# How a derivatives venue is scored here

Margin Brief | The Margin Brief desk | Updated: 24 September 2026

## In short

This is the only page on this site that holds the weights, and every table in the brief is built from
them. Six criteria were settled on 22 September 2026, before a single venue had been read against them, and
nothing has moved since. Thirty of the hundred points go to the depth standing behind an exit, because
that is what a margined position is closed into. Each criterion below is given in words and then worked
through on a reading actually taken, so that any total on the site can be redone by hand.

## What the six criteria weigh

The six weights add to a hundred and are the same on every page of this brief. Which venues appear
on a page changes from page to page; what any of them is worth does not.

| Criterion | Weight | What the criterion reads |
|---|---|---|
| Exit liquidity | 30 | Dollars resting within 10 basis points of the mid on the BTC perpetual, bids and asks together: the median of this desk's own readings, and the money an exit is sold into. |
| Collateral and settlement | 22 | What the venue's own documents say becomes of posted collateral: how a trade reaches a chain, whether a withdrawal works without the operator, who audited the contracts holding the deposit, and what opens the account. |
| Cost per round trip | 20 | Two taker fills at the entry tier on the BTC perpetual, one to open and one to close, counted in basis points of the notional they are charged on. |
| Loss record and audits | 14 | The public record standing behind a venue: audit reports on file, a bounty anyone can report into, and whether user money went missing in the last 24 months. |
| Contracts and asset classes | 9 | How far one collateral balance reaches: the perpetual markets on the venue's own list, and how many of the asset classes among them are something other than crypto. |
| Getting a position on | 5 | What a venue asks for before it will accept an order, read off its own onboarding pages and terms, and restated out of ten. |


Why each sits where it does:

- **Exit liquidity.** A margined position is closed at market when it moves against the trader, so the depth resting within 10 basis points of the mid decides what the exit actually costs; it is the only criterion measured by us rather than read off a document, and it is the hazard this domain exists to describe.
- **Collateral and settlement.** Posted margin is money handed over before anything goes wrong; how often the venue writes trades or positions to a chain, whether a withdrawal that bypasses the operator is documented, and whether the contracts holding deposits were audited decide what is recoverable if the venue stops.
- **Cost per round trip.** Fees are charged on notional, not on margin, so at ten times leverage a taker round trip costs ten times more against the collateral than the headline rate reads; ranked third because a basis point is recoverable and a failed exit is not.
- **Loss record and audits.** Audit count, a published bug bounty and whether user funds were lost in the last 24 months are the only evidence available about a venue's behaviour under stress, and they are history rather than a promise.
- **Contracts and asset classes.** How many perpetual contracts, and how many asset classes beyond crypto, can be margined from one collateral balance decides whether exposure can be moved without moving venue; useful, but it changes the opportunity set rather than the risk of a position.
- **Getting a position on.** What a venue asks for before it lets an order through, and how wide its restricted list runs, decides whether the trader can trade at all, but it says nothing about how the position behaves once open, so it carries the least weight.

## How six marks become one total out of 15

Each criterion is marked out of ten on its own terms, and the six marks are then combined in the
proportions in the table above. The combined mark, still out of ten, is restated on this brief's own
scale, which runs from 0 to 15 and is printed to one decimal place. Nothing else
enters a total: no adjustment, no editor's mark, no allowance for a venue being new or large.

Worked through on edgeX as it was read on 18 September 2026:
6.9 for exit liquidity,
10.0 for collateral and settlement,
6.5 for cost per round trip,
7.0 for the loss record,
8.0 for contracts and asset classes and
8.0 for getting a position on. Taken in the proportions above,
those six come to 7.7 out of
ten, and on a scale of 0 to 15 that is 11.5.

## Each criterion in words, with a reading worked through

### Exit liquidity, 30 points

Ten thousand dollars of resting depth scores nothing; every tenfold step above it is worth two more
points, and the mark stops at ten. A hundred thousand is worth two, a million four, ten million six.

The edgeX book held a median $29.3 million in the polling window that closed on
22 September 2026, close to
three thousand times the floor, and the criterion returns
6.9. Paradex held $55.00 in the same window,
about three and a half times the floor, and returns 0.0.

