OAK does not exist in a vacuum. This file documents the public frameworks, datasets, and bodies of work that OAK draws on, sits alongside, or differentiates from. Future contributors are expected to update this file when they introduce a citation or mapping that materially changes how OAK relates to a referenced project.

Adversary-behavior taxonomies

OAK uses a tactics-as-columns / techniques-as-cells matrix structure, which is a well-established pattern in adversary-behavior taxonomies broadly. The matrix shape itself is not unique to any single project — it reflects how adversary behaviour is most readably catalogued for defenders.

OAK's contribution sits in what the matrix catalogues, not the matrix shape: crypto's threat surface (public-ledger finality, pseudonymous adversaries, smart-contract attack surfaces, money-laundering as part of the attack chain rather than aftermath) doesn't fit cleanly into general IT-security frameworks. OAK is built from scratch for this segment.

What OAK does differently from general adversary-behavior frameworks:

Scope. On-chain-specific, with explicit chain-coverage labelling (EVM / Solana / cross-chain / etc.) per Technique.
Loss + recovery + attribution-strength layer. Every worked example carries dollar-loss, laundering rail, recovery outcome, and explicit attribution-strength (confirmed / inferred-strong / inferred-weak).
Laundering pipeline as first-class. Laundering rails (mixers, CEX deposit-address layering, cross-chain bridge laundering, NFT wash-laundering, privacy-chain hops) are catalogued as Tactic-level, not aftermath.
License. Content under CC-BY-SA 4.0, tooling under MIT — fully redistributable and adaptable.
Governance. v0.1 single-maintainer with PR-based community contribution; co-maintainers invited by v0.5; long-term sustainability and incorporation patterns under research per ROADMAP.

OWASP Smart Contract Top 10

The closest existing public framework, by audience and intent, is the OWASP Smart Contract Top 10, developed under the OWASP Foundation's Smart Contract Security project (2025/2026 edition: [owaspscstop10]). The 2025/2026 list is derived from ~122 deduplicated smart-contract incidents totalling ~$905.4M in tracked losses, with categories including access-control vulnerabilities, price oracle manipulation, and flash-loan attacks.

OAK and OWASP SC Top 10 are orthogonal and complementary. Specifically:

OWASP SC Top 10 occupies the contract-vulnerability layer: what makes a contract exploitable.
OAK occupies the operator-behaviour layer: what an adversary does with a contract once deployed, what they do to capture holders, how they extract value, how they launder.
The two layers overlap at oracle manipulation and flash-loan attacks (OAK plans to add these as Techniques per TAXONOMY-GAPS.md), but OWASP describes them as protocol-vulnerability classes whereas OAK describes them as adversary behaviours with attribution and mitigation framing.

A contributor mapping their detector should expect to claim coverage in both frameworks, not one or the other. v0.5 of OAK plans to publish a cross-walk between OAK Techniques and OWASP SC Top 10 categories so vendors can produce coverage statements without translation overhead.

OWASP SC Top 10 also provides an indirect prior-art baseline for how OAK's coverage matrix should evolve: OWASP's empirical sourcing methodology (deduplicate, categorise, count) is consistent with what COVERAGE.md aspires to at v0.5.

Academic taxonomy: SoK papers

The single most important academic prior-art reference for OAK is [zhou2023sok] — SoK: Decentralized Finance (DeFi) Attacks by Liyi Zhou, Xihan Xiong, Jens Ernstberger, Stefanos Chaliasos, Zhipeng Wang, Ye Wang, Kaihua Qin, Roger Wattenhofer, Dawn Song, and Arthur Gervais (arXiv:2208.13035, latest revision 2023). The paper systematises 7 prior surveys, 29 security tools, and 42 attack papers, documenting ~$3.24B in DeFi-protocol-layer losses from April 2018 through April 2022.

OAK uses Zhou et al. 2023 as the canonical academic taxonomy of DeFi-protocol-layer attacks — a category that OAK v0.1 specifically does not cover. When OAK v0.3 introduces a Smart-Contract Exploit Tactic, Zhou et al.'s threat-model taxonomy (utilities, goals, knowledge, capabilities) will be the canonical reference frame for the new Techniques.

