Datasets:

Farseen0
/

scar-eval

query large_stringlengths 39 1.7k	ground_truth_code large_stringlengths 26 116k	severity large_stringclasses 3 values	vuln_type large_stringclasses 95 values	report_name large_stringclasses 16 values	audit_firm large_stringclasses 7 values	report_date large_stringclasses 10 values	source large_stringclasses 10 values
HIGH severity: 18_deploy_RollupRevenueVault.ts – Deployment Script Leaves Contract Uninitialized; fallback Does Not Enforce msg.value > 0. The deployment script for the RollupRevenueVault contract attempts to call a non-existent zero-parameter initialize() function, which causes the call to be routed to the fallback fu...	// SPDX-License-Identifier: Apache-2.0 OR MIT pragma solidity 0.8.30; import { AccessControlUpgradeable } from "@openzeppelin/contracts-upgradeable/access/AccessControlUpgradeable.sol"; import { IERC20 } from "@openzeppelin/contracts/token/ERC20/IERC20.sol"; import { L2MessageService } from "../messaging/l2/L2MessageS...	HIGH	CWE-691\|CWE-705\|CWE-584	Linea - Burn Mechanism _ Consensys Diligence.md			FORGE-Curated
MEDIUM severity: RollupRevenueVault - Deployment and Initialization Flow. The RollupRevenueVault contract implements two initialization functions: an initial 'initialize' with the 'initializer' modifier and a 'reinitializer(2)' function intended for upgrades. However, the use of a reinitializer that becomes immediately...	// SPDX-License-Identifier: Apache-2.0 OR MIT pragma solidity 0.8.30; import { AccessControlUpgradeable } from "@openzeppelin/contracts-upgradeable/access/AccessControlUpgradeable.sol"; import { IERC20 } from "@openzeppelin/contracts/token/ERC20/IERC20.sol"; import { L2MessageService } from "../messaging/l2/L2MessageS...	MEDIUM	CWE-691\|CWE-362	Linea - Burn Mechanism _ Consensys Diligence.md			FORGE-Curated
HIGH severity: Potential for Arbitrary Application in Message Verification. This vulnerability allows an attacker to manipulate the application identifier during message verification. The issue arises because the broadcast message does not include the application address in its hash, despite the `source` field in the p...	// SPDX-License-Identifier: MIT pragma solidity ^0.8.26; import { IAttester } from "../../../interfaces/IAttester.sol"; import { LibAddress } from "../../../libs/LibAddress.sol"; import { MessageEncodingLib } from "../../../libs/MessageEncodingLib.sol"; import { BaseInputOracle } from "../../../oracles/BaseInputOracle...	HIGH	CWE-693\|CWE-345\|CWE-346	OIF Broadcaster Audit.md			FORGE-Curated
HIGH severity: Prover Copies Cannot Be Updated. The `updateBlockHashProverCopy` function in the `Receiver` contract fails to update prover copies due to a revert when attempting to read the version from the zero address. This occurs because the `_blockHashProverCopies` mapping is initialized to address(0), and the func...	// SPDX-License-Identifier: MIT pragma solidity ^0.8.27; import {IReceiver} from "./interfaces/IReceiver.sol"; import {IBlockHashProver} from "./interfaces/IBlockHashProver.sol"; import {IBlockHashProverPointer} from "./interfaces/IBlockHashProverPointer.sol"; import {BLOCK_HASH_PROVER_POINTER_SLOT} from "./BlockHashP...	HIGH	CWE-703\|CWE-755\|CWE-248	OIF Broadcaster Audit.md			FORGE-Curated
MEDIUM severity: Custom Fees Settable up to 100% in fulfillFastRedeems Can Seize Redemptions. The fulfillFastRedeems function allows a caller with the FULFILL_MANAGER_ROLE to specify a customFees array. While the code checks that the fee does not exceed 100% (feeBps <= BASIS_POINTS), it lacks a stricter, more reasonabl...	// SPDX-License-Identifier: BUSL-1.1 // Terms: https://liminal.money/xtokens/license pragma solidity 0.8.28; import {IERC20} from "@openzeppelin/contracts/token/ERC20/IERC20.sol"; import {IERC20Metadata} from "@openzeppelin/contracts/token/ERC20/extensions/IERC20Metadata.sol"; import {SafeERC20} from "@openzeppelin/c...	MEDIUM	CWE-707\|CWE-20\|CWE-1284	cantina_liminal_oct2025.pdf	Cantina	2025-10-12	FORGE-Curated
MEDIUM severity: Unsafe Type Conversion for Refund Recipient Can Lead to Funds Lost to 0xdead Address. In the lzCompose function's catch block, the logic to determine the refund recipient contains an unsafe type conversion. The code extracts a bytes32 value (extractedRecipient) and checks if it is non-zero before conve...	// SPDX-License-Identifier: BUSL-1.1 // Terms: https://liminal.money/xtokens/license pragma solidity 0.8.28; import {SafeERC20, IERC20} from "@openzeppelin/contracts/token/ERC20/utils/SafeERC20.sol"; import {ReentrancyGuardUpgradeable} from "@openzeppelin/contracts-upgradeable/utils/ReentrancyGuardUpgradeable.sol"; i...	MEDIUM	CWE-707\|CWE-20\|CWE-179	cantina_liminal_oct2025.pdf	Cantina	2025-10-12	FORGE-Curated
