---
language:
- "py"
license:
- "mit"
pretty_name: "W3SA Solidity Access Control Benchmark"
---
# W3SA - Solidity Repo Benchmark
This benchmark evaluates audited EVM projects containing smart contracts written in Solidity. It’s designed to assess the vulnerability detection capabilities of AI models and static analysis tools. The benchmark tests the ability of different models to detect high and medium severity vulnerabilities. Our model ALMX 1.5 demonstrates a significant advantage by catching up to 5x more high-severity vulnerabilities than traditional static analysis tools like Slither, with a slight improvement in performance on Medium Severity bugs and while outputting one fourth of the findings, reducing overall noise. 

The benchmark covers 8 audited Solidity projects sourced from diverse origins such as Code4rena competitions, open datasets, and audit reports from various Web3 security firms.

## Project Statistics

| Project                 | Platform                | Scanned Files | Total files | Cri-sev | High-sev | Med-sev | Low/Info-Sev |
|-------------------------|-------------------------|---------------|-------------|---------|----------|---------|--------------|
| Connext                 | Eth, Optimism, Polygon  | 1             | 6           | 0       | 5        | 2       | 5            |
| Sandclock               | Eth                     | 11            | 25          | 0       | 5        | 15      | 5            |
| InsureDAO               | Eth, Arbitrum, Optimism | 10            | 30          | 0       | 13       | 9       | 0            |
| algebra                 | Base                    | 5             | 175         | 0       | 5        | 4       | 16           |
| arcadia_finance         | Base                    | 21            | 409         | 0       | 1        | 7       | 4            |
| seamlessfi              | Base                    | 8             | 57          | 2       | 0        | 5       | 19           |
| gamma                   | Eth                     | 5             | 19          | 0       | 8        | 2       | 0            |
| Beebots                 | Eth                     | 1             | 1           | 0       | 6        | 3       | 7            |
| Total ground truth bugs |                         | 62            | 722         | 2       | 43       | 47      | 56           |

<!-- ![Alt Text](./resources/benchmark.png) -->
<img src="./resources/benchmark.png" alt="Solidity Codebase Benchmark" style="border: 2px solid black; border-radius: 15px;" width="500">

ALMX-1.5 was able to capture 21.42% of High Severity bugs, against 14.28% captured by o1 and 4.76% captured by Slither. 

## Repo Structure

```
.
├── README.md
├── benchmark
│   ├── config
│   ├── ground_truth
│   └── repositories
├── bm_src
└── resources
..
```
## Set up
- Install `uv` package manager if not yet available
- Run `uv sync`

## Run an experiment
- Set your `OPENAI_API_KEY` or `ANTHROPIC_API_KEY` as environmental variable
- Launch your experiment by running:
```bash
uv run experiment.py --model o3-mini
```
A `tqdm` progress bar will pop up and, at the end of the experiment, the results metrics will be printed out.

## Contact Us 
For or questions, suggestions, or to learn more about Almanax.ai, reach out to us at https://www.almanax.ai/contact