The Weekly Take: Crypto’s Most-Quoted Number Is the One Nobody Can Check

This week we published two measurements of concentration. One of them we stand behind to the second decimal place. The other we withdrew in public, in a piece that exists mainly to explain why it could not be completed. Here is the moment it came apart: in the exchange-volume ranking the industry quotes every day, BTCC came second and Coinbase — a listed company that files audited transaction revenue with the SEC four times a year — came twelfth. The difference between the two pieces was not effort and it was not method. It was whether a stranger who wanted to prove us wrong could recount the underlying number. Bitcoin’s can be recounted. Exchange volume cannot. And exchange volume is the number the industry quotes.

What actually happened this week

Over the seven days to 4 August 2026, Bitcoin produced 1,060 blocks and three mining pools built 60.75% of them. Foundry USA alone assembled 26.04%, F2Pool 17.83%, AntPool 16.89%. The top five reached 76.89%. Widen the window to a trailing year — 52,437 blocks, 33 distinct pools — and the top three still hold 58.34%, with Foundry at 28.90%. The Herfindahl index reads 1,536 over the year against 1,484 over the week. Whatever else moved in Bitcoin mining across twelve months, this did not.

On the same day we tried to do the equivalent for exchanges. Sixty-one centralized venues reported $33.32 billion of spot volume to DefiLlama over twenty-four hours, and Binance came out at 17.9% of it. That came off DefiLlama’s centralized-exchange endpoint on the evening of 4 August 2026, and the window is a rolling twenty-four hours ending at the moment of the pull, not a calendar day. We re-ran the same endpoint on 6 August and got sixty-one venues again, $31.44 billion, Binance at 17.3%, an index of 630. Nobody can reproduce our exact figure, including us. The index read 641 on the day, which in any other industry would describe a comfortably competitive market. Then you read down the ranking: BTCC second at 10.5%, Pionex third at 7.5%, Coinbase twelfth at 2.8%. We published that as a failure, because that is what it is.

On that 641. It is not our estimate of exchange concentration and it should not be cited as one, by anyone, including us. It is what the standard Herfindahl calculation returns when you feed it inputs nobody can verify. We are printing it to show what the method does with self-reported data, not to characterize the market. If you meet it quoted as a concentration figure, the quotation is the error this piece is about. And do not read it against the 1,484 above. That index covers seven days of block production; this one covers twenty-four hours of self-reported volume. Shorter windows are noisier, and at the same underlying concentration a one-day snapshot will generally return a lower index than a seven-day one. The gap between the two figures is therefore part window length and part input quality, and it is not evidence that mining is more concentrated than exchange volume. We have not measured that, and on this data we cannot.

The argument

The distinction here is not between good data and bad data. It is between a measurement and an assertion, and the two are not the same class of fact.

Block attribution is a measurement. The coinbase tag is written into the chain, the chain is public, and anyone who disputes our 60.75% can recount it this evening and tell us we are wrong. That possibility is what makes the number worth printing. Exchange volume is an assertion. It is a figure a private company publishes about itself, with a direct commercial interest in the answer, no audit, no agreed definition of what counts as a trade, and no consequence whatsoever for being wrong. Aggregating sixty-one assertions does not produce a measurement. It produces a tidier assertion.

There is a third number from this week that shows what the confusion costs. Decentralized exchanges settled $6.53 billion across 1,291 protocols on 4 August, which is 16.4% of the combined total. We published that as a floor rather than an estimate, and the reason is arithmetic. You cannot compute a clean ratio when the denominator contains a quantity nobody can check. The true on-chain share is 16.4% or higher and we cannot tell you how much higher, because the other side of the fraction is self-reported. One qualification we owe that piece. Recountable is not the same as clean: on-chain volume can be inflated by wash trading, and while anyone is free to go and identify it, we have not. If a material share of that $6.53 billion is wash volume, the floor moves down. It is arithmetic plus one assumption, not arithmetic alone, and we should have said so the first time.

The second-order effect

Self-reported volume is not a curiosity confined to league tables. The aggregate it belongs to is the input to exchange-eligibility screens run by index providers, to venue selection for institutional execution, and to the market-share framing that accompanies any policy argument about whether crypto trading is concentrated. We are not going to name a specific index rulebook or a specific consultation document that cites this week’s figure, because we have not traced it downstream and cannot show you one — and a piece about unverifiable assertions is the wrong place to start making them. The narrower claim is enough. Numbers of this kind circulate continuously, stripped of their qualification, and the qualification is the only part that matters.

Meanwhile the concentration that can be measured attracts almost no attention. Three organizations assembling a supermajority of Bitcoin’s blocks, every week, for a year, is a more serious structural fact than any exchange ranking published this quarter. It generates no coverage because it arrives without a press release attached.

The steelman

The strongest objection to our mining piece is that pool concentration is not miner concentration. A pool is a payout coordinator; hashrate can leave in an afternoon, and Stratum V2 lets individual miners build their own block templates. That is true, we said so at the time, and it is the reason we did not use the word capture. But it only half works. Exit is cheap in principle and rarely exercised in practice, and the trailing-year figures show the same three names in the same order. “The hashrate could leave” is a claim about capability, not about who is ordering transactions this week.

The objection on exchange data is weaker. It runs that self-reported volume is directionally useful even if the levels are wrong. Directional usefulness requires the bias to be consistent, and there is no reason to believe the incentive to inflate is uniform across sixty-one venues with sixty-one different business models, listing regimes and regulators.

What would change our mind

On exchanges: a venue submitting to a volume audit under a published standard, with a named auditor and a defined trade taxonomy that says explicitly what is excluded. Proof of reserves does not qualify. It is a balance-sheet snapshot, and volume is a flow. On mining: evidence that Stratum V2 template construction has become both observable and dispersed, so that the pool share and the ordering share stop being the same figure. Both are checkable, both re-run on our quarterly schedule, and either one would move us.

Where this leaves us

The rule we apply here is short, and we would suggest it more widely. If a number cannot be recounted by somebody who wants to prove you wrong, it is not a measurement. Publish it as a claim, say whose claim it is, or leave it out. We would rather run a piece explaining why we could not count something than a table that looks authoritative and is not, and this week we ran both, deliberately, side by side.

Standing reference: how exchanges work and what proof of reserves does not prove.

Comments are closed sitewide. If you can point to a volume audit, an index rulebook or a consultation document that would change either conclusion above, write to corrections@coinagereport.com. We will publish the correction, with your name on it if you want it there.