MEDIUM severity: Critical Oracle Parameters Lack Timelock Protection. The setPriceId and setPriceIds functions in PythPriceOracle.sol, which configure Pyth price feed IDs and decimals, are only protected by PRICE_MANAGER_ROLE and lack timelock protection. A compromised or malicious price manager could instantly change ...	// SPDX-License-Identifier: BUSL-1.1 // Terms: https://liminal.money/xtokens/license pragma solidity 0.8.28; import {AccessControlUpgradeable} from "@openzeppelin/contracts-upgradeable/access/AccessControlUpgradeable.sol"; import {IPriceOracle} from "./interfaces/IPriceOracle.sol"; import {IERC20Metadata} from "@open...	MEDIUM	CWE-693	cantina_liminal_oct2025.pdf	Cantina	2025-10-12	FORGE-Curated
MEDIUM severity: Oracle Manipulation Enables Attackers Unfair Share Minting and Inflated Redemptions. The protocol uses spot prices from PythPriceOracle for asset valuation, making it vulnerable to price manipulation via flash loans or other means. Attackers can exploit this during deposits (to mint excessive shares) o...	// SPDX-License-Identifier: BUSL-1.1 // Terms: https://liminal.money/xtokens/license pragma solidity 0.8.28; import {AccessControlUpgradeable} from "@openzeppelin/contracts-upgradeable/access/AccessControlUpgradeable.sol"; import {IPriceOracle} from "./interfaces/IPriceOracle.sol"; import {IERC20Metadata} from "@open...	MEDIUM	CWE-707\|CWE-20	cantina_liminal_oct2025.pdf	Cantina	2025-10-12	FORGE-Curated
MEDIUM severity: Timelock Admin Can Instantly Change Delays, Bypassing Governance Safeguards. The VaultTimelockController's setFunctionDelay and setDefaultFunctionDelay functions are protected only by DEFAULT_ADMIN_ROLE without any timelock, allowing immediate changes to delay periods. This undermines the entire purpos...	// SPDX-License-Identifier: BUSL-1.1 // Terms: https://liminal.money/xtokens/license pragma solidity 0.8.28; import {TimelockControllerUpgradeable} from "@openzeppelin/contracts-upgradeable/governance/TimelockControllerUpgradeable.sol"; /** * @title VaultTimelockController * @notice Enhanced timelock controlle...	MEDIUM	CWE-693	cantina_liminal_oct2025.pdf	Cantina	2025-10-12	FORGE-Curated
MEDIUM severity: Privileged Role Can Arbitrarily Set NAV and Drain Instant Redeemable Funds Due to Circumventable & Missing Safeguards. The NAVOracle contract allows the VALUATION_MANAGER_ROLE to call setTotalAssets and directly overwrite the Net Asset Value (NAV). While a maxPercentageIncrease check exists, it can be ...	// SPDX-License-Identifier: BUSL-1.1 // Terms: https://liminal.money/xtokens/license pragma solidity 0.8.28; import {AccessControlUpgradeable} from "@openzeppelin/contracts-upgradeable/access/AccessControlUpgradeable.sol"; import {Math} from "@openzeppelin/contracts/utils/math/Math.sol"; import {IERC20Metadata} from ...	MEDIUM	CWE-693	cantina_liminal_oct2025.pdf	Cantina	2025-10-12	FORGE-Curated
MEDIUM severity: Incorrect Performance Fee Calculation. 1. Description: The collectPerformanceFee function in the FeeManager contract fails to differentiate between actual yield generated by the vault's investment strategies and new capital deposited by users. This leads to inaccurate performance fee calculations. ...	// SPDX-License-Identifier: BUSL-1.1 // Terms: https://liminal.money/xtokens/license pragma solidity 0.8.28; import {AccessControlUpgradeable} from "@openzeppelin/contracts-upgradeable/access/AccessControlUpgradeable.sol"; import {ReentrancyGuardUpgradeable} from "@openzeppelin/contracts-upgradeable/utils/ReentrancyG...	MEDIUM	CWE-682	cantina_liminal_oct2025.pdf	Cantina	2025-10-12	FORGE-Curated
MEDIUM severity: High-Water Mark Logic Penalizes Performance Fee Collection. The collectPerformanceFee function in FeeManager updates the lastSupplyForPerformance state variable before minting fee shares. This causes the high-water mark to be set pre-dilution, so the next fee calculation interprets the dilution as a pe...	// SPDX-License-Identifier: BUSL-1.1 // Terms: https://liminal.money/xtokens/license pragma solidity 0.8.28; import {AccessControlUpgradeable} from "@openzeppelin/contracts-upgradeable/access/AccessControlUpgradeable.sol"; import {ReentrancyGuardUpgradeable} from "@openzeppelin/contracts-upgradeable/utils/ReentrancyG...	MEDIUM	CWE-691\|CWE-670	cantina_liminal_oct2025.pdf	Cantina	2025-10-12	FORGE-Curated

End of preview. Expand in Data Studio

SCAR Evaluation Set (development split)

838 audit-finding queries paired with the vulnerable Solidity snippet each finding refers to, the development split of the SCAR benchmark for retrieving vulnerable smart-contract code from natural-language audit findings. Queries are severity-prefixed findings ("SEVERITY: [title]. [description]"); scoring is over the full 231,269-contract corpus with this split's gold documents appended.

