Proof of reserves is an attempt to answer, cryptographically, the question a customer cannot otherwise ask: does this exchange actually hold what my account balance says it does? It became near-universal after 2022, and it is a real improvement on taking a company's word for it. It is also, in the form most exchanges publish, half of an answer.
The mechanism has two parts. The asset side is straightforward: the exchange publishes addresses it controls and proves control by signing a message with the corresponding keys, so anyone can total the holdings on-chain. The liability side is the harder half, and the standard approach is a Merkle tree. Every customer balance becomes a leaf, the tree is hashed up to a single root, the root is published, and each customer is given the path proving their own balance is included in that root without revealing anyone else's. If total assets meet or exceed the total the root commits to, reserves cover liabilities.
The first limitation is what a Merkle tree of customer balances is not: a complete liability statement. An exchange can hold enough to cover customer deposits and still be insolvent, because it owes money elsewhere — loans taken against the reserves, obligations to affiliates, credit extended to a related trading firm. None of that appears in the tree. Reserves against customer balances alone is a solvency claim with the awkward part omitted.
The second is timing. A proof is a snapshot at a block height. Assets borrowed for the morning of the snapshot and returned in the afternoon satisfy it perfectly, and this has been documented. Frequent, unannounced or continuously updated proofs are much harder to game than a quarterly one on a known date.
The third is participation. The inclusion proof only works if customers check it, and almost none do. A customer whose balance was quietly omitted from the tree would find out by verifying and in no other way. An attestation nobody tests is a published number.
What raises the quality of one materially: a reputable firm attesting to the liability side rather than the exchange asserting it; negative net-equity checks that account for borrowings; frequent proofs; and a full financial audit alongside, which is a different exercise examining controls over time rather than a position at an instant. The honest summary is that proof of reserves makes one specific kind of fraud much harder and does not establish that a venue is solvent. It is a reason to trust an exchange more than one that publishes nothing, and not a reason to treat an exchange balance as custody.