# Five criteria, a hundred points, one score out of six

The Leverage Ladder desk · Updated: 24 September 2026

Five criteria, fixed on 18 September 2026 and unmoved since, decide the score printed beside every venue
on this site. They were written down before a venue had been read against them, and they are the same
on every plate: what a plate is about decides which venues appear and which figure sets the sequence,
never what a criterion is worth. Every plate that carries a ranking prints the weights above its table
and points back here.

## The five criteria and what each one is worth

Venues with a central limit order book that publish a leverage ladder for the BTC perpetual and expose a public book we can read. Oracle-priced and pool-filled venues are named where they matter but carry no score, because the liquidity criterion measures a book they do not have. The score answers what happens to a levered position once it is open: how far the book carries it, what the round trip costs on the levered notional, who holds the collateral while the position is live, and what the venue's record says. It does not rank the cap itself. The cap, and the position size that cap still holds to, are printed beside every ranking and set the order of the plates built around them, so the two questions stay apart and neither hides inside the other.


| Criterion | Points of a hundred | What it measures |
|---|---|---|
| Depth under the position | 35 | What is standing in the BTC perpetual book inside ten basis points of the mid price, bids and asks added together, in dollars — the median of a day of our own reads rather than any one of them. |
| Cost on the levered notional | 25 | The entry-tier taker fee paid twice, once to get into the BTC perpetual and once to get back out, read in basis points of the position rather than of the margin behind it. |
| Where the fill is recorded | 20 | What a venue's own architecture, withdrawal and audit pages say about a fill — how much of it a chain ever sees, whether the collateral can be taken back with the operator gone, and what has been audited. |
| What the record shows | 15 | Three things a venue either has or has not — audit reports it has published, a bug bounty anyone may claim against, and twenty-four months of public record with no loss of user funds in it. |
| How far the ladder reaches | 5 | How long the market list runs and how far past crypto it goes — perpetual contracts counted on the venue's own interface or documentation, with the asset classes standing beside them. |


The cap carries nothing at all. It is a number a venue publishes rather than a measurement anyone
took, and it sets the sequence on the plates built around it instead.

## What a mark out of ten means, criterion by criterion

Every criterion is marked out of ten before any weight is applied, and the share in each heading is
what that mark is worth once the five are put together.


### Depth under the position, 35 points of the hundred

What is standing in the BTC perpetual book inside ten basis points of the mid price, bids and asks added together, in dollars — the median of a day of our own reads rather than any one of them.

Read it as a ladder of decades. The bottom step is ten thousand dollars resting inside the band and pays nothing at all; each step above it multiplies the depth by ten and adds two points, so a hundred thousand stands at two, a million at four and ten million at six. Ten points is where the ladder ends.


Worked through: edgeX held a median $29.2 million of resting bids and asks within ten basis
points of the mid price on 22 September 2026. That lands a little under half
a step past ten million, so edgeX takes 6.9 of ten.

At a hundred times margin the liquidation price sits roughly one percent from entry, so what stands between the fill and that price decides the outcome more than the cap does. It carries the heaviest weight here because it is the one figure a levered trader cannot read off a fee page, and because the twenty-five points the cap itself would have carried were moved to the thing the cap acts on.



### Cost on the levered notional, 25 points of the hundred

The entry-tier taker fee paid twice, once to get into the BTC perpetual and once to get back out, read in basis points of the position rather than of the margin behind it.

The mark falls half a point for every basis point the round trip costs past the second, so a trip of two basis points or less keeps the full ten and a trip of twenty-two keeps none of it.


Worked through: Aevo charges 0.080% to take, so entering and leaving a position costs
sixteen basis points of its value. Fourteen basis points past the free two cost seven points, which
leaves 3.0 of ten.

Fees are charged on the levered notional and not on the margin, so leverage multiplies them directly: at a hundred times margin a nine basis point round trip is nine percent of the money behind the position, before any funding is paid. It sits below liquidity because the spread between venues on fees is narrower than the spread between them on depth.



### Where the fill is recorded, 20 points of the hundred

What a venue's own architecture, withdrawal and audit pages say about a fill — how much of it a chain ever sees, whether the collateral can be taken back with the operator gone, and what has been audited.



- a chain receives the trade itself, whether batched by a rollup or proved, 4 points
- a chain receives the standing position now and then, and never the trade, 2 points
- the terms describe a way out that still works when the operator answers nothing, 3 points
- a published audit covers the contract the collateral sits in, 2 points
- no exchange account to open, a wallet signature being the whole of it, 1 point


Worked through: EVEDEX writes position data to Arbitrum in batches rather than trade by trade, worth
two points; the deposit contract is covered by a published audit, worth two more; an account opens
from a wallet signature, worth one. No withdrawal that works without the operator is documented, so
nothing is added for one. Five of ten.