Evaluation protocol (v2, September 2026)

Splits. scar-eval (838 queries) is the development split. scar-eval-2026-holdout (764 queries; Solodit findings ingested after March 2026, audit dates May 2025 to May 2026, disjoint from every training pool by exact-hash and near-duplicate gates) is the held-out split. The held-out split is private at the time of writing; release is planned with the paper.
Training pool. Models scored on either split are trained on scar-pairs-clean (6,202 pairs, revision ec2750c), which is deduplicated against scar-eval at the document level. scar-pairs (7,552 pairs) and scar-pairs-extended (11,961 pairs) are the v1 pools: they share 800 (query, positive) pairs and all 838 positives with scar-eval, so a model trained on them is not scored on scar-eval.
Scoring. R@k over the full corpus (scar-corpus, 231,269 contracts) with the split's gold documents appended (development pool 232,025 documents, held-out pool 232,033). Rank = 1 + the number of non-gold documents scoring at or above the gold document, so ties count against the system; float32 scores; one rule for every row.
Superseded numbers. Tables shown on these pages before September 2026 were computed under the v1 protocol, before the training pool was deduplicated against the development split. They are superseded and no longer shown.

Overlap statement for this split against scar-pairs-clean (revision 8fa29c3, exact match after whitespace normalisation): 0 shared queries; 0 training positives or hard negatives equal to a gold document of this split.

Results (v2 protocol, 8 September 2026)

System	Trained on this task	dev R@10	dev R@100	held-out R@10	held-out R@100
Qwen3-Embedding-4B (512 tokens)	no	0.721	0.820	0.762	0.855
SPLADE-code-0.6B, top-400 terms	no	0.735	0.833	0.827	0.889
SPLADE-code-0.6B, full vocabulary	no	0.696	0.784	0.784	0.851
SPLADE-code-0.6B, top-60 terms	no	0.670	0.773	0.750	0.840
jina-code-1.5B (512 tokens)	no	0.650	0.776	0.689	0.796
Qwen3-Embedding-0.6B (512 tokens)	no	0.655	0.769	0.651	0.780
Qwen3-Embedding-0.6B (256 tokens)	no	0.634	0.748	0.649	0.783
Dense reference (Qwen2.5-Coder-1.5B + LoRA, dense head)	`scar-pairs-clean`	0.662	0.770	0.486	0.617
SCAR sparse (same backbone, trained sparse projection)	`scar-pairs-clean`	0.652	0.765	0.344	0.511
BM25	no	0.310	—	0.052	—

Notes. Stock retrievers are used as released; document text is capped at 6,000 characters and then truncated to each model's token limit. The two trained references use the same backbone, InfoNCE with τ = 0.1, a 25-epoch schedule on scar-pairs-clean@ec2750c and the checkpoint at 25 % of that schedule; one seed is shown, and over three seeds under the earlier evaluator the means are SCAR 0.634 ± 0.016 / 0.375 ± 0.026 and dense 0.656 ± 0.008 / 0.513 ± 0.023 (dev / held-out). BM25 rows come from the earlier evaluator and will be re-scored under the v2 rule. The held-out split separates systems that the development split does not; the paper in preparation analyses that gap.

Source Breakdown

Source	Pairs
Solodit	228
msc-audits-with-reasons	178
msc-smart-contract-auditing	131
FORGE-Curated (original)	91
DeFiHackLabs	67
FORGE-Artifacts	59
FORGE-Curated-v2	36
EVuLLM	32
SmartBugs-Curated	14
GitmateAI	2
Total	838

Usage

from datasets import load_dataset
ds = load_dataset("Farseen0/scar-eval", split="train")

🗂️ All SCAR artifacts (Collection)
scar-pairs-clean — 6,202 training pairs, the pool for every model scored on this split (private at the time of writing)
scar-eval-2026-holdout — 764-query held-out split (private at the time of writing)
scar-corpus — 231,269-contract document pool
scar-pairs — 7,552-pair v1 pool (provenance)
scar-weights — research checkpoints

Paper

Paper in preparation (2026). Code: github.com/FarseenSh/scar-retrieval. Model: Farseen0/scar-weights.

Citation

@misc{shaikh2026scar,
  title  = {SCAR: a benchmark and analysis for retrieving smart-contract audit findings},
  author = {Shaikh, Farseen},
  year   = {2026},
  note   = {Paper in preparation},
  url    = {https://github.com/FarseenSh/scar-retrieval}
}