CEX Perp Volume vs Observable Depth, Sep 2026

What this report measures

PerpFinder collects two independent measurements every six hours. The first measurement is the 24-hour perpetual volume that each venue reports on its own ticker endpoint. The second measurement is the visible order book of the venue, walked by the PerpFinder cost engine at standardized order sizes. The volume-quality dataset stores both measurements together with the sweep clock. The full rule set is on the volume-quality methodology page.

This report uses three measurement windows. It uses the sweep stored at 2026-09-04 18:00 UTC, the 120 sweeps stored between 2026-08-06 00:00 UTC and 2026-09-04 18:00 UTC, and two live cost-engine walks made on 2026-09-04. Each table names its own instant, because the three clocks are different.

The dataset covers the flagship linear BTC and ETH perpetual contract on each venue. Spot markets form a separate dataset and never enter these rows. Every stored row used here carries volume-quality methodology version 2. Version 1 rows stay under their own version and are never merged into this series.

What the ratio does and does not mean

The reported-to-support ratio divides the reported BTC and ETH perpetual volume of a venue by the largest standardized order that its book absorbed in the same sweep. A high ratio means the venue reported a large amount of turnover next to a small measured order size.

A low or high ratio is a measurement of visible depth against reported turnover. It is not a finding of misreporting. This report does not claim that any venue misreports its volume, and the dataset makes no such claim either.

Ordinary market structure produces high ratios. Market makers recycle the same inventory many times each day, so real turnover can exceed the depth that is visible at any single instant. A venue can also keep size off the visible book and add it when a large order arrives. The ratio therefore measures what a taker can see, and not what a venue can eventually fill.

Observation 1: the standardized ladder stops at $1 million

All 20 venues in the sweep absorbed the top rung of the standardized ladder, so each one recorded a largest sufficient fill of $1,000,000. Across the 30-day history, 2,373 of 2,400 stored venue rows carry the same $1,000,000 figure, which is 98.9 percent.

The denominator of the ratio is therefore almost constant. In this sweep the ratio ranks venues by reported main-market volume alone, from 406 at Phemex to 30,822 at Binance, with a median of 7,649. A reader who wants a depth ranking must use the larger order sizes in the cost comparison tool.

Five of the 20 venues produced no ratio at all. Those venues publish no per-symbol breakdown that the feed can use, so their main-market volume stays null and their data confidence is 80 instead of 100. A missing value stays null and never becomes zero.

Observation 2: depth separates the venues at larger sizes

The cost engine walked the BTC perpetual book at two larger sizes on 2026-09-04. At $10,000,000, 16 of the 20 venues filled the order inside the price band of the engine. Two venues did not hold enough visible depth, and two more reached the end of the depth page that their API returns.

At $50,000,000 the picture changes. Seven venues filled the order, six ran out of visible depth, and seven reached their API depth limit. The venues that did not fill $50,000,000 reported $158.6 billion of 24-hour volume, which is 61.3 percent of the sweep total. At $10,000,000 the equivalent share is 10.0 percent.

An API depth limit and a shallow book are different results, so this report keeps them apart. A venue that reaches the end of its depth page may hold more size than the walk can measure.

Observation 3: a single sweep does not describe a venue

The stored history holds 120 sweeps between 2026-08-06 00:00 UTC and 2026-09-04 18:00 UTC. Fifteen venues produced a ratio in every one of those sweeps. No venue kept its ratio inside a 2× band across the window.

The tightest venue is Toobit, which moved between 3,342 and 21,478, a factor of 6.4. The widest venue is Bitmart, which moved between 743 and 34,449, a factor of 46.4. The median venue moved by a factor of 24.3 between its own minimum and its own maximum.

The practical rule follows from those numbers. Use the median across many sweeps to compare venues, and treat any single sweep as one moment in a moving series.

Observation 4: reported volume above the 3× spike rule

PerpFinder flags a venue when its reported 24-hour volume passes 3× the median of its own previous seven captured days. A flagged venue stays visible and leaves the headline market total until the figure holds.

In the 30 daily captures from 2026-08-06 to 2026-09-04, 24 venue-days passed the rule across 14 venues. The reported-volume feed carries the centralized venues plus Hyperliquid and dYdX, so both venue types can appear. A cluster of 21 venue-days falls on 2026-08-20, 2026-08-21 and 2026-08-22, which points to a market-wide move rather than a single-venue event.

Three venue-days stand on their own: WEEX on 2026-09-04 at 11.9×, WhiteBIT on 2026-08-16 at 3.9×, Hyperliquid on 2026-08-25 at 3.7×. WEEX carried the live flag at 2026-09-04 21:22 UTC. A flag marks a figure that needs a second day to confirm it, and nothing more.

Limitations

Each venue API returns a limited depth page. When a walk reaches the end of that page, the row records the limit, and the venue may hold more size than the measurement shows.

The engine walks the buy side only. A one-sided walk does not describe the sell side of the same book, and the two sides can differ.

The sweep cadence is six hours, and each cost-engine walk is one instant. Depth changes between calls, and a later call returns different numbers.

Reported volume comes from the ticker endpoint of each venue. PerpFinder normalizes the unit and the market type, and does not audit the venue figure. The order-book series starts on 2026-07-16, so 30 days is the longest window this report can use.

How to reproduce these numbers

Every number comes from public endpoints that need no key. The sweep table comes from https://perpfinder.com/api/data/volume-quality. The 30-day table comes from https://perpfinder.com/api/data/volume-quality-history?days=30. The fill table comes from https://perpfinder.com/api/data/slippage?asset=BTC&size=10000000 and the same endpoint with size=50000000. The spike table comes from https://perpfinder.com/api/data/volume and the stored daily captures.

Those endpoints serve the current state, so a later call returns later numbers. Each table on this page carries a CSV with the exact rows measured for this report, and the fact block carries a CSV of every headline figure.

Next update

PerpFinder builds each report by hand and keeps no fixed publication schedule for this series. This page is updated when the next report is published. The sweeps continue every six hours in the meantime, and the API endpoints stay current between reports.