- JSONL loader with capability/scenario filtering and Pydantic validation - Sample-to-EvalSample conversion for eval scenario (strips Gold_Trace) - 6 fake samples across 5 capabilities with realistic research tasks - Each sample has complete 7-tuple structure with gold traces and negative cases Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> |
||
|---|---|---|
| airbench | ||
| data/fake | ||
| docs | ||
| requirements.txt | ||
| setup.py | ||