README.md

---
language:
  - en
license: mit
pretty_name: "SPECA Vulnerability Reports"
tags:
  - "security"
  - "smart-contracts"
  - "vulnerabilities"
  - "audit"
---

# SPECA Vulnerability Reports

A normalized corpus of public smart-contract audit findings, curated by
the [SPECA](https://github.com/NyxFoundation/speca) project. Each row is
one expert-judged audit issue, deduplicated across public competitive-audit
platforms.

The dataset is split into one **config per security domain** (`defi`,
`lending`, `oracle`, …). HuggingFace auto-detects each `<domain>/`
folder as a config; pick the domain you want at load time.

## Loading

```python
from datasets import load_dataset

# A specific domain (recommended).
ds = load_dataset("NyxFoundation/vulnerability-reports", "defi", split="train")

# Or, from inside the SPECA repo:
from scripts.datasets.load import load_findings
df = load_findings(domain="defi").to_pandas()
```

## Schema

| Field | Type | Description |
|---|---|---|
| `id` | string | Stable key, `<platform>:<contest-slug>:<issue_id>`. |
| `source_platform` | string | One of `code4rena`, `sherlock`, `codehawks`. |
| `contest` | string | Slugified contest identifier. |
| `issue_id` | string | Platform-local issue / finding id, `#`-stripped. |
| `severity` | string | `High` / `Medium` / `Low` / `Info`. |
| `title` | string | Verbatim issue title from the upstream platform. |
| `description` | string | Verbatim issue body. |
| `source_url` | string | Best-effort upstream link (deterministic for code4rena; may be empty for others — check the scraper output if you need a guaranteed URL). |
| `domain` | string | Matches the config name (`defi`, `lending`, …). |
| `scraped_at` | string | ISO 8601 UTC timestamp of the underlying scrape that produced this row. |

Per-build state for each domain — row count, scrape time, source CSVs,
`speca_commit` — lives in `<domain>/manifest.json` rather than this card,
so re-publishing one domain doesn't churn another's metadata.

## Provenance

All findings originate from public competitive-audit platforms:

- [Code4rena](https://code4rena.com/) — contest reports published openly
  on GitHub under the `code-423n4` org.
- [Sherlock](https://audits.sherlock.xyz/) — contest reports published
  via Sherlock's public site and their `sherlock-audit` GitHub org.
- [CodeHawks](https://codehawks.cyfrin.io/) — contest reports published
  on the CodeHawks platform.

This dataset redistributes the publicly accessible finding metadata
(title, description, severity, source URL). The `source_url` column
points back to the upstream record. If you are the original author of a
finding and want it removed, please open an issue on the
[SPECA repository](https://github.com/NyxFoundation/speca/issues).

## How this dataset is built

The build pipeline lives in
[`scripts/datasets/`](https://github.com/NyxFoundation/speca/tree/main/scripts/datasets)
of the SPECA repo:

1. `scripts/scrape_*.py` pulls raw data from each platform into CSVs
   under `benchmarks/data/defi_audit_reports/`.
2. `scripts/datasets/build_derived.py` normalizes those CSVs (and the
   curated `csv/similar_audit_findings.csv`) into the schema above and
   emits a parquet + manifest under `dist/datasets/<domain>/`.
3. `scripts/datasets/publish_hf.py` pushes the parquet plus this README
   to `NyxFoundation/vulnerability-reports` on the `main` branch, scoped to the published
   domain's folder.

The `.github/workflows/datasets-publish.yml` workflow runs the last two
steps on a self-hosted runner via `workflow_dispatch`.

## License

The schema, build code, and dataset card are MIT-licensed. Original
finding text is reproduced for research / educational use under the
public terms of each upstream platform; consult the source URL for the
authoritative version.