A levered position is an open exposure to the operator for as long as it is live, so whether every trade settles on-chain and whether money can be pulled out without asking decides what survives an outage that starts while the position is still open. It sits below cost because decentralisation is not the question a reader of this domain arrived with.



### What the record shows, 15 points of the hundred

Three things a venue either has or has not — audit reports it has published, a bug bounty anyone may claim against, and twenty-four months of public record with no loss of user funds in it.



- published security audits, 2 points each, counted to a maximum of two audits
- a public bug bounty programme, 3 points
- no loss of user funds in the last twenty-four months, 3 points


Worked through: Aster has seven audits on the record, of which only the first two count, worth four
points; a public bounty programme paying up to $200,000, worth three; and no loss of user funds in
the last twenty-four months, worth three more. Ten of ten.

Leveraged accounts are the first casualty of an incident: a halted chain or a stale oracle liquidates them before it touches anyone else. Fifteen points rather than twenty because the record is documentary and backward-looking, not a measurement of the venue as it trades today.



### How far the ladder reaches, 5 points of the hundred

How long the market list runs and how far past crypto it goes — perpetual contracts counted on the venue's own interface or documentation, with the asset classes standing beside them.

Six points at most for the length of the list, all six of them at five hundred markets, and every tenfold step in the count is worth the same share of that six. Past crypto the counting is plainer: a point for each asset class, and four is as far as that goes.


Worked through: Paradex lists 63 perpetual markets, which is four points on
that curve, and three asset classes beyond crypto at one point each. Seven of ten.

A tiebreaker on this domain rather than the question. A trader running a hundred times margin needs depth on a handful of pairs, not five hundred listings, and breadth only matters where the ladder reaches past crypto. Access is left out of the formula altogether: it decides whether an account can be opened, not how a levered position behaves once it is.

## From five marks to one score out of six

The five marks are weighted by the shares in the table above, which add to a hundred, and the
weighted average is then read on a scale running from zero to six. Six rather than ten because a
scale out of ten invites a reader to treat 4.97 as a percentage of something, and nothing on these
plates is a percentage of anything.

Worked through, in words rather than as an expression. Aster's five marks are
6.3 for the depth under a position,
7.0 for cost on the levered notional,
7.0 for where its fills are recorded,
10.0 for what its record shows and
10.0 for how far its ladder reaches. Weight each mark by
its share of the hundred, add the five results together and divide by a hundred: the weighted
average is 7.36 out
of ten. Six tenths of that is 4.42, the
figure printed beside Aster wherever it appears.

Two decimal places, because one does not separate neighbours: Extended and edgeX sit a hundredth
apart. The marks in the table below are printed to one decimal place while the total behind them is
worked out from the unrounded marks, so a total rebuilt by hand from the printed marks can land a
hundredth away from the printed total. Where the two differ, the printed total is the one the
plates use.

Ties are settled by rule and never by preference. Two venues whose totals match to the hundredth
share a place, and the plate marks both as tied. On a plate ordered by a leverage cap, two venues
printing the same cap are separated by the notional that cap holds to, and only if that is equal as
well by the overall score. On a plate ordered by cost, two venues charging the same maker and taker
fee share a place and are printed in alphabetical order.

## Which venues are ranked, and which are named without a place

Ten venues carry a score. Each runs a central limit order book, publishes a leverage figure for the
BTC perpetual and exposes a public book that could be read from outside without an account and
without a trade.

Six of those ten are ranked. Every plate here is sorted on the size a rung holds to, so a venue
takes a place only where it publishes both halves of its top rung: the multiple, and the notional
that multiple survives. Lighter, Paradex, dYdX and Aevo publish the first and not the second — two
percent of initial margin, or five, with no step saying where the number stops — so they keep a
score, stand in the table below, and are ordered on no plate. Four venues advertise a maximum they
never size, and this is the site that says which. Nothing in the arithmetic moved to make that
happen: no weight, no key, no scale, and the cap itself is still worth nothing.

Six more are named on the full list with the score column empty. GMX, Jupiter Perps, gTrade and
Ostium price a trade against an oracle and fill it from a pool or a vault, so there is no resting
book to read and the criterion carrying the largest share of the total would have nothing to
measure. Scoring them here would mean giving them a mark on a figure taken from somebody else.
Orderly is one book shared behind several front-ends, so a read of it describes a front-end rather
than the venue. Drift has been paused since the April 2026 exploit and its fork trades under
another name, so there is no current venue to measure at all.

An empty score column is not a verdict. An oracle venue with a deep pool behind it can be a sounder
place to carry a levered position than a thin order book with a high cap; this series simply has no
instrument that measures the two on one axis, and inventing one would be worse than leaving the
column empty.