Other academic prior art used directly in OAK Technique pages:

[torres2019] — honeypot detection (foundational for OAK-T1.001).
[victor2021washtrade] — wash-trade detection (foundational for OAK-T3.002).
[xia2021mintdump] — Uniswap scam-token characterisation (foundational for OAK-T5.003).
[daian2019flashboys] — MEV characterisation (foundational for OAK-T5.004).
[liu2025sybil] — sybil-detection methodology (with airdrop-vs-launch caveat, applied to OAK-T3.001 and T8.001).
[karbalaii2025], [bolz2024] — pump-and-dump cohorts (foundational for OAK-T3.003).
[solrpds], [tmrugpull2026], [frp2025] — rug-pull datasets (foundational for OAK-T2.001 / T2.002 / T5.001 / T5.002).
[tsuchiya2025poisoning] — address poisoning at scale (driver for the missing T4 sub-Technique called out in TAXONOMY-GAPS.md).

Industry intelligence sources

OAK draws on industry reports for incident attribution and aggregate context. These are not academic in the peer-reviewed sense, but they are the canonical operational record of the events that OAK Techniques describe:

Chainalysis — [chainalysis2025rug] (2025 Crypto Crime Report market-manipulation chapter), [chainalysis2021scams] (2021 rug-pull retrospective), [chainalysis2024dprk] (DPRK attribution scale).
TRM Labs — [trmsquid2021] (SQUID-cluster cross-incident attribution).
SlowMist — [slowmist2024report] (2024 Blockchain Security and Anti-Money Laundering Annual Report).
EigenPhi — [eigenphijared2023] (sandwich-MEV operator profiling).
Elliptic — [ellipticronin2022] (Ronin Bridge forensic write-up).
rekt.news — [rektcurve2022] (Curve DNS hijack write-up).

OAK treats these as primary sources when they document specific incidents with public on-chain artefacts. Vendor reports that re-state aggregate statistics without primary-source attribution are not citation candidates.

Regulatory and government sources

OAK preferentially cites regulatory and government records when they are available, because they are the strongest possible source for claims about specific incidents:

U.S. SEC — [secsafemoon2023] (SafeMoon civil enforcement action and complaint).
U.S. CFTC — [cftcmango2023] (Mango Markets oracle-manipulation enforcement).
U.S. Treasury (OFAC) — [ofac2022tornado] (Tornado Cash SDN designation, August 2022; partially delisted 2025 following Fifth Circuit ruling).

Out-of-scope-at-v0.1 government sources OAK has not yet pulled but plans to in v0.2: FBI IC3 cryptocurrency annual reports, FinCEN advisories, NCSC (UK) cryptocurrency alerts, CISA DPRK-related alerts.

Vendor-side frameworks not yet pulled

Several vendor-published frameworks and methodologies are on OAK's radar but not yet integrated as references at v0.1. They will be pulled into v0.2's PRIOR-ART expansion as the framework matures and as vendor mappings begin landing via PR:

Halborn's own attack methodology and incident retrospectives.
Trail of Bits' security frameworks and audit-finding retrospectives.
OpenZeppelin's smart-contract security frameworks.
PeckShield, CertiK Skynet, Beosin annual retrospectives.
Smart Contract Security Verification Standard (SCSVS).

OAK does not endorse any of these by listing them; the policy in CONTRIBUTING.md is that vendors submit Reference-implementation entries via PR, with public-verifiability requirements. Vendor frameworks are listed here only as prior art for the methodology layer, not as endorsed coverage providers.

Audit-competition findings as a data source — evaluated, not integrated

Accepted findings from competitive-audit platforms (Code4rena, Sherlock) are periodically proposed as an "empirical validation set" for OAK. The idea is intuitive: OAK catalogues what attackers did, audit findings catalogue what reviewers caught, so the overlap should validate the taxonomy. OAK evaluated this in July 2026 and decided against integration. The findings are recorded here so the evaluation is not repeated.

What is publicly available. The data is genuinely open and reasonably structured — no vendor or API is required:

github.com/code-423n4 — 376 *-findings repositories, one per contest.
github.com/sherlock-audit — 230 *-judging repositories.
Each C4 findings repo carries data/<handle>-<issueId>.json per submission (contest, handle, risk where 3 = High and 2 = Med, title, issueId, issueUrl), plus report.md — the curated, judged, already deduplicated final report. Work from report.md, not from raw issues.
Judging outcomes live in issue labels: satisfactory, unsatisfactory, duplicate-<primaryId>, confirmed for report, downgraded by judge, disagree with severity.

