Abstract
Protocols, researchers, contest platforms and bug-bounty programmes all ask the same question before spending review money: which of the code that is live on mainnet today has no public audit behind it? Answering it by hand means resolving proxies and facets, listing every callable function, finding every report, and reading each report's scope pages. Few teams do it, and nobody publishes the result in a form others can reuse.
The audit-gap map does this for the ten largest DeFi and infrastructure protocols on Ethereum, pinned to block 26,392,435. Every one of 656 callable functions gets one of four statuses, and every "In audit scope" claim carries the page, the exact quote, the quote's hash and the SHA-256 of the report it came from. The method is deterministic; AI only proposes candidate citations, and a validator records eight gates for every claim and never upgrades a failed or blocked one to coverage.
Today every one of the 656 functions has an answer: 584 sit in a cited audit scope and 72 have no audit found, each with the list of reports we read. The largest gap is Pendle's RouterV4 (58 functions): none of the 23 audit PDFs in Pendle's repository names the router files. This grant funds the open dataset and generator, expansion to 25 protocols with a refresh on every upgrade, contest-ready scope sheets, and adoption by the platforms that decide what gets reviewed next. The audits themselves stay paid by protocols.
1. The problem
Audit coverage decays quietly. A protocol is audited at launch; then it adds an extension, a new facet, a periphery router, a library upgrade. Each change may be reviewed by a different firm with a different scope, or not at all. The protocol's audits page lists the reports, but does not say which of today's deployed functions none of them cover.
- Contest and bounty platforms scope reviews from what the protocol tells them, not from a map of what is uncovered.
- Researchers looking for high-value unreviewed code must rebuild the same analysis for every protocol.
- Protocols often do not know their own gaps, especially in code inherited through libraries or added by facets.
- Users and wallets see "audited by X" for a protocol as a whole, regardless of which function they call.
2. Four statuses, defined strictly
| Status | Rule | How to read it |
|---|---|---|
| Byte-for-byte match | An exact-build record exists for this address: the deployed runtime equals a rebuild of the commit a report names. | The code is the reviewed code. Scope still needs human review. |
| In audit scope | A collected public report names this function's source file. The binding cites page, quote and hashes. | A reviewer was asked to look at this file. The deployed code has not been compared with it. |
| No audit found | We hold the protocol's reports with extracted file lists, and none names this file. | A candidate gap. A report we have not collected may still cover it. |
| Not checked | The source path is unknown, or the protocol has no extracted file list. | No conclusion either way. |
The rules are applied in that order by one function (functionEvidence in lib/site/evidence/callEvidence.ts), and the same function feeds the explorer, the demos and this paper. Bindings that come only from a finding in a change review (for example a "Files:" line in a Lido or EigenLayer upgrade review) are kept and labelled "File named in a finding of a change review", so a reader can discount them.
3. How the map is built
3.1 The pinned catalog
- Pin a block. All reads use block 26,392,435. Its full header (including hash) is stored in
pinned-state.json, so any later reader can check they are looking at the same chain state. - Resolve deployments. For each protocol, the entry contracts are listed and followed through proxies and facets to the implementation that runs (Pendle's RouterV4 resolves to seven facets; Aave's Pool to its implementation revision 12).
- List functions. Every external and public function of every resolved contract is recorded with its signature, selector, mutability and inputs.
- Locate source. Each function is mapped to its source file using the verified source of the exact deployed bytecode (source-registry snapshots and checksum-pinned compiler ASTs).
- Derive dependencies. Call dependencies between functions are taken from the compiler's output, so the explorer can show what a function reaches. This is static analysis, not a live trace.
Each protocol's catalog is one JSON file (public/evidence/catalog/<protocol>-<build>.json) with snapshotBlock, snapshotHash, contracts, functions, nodes, bindings and reports. A manifest lists every file with its hash.