## The scores as they stand

Every criterion has a column of its own, headed by the short form the plates use — depth, notional
cost, settlement, record, breadth — so a total can be rebuilt without taking anything here on
trust. The six the register ranks stand below, each with the column that produced every part of its total. The other four carry a score too — it is the size a rung holds to that they do not publish, which is what keeps them off the plates rather than off the formula.

| Venue | Depth of 10 | Notional cost of 10 | Settlement of 10 | Record of 10 | Breadth of 10 | Score of 6 |
|---|---|---|---|---|---|---|
| Extended | 5.8 | 8.5 | 7.0 | 10.0 | 9.6 | 4.51 |
| edgeX | 6.9 | 6.5 | 10.0 | 7.0 | 8.0 | 4.50 |
| Aster | 6.3 | 7.0 | 7.0 | 10.0 | 10.0 | 4.42 |
| Hyperliquid | 5.8 | 6.5 | 7.0 | 7.0 | 9.6 | 3.96 |
| ApeX Omni | 4.5 | 6.0 | 10.0 | 7.0 | 7.7 | 3.90 |
| EVEDEX | 6.6 | 6.5 | 5.0 | 7.0 | 7.8 | 3.82 |


Ladders, fees, market counts, asset classes, audits and bounty pages read on 18 September 2026;
depth measured over the window dated in the last table on this page.

## Which figure sets the order on each plate

The weights never change between plates. The sort does, and each plate says in its own opening line
which figure it follows.

| What sets the sequence | Plates ordered by it |
|---|---|
| The top rung a venue prints, then the position size that rung survives | Home · The ladder · Every venue · By the rung |
| What stood in the BTC book, ten basis points either side of the mid price | At the book · Above the cap |
| What $10,000 costs to put on and take off again at a new account's taker rate | Cost · At 100x |


Two figures each set the order on more than one plate. Measured depth orders two, and both land on
the same five venues, because every ranked venue above forty times margin belongs on both. Cost
orders two more: one takes the venues reaching a hundred times margin, the other the whole
register. The ladder orders four, because the ladder is what this series is for, and one of the four
is the front page, where it sets the table of caps. The overall score sets the sequence on no plate
in the list above: where a plate sorts on a measurement the score rides beside it as a column and is
reconciled in the words over the table, and on the plate named for the rung it is written out in
full. It orders two tables in the series and no more — the one earlier on this page, and the board
that stands under the caps on the front page, which says in its own opening line that the formula
put it in that order.

The two head-to-head plates and the explainer rank nothing. A head-to-head sets its two venues side
by side in the order its title names them, prints both scores and puts neither above the other; the
explainer names venues where a figure needs one and orders none of them.

## Where every number comes from, and when it was read

| Source | What it gives | Read on |
|---|---|---|
| Each venue's own margin schedule, trading terms and documentation | ladders, notional tiers, fees, market counts, asset classes, settlement and withdrawal mechanics | 18 September 2026 |
| Our own reads of each public BTC perpetual order book, every ten minutes | median depth within ten basis points, median spread, price impact at three order sizes | 22 September 2026 |
| Published audit reports, bounty pages and public incident records of the past twenty-four months | the security record | 18 September 2026 |
| [CoinMarketCap, DEX derivatives ranking](https://coinmarketcap.com/rankings/exchanges/dex/?type=derivatives) | open interest and 24-hour volume, each as the venue itself reports it | 22 September 2026 |

Every median printed on a plate carries the number of reads behind it, and that number is not the
same on every venue. Reported volume is never treated as evidence that a book exists: it is printed
as a venue's own claim and nothing more. Depth, spread and price impact are ours, and none of the
three is ever taken from a third party.

Placement on this site is paid for. What the payment does not reach is the arithmetic above: the
weights were fixed on 18 September 2026, before a venue had been read against them, and the sequence on
every plate follows from those weights and from figures that each carry a source and a date. A
figure that can be disproved goes to corrections@aurascopeapp.com with the address of the plate
and a link to the page that shows the current number.

## Plates in this series

- [The ladder](https://aurascopeapp.com/best-crypto-leverage-trading-platform)
- [At 100x](https://aurascopeapp.com/best-crypto-exchange-for-100x-leverage)
- [Every venue](https://aurascopeapp.com/leverage-trading-platforms-list)
- [Method](https://aurascopeapp.com/method)
- [Cost](https://aurascopeapp.com/crypto-leverage-trading-fees-compared)
- [At the book](https://aurascopeapp.com/highest-leverage-crypto-exchange)
- [About](https://aurascopeapp.com/about)

The Leverage Ladder desk, 24 September 2026. Corrections: corrections@aurascopeapp.com.

A plate is paid for.