A Merkle-tree snapshot is evidence, not a solvency statement. Here is exactly what an exchange proves when it publishes reserves, and the three gaps every programme leaves open.
Nadia OkoroPolicy & Consumer Editor · Regulation, tax, stablecoins and the products retail users hold
Proof of reserves demonstrates that an exchange controlled a set of wallets at a moment in time. It does not show what the exchange owes, whether the assets were borrowed for the snapshot, or whether they are still there today. A reserve attestation is only a solvency statement when it covers liabilities as well, which most do not.
After FTX collapsed with a hole where customer assets should have been, every large exchange published proof of reserves within about six weeks. The pages look reassuring: a total, a Merkle root, a verification tool, sometimes an accountant's logo. Most of them prove considerably less than readers assume, and the gap between what is proved and what is inferred is where the next failure will live.
The standard implementation does two things. It publishes a list of wallet addresses and signs a message from each to show the exchange controls the keys. And it builds a Merkle tree of customer balances, publishes the root, and gives each customer a proof they can check to confirm their own balance was included in the total.
Together those establish one narrow fact: at the moment of the snapshot, the exchange controlled wallets holding X, and the customer balances it claimed to owe summed to Y. If X is greater than or equal to Y, the exchange was covered for those specific assets at that specific instant.
That is genuinely useful. Before 2022 the industry offered nothing at all, and an exchange that publishes reserves monthly is measurably more accountable than one that does not. It is also three steps short of a solvency statement.
The Merkle tree covers what the exchange owes its customers. It says nothing about what the exchange owes anyone else — loans from lenders, obligations to a sister trading firm, bond issues, unpaid tax, litigation provisions. An exchange can hold every customer coin and still be insolvent because of debts that never appear in the tree.
This is exactly the FTX failure mode. Customer assets moved to Alameda; the shortfall was a liability to customers created by lending their assets to a related party. A reserves-only proof published the week before would not have flagged it, because the wallets and the tree address a different question.
The programmes that close this gap say so explicitly: they cover liabilities alongside assets and are conducted with an independent accounting firm. Kraken is the clearest example among the venues we rate, which is one reason it carries the highest security mark in the exchange ratings.
A proof covers one block height. Assets present at that height can leave at the next one. Nothing in the mechanism prevents an exchange from borrowing assets shortly before the snapshot, signing, and returning them afterwards — a practice traditional finance calls window dressing and has spent decades writing rules against.
There is a partial defence. If snapshots are frequent, unannounced in advance, and taken at block heights the exchange cannot choose, borrowing for each one becomes expensive and visible on-chain. Monthly publication on a fixed cadence, as OKX has run for several years, is meaningfully harder to game than an annual page refreshed when convenient.
Most programmes cover a handful of major assets — bitcoin, ether, a couple of stablecoins — because building the tree for hundreds of long-tail tokens is work with little marketing value. If you hold something outside that set, the proof does not cover your balance, and the exchange's total tells you nothing about whether your specific asset is fully backed.
Check the asset list before you take comfort from the headline number. On several venues, the covered set is well under half of what the exchange lists for trading.
Start with the date and cadence. A page updated monthly for two years is a different signal from one published once in 2023. Then look for the word liabilities: if it appears only as a description of the customer-balance tree, the programme covers assets and customer claims, not total obligations.
Check who verified it. An independent accounting firm performing an agreed-upon-procedures engagement is a stronger claim than a self-published tree, and both are weaker than a full financial audit — which almost no exchange provides. Then check which assets are covered, and whether your own balance verifies through the tool rather than assuming it does.
Finally, note what the exchange does not publish. A venue that has never disclosed an ownership structure or filed accounts anywhere is asking you to trust a snapshot as a substitute for everything else it will not show you.
Treat reserves as one input among several: incident history, regulatory standing, disclosure quality, and whether the entity you contracted with is the one holding the coins. Our exchange rubric weights security and custody at 25% and regulatory standing at 20% precisely because a reserve page cannot carry that judgement on its own.
The practical conclusion has not changed since 2022: an exchange is a custodian, proof of reserves is partial evidence about a custodian, and the only position that removes custodial risk entirely is self-custody.
Every major stablecoin publishes reserve reports. Almost none of them are audits, and the difference decides what the document is worth.
Providing liquidity looks like free yield until you compare it against simply holding the two assets — the gap between those outcomes is impermanent loss, and it's bigger than most LPs realise.
Creator royalties were once pitched as NFTs' killer feature over traditional art. A few years and one marketplace price war later, most of that promise has quietly evaporated — here's how NFT royalties actually work and why enforcement collapsed.