“Audited” is the most overloaded word in crypto disclosure. It is used for a code review, for an accountant’s signed statement, and for a snapshot of an exchange’s addresses — three documents produced by three different professions, answering three different questions. None of them proves solvency, and only one of them looks at code at all.

The confusion is not entirely accidental. Each document represents real work, and each is expensive enough that publishing one is a genuine signal of effort. The problem is that the market reads all three as “someone checked”, when the checks cover different things, were performed to different standards, and in some cases were never designed to answer the question readers ask.
- Three document types share one word. Code audit, attestation and proof of reserves are not versions of each other.
- Scope decides meaning. A report is only as strong as the list of things it examined, which is usually on page two.
- Point-in-time is not continuous. A snapshot taken at one block height says nothing about the other 8,759 hours of the year.
- Assets are easier than liabilities. Addresses can be signed for; a complete liability figure requires the issuer’s own records.
Three documents, one word
Before comparing quality, separate the categories. A code audit examines software. An attestation examines controls or figures against agreed criteria. A proof-of-reserves report examines asset holdings at a moment in time. They are commissioned for different reasons, signed by different people, and read by different audiences.
| Document | Issued by | What it covers | What it cannot prove |
|---|---|---|---|
| Smart contract code audit | A security firm | Specific code at a specific commit, against a stated threat model | Deployment configuration, key custody, operator behaviour, economic design |
| Attestation report | An accounting or assurance firm | Controls or balances against agreed criteria, for a date or a period | That the criteria are the ones you care about, or that the business is sound |
| Proof-of-reserves snapshot | An assurance firm, sometimes only the exchange | Assets controlled at a block height, sometimes matched to a liability total | Completeness of liabilities, encumbered assets, anything outside the snapshot |
| Bug bounty programme | The project itself, via a platform | Incentives for outside researchers to report flaws | That anyone looked, that findings were fixed, or that coverage is broad |
Notice that the fourth row is not an audit and is often described as one. A bounty is a standing invitation. It produces reports, not an opinion, and the public usually sees only the ones the project chooses to disclose.
Code audit: the scope is the whole argument
A smart contract audit is a review of source code by people who look for ways to break it. The useful part of the report is not the badge on the website. It is the scope section, the findings list and the remediation record.
Commit hash and scope
Audits are performed against a specific version of the code, usually identified by a commit hash. That hash is the anchor: it tells you exactly what was reviewed. If the deployed contract differs from the audited commit, the audit describes software that is not running.
The scope section also lists what was excluded. Common exclusions include third-party libraries, deployment scripts, front-end code, off-chain infrastructure and the admin interface. Those exclusions are reasonable, and they are also where incidents tend to happen. A contract that is flawless but whose upgrade authority sits with a single key has a security model that the audit did not assess.
Findings, severity and remediation
A credible report lists findings by severity — critical, high, medium, low, informational — and states for each one whether it was fixed, acknowledged or disputed. A report showing zero findings across a large codebase is unusual, and it is more often a sign of a narrow scope than of perfect code.
Look for the re-test. Fixes introduce new code, and new code is unreviewed until someone looks at it. A report that ends with “all critical and high findings resolved” without a second pass is describing an intention as much as a fact. The strongest reports include a remediation commit and a short verification of each fix.
What a code audit never covers
It does not cover the deployment. Proxy admin keys, upgrade permissions, pause functions and ownership transfer all live outside the reviewed source, or inside it but configured after deployment. It does not cover key management, which is why the difference between a hot wallet and a properly separated cold setup matters far more than most audit badges suggest; our guide to cold storage for a large balance covers that layer.
It does not cover the economic design. A contract can be implemented exactly as specified and still be unprofitable or exploitable when the specified mechanism meets real markets. Auditors check whether the code does what it says. They do not usually check whether what it says is a good idea.
Attestation: a signature about controls
An attestation is a written conclusion by a qualified third party about a subject matter, measured against criteria. In practice, for crypto businesses, it usually concerns controls: whether customer assets are segregated, whether withdrawals require two approvals, whether access to production systems is logged and reviewed.
Type I and Type II, and why the difference matters
Under the common SOC framework, a Type I report describes whether controls were suitably designed as of a specific date. A Type II report covers whether those controls operated effectively across a period, typically six to twelve months. The first is a photograph of the design; the second is a record of behaviour.
Most crypto attestations published for marketing purposes are point-in-time. That is not worthless — it shows a firm looked at something on a given day — but it cannot tell you whether the controls held during the week of a withdrawal surge. When a report covers a period, read the exceptions. An unqualified opinion with a list of exceptions is a more honest document than a clean one with no detail.