3.2 Collecting reports
Reports are collected from public sources only: protocol audit folders on GitHub, firm publication pages and their feeds. Each document is downloaded over HTTPS, and its SHA-256 and origin are recorded. Gate G0 passes only when the origin publisher is proven; otherwise it stays blocked (section 4). 20 distinct reports are held today: 10 from the original catalog and 10 added on 13 November 2026 because they name files that the twelve recorded transactions actually executed. The 13 November pass produced 15 supplement entries; 5 of them re-read a report the catalog already held, to cite files its original extraction missed.
3.3 Binding a report to a file
A scope binding says: report R names source file F. It is stored with the evidence that lets anyone check it:
{
"protocolId": "compound",
"sourcePath": "contracts/Comet.sol",
"file": "Comet.sol",
"reportId": "<report id>",
"citation": {
"documentSha256": "<SHA-256 of the retrieved PDF>",
"locator": "page:4;section:Scope",
"page": 4,
"quote": "<exact text from the report naming the file>",
"quoteSha256": "<SHA-256 of the quote>"
}
}File paths are matched by suffix on a path-segment boundary and case-sensitively, so Comet.sol never matches CometExt.sol. A reviewed version that differs from the deployed version is kept visible rather than silently matched. The full set is published at https://proofofaudits.com/evidence/supplements/scope-bindings.v1.json.
4. Eight gates every claim must pass
Each collected report becomes an evidence candidate, including any citation proposed by AI, and a deterministic validator checks it against eight gates. Each gate ends as passed, failed, blocked or not applicable. Blocked means a known missing proof is recorded as a blocker: it is not a failure, and it is never upgraded to coverage. A passing run means the candidate is consistent and honest about its gaps; it does not by itself verify a build or a signature.
| Gate | Checks | Example failure codes |
|---|---|---|
| G0 Source | The document came from a safe HTTPS origin, its hash is real and matches, the publisher matches the firm. | origin_url_unsafe, document_hash_mismatch, publisher_mismatch |
| G1 Citation | Every claim has a locator, page and exact quote, and the quote appears in the document text. | quote_not_in_text, quote_hash_mismatch, page_mismatch |
| G2 Commit | The reviewed commit is a full immutable hash in a named repository, cited from the report, or marked unresolved with a reason. | commit_not_immutable, commit_uncited |
| G3 Build | Compilation is deterministic with a pinned compiler and dependencies. | compiler_not_pinned, build_without_source |
| G4 Runtime | Build-to-runtime match verified independently; only documented metadata, link and immutable transforms allowed. | match quality recorded |
| G5 Scope | Inclusions, exclusions, method limits, findings and fix rechecks considered. | scope recorded per claim |
| G6 Attestation | A firm signer is authenticated before any firm-attested claim. Report attribution is not a signed attestation. | no firm attestation without a signature |
| G7 Change | A change or revocation supersedes derived coverage atomically, keeping the original evidence. | supersession recorded |
| Gate | Result across the reports |
|---|---|
| G0 | 1 passed, 14 blocked |
| G1 | 15 passed |
| G2 | 5 passed, 10 blocked |
| G3 | 15 blocked |
| G4 | 15 blocked |
| G5 | 8 passed, 7 blocked |
| G6 | 15 blocked |
| G7 | 15 not applicable |
Read this table as the honest state of the evidence. G1 passes for every report: every quote was found in the document and its hash matches, which is what an "In audit scope" claim rests on. G3, G4 and G6 are blocked for every report, because none of these candidates claims a byte-for-byte build or a firm signature, so none of them can produce a release on its own. The per-report matrix is public at https://proofofaudits.com/methodology#gates.
