What crypto proof of reserves can establish

A crypto exchange’s reserve disclosure raises two questions: Do the assets shown in a customer’s account actually exist, and can they be withdrawn when needed? A strong technical answer to the first question is possible. Answering the second requires knowing who has rights to those assets, which obligations they must cover, and whether the withdrawal system works.

In its October 30, 2026 announcement, Binance reported that its internet-connected hot wallets had been compromised and that it had paused withdrawals during its security review. Its subsequent reserve and withdrawal announcements brought this distinction back into focus. Binance’s October 30 announcement (opens in a new tab). I am not making a judgment about the company’s solvency here. I am examining which conclusions the available forms of evidence can support. The same framework can be applied to other crypto exchanges that hold assets for their customers.

The short answer

A proof of reserves with a clearly defined scope, successfully verified, can provide a meaningful comparison between assets and included customer balances at a particular moment. Depending on its design, it can also establish that records reconcile with the totals and satisfy specified mathematical conditions. That is stronger evidence than publishing a wallet address or a high reserve ratio alone.

The result is nevertheless bounded by the information included in the proof. Whether all economic obligations are covered, whether assets can be used for customers without competing creditor claims, and whether a customer can withdraw later are separate questions. They require accounting records, evidence of rights to assets, and current operational information.

I therefore apply four checks when reading a reserve ratio: Which assets were measured, and when? Which obligations were included? Who controls the assets, and what rights attach to them? Which assets can customers actually withdraw, and on which networks? Keeping these questions separate preserves the value of strong evidence without asking it to support a broader conclusion than it can establish.

The strongest counterargument

One prominent criticism of proof of reserves is that it looks only at assets and ignores what an exchange owes its customers. That is not true of every system. Modern methods can incorporate customer liabilities into the proof.

In Binance’s described system, for example, a Merkle tree combines many account records under a single verifiable summary. Customers can use their supplied records to check whether their accounts were included in a selected liabilities report. A zero-knowledge proof, using a method called zk-SNARK here, tests specified computational conditions without disclosing private account data. The company’s technical explanation includes consistency between included balances and totals, and a requirement that each account’s total net value be nonnegative. Binance’s proof-of-reserves methodology (opens in a new tab), technical explanation (opens in a new tab).

That net value is measured in dollars across different assets. An individual asset balance may be negative because of borrowing. A nonnegative total net value therefore does not mean every account has a positive balance in every crypto asset.

This technical counterargument deserves serious weight. Successful verification produces real information within a defined scope. But satisfying a computational condition and establishing that all relevant obligations entered the system in the first place are different checks. Reading the precise claim being tested is more useful than relying on the technology’s name.

What would change the conclusion

A current proof successfully verified for specified assets and customer liabilities would strengthen the assessment. An independent examination reconciling completeness against relevant books and external obligations, documentation of asset availability, and functioning withdrawals would narrow the uncertainty further.

If a more comprehensive method independently establishes all relevant obligations and enforceable rights to assets, the limitations I describe would narrow for that scope. I am not arguing that a reserve proof can never be sufficient, regardless of its design. The actual scope of the evidence determines which questions it answers.

Conversely, failed verification, a demonstrated omission of a customer record that should have been included, or reliable documentation of an undisclosed claim on reserves would change the assessment. These are conditions that would alter the conclusion, not findings I am making about Binance. A new withdrawal problem would also change the assessment of access; it would not, by itself, prove that the company’s total assets were worth less than its liabilities.

What is in the reserve ratio's denominator?

A reserve ratio depends as much on the obligations below the line as on the assets above it. Before interpreting “more than 100%,” a reader needs to know which customers, products, assets, and company are covered. Services operating under the same brand cannot simply be assumed to belong to the same legal entity.

Consider a wholly hypothetical example. An exchange holds 100 BTC in reserves and owes customers a total of 100 BTC payable from that same reserve pool. If the report includes only 90 BTC of those customer claims, dividing reserves by reported liabilities gives approximately 111.1%. Using all the customer claims produces a ratio of 100%. The first calculation is arithmetically correct; its scope is narrower. These are not Binance figures, and the example does not allege that any customer was omitted from a report.

Finding one’s own record is an important check. It does not establish that every other customer was included. Provisions, an original study of cryptographic solvency proofs, explains how an omitted customer can detect the omission in its design by checking. Detection depends on at least one affected customer doing so. The study is not an examination of Binance’s system; it illustrates the distinction between individual verification and completeness of the entire set of records. Dagher and colleagues, Provisions (opens in a new tab).

Customer balances are also different from the company’s full set of liabilities. Corporate solvency and the adequacy of segregated customer assets are separate assessments. Adding every corporate liability to the denominator for customer reserves would not necessarily be correct; it is necessary to establish which claims attach to which asset pool. A report that measures customer obligations well cannot be expected to answer questions about other corporate or contingent liabilities outside its scope. The approach in my guide to reading financial statements, which considers balance sheets alongside cash flows, is useful here: The existence of an asset does not automatically explain which payment obligation it can meet.

