File size: 1,306 Bytes
29a12ae | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 | # Benchmark Data Card
The detailed StableEval Arena data card is maintained at:
`data/benchmark/data_cards/benchmark_data_card.md`
It documents source APIs, asset coverage, time window, frozen-window construction, label definitions, feature fields, leakage controls, limitations, and redistribution caveats.
## Release Scope
The GitHub release contains two benchmark blocks:
| Block | Contents |
|---|---|
| `StableEval-120-Enriched` | 120 stress-enriched validation cases, prompt packets, six-system predictions, metrics, ablations, and freeze manifest |
| `StableEval-507-Natural` | 507 natural-distribution cases, labels, prompt packets, baselines, four native/framework-level agents, two adapter-based `-Claw` extensions, audits, and v6 leaderboard |
## License And Redistribution Notice
Repository code, scripts, prompts, and documentation are licensed under Apache-2.0 unless otherwise stated. Benchmark data and result artifacts may include processed or derived market-data artifacts and LLM-generated outputs; these artifacts are included for academic reproducibility and may remain subject to the original providers' terms. Raw provider/vendor snapshots are not redistributed.
For the Hugging Face package, use:
```bash
python scripts/release/prepare_huggingface_dataset.py --root .
```
|