What “unqualified” means, and what it does not
An unqualified, or clean, opinion means the subject matter met the criteria in all material respects. It does not mean the criteria were demanding, that the business is profitable, or that the client will still exist next year. Assurance firms opine on the engagement’s terms. If the terms are narrow, a clean opinion is a narrow statement.
The criteria are usually described in the report’s opening pages. Read them before the conclusion. A report that attests to “the existence of the balances shown in management’s schedule” is attesting to a schedule, not to the completeness of the schedule.
Proof of reserves: assets at one moment
A proof-of-reserves report is an attempt to show that an institution holds the assets it says it holds. The common method is to combine a list of addresses controlled by the institution with a signed message proving control, then have a third party confirm the balances at a specific block height.
What the snapshot contains
A good report states the block height, the total asset figure, the list of addresses or a commitment to them, and the party that performed the check. Where liabilities are included, there should be a total customer liability figure and a description of how it was derived — usually a Merkle tree of account balances, with the root published so users can check their own inclusion.
Each of those elements can be missing. Without a block height, the figure has no anchor. Without address control, the assets may belong to someone else. Without a liability total, the report is an asset inventory, not a solvency statement, and the two are routinely confused in headlines.
Liabilities are the hard part
Assets are visible on a public ledger. Liabilities are an internal database, and their completeness depends on the institution’s own records. A snapshot that shows $2 billion of assets against $1.8 billion of customer balances proves a surplus only if every balance is in the list.
Several things sit outside that list even in a good implementation. Borrowed funds may be counted as assets while the corresponding loan is not counted as a liability. Assets may be pledged as collateral elsewhere. Balances may exist at related entities that are not in scope. And the whole picture is frozen at one height: deposits made after it, withdrawals before it, and positions opened on derivatives are all invisible.
This is why a reserve snapshot should be read as a test of one claim — “at block height X, controlled addresses held at least the stated assets” — rather than as a statement about the business. The distinction is close to the one that matters when an intermediary fails, which our note on exchange insurance funds approaches from the loss side rather than the disclosure side.
A snapshot proves a balance at an instant. Solvency is a relationship between assets and liabilities that has to hold at every instant, including the ones nobody photographed.
What all three leave out
Even when all three documents exist and are well made, a set of risks remains outside their scope. Naming those gaps is more useful than arguing about which badge is better.
Operational security
None of the three typically examines how signing keys are generated, stored and used. A multisig with a sound threshold can still be defeated by three signers approving a malicious transaction on a compromised machine. Our explainer on multi-signature wallets describes the mechanism; the audit usually stops at the contract that enforces it, not at the people operating it.
Governance and upgrade authority
Who can change the rules after the audit? If a single key can upgrade the contract, the audited behaviour is one transaction away from being replaced. Reports sometimes mention this in a “centralisation” section. It deserves more weight than it usually gets, because it is the difference between a fixed system and a system whose current state happens to be audited.
Business continuity
No audit report tells you whether the entity can survive a bank run, a counterparty default or a regulatory freeze. Those are questions about capital, liquidity and legal structure. They are the questions that matter most in a failure, and they are the ones least likely to appear in a document with the word “audit” in the title.
How to read an audit page in ten minutes
Most venues publish a page of badges. The page is designed to end the inquiry. Read it in the opposite spirit: as a list of documents whose first pages decide whether the rest is worth reading.
- Identify the document type. Code audit, attestation and reserve snapshot are not interchangeable, and the page often blurs them.
- Find the date and the period covered. A report older than the current codebase or deployment is history.
- Read the scope, including exclusions, before the conclusion.
- Check the commit hash or block height against the system you actually use.
- Look for findings and remediation, not just a summary letter. A report with no findings should make you check the scope again.
- Confirm who issued it and whether that firm is named, contactable and independent of the project.
- Ask what the document does not cover, and whether that gap is the risk you care about.
Then write one line for your own records: what was checked, by whom, as of when, and what it left out. That line is more valuable than the badge, because it survives the marketing refresh that replaces the badge with a newer one.
The word “audited” is not going to be retired. It is too useful to marketing and too vague to regulate out of existence. The realistic response is to translate it every time you see it: audited what, by whom, against which criteria, for which period, and with what excluded. When a project can answer those five questions without a press release, the document is probably worth reading. When it cannot, the badge is decoration.

0 comments
No comments yet — ask a question and we will answer it here.