Cohorts · positioning · activity · quality

Wallet cohorts, with their validation status

Cohorts of Hyperliquid accounts defined by rules we publish. Each response states the rule, the formation date, its coverage, and whether the rule has passed our pre-registered out-of-sample persistence test. Until it passes, a skill cohort is shown as past performers (unvalidated); size, vault and seed cohorts are descriptive. Nothing on this page is a win-rate, a profit figure or a price call.

Loading
Cohorts
—
rules published
Skill cohorts validated
—
pre-registered test
Validation windows
—
accrued of required
Coverage status
—
computed when served
Latest formation
—
UTC

Cohorts and their rules

CohortLabelClass RuleMembersFormationValidation
Loading…

Positioning — BTC and ETH

Trader+
Loading…

Cohort rows appear here only where cohort disclosure is enabled; otherwise the table carries the everyone row (all tracked accounts). Whenever membership is not published, the response’s disclosure_control field states the rule in force.

Validation ledger

StudyRuleStatus WindowsDecision
Loading…
Definitions
Positioning
What a cohort holds, where cohort disclosure is enabled: long, short, net and gross USD per symbol, from a complete sweep of the tracked accounts. Otherwise the everyone row is the one served. Counts report observed addresses and a conservative independent-participant estimate.
Activity
What wallets did in an interval: open, add, reduce, close, flip, deposit, withdraw and transfers. A price move is never an activity event; mark-to-market change is reported beside the price-neutral net change, never summed into it.
Quality
Accounting, risk, evidence and specialization reconstructed from a wallet’s own fills, funding and ledger updates. Which cohorts the wallet belongs to, and why it qualifies for them, are shown only where cohort membership is published; elsewhere those fields say withheld. No names and no off-chain identity; an address can be suppressed on request.
Skill cohort
The top or bottom decile of a published skill rule at its formation date. Labelled past performers (unvalidated) until the pre-registered out-of-sample study passes.
Descriptive cohort
Accounts grouped by size, vault leadership or a seed list. Labelled descriptive. These are never tested as skill and never named as such.
Performance on the observed venue
Every per-wallet number covers Hyperliquid perps on the first perp dex only. Spot, other perp dexes and other venues are not observed, so a wallet's whole book can differ.
Zero, missing, stale
Every row carries a state. Observed means measured, and a measured zero is an observed 0. Missing is a blank with a stated reason, never a zero. Stale means present but older than its bound.
Coverage status
complete, partial, stale, missing or unreadable, computed when the response is served, so a cached copy turns stale on its own.
Validation status
descriptive → accruing → interim read → validated or failed; a validated rule that fails two consecutive re-tests becomes lapsed. Failed studies stay published.
Limits, stated before you rely on anything here

One venue. Only Hyperliquid perps are observed. A wallet that hedges elsewhere looks one-sided here.

Truncated history. Hyperliquid serves only each wallet's 10,000 most recent fills, so very active wallets carry history_truncated, their evidence is graded insufficient, and they stay in every universe and denominator rather than being dropped.

Not validated yet. The persistence test accrues one weekly window at a time and decides only after the pre-registered number of windows shown in the ledger above. Until then this is a description of past performance, not evidence of skill.

Independence is an estimate. Addresses are merged only on ownership evidence (sub-accounts, direct transfers), so the estimate never overstates independence; trading together is never treated as common ownership.

On-chain activity (BSC, Avalanche). Swap legs are stored without a direction, so no buy or sell is claimed. The honeypot probe is an estimate from an AMM round trip on our own node that does not execute the token's transfer hooks. No safety score, rating or recommendation is published.