# How we rate decentralized exchanges

Method sheet | Chain Sheet | Updated: 24 September 2026

Five criteria share a hundred weight points between them. They were written down on
22 September 2026, before a single venue had been scored, and every sheet
published here has used them since. A total runs from 0 to 12 and is printed to
1 decimal place. What follows is all of it: what each criterion reads, one
score worked through from its inputs to the figure in a venue's row, the inputs behind every other
total, and the things the method is deliberately not allowed to read.

Sections on this sheet: The five criteria and their weights · One score worked through · Where this sheet draws the line on an incident · Every input behind a score · What the weights leave out · Which venues carry a score · What decides which venues appear, and in what order · Equal totals and rounding · Revisions.

## The five criteria and their weights

| Criterion | Weight | What it reads | How that becomes points out of ten |
|---|---|---|---|
| Settlement and custody | 30 | Read off the venue's own architecture, custody and withdrawal pages, and off the audit reports for the deposit contracts — marked line by line, never measured. | the chain ends up holding every fill, batched by a rollup or carried in a validity proof — 4; the chain ends up holding position balances only, written at intervals — 2; the documentation sets out a withdrawal the trader can drive through with the operator silent — 3; the contracts a deposit lands in carry an outside audit — 2; the account is a wallet signature and nothing more — 1 |
| Measured book depth | 25 | The only criterion here that is measured rather than read — dollars resting within 10 basis points of mid on the BTC perpetual, bids and asks together, given as the median of our own readings. | The column is read as a ladder of tenfolds. Ten thousand dollars resting is the bottom rung and scores nothing; each tenfold above it adds two points, which puts a hundred thousand at two, a million at four and ten million at six. The ladder stops paying at ten. |
| Round-trip taker cost | 18 | Entering a BTC perpetual at the entry-tier rate and leaving it again — the taker fee counted twice, filed in basis points. | Half a point comes off the mark for each basis point the round trip costs above two. A round trip of two or less therefore keeps the full ten, and one of twenty-two keeps nothing. |
| Audit and incident record | 15 | Three documentary lines — the audit reports a venue publishes, the bounty page it keeps open, and what the public record holds against it over 24 months. | a published audit report, two points apiece, the count stopping at the second — 4; a bounty page outside researchers can file against — 3; the past 24 months clear of any loss of user funds — 3 |
| Markets and asset classes | 12 | How much a trader can keep on one venue — perpetual contracts counted off its own interface or documentation, and the asset classes standing behind them outside crypto. | Six of the ten points sit on the contract count and are paid on a curve: five hundred perpetuals takes the whole six, fifty takes a little under four. The other four are handed out one at a time, one per asset class outside crypto — shares, currencies, commodities, indices, pre-listing contracts — and four is where the handing out stops. |


Two of the five are awarded from a list. Settlement and custody, and the audit and incident
record, hand out the points written against each line, and the lines add to ten: a venue that
does everything on its list takes ten, a venue that does none of it takes nothing. The other
three read a figure and turn it into points on a stated curve. Each criterion is marked out of
ten on its own terms, the five marks are combined in proportion to the weights above, and the
result is stretched onto the scale of 12 that every sheet prints.

The order of the weights is a judgement, and it is the same on every sheet here. Settlement and
custody comes first because the word decentralized is a claim about where a trade lands and who
can stop a withdrawal, not about how a trader signs in. Measured depth comes second because a
book with nothing resting in it cannot be the answer to where to trade on-chain, whatever the
documentation says. Round-trip cost comes third: it still separates these venues by several basis
points, which is real money on size, but it is also the figure a venue can change in an afternoon.
The audit and incident record comes fourth, because on a venue signed into with a wallet the
contract is the counterparty and no helpdesk reverses a loss. Markets and asset classes comes
last, because listing more contracts does not make a venue more decentralized.

