How the screen works — and what it does not do.

The criteria, as applied today

These values are read live from the same configuration the pipeline uses, so this page cannot drift from the screen. Thresholds change rarely, never for a single vault, and every change is recorded in the change log.

Two further checks are applied before the thresholds: the vault must be open to outside depositors (sub-vaults that a parent vault allocates capital to are listed under their parent, not on their own), and it must have moved in the last 30 days — a vault whose share value has barely changed in the trailing 30 days is dormant, whatever its history says.

One rule guards against flicker: entry always requires the full threshold, but a vault already on the screen leaves it only when it misses a threshold by a clear margin. Without this, a vault sitting exactly at a limit would enter and leave with every run, and each move would look like news.

Not appearing is not a judgement. Many good vaults are simply too young, too small, or had one bad month. They reappear the day they qualify.

Why our returns differ from the platform's numbers

Most platforms show either the vault's account value over time, or a 30-day return multiplied by twelve. Account value includes deposits and withdrawals, so a vault that attracted new capital looks like it earned it, and a vault that saw a large withdrawal looks like it crashed. A thirty-day annualisation turns one lucky month into a headline.

We measure what a single share would have earned. For each interval, the vault's profit or loss is divided by the capital it managed just before, and those interval returns are compounded. Deposits and withdrawals drop out. Where a platform already publishes a share price, we use it directly. This is the same convention used for mutual funds, and the one in which two vaults can be compared on equal terms.

What goes into the score

The score is a number from 0 to 100 that summarises return per unit of risk. What goes in is listed here; how it is weighted and capped is ours and is not published.

• Return set against downside volatility, total volatility and maximum drawdown, each computed on daily returns so that vaults sampled at different frequencies are measured the same way. Drawdown uses every observed point.
• Ceilings for tail risk: deep drawdowns cap the score, with the cap phasing in gradually rather than jumping at a line.
• A plausibility ceiling: very high annualised returns lower the score rather than raise it. Extreme returns are treated as a warning, not an achievement. Very short histories are not annualised at all.
• A smoothness check: a curve that only ever goes up, with almost no variance and almost no losing days, is capped until explained. Real strategies have bad days.
• Short histories pull the score towards neutral rather than rewarding or punishing it.
• Lock-ups and performance fees apply small deductions.

Exact bands, weights and cut-offs are deliberately not published, so that a curve cannot be shaped to the score.

The score is one criterion among several. The list opens sorted by score; every column can be re-sorted, and the sort never changes who is on the list.

Three questions the platform pages don't answer

For every vault we publish the same three verdicts, derived from whatever the platform exposes: who runs it (protocol pool or operator, the operator's own stake, the fee), time to your money (days to withdraw normally and under stress, from platform rules and current withdrawable capital), and what it is right now (net direction, gross exposure relative to capital, concentration). Where a platform does not publish an input, the verdict says so rather than guessing. Raw platform fields are never shown; only the classification is.

The two comparisons

Each vault is set against two alternatives over exactly its own history: holding bitcoin, and lending a stablecoin at the largest USDC lending market. The first shows what the vault gave up in a bull market; the second shows whether it earned more than a plain stablecoin lending rate at all. A vault that beats neither has, so far, not been paid for its risk.

Market-wide stress

We keep our own continuous record of liquidations, open interest and funding across the major perpetual markets — Binance, Bybit, OKX and Hyperliquid. Venues differ in how completely they publish liquidations, so the market regime is computed on a single fixed reference series and a fixed basket of large markets, so that a day last year and a day today mean the same thing; the headline figure is the sum across venues. Days in the top few percent of liquidation volume over the trailing year are marked as stress days; the label means "extreme relative to the past year", not a fixed count. For each vault we report the average return on those days, the return in the days after them, the worst stress day, and returns split by regime. Where a platform's long-run series has gaps, the return across a stress day is measured over the shortest window that spans it, and such windows are counted and labelled. A figure is shown once enough stress events are covered; until then the page says how many are. These are descriptive statistics, computed identically for every vault; they are never used to rank.

The bitcoin comparison in detail

Each vault is compared with holding bitcoin over exactly the vault's own history, so a vault that launched in a bull market is not flattered by a longer bitcoin window. We show annualised return, maximum drawdown and correlation. A vault that trails bitcoin while drawing down a tenth as much is a different kind of exposure, not simply a weaker one; the comparison shows both sides.

Where the data comes from

• Hyperliquid — public vault API; share value reconstructed from profit and account history.
• Paradex — public vault API; share price published directly.
• dYdX — public indexer; equity and profit per tick.
• GMX — market state from the public API; GLV prices from the oracle feed, GM pool prices computed from pool value and supply (an approximation that ignores open trader profit and loss, labelled as such).
• Jupiter, Gains — share prices read from the protocols' own endpoints and contracts.
• Bitcoin — daily closes from public exchange data.

Where a platform's history is shorter than ours, we extend it with our own daily observations, which began in September 2026 and are never edited after the fact.

What the daily hash proves

Every snapshot is stored in a canonical form (sorted keys, no whitespace, UTF-8) and hashed with SHA-256 over the file exactly as served. The hash is published in the footer and kept for every day. Anyone can download a day's snapshot, compute its hash with standard tools and compare; the Verify page does it in your browser. This shows that the numbers were not changed after publication. It does not prove they were correct; if we find an error, we publish a correction as a new snapshot rather than editing the old one, and list it under Corrections.

Consistency, cross-checks and the change log

Beyond headline return we report how it was earned: rolling 30- and 90-day returns, the share of positive months, the best and worst month, and how much of the total came from the single best month. Where a platform publishes a comparable figure — Paradex's own drawdown, GMX's realised APY by period — we show ours next to theirs and flag disagreement beyond tolerance rather than choosing one. Every run is compared with the previous one; vaults entering or leaving the screen, changes of exposure style, and market regime changes are appended to a log that is never edited.

Depositor concentration excludes known contracts — vault-of-vaults, staking and router contracts hold shares on behalf of many users and would otherwise appear as a single large holder.

What the tags mean

Short labels next to a vault or a value carry a fixed meaning throughout the site.

Conflicts of interest and sponsorship

We do not operate vaults and do not accept payment from vault operators. No vault can pay to be listed. During the current free phase nothing on this site is for sale; should a labelled sponsor slot be offered later, it will never be sold to a vault operator and will have no influence on any figure. The people behind PerpVault trade on some of the covered exchanges and may hold positions in vaults that appear here; any such position is listed below. Thresholds are fixed in advance and applied by an automated pipeline, so an individual vault cannot be favoured or penalised by hand.

Positions in listed vaults: none declared at this time.

Partner placement: the KDA community placement on the front page is unpaid; we receive nothing for it. The publisher of this site holds KDA; more features built with the Kadena Community Edition will follow. Neither has any influence on which vaults appear or on any figure.

This site provides data, not advice. Nothing here is a recommendation to deposit into, or withdraw from, any vault.