Analytics uses the same qualified universe as the rest of the site: vaults at least 30 days old, above $10K TVL, open to deposits and standalone. HLP is excluded entirely because it is protocol liquidity rather than a manager expressing a view, and a configurable denylist allows other vaults to be excluded later.

Once per day the qualified universe is ranked by HV Score and that ranking is frozen for the following 24 hours as Top 10, Top 25, Top 50 and All Qualified. Every historical chart uses the cohort that existed at that moment, so a vault that entered the Top 50 on 10 September does not appear in Top 50 history for earlier days. Membership snapshots before 15 September 2026 were reconstructed from the earliest stored scores, which is stated here rather than hidden.

Each position is converted to signed notional (position size × mark price; long positive, short negative) and divided by the vault's TVL, using the same canonical TVL definition as the rest of the site. Exposure above 100% is valid and never capped: it simply means the manager is leveraged.

Analytics measures manager opinion, so vaults are grouped by their leader address where one is known, otherwise by vault address. When one manager runs several included vaults their TVL and signed positions are combined first, and the resulting portfolio counts once.

Manager view gives every manager equal weight after deduplication and answers: what does the typical manager believe? Capital view divides aggregate signed notional by aggregate TVL and answers: where is the capital actually positioned? They measure different things and are never merged.

A manager counts as participating in an asset when exposure is at least 0.5% of their TVL. Dust below that threshold still counts in raw position accounting but not in participation, breadth or conviction.

Directional ratio is absolute net exposure divided by gross exposure, between 0 and 1. It is evaluated on a rolling 30-day window rather than on a single snapshot: below 0.2 is non-directional, up to 0.4 is mixed, up to 0.65 is directional and above that is strongly directional. A vault whose median gross exposure is below 10% is marked low activity instead, and a vault with fewer than 5 days of history or 72 valid snapshots stays unclassified. Classifications are stored per day and never applied backwards.

A larger dollar position does not mean the manager bought. Positioning changes are computed from coin-size differences between snapshots, priced at the current mark, so pure mark-to-market moves are excluded. A manager counts as active when at least one position size changes by 0.5% of TVL, and portfolio-level risk changes use a 5 percentage point threshold.

Deposits and withdrawals are derived per vault-day as the part of the equity change that the vault's own time-weighted performance cannot explain. Flows are attributed to the cohort the vault belonged to on that day, so past Top 50 flows never use today's Top 50. Moves smaller than 0.1% of vault equity are treated as rounding noise.

Position snapshots are taken hourly and never overwritten. A snapshot with zero or missing TVL, or with corrupt position data, is excluded rather than replaced by zero. If an hourly snapshot is missing, the newest earlier snapshot within 90 minutes is used; otherwise that manager does not contribute to that timestamp, which is why chart tooltips show the real manager count. Aggregates are recomputed after each hourly snapshot, and the dashboard shows a delay warning when the newest aggregate is older than expected.

There is no smart-money score, no 0–100 sentiment index, no prediction and no buy or sell signal. Analytics reports what managers did, never why they did it or whether they were right. Nothing here is investment advice; HyperliquidVaults is an independent analytics site and is not affiliated with Hyperliquid.