## One score worked through

ApeX Omni scores 8.6. A weighted total is worth nothing to a reader who cannot
take it apart, so here is that one, criterion by criterion, on the inputs printed in its own row.

| Criterion | What the inputs say | Points out of ten | Weight | Contribution |
|---|---|---|---|---|
| Settlement and custody | Matched trades are proven to contracts in a rollup batch, the documentation describes a withdrawal forced through those same contracts, the contracts holding deposits are audited, and an account opens on a wallet signature. | 10.0 | 30 | 300 |
| Measured book depth | A median of $1.7 million resting within 10 basis points of the mid price on the BTC perpetual, across our snapshots. | 4.5 | 25 | 112.5 |
| Round-trip taker cost | 0.050% to take liquidity, which is ten basis points in and out again. | 6.0 | 18 | 108 |
| Audit and incident record | 4 published audits, which the list counts to its limit of two; no public bug bounty; no loss of user funds in 24 months. | 7.0 | 15 | 105 |
| Markets and asset classes | 125 perpetual contracts, and 3 asset classes beyond crypto. | 7.7 | 12 | 92.4 |


The five contributions come to 717.9. Spread across the hundred weight
points that is 7.179 out of ten, and on the scale of
12 every sheet prints, 8.6 — the figure in the row. Marks are
printed to one decimal, so the contributions above are rounded as well; the arithmetic behind the
sheets keeps the unrounded marks, which moves this total by less than a hundredth.

## Where this sheet draws the line on an incident

One line of the audit and incident record turns on this phrase, so it is defined here rather than
left to judgement. A loss of user funds means money taken in an exploit, money lost in a venue-run vault through market manipulation, or money lost because the venue stopped working, and not repaid in full. Liquidations caused by ordinary market moves are not a loss under this definition.

On 18 September 2026 the definition was met by 3 of the 10
scored venues. Each episode is described in the venue's own row on the sheet where it appears,
together with what the venue did about it afterwards.

## Every input behind a score

Every figure the method reads. Venues are listed alphabetically: this is a register, and the
sheets are where venues are placed against one another. Chain record is what a venue writes to a
chain; forced exit is a withdrawal its documentation says works without the operator; loss is a
loss of user funds inside 24 months, on the definition above; round trip is twice the entry-tier
taker rate. Documents were read on 18 September 2026 and the books from 20 September 2026 into 21 September 2026, a reading every ten minutes. Two lines
of the settlement and custody list have no column here, because every scored venue earns them:
each opens on a wallet signature, and each holds deposits in contracts that carry an audit.

| Venue | Chain record | Forced exit | Audits | Bounty | Loss | Median depth | Round trip | Markets | Non-crypto |
|---|---|---|---|---|---|---|---|---|---|
| Aevo | every trade | not described | 5 | no | yes | $2.0 million | 16.0 bps | 102 | 5 |
| ApeX Omni | every trade | yes | 4 | no | no | $1.7 million | 10.0 bps | 125 | 3 |
| Aster | every trade | not described | 7 | yes | no | $14.4 million | 8.0 bps | 578 | 5 |
| dYdX | every trade | not described | 6 | yes | yes | $829,432 | 10.0 bps | 78 | 1 |
| edgeX | every trade | yes | 6 | no | no | $29.1 million | 9.0 bps | 172 | 3 |
| EVEDEX | positions, periodic | not described | 2 | no | no | $18.7 million | 9.0 bps | 52 | 5 |
| Extended | every trade | not described | 2 | yes | no | $7.6 million | 5.0 bps | 325 | 5 |
| Hyperliquid | every trade | not described | 2 | yes | yes | $8.7 million | 9.0 bps | 324 | 4 |
| Lighter | every trade | yes | 9 | no | no | $16.5 million | 0.0 bps | 214 | 5 |
| Paradex | every trade | not described | 5 | yes | no | $32,552 | 0.0 bps | 63 | 3 |


And the marks those figures produced, each out of ten, in the order the criteria are set out
above, with the total they make on the scale of 12.

| Venue | Custody | Depth | Cost | Record | Coverage | Total |
|---|---|---|---|---|---|---|
| Aevo | 7.0 | 4.6 | 3.0 | 4.0 | 8.5 | 6.5 |
| ApeX Omni | 10.0 | 4.5 | 6.0 | 7.0 | 7.7 | 8.6 |
| Aster | 7.0 | 6.3 | 7.0 | 10.0 | 10.0 | 9.2 |
| dYdX | 7.0 | 3.8 | 6.0 | 7.0 | 5.2 | 7.0 |
| edgeX | 10.0 | 6.9 | 6.5 | 7.0 | 8.0 | 9.5 |
| EVEDEX | 5.0 | 6.5 | 6.5 | 7.0 | 7.8 | 7.6 |
| Extended | 7.0 | 5.8 | 8.5 | 10.0 | 9.6 | 9.3 |
| Hyperliquid | 7.0 | 5.9 | 6.5 | 7.0 | 9.6 | 8.3 |
| Lighter | 10.0 | 6.4 | 10.0 | 7.0 | 9.2 | 10.3 |
| Paradex | 7.0 | 1.0 | 10.0 | 10.0 | 7.0 | 7.8 |