5. Results at block 26,392,435
| Protocol and contracts | Functions | In audit scope | No audit found | Not checked | Reports |
|---|---|---|---|---|---|
| Aave V3 · Pool revision 12 | 75 | 73 | 2 | 0 | 6 |
| Uniswap V4 · PoolManager | 33 | 31 | 2 | 0 | 2 |
| Compound III · USDC Comet | 92 | 92 | 0 | 0 | 2 |
| Morpho Blue · core | 28 | 28 | 0 | 0 | 1 |
| Lido stETH core · revision 4 | 93 | 93 | 0 | 0 | 1 |
| Safe Smart Account 1.4.1 | 32 | 32 | 0 | 0 | 2 |
| Pendle V2 · RouterV4 + 7 facets | 58 | 0 | 58 | 0 | 1 |
| Sky sUSDS · pinned implementation | 51 | 51 | 0 | 0 | 1 |
| Euler V2 · EVC + EVault implementation | 141 | 141 | 0 | 0 | 2 |
| EigenLayer 1.13.0 · StrategyManager | 53 | 43 | 10 | 0 | 2 |
| All ten protocols | 656 | 584 | 72 | 0 | 20 |
Each protocol can be opened in the explorer at https://proofofaudits.com/explorer/functions, where each function opens with its source file, the citation behind its status and its dependencies.
| Protocol | Largest file with no audit found | Functions in that file | Gap files in protocol |
|---|---|---|---|
| Pendle | lib/pendle-core-v2/contracts/router/ActionMiscV3.sol | 20 | 9 |
| EigenLayer | lib/openzeppelin-contracts/contracts/proxy/transparent/TransparentUpgradeableProxy.sol | 5 | 3 |
| Uniswap | lib/v4-core/lib/solmate/src/auth/Owned.sol | 2 | 1 |
| Aave | @aave/core-v3/contracts/dependencies/openzeppelin/upgradeability/InitializableUpgradeabilityProxy.sol | 1 | 2 |
Two readings matter. First, most functions of the large protocols do sit in a cited scope, which is good news that nobody had measured. Second, the gaps cluster: Pendle's router facets hold 58 of the 72 functions with no audit found. That is exactly where a contest or a bounty should look first.
| Protocol | Firm | Report | Date | SHA-256 (prefix) |
|---|---|---|---|---|
| Aave | MixBytes | Aave V3.7 security audit | 30 March 2026 · report date | 1eabc75a22ae… |
| Aave | ABDK Consulting | Aave v3.5 code audit | 18 July 2025 · report date | 5b10473c8619… |
| Aave | Blackthorn | Security Review For Aave v3.4 | 27 May – 8 June 2025 · review period | e3dd14a1d112… |
| Aave | Certora | Aave v3.3 formal verification and manual review | 24 October – 7 November 2024 · review period | fd4d5727584f… |
| Aave | ABDK Consulting | Aave V3 smart contract audit | 26 January 2022 · report date | 39d229b72646… |
| Aave | Trail of Bits | Aave V3 Security Assessment | 29 November 2021 · report date (fix log 21 January 2022) | 5e1e02bfa049… |
| Uniswap | OpenZeppelin | Uniswap V4 core audit | 17 July 2024 · report date | a47a75608486… |
| Uniswap | OpenZeppelin | Uniswap V4 Periphery and Universal Router Audit | 5 September 2024 · report date | ee86897a2163… |
| Compound | OpenZeppelin | Compound III audit | 14 June 2022 · delivered; published 20 July 2022 | e88e025aa58a… |
| Compound | OpenZeppelin | Comet Asset Extension Audit | 16 January 2025 · report date | 84bb30834633… |
| Morpho | OpenZeppelin | Morpho Blue audit | 13 October 2023 · report date | a47b50009f40… |
| Lido | Certora | Lido V3 audit | January 2026 · report month only; draft | 9ca94dbf2936… |
| Safe | Ackee Blockchain | Safe smart account 1.4.0 audit | 28 March 2023 · report date (Safe 1.4.0) | bd64303f4ae9… |
| Safe | Certora | Safe Smart Account v1.3.0 security review | 20–29 July 2026 · review period | 9b3bd212e97d… |
| Pendle | ChainSecurity | Pendle V2 code assessment | 15 August 2024 · report date (V2 core scope) | 63e08e69377b… |
| Sky | ChainSecurity | Savings USDS code assessment | 30 September 2024 · report date | 5a3903cb7bbb… |
| Euler | ChainSecurity | Ethereum Vault Connector code assessment | 6 May 2024 · report date (EVC scope) | 3e8239617f97… |
| Euler | ChainSecurity | Code Assessment of the Euler Vault Kit Smart Contracts | 18 June 2024 · report date | c90719f0efb4… |
| EigenLayer | Certora | EigenLayer redistribution slash delay v1.13.0 | 11–12 May 2026 · review period; draft | 2793a930a796… |
| EigenLayer | Certora | EigenLayer Slashing (V1.0.0) formal verification%20-%20Certora%20-%20Feb%202025.pdf) | 10 January – 11 February 2025 · review period | 430ef6ccc1bd… |
6. Who uses it and how
| User | Use | What they get |
|---|---|---|
| Contest and bounty platforms | Scope the next review | A list of deployed functions with no public audit, grouped by file, with the source location and dependencies. |
| Protocols | Plan audits after upgrades | Their own gap list, refreshed when an implementation changes. |
| Security researchers | Find unreviewed high-value code | Ranked gaps across protocols, without redoing the analysis. |
| Wallets and explorers | Show per-function coverage | The same statuses Proof Before Sign shows at signing time (Initiative 1). |
| Auditors | Prove their scope | A public, cited record of which deployed functions their reports cover. |
7. Known limits
- Scope evidence is file-level. A report naming a file does not prove every function in it was reviewed.