Why headline finding-counts overstate the corpus. Raw submission counts are dominated by noise and duplicates. Measured on one contest (2023-10-zksync-findings): 1,142 issues total, 86 satisfactory, 788 unsatisfactory — roughly an 8% acceptance rate on raw submissions, and the accepted set still contains mutual duplicates that collapse via the duplicate-<id> labels. Any figure quoted as "N accepted findings" is therefore a count of submissions, not of distinct validated vulnerabilities; the distinct-bug count is substantially lower.

Why the mapping does not pay off. Code4rena tags findings with an ### Assessed type field, which is the natural join key to an OAK Technique ID. Sampled across five contests (76 classified findings), the distribution was: Context 28 (37%), Math 6, Invalid Validation 5, Access Control 5, Timing 4, DoS 4, en/de-code 3, Error 3, ERC20 3, Under/Overflow 2, Reentrancy 2, Governance 2, ERC4626 2, and singletons (MEV, Rug-Pull, Uniswap, Library, Payable, Decimal, Other). Three problems follow:

The plurality category, Context, carries no classification information. Only ~15–20% of the sample maps cleanly onto an OAK Technique, and every one of those (Reentrancy → T9.005, Access Control → T9.004, Governance → T9.003) is a Technique OAK already documents and already anchors to realised incidents with losses.
The axes differ. Assessed type classifies the nature of a code defect; OAK classifies adversary behaviour. These are adjacent but not joinable without a lossy hand-mapping.
The corpus overlap is one Tactic of fourteen. Competitive audits speak to T9 (Smart-Contract Exploit). They have no analogue for T7 Laundering, T11 Custody/Signing, T5 Value Extraction, T8 Operator Continuity, or T4 Access Acquisition — key compromise, quorum capture, package-registry supply chain, and mixer routing are not things a contest can find.

The structural blocker. Audit findings are pre-deployment defects that were caught; they never became incidents. CONTRIBUTING.md requires a publicly verifiable incident for a worked example, and tools/check_backlinks.py enforces zero Techniques without a canonical example anchor. A Technique minted from an audit finding would have no anchor and would fail CI. This is not a matter of editorial preference — the corpus cannot accept this class of input as-is.

What remains worth stating, and it needs no external data. The useful claim runs the other way: competitive audit does not reach most of where realised losses occur, and OAK can evidence that from its own corpus — T9 is a minority of documented examples, while T7, T11, T5, and T8 together dominate. A contributor wanting to pursue the forward direction anyway should scope it to satisfactory + risk: 3 findings from report.md, dedupe on duplicate-<id>, and treat the result as a prevalence signal about pre-deployment defect classes — not as validation of OAK's incident corpus.

On vendor offers. Commercial re-packagings of this data appear periodically, sometimes as GitHub issues offering "free" access. The underlying records are public; the policy in CONTRIBUTING.md applies unchanged, and paid analysis services are not reference_implementations candidates.

How OAK positions itself

The 30-second elevator pitch with prior art surfaced:

OWASP Smart Contract Top 10 is the closest existing framework; it occupies the contract-vulnerability layer; OAK occupies the orthogonal operator-behaviour layer; the two are complementary, not competing.
Zhou et al. 2023 SoK is the academic taxonomy of DeFi-protocol-layer attacks; v0.3 of OAK will build on it for the Smart-Contract Exploit Tactic.
Chainalysis / TRM / SlowMist / EigenPhi / Elliptic / rekt.news are the industry intelligence sources OAK draws on for incident-level attribution.
U.S. SEC / CFTC / Treasury records are preferentially cited for incidents with regulatory action because they are the strongest available source.

OAK adds: a stable identifier system for vendor-coverage claims; a defender-perspective per-Technique template that is shorter and more contributor-friendly than the academic SoK; and an explicit honest-coverage convention (COVERAGE.md) that comparable frameworks do not formally publish.

How this file evolves

When OAK adds a new Tactic or major axis (Threat Actors, Data Sources), the corresponding section in this file is expanded with the prior-art references that informed the new content. When a vendor or framework is integrated as an authoritative reference (rather than as prior art), it migrates to the appropriate Technique page or to a new IMPLEMENTATIONS.md registry document.

OAK — Prior Art and Related Frameworks