configured directory floor and $500M; normalize log10 to [0,1]. Floor returned in ranking response.

eligibility

Active directory vault, fresh snapshot confidence >=0.7, recorded risk evidence, and available 30d PnL return. No APR fallback. Sort descending, address ascending for ties; limit after sort.

universe

Up to 500 active Hyperliquid directory vaults by descending TVL, subject to the configured TVL floor; candidates_examined is returned. Unscored candidates are excluded, not assigned zero.

meaning

Research ordering, not Sharpe, expected return, or a recommendation. Verified/estimated rankings retain their separate legacy APR-based formulas; they are not this v3 ranking.

Entry

version

entry-score-v1

start

50

upnl

u<-0.05: -min(20,abs(u)*200); else u<-0.02: -abs(u)*100; else u>0.02: +min(10,u*100)

leverage

L>5: -min(15,(L-5)*3); else L>3: -(L-3)*2; else L<2: +5

concentration

top1>0.8: -10; else top1>0.5: -5

liquidation

elevated: -10; otherwise no adjustment

flow

f=net_flow_7d/equity; f<-0.1: -min(15,abs(f)*50); else f<-0.03: -5; else f>0.05: +5

whale outflow

whale_outflow_7d < -50000: -5

realized pnl

p=realized_pnl_30d/equity; p>0.05: +min(15,p*100); else p>0: +5; else p<-0.05: -10

rounding

Clamp round(score) to [0,100], ties to even. >=70 Good Entry, >=40 Neutral, otherwise Avoid. Missing inputs skip their adjustment; ratios need positive equity.

publication gate

Fresh position snapshot, trusted PnL history with 30d return, finite score, known position scope. HLP consolidated scope remains unverified, so no HLP entry score.

limitations

Heuristic, not a profitable-entry prediction. Rolling flow estimates are not a wallet-transfer ledger. Entry inputs do not yet have raw position-source receipts.