- "No audit found" depends on the reports collected. A missing private or unpublished report produces a false gap; the map says "no audit found", not "unaudited".
- Code outside the ten protocols is answered contract by contract from its own public reports (for example ChainSecurity for Permit2). USDC has no public report: Circle says Halborn audited v2.2, but the report is not published, so it shows "No audit found".
- The dependency view is static and pinned to one block.
- Contests are not listed publicly on the site yet.
8. What the grant funds
A public, continuously updated gap map: which deployed code sits outside every public report's scope, re-checked when contracts upgrade. Proposed ask: $60,000 over 9 months, of which $20,000 (one third) is adoption.
| # | Milestone | Budget | Pass condition for the independent reviewer |
|---|---|---|---|
| 1 | Open dataset and generator for the ten protocols, with the method documented. Code under Apache-2.0, data under CC BY 4.0. | $15,000 | A third party regenerates the map from the published scripts and gets the same statuses. |
| 2 | Expand to 25 protocols ranked by value held on Ethereum; refresh automatically when an implementation changes. | $15,000 | Coverage list and upgrade refreshes visible publicly with dates. |
| 3 | Per-gap scope sheets a contest or bounty can use directly (files, functions, dependencies, cited reports). | $10,000 | Sheets published for every protocol. |
| 4 | Adoption: at least 2 contest platforms, bounty programmes or protocols use the map to scope a review. | $20,000 | Public reference from each adopter. |
Outreach targets (none committed yet): Code4rena, Sherlock, Cantina and Immunefi among platforms; Compound and Pendle first among protocols, since they hold the largest gaps.
9. Check it yourself
- Download the bindings: scope-bindings.v1.json.
- Pick any binding, download the report from its URL, and check
sha256sumequalsdocumentSha256. - Open the cited page and confirm the quote is there; hash the quote and compare with
quoteSha256. - Open the function in the explorer and confirm the source path matches the verified source of the deployed contract.
10. Team, reviewer and co-funding
- Team. Proof of Audits, led by its founder, Veer (veer@proofofaudits.com). The map, the explorer and the 20-report evidence set described here are live and checkable.
- Independent technical reviewer. To be agreed with TheDAO Security Fund before milestone 1 starts, with no ties to Proof of Audits, the ten protocols or the cited firms.
- Co-funding. None committed as of 13 November 2026. We will ask the protocols with the largest gaps and the contest platforms that would use the map to pledge about 15%.
References
- Sourcify: verified source for deployed bytecodehttps://sourcify.dev/
- Scope bindings (raw data)https://proofofaudits.com/evidence/supplements/scope-bindings.v1.json
- Explorer dataset manifesthttps://proofofaudits.com/evidence/catalog/manifest.json
- Methodology and gate resultshttps://proofofaudits.com/methodology
- TheDAO Security Fundhttps://thedao.fund/