### Collateral and settlement, 22 points

Four things, all answerable from documents a reader can open. Writing every trade to a chain, whether
in a rollup batch or inside a validity proof, is worth four points; writing only position snapshots
from time to time is worth two. A withdrawal the venue documents as working without its operator adds
three. Deposit contracts audited by somebody else add two. An account opened with a wallet signature
rather than a registration with the venue adds one.

Lighter proves each batch on Ethereum, documents a priority exit queue there for the case where its
sequencer stops answering, has had the contracts holding deposits audited and opens an account from a
signature: 10.0. Extended settles every trade on Starknet and
matches Lighter on audited deposit contracts and on sign-in, but documents no route out that does not
need the operator, so it stops at 7.0.

### Cost per round trip, 20 points

The taker rate a new account pays on the venue's own interface, charged twice — once to open, once to
close. Two basis points or less across that pair scores ten; every further basis point costs half a
point, and the mark is exhausted at twenty-two. Cashback, rebates and volume tiers are left out,
because a table cannot show whether a given reader would earn them.

Hyperliquid's entry tier is 0.015% maker / 0.045% taker, which is nine basis points across an opening
and a closing trade, and returns 6.5. Lighter's standard
account is charged nothing at either end, so its round trip is zero and it takes the whole ten.

### Loss record and audits, 14 points

Three published facts, and no judgement about a venue's character. Independent security audits are
worth two points each and stop counting at two, so four is the most any venue takes for them. A
public bug bounty adds three. No loss of user funds in the last 24 months adds three.
Money counts as lost when an exploit takes it, when a vault the venue itself runs is manipulated out of it, or when an outage swallows it, and in each case it was not paid back in full. A position the market closed out is not counted here: that is the contract doing what it says.

Extended has two audits behind the contracts it uses, runs a bounty paying up to $500,000 for a
critical report and lost no user money inside the window:
10.0. Aevo has five audits on file, which still counts as the
same four, publishes no bounty, and loses the third component to a December 2025 exploit of vaults
the same team ran earlier — a secondary report, marked as one wherever it appears:
4.0.

### Contracts and asset classes, 9 points

Two parts. The number of perpetual markets a venue's own interface or documentation reports earns up
to six points on a curve that reaches its ceiling at five hundred markets, so fifty markets is worth
about 3.8 and a hundred about 4.4. Each asset class beyond crypto that can be margined from the same
balance adds a point, counted to a maximum of four.

Aster reports 578 perpetual markets and five asset classes besides crypto, one
more than the criterion counts, so it reaches both ceilings and returns
10.0. Paradex reports 63 markets and three
classes beyond crypto, and returns 7.0.

### Getting a position on, 5 points

Five points before scaling, then doubled so this criterion is marked out of ten like the rest. No
identity verification required before an order goes through is worth two. An account opened with a
wallet signature is worth one. A restricted list confined to sanctioned jurisdictions and one or two
others is worth two, and a list running wider than that is worth one.

Every venue scored in this brief asks for no identity documents and opens from a wallet, and every
one of them publishes a restricted list running wider than sanctioned jurisdictions alone. That is
four points before scaling and 8.0 after it, the same figure
for all eight. A criterion that separates nobody in this field is a criterion that should carry
almost no weight, which is why this one carries five.

## What has a book to read, and what does not

The scope was settled with the weights. What is in it: Venues that run a central limit order book for perpetual futures, whether the book is matched off-chain or held on a chain, and that this brief has taken into its standing coverage: eight of the sixteen on file, deliberately including the thinnest books rather than dropping them. Coverage is an editorial decision, closed with the weights and listed on this page; it is not part of the arithmetic. ApeX Omni and dYdX are on file and outside it in this edition, read and quoted wherever they bear on a point. Oracle-priced and pool-priced venues are listed with their facts but left unscored, because the depth criterion measures a book they do not have.

Six venues on file are therefore listed with their facts and left without a total. GMX, gTrade,
Ostium and Jupiter Perps price every position from an oracle and settle it against a pool or a vault,
so there is no resting depth to read and the heaviest criterion has nothing to measure. Drift's own
programme has been paused since 1 April 2026, and the fork that carries its documentation trades
4 markets. Orderly runs one book behind several front-ends rather than a venue
of its own, and it was not among the books this desk polled.