Controlling a wallet and being able to use its assets

Demonstrating the ability to use an address’s key provides meaningful evidence of technical control over its assets. Economic ownership is a broader question. Assets could be borrowed or subject to another creditor’s claim. These are possibilities to examine separately from a control proof, rather than conditions to assume about any particular exchange. The Provisions study explicitly discusses this boundary between control and ownership.

For the customer, the next question is: Which company owes me, and can the reserves shown be used to meet that claim? Answering it requires knowing the contracting entity, relevant terms, and applicable law. An account record’s inclusion in a tree does not settle the customer’s legal position. I draw no conclusion here about a customer’s creditor priority in any specific jurisdiction.

A 2023 advisory from the Office of the Investor Advocate at the U.S. Public Company Accounting Oversight Board, or PCAOB, stresses that third-party proof-of-reserves engagements are not financial-statement audits under PCAOB standards and may leave gaps concerning liabilities, borrowed assets, or customer rights. It expresses office staff views; it is not a Board rule or a finding that every current cryptographic design has the same shortcomings. PCAOB investor advisory (opens in a new tab).

A third party’s name on a report is therefore not enough. What it examined, which standard it used, and what it excluded determine how the result should be read.

Keeping Binance's announcements on their own timelines

Following the security incident, Binance stated that its new reserve snapshot was taken at 00:00 UTC on October 31, 2026, and that it expected the full zero-knowledge proof to be ready by November 3 UTC. The first was a measurement time; the second was the company’s completion estimate. Later developments cannot be projected backward as though the same conditions already held when the snapshot was taken. October 31 announcement (opens in a new tab).

In its November 1 announcement, the company reported that approximately 95% of the incident’s impact had been covered and shared its phased withdrawal reopening plan. That percentage was neither the share of all customer liabilities backed nor the share of stolen assets recovered. The plan did not establish that services had actually resumed across every asset and network. November 1 announcement (opens in a new tab).

Binance Japan’s update as of 16:00 Japan Standard Time on November 2 reported that withdrawals had fully resumed for some tokens while remaining partly available or unavailable for others; its scope is Binance Japan and cannot be extended to all Binance services worldwide. Binance Japan’s fourth notice (opens in a new tab).

These announcements show how the company reported progress. They do not constitute an independent solvency assessment. A company with adequate assets can pause withdrawals for security reasons; publishing a timetable does not, by itself, establish that a customer’s transaction went through.

Four checks from reserves to access

I use the following framework to identify which question the available information answers, rather than to assign an exchange a score. Strong evidence in one row does not automatically fill a gap in another.

Check	Information to look for	Limit of the conclusion
Snapshot assets	Snapshot date, covered tokens, and verified amounts	Does not establish later balances.
Liabilities	Included accounts, products, obligations, and completeness checks	Inclusion of one’s own account does not establish all obligations.
Control and rights	Address control, the entity with rights to assets, and restrictions on their use	Use of a key does not prove all economic rights.
Actual access	Current availability and completed transactions by entity, token, and network	One successful transfer does not guarantee future access.

Asset composition belongs within these checks. Total dollar value and available units of a requested token are different measures. In a report that combines different assets, prices, borrowing, and netting rules can affect the result. Comparisons of balances in the same asset must preserve the report’s definitions; a shortfall cannot simply be inferred by comparing different measures.

Holding a stablecoin adds another relationship. Access to an exchange balance should be examined separately from the stablecoin issuer’s reserves and redemption mechanism. The issuer layer discussed in my guide to stablecoin demand and U.S. Treasury bills does not replace the exchange’s custody and withdrawal layer. Strong issuer reserves do not, by themselves, establish that a particular exchange account is accessible.

The developments I would monitor follow from these four checks: What date does the new proof cover, and what scope was verified? What further examination addresses rights and liabilities? At which entities and on which networks are services actually working? This makes it possible to read a new announcement for the uncertainty it resolves, rather than just the size of the reserve figure it displays.

Methodology and limitations

Evidence cutoff: 2026-11-02T14:12:36+03:00. This guide considers original company announcements alongside technical documentation, original cryptographic research, and an investor advisory with a defined scope. Company-reported conditions are not treated as independently verified financial outcomes. The hypothetical BTC calculation illustrates only the effect of the denominator.

This research did not verify the current full proof’s publication or successful verification, completeness of all liabilities, rights attaching to reserves, or current withdrawal access for every customer. These uncertainties do not establish that the relevant evidence is absent. No personal account was accessed and no withdrawal was attempted. The guide offers no conclusion about an exchange’s solvency, safety, or customers’ legal rights.