Proof of Reserves: What It Shows and What It Cannot

A custodian publishes proof that it controls a stated quantity of assets. That is a real, checkable claim, and it is half of a balance sheet. Solvency requires assets to exceed liabilities, and liabilities — what the custodian owes its customers — are the half that is hard to prove and usually absent.

The assets half, and how it is proven

Demonstrating control of on-chain assets is straightforward: publish the addresses and sign a message from each, proving possession of the keys at a moment in time.

The weaknesses are known. Borrowed assets can be moved in for the snapshot and returned afterwards, so a point-in-time proof shows a moment rather than a state. Repeating the exercise unpredictably raises the cost of that considerably, which is why frequency matters more than any single publication.

And addresses can be shared. Two custodians proving control of the same assets both produce valid proofs, which is why an attestation covering the completeness of the address set — that these are all the addresses and they are exclusively ours — is worth more than the signatures alone.

The liabilities half, and the Merkle approach

Proving what a custodian owes is harder, because it means proving a total over private customer balances without publishing anyone's balance.

The Merkle tree approach solves part of it. Each customer balance is hashed into a tree, the root is published, and every customer receives a proof letting them verify their own balance is included in the total. If enough customers check, the custodian cannot understate the total without being caught by whoever it excluded.

This is a genuine improvement over an accountant's letter, because verification does not require trusting the accountant. It has two gaps.

It relies on customers checking. A published root nobody verifies proves nothing, since an omitted customer only discovers the omission by looking.

And it does not cover off-platform liabilities. Loans taken by the custodian, obligations to counterparties, and money owed to anyone who is not a customer with a balance are outside the tree entirely. A custodian can be fully reserved against customer balances and insolvent because of debts the scheme never touched.

Why the strongest version is still not an audit

The best available arrangement — recurring proof of both halves, verified by customers, with an accountant attesting to address completeness — is materially stronger than a letter. It still is not an audit.

An audit examines financial statements as a whole: liabilities of every kind, related-party transactions, controls, going-concern assessment. Proof of reserves examines one line and a moment. Both are useful; only one of them would have caught the failures people invoke proof of reserves to prevent.

This is why no custodian in our exchange ratings scores at the top of the custody pillar on reserve proof alone, and why a publicly listed company filing audited accounts is treated as a different class of evidence.

Reading a scheme

• Does it cover liabilities as well as assets, or only assets — this is the first question and it eliminates most schemes.
• Can you verify your own balance is included, and have you actually done it.
• How often is it published, and on a predictable schedule or an unpredictable one.
• Does an accountant attest that the address set is complete and exclusive, or only that the addresses hold what is claimed.
• Are off-platform borrowings disclosed anywhere, since they sit outside every proof-of-reserves scheme.

The realistic conclusion is that proof of reserves raises the cost of a specific deception and does not establish solvency. It is worth demanding, worth checking, and worth not mistaking for the thing it is named after.