Brierz rates the records of Hyperliquid vaults, so luck does not pass for skill.

What Brierz is

A vault’s return tells you what happened. It does not tell you whether the record is evidence of skill or a run of luck. Brierz asks that second question of every Hyperliquid vault and answers with a score from 0 to 100, a 90% range and a tier, or with a reason when the record can’t support an answer yet.

Most records can’t: 2,674 of the 2,939 vaults we check are not rated. We say so instead of guessing.

The engine

At the core of Brierz is a calculation engine. It replays a trading record trade by trade and asks one question: does the record hold evidence of skill, or could luck explain the result? It merges repeated orders into decisions, so splitting one idea into a hundred fills adds no evidence.

It measures return for the risk taken, consistency over time, position sizing and downside protection on fixed scales, and puts a 90% range around its answer, so a thin record cannot pretend to be a strong one.

The engine reads a record, not a venue. It needs a public trading archive it can replay, and nothing else. Hyperliquid is where we run it today, because its archive is public and complete since 25 May 2025. Vaults are what we publish: 2,939 checked, 265 rated. Trader wallets are already rated with the same engine and not published yet. Other venues can follow wherever such a record exists.

The skill test

50 means no evidence either way. A record earns a score above 50 when its results are too good, for too long, across too many decisions, to be luck; it falls below 50 when the risk taken was not paid for. Most records cannot answer the question yet, and we say so: 2,674 of 2,939 vaults are not rated, each with its reason.

The test is itself tested. Before publication every score is recomputed from its stored inputs and must come out the same; every method version is run against a fixed set of reference cases before release, and the stored inputs are recomputed under the old and the new version so every difference is reviewed. The method page sets out the controls.

Brierz Lab

The Lab is our test bench, built on the same engine. It treats a trading strategy the way the ratings treat a vault. Strategies run on Hyperliquid’s own market data with real fees, funding and slippage. Every result is checked for look-ahead, re-run with execution realism and walk-forward, and comes with an estimate of how likely it is to be overfit. Then the strategy is frozen and followed on data it has never seen.

Strategies start from templates, from rules, from Pine Script, or from a wallet’s real trades replayed as a strategy. The Lab is in closed beta while we use it to test the method itself.

Independence

No vault asks to be rated and no vault pays for it. Nobody can pay to change a rating, to be re-rated or to be left out. The method is versioned and fixed within a version; every change and every correction is dated in the method’s change log. Vault leaders can reply, and a verified reply will be shown next to the rating.

Data

Public data only: the Hyperliquid API and its trade archive since 25 May 2025. Returns are the value of $1 since the vault’s start, after fees and funding; depositors’ money moving in and out is left out, so inflows never look like profit. Vault names and descriptions come from Hyperliquid and are not checked by us.

What we don’t do

• We don’t manage money, take deposits or run vaults.
• We don’t recommend where to deposit. A rating describes the past; it is not a forecast and not investment advice.
• We don’t rate people or wallets. Vaults only, for now.
• We don’t publish the method’s internals, so that nobody can trade against the rating instead of trading well.

Who

Brierz is built by a small independent team. Write to [email protected] for anything: a question about a rating, a correction, a story you’re writing, a product you’re building.

Citing a rating

Name the vault and its address, the tier, the score and the 90% range, the method version and the data date, and link the page. For example:

Cold Process (0xc3b1…4b11): Above average, score 70.9, 90% range 58.4–74.1. Brierz, method 0.9, data as of 23 Oct 2026, brierz.com/v/0xc3b1…4b11