README.md

---
license: cc-by-nc-4.0
language:
- en
- et
task_categories:
- visual-question-answering
- question-answering
tags:
- physics
- olympiad
- multimodal
- evaluation
- audited
- novel-source
size_categories:
- n<1K
configs:
- config_name: default
  data_files: eval_olympiad_v2.jsonl
---

# eval_olympiad_v2

A 500-problem 99.8%-novel-source open-ended physics olympiad evaluation set, two-stage audited (5-gram Jaccard ≥ 0.4, mxbai-embed-large cosine ≥ 0.85) against PhyX, MMMU-Pro Physics, OlympiadBench-Physics, and UGPhysics-Train.

## Sources
- **Estonian Physics Olympiad** — with native 1–10 organizer-issued difficulty labels and 201 EN/ET bilingual subset
- **Kevin Zhou's olympiad handouts** — with native 1–5 point values
- Seven international olympiad scrapes (USAPhO, IPhO, APhO, EuPhO, OPhO, PUPC, PhysicsBowl)

## Schema (per record)
- `messages`: list of `{role, content}` (chat-format problem statement; may include `<image>` tags)
- `solution`: gold solution / final answer (string; may include LaTeX)
- `source`: canonical source identifier (e.g., `estonian-2007-lahg-07`)
- `metadata`: dict with `competition`, `year`, `round`, `problem_number`, `difficulty_native` (Estonian), `language`, `image_paths`

## License
**CC BY-NC 4.0** — non-commercial; attribution required. Olympiad source materials are typically released by their organizers under non-commercial / educational-use terms; this redistribution honors those constraints. Cite the original olympiad organizers in addition to this dataset.

## Citation
Anonymous NeurIPS 2026 Datasets and Benchmarks Track submission.