Datasets:
Formats:
parquet
Sub-tasks:
closed-domain-qa
Size:
1K - 10K
Tags:
forecasting
event-prediction
benchmark
llm-evaluation
contamination-free
prospective-evaluation
License:
| cff-version: 1.2.0 | |
| message: If you use this dataset, please cite it as below. | |
| title: "WorldCup Arena — frozen prediction archive" | |
| abstract: >- | |
| The complete frozen record of a prospective, leakage-free evaluation of six | |
| frontier LLMs over the 2026 FIFA World Cup: 4,494 scored predictions, 104 | |
| official results, 46 timestamped snapshots of 48 team dossiers, 848 raw model | |
| responses and 171 byte-identical archived prompts. | |
| type: dataset | |
| license: CC-BY-4.0 | |
| year: 2026 | |
| authors: | |
| - family-names: Wang | |
| given-names: Zhenran | |
| - family-names: Bian | |
| given-names: Zhonghan | |
| - family-names: Li | |
| given-names: Jinsong | |
| - family-names: Qi | |
| given-names: Zhangyang | |
| keywords: | |
| - forecasting | |
| - llm-evaluation | |
| - benchmark | |
| - prospective-evaluation | |