---
license: mit
task_categories:
- question-answering
- text-generation
language:
- en
tags:
- game-theory
- reasoning
- connect4
- synthetic-dataset
- logic
pretty_name: Connect4 Reasoning Task
size_categories:
- n<1K
---

# Dataset Card for Connect4 Reasoning Task

## 1. Dataset Summary
This dataset is dynamically constructed using the **GAMEBoT** framework in the **Connect4** game. It is designed to evaluate Large Language Models' (LLMs) ability in symbolic reasoning, board state parsing, and lookahead planning. 

By serializing 6 &times; 7 grid states into text-based formats, this dataset challenges models to identify winning topologies in a deterministic environment with a state-space complexity of approximately 4.5 &times; 10<sup>12</sup>, ensuring a robust evaluation against data contamination.

## 2. Construction Methodology
- **State Generation**: Two randomized agents simulated gameplay to generate a diverse distribution of board states, covering both balanced and critical tactical scenarios.
- **Filtering and Balancing**: 
  - Duplicate states were removed.
  - To prevent label imbalance (where LLMs might score high by simply predicting "no winning moves"), states with empty answers were downsampled to a **maximum ratio of 20%**.
- **Ground Truth**: All labels were verified using a perfect solver integrated within the GAMEBoT engine.

## 3. Evaluation Protocol
For each instance, the model is prompted to solve two sub-problems:
1. **Self-Winning**: "Are there any potential winning moves to form 4-in-a-row for you? Output all winning moves."
2. **Opponent-Winning**: "Are there any potential winning moves to form 4-in-a-row for your opponent? Output all winning moves."

The evaluation extracts the final answer from the model's **Chain-of-Thought (CoT)** and compares it against the engine-validated ground truth.

## 4. Citation
If you find this dataset helpful in your research, please cite the following work:

```bibtex
@inproceedings{lin2025gamebot,
  title={GAMEBoT: Transparent assessment of LLM reasoning in games},
  author={Lin, Wenye and Roberts, Jonathan and Yang, Yunhan and Albanie, Samuel and Lu, Zongqing and Han, Kai},
  booktitle=ACL,
  year={2025}
}