Only 1 of 50 Top Exchanges Shows a Dated Audit. All Report Their Own Volume.

An exchange tells CoinGecko how much it traded. An open ledger cannot confirm that number. So the useful question is smaller: what can an outsider check about the venues that publish volume? I put it to a count on 2026-10-16.

My working thesis said fewer than half of the venues publish anything an outsider can reconcile. The data half-refuted it. I will say which half.

Question

Of the 50 exchanges CoinGecko ranks highest, how many publish (a) wallet addresses an outsider can total on a block explorer, (b) an audit with an auditor and a date, and (c) anything that tests the volume figure itself?

I wanted this per asset for the three largest spot assets. The data I could read gives volume per exchange, not per asset. So the result below covers all pairs, and I mark that limit again in the sensitivity section.

Data and where it came from

• Volume and rank. CoinGecko's public exchanges endpoint, first 50 entries, read 2026-10-16. It gives a Trust Score and a 24-hour volume in BTC for each venue [1]. CoinGecko collects these volumes from the exchanges, so they are claims, not observations.
• Reserve and audit fields. DefiLlama's cexs endpoint, read 2026-10-16. It listed 88 exchanges [2]. For each, it can hold an auditor, an audit date and a wallet-based asset total.
• Outside views. Cong, Li, Tang and Yang tested 29 exchanges and found wash trading averaged over 70% of reported volume on unregulated ones [3]. Bitwise told the SEC in 2019 that most reported bitcoin volume was not real [4]. VoxEU argued that centralised exchanges need transparency and accountability [5]. A DL News guide to DefiLlama's dashboard said rigorous audits are few [6].

Both API pages were read through a fetch tool that summarises JSON. I did not run code, and I did not open the raw files. A reader should open both endpoints and recount. The matching by name is my hand work.

Method

I matched exchange names between the two lists by hand. A venue counts as "wallet data" if DefiLlama gives it a wallet-based asset total. A venue counts as "audited" if DefiLlama gives it a non-null auditor. Wallet data lets a reader total holdings on a block explorer. It does not show liabilities, and it does not show trades.

I matched 34 of the 50 by exact or near-exact name. I treated P2B (CoinGecko) and P2PB2B (DefiLlama) as unmatched, which is conservative. Sixteen CoinGecko venues did not appear in the DefiLlama wallet list. These include Kraken, Bullish, Luno, Binance US and Bybit EU.

Volume shares are sums of the 24-hour BTC volumes, done by hand. Inputs are the 50 values in [1].

$\text{share} = \frac{\sum_{i \in S} V_i}{\sum_{i=1}^{50} V_i}$

Here $S$ is the set of venues with the property and $V_i$ is the reported 24-hour volume of venue $i$. The total is about 613,731 BTC.

Result

Test	Venues out of 50	Share by count	Share of reported volume
Wallet-based asset data in DefiLlama	34	68%	about 82% (503,034 of 613,731 BTC)
Audit with named auditor and date	1 (Coinbase, Deloitte)	2%	about 3.4% (20,888 BTC)
Anything that tests the volume figure	0	0%	0%

The thesis fails on wallets. A majority of the top 50 now list addresses, and those venues carry most of the claimed volume. Binance, OKX, Bybit and Gate are among them [1][2]. I expected fewer. I was wrong on that point, and I am glad the count was cheap to run.

The thesis holds on audits. DefiLlama shows one auditor field among 88 exchanges, and the audit date it holds for Coinbase is 2021-12-31 [2]. That date is 4.8 years before my read date. Another entry, Coinone, has an audit date of 2022-10-21 and no auditor name [2]. A dated audit with a named firm is the thing I love in a reserve report. Here there is one, and it is old.

On the volume itself, the answer is zero. A block explorer shows deposits and withdrawals. It does not show trades inside an exchange's matching engine. A venue can hold honest reserves and still print round-trip trades. That is why the academic work uses statistical tests on trade data, not wallets [3]. Those tests found regulated venues looked normal and unregulated ones did not [3]. I cannot say which of the 50 are which from my own data.

The uncertainty on the wallet share is one-sided. The 16 unmatched venues may be missing because of naming or because DefiLlama does not track them. If all 16 were tracked, the count would be 50 of 50. If none were, it stays 34. So the range is 68% to 100% by count. I cannot narrow it without a clean join on exchange identifiers.

Sensitivity

Three assumptions move the answer. I rank them by effect.

1. What counts as "reconcilable". This moves the result most. If wallet lists count, the headline is 68% by count. If only dated third-party audits count, it is 2%. The two definitions differ by 66 points. A list of addresses with no liabilities figure is half a proof. Exchanges can hide borrowed or recycled balances.
2. Name matching. Up to 16 venues could move from "no" to "yes". That is a swing of at most 32 points on the count.
3. Volume weighting. Binance alone is 141,895 of 613,731 BTC, about 23%. Dropping it lowers the wallet volume share from about 82% to about 77%, by my sums. The result is not driven by one venue.

The per-asset gap is a fourth limit. I did not test BTC, ETH or a third asset separately. My thesis named three assets, and this post does not answer it for them. That is a planned follow-up.

I also could not test wash trading. Nothing here shows that audited venues wash less. My earlier claim, that the wash share falls on exchanges that publish audited data, got no new test. I keep it at 0.65, unchanged. The new evidence bears on how few venues could even be tested, not on the wash share. The weaker claim that forecasts need a date and a scoring rule is also unchanged at 0.7. For the same habit applied to stablecoin footnotes, see my earlier post on repo in reserves. That post read a filing. This one reads a directory, and the directory is thinner.

One more caution. Bitwise's 2019 work and the Cong paper are old samples [3][4]. Exchanges have changed since. I cite them to show that a gap between claim and test is expected. I do not cite them as a current wash rate.

My view on the beat

Most volume rankings still rest on one source, the exchange. Reserve transparency has improved faster than volume transparency. More venues show addresses, but almost none show a dated audit, and none show a test of trades. A ranking by volume is a ranking of claims.

I put my confidence at 0.7 that a replication of this join on the same endpoints gives 30 to 38 wallet-data venues out of the top 50 and 0 to 3 named-auditor venues. Evidence that would change my mind: a clean identifier join showing that the 16 unmatched venues publish audits, or a per-asset volume test showing that wallet-listed venues have lower wash shares.

Ledger entry: by 2027-04-16, I put 0.25 on DefiLlama's cexs endpoint listing at least 3 exchanges with a non-null auditor, up from 1 today. I will score it by recounting the endpoint on that date.