## What the weights leave out

**Access.** There is no criterion for how a venue lets a trader in, because it would not have
separated anybody. Every venue scored here opens on a wallet signature, and none of them asked
for an identity document before a first trade in the pages read on 18 September 2026;
Aster and Extended reserve a right in their terms to verify or to
refuse, and that is as far as the difference goes. A criterion on which every scored venue marks
alike adds weight to each total and moves none of them relative to another. Where onboarding does
separate venues — an account that opens on an email address rather than a wallet, a deposit that
passes automated screening before it lands — the sheet says so in words, in the venue's row and
in its card, which is where a difference of that shape belongs.

**Reported volume and open interest.** Printed on the sheets where they help and scored nowhere:
a venue reports them about itself, and reported turnover is cheap to manufacture in a way that
resting depth is not.

**Leverage ceilings.** Printed too, and scored nowhere either. A higher ceiling widens the range
of ways to lose the margin behind a position; it says nothing about the venue underneath it.

**Incentives.** Points programmes, token emissions and rebates earned on volume pay some traders
and not others, and they change faster than a sheet is rebuilt.

**The interface.** No account was opened anywhere, so nothing here has a view on how a venue
feels to use.

## Which venues carry a score

Venues that run a central limit order book, whether the book sits off-chain with on-chain settlement or inside the venue's own chain. Pool- and oracle-priced venues are listed with their facts but carry no score, because the depth criterion measures resting orders they do not have.

Leaving them unscored is not a dismissal. A pool-priced venue answers some of the five criteria
readily — its contracts are audited or they are not, its markets can be counted — and cannot
answer the depth criterion at all, because a measurement of resting orders has nothing to read
where no order rests. A zero there would be a finding nobody made, and a pass would leave the
venue judged on four criteria against venues judged on five. So its facts go on the sheet, the
score column stays empty, and the row carries the reason.

## What decides which venues appear, and in what order

What reaches a sheet is decided by that sheet's condition and by nothing else — venues that
settle every trade on a chain, venues that charge nothing to take liquidity, venues an
order-book trader might move to. What arranges the rows depends on what the sheet is about: a
sheet built around one figure is ordered by that figure — one we measured, one off a rate card,
one standing on an outside register — and every other sheet is ordered by the total.

| Sheet | Ordered by |
|---|---|
| Front sheet, Best decentralized exchanges, Venue list, Perpetuals for swap traders | The weighted total, highest first |
| Skynet order | The standing score on an outside register — CertiK's Skynet figure, every venue read on one day |
| Settlement and custody | The settlement and custody mark, the total settling equal marks |
| Top venues by depth | The depth reading — dollars resting within 10 basis points of mid |
| Fees compared | The round-trip column — what taking liquidity in and out again costs on $10,000 |


On a sheet ordered by one figure the total stays in the table and settles equal values, and a
figure borrowed from an outside register orders the rows without earning a point anywhere. The
criteria, the weights and the marks are identical on every sheet: a condition changes the company
a venue keeps, never its score.

## Equal totals and rounding

A total is rounded once, at the end, from the unrounded marks, so two venues whose inputs differ
a little can still print the same figure. Venues whose totals are equal after rounding to one decimal share a place and are marked with an equals sign. Inside a shared place the order is alphabetical and means nothing.

## Revisions

The method carries a version number. Version 1.0 was fixed on
22 September 2026, and no criterion, weight, list line or curve in it has
changed since. When one does, the number moves with it and every sheet is rebuilt on the new
version rather than left half on the old.

Inputs move more often than the method does, and the register above is rebuilt whenever a sheet
is, so a score and the figures behind it were always read on the same day. A revision that
changes a placing is noted on the sheet it changes.

Compiled by Chain Sheet editors and revised 24 September 2026. Corrections, quoting the sheet and the document that disagrees: sheet@echoblock.org.

Publication of this sheet is bought.