Srivatsormylord/mafia-training-scenarios-claude

---
license: mit
task_categories:
- text-generation
- question-answering
language:
- en
tags:
- mafia
- game
- social-deduction
- strategy
size_categories:
- 1K<n<10K
---

# Mafia AI Training Dataset

A comprehensive dataset of **9950 scenarios** for training AI to play the social deduction game Mafia with 6 different roles.

## Dataset Description

**Total Scenarios:** 9950

**Roles Covered:**

- 🃏 **Jester** (2,039 scenarios) - Win by getting eliminated
- 🔫 **Mafia** (1,847 scenarios) - Eliminate town members
- 🔍 **Sheriff** (1,530 scenarios) - Investigate players
- 🔇 **Silencer** (1,341 scenarios) - Prevent players from talking
- 💉 **Doctor** (1,318 scenarios) - Protect players from elimination
- 🎭 **Vigilante** (1,000 scenarios) - Game role
- 🧠 **Analyst** (875 scenarios) - Gather information and analyze patterns

## Format

Each example contains:
- `