Scoring them anyway would mean entering a zero where a measurement is missing, which reads as a
finding and is not one.

## How the order-book readings were taken

Depth, spread and price impact all come from one window. Public order-book endpoints were read from
a single machine over 122 polls between 21 September 2026 and 22 September 2026, on each venue's BTC perpetual contract. None
of it required an account: these endpoints are open to anyone, so nothing was deposited and no order
of ours reached a book.

Three quantities were taken from each snapshot, and each figure printed on the site is a median over
every snapshot that returned a usable book — not an average, and never a single reading:

- **Depth.** Dollars of resting orders inside a band 10 basis points wide on each side of the mid
  price, the two sides added together.
- **Spread.** The distance between the highest resting bid and the lowest resting ask, divided by
  the mid price.
- **Price impact.** The distance between the volume-weighted price at which a market order of a
  stated size would have filled and the mid price at that moment, taken on a buy and on a sell and
  averaged. Where the book could not fill the order at all, the page says it was not filled instead
  of printing a number.

121 snapshots stand behind a typical venue's medians. Two venues in this
edition answered far fewer, Lighter and Extended at 50 attempts of
122, and the readings tables print the count beside every median so
a thinner basis is visible. A reading describes one book over the window named beside it, not a
property the venue owns; the same book read in a rougher hour would give a different number.

## Measured, reported, and why the two are never mixed

Two kinds of figure appear in this brief, and they are kept apart deliberately. Depth, spread and
price impact are measurements of a public book, taken by the method set out above and printed with
the window they were taken in. Everything else belongs to somebody else and is reproduced with the
name of the source and the day it was read.

| Source | What it supplies | Read on |
|---|---|---|
| Each venue's own documentation, terms and margin schedules | fees, market lists, leverage ladders, settlement mechanics, withdrawal routes, restricted jurisdictions | 18 September 2026 |
| The auditors' own reports, the venues' security pages, and CertiK Skynet where a venue publishes nothing of its own | audit counts and their scope, bug bounty programmes and their limits | 18 September 2026 |
| CoinMarketCap, decentralized exchange derivatives ranking | open interest and 24-hour volume, each as the venue itself reports it | 22 September 2026 |
| Venue statements, and news reports named as secondary where no statement exists | the public record of outages, exploits and repayments | 18 September 2026 |
| The order-book reading behind this brief, method above | depth, spread and price impact | 22 September 2026 |

Only the last row reaches a total. Open interest and 24-hour volume are printed where they bear on a
point and score nothing, because what a venue reports as traded is not a reading of what is resting
in its book. No measurement is used to correct somebody else's figure, and no reported figure is
ever printed here as though it had been measured.

## Level totals, rounding, and what a tenth of a point is worth

Two venues that round to one total hold the place jointly rather than being separated by hand. Where that happens a table shows the same place twice and says so, and
the sequence inside a shared place is alphabetical and should not be read as an order.

One decimal place is finer than the evidence behind it. A tenth of a point separates venues whose
measured books differ by a few per cent, or whose fee schedules differ by a fraction of a basis
point, and it is better read as level than as ahead. The differences worth a reader's attention on
this site are whole points, and they come almost entirely from the two heaviest criteria.

## What is bought on this site, and what is not

A methodology is worth reading only if it says who paid for it. This brief exists because a venue
named in it paid for it to be published, and nothing here is offered as independent judgement — not
this page, not a table, not a total.

What the money does not reach is the arithmetic above. Every weight on this page predates the first
measurement taken for the brief, and one set of them runs every table in it. Nor does it buy a mark
on the page: no highlighted row, no badge, no colour kept for one venue, no ordering done by hand. A
venue's place is whatever the six criteria return, and every venue in a comparison, the one at the
head of it included, is given at least two documented shortcomings, each taken from that venue's own
documents or from the record of what has happened to it. A different set of weights would give a
different order — which is why this set is dated and printed ahead of the tables it produces.

This page belongs to a paid placement. Every figure on this site carries the date it was taken and the source it came from, the order of every table comes from one published formula, and nothing here is advice or cover against a loss.

The Margin Brief desk, 24 September 2026. Corrections: editorial@insurancerefocused.com.

Placement in this file is paid for; the order of every table follows the published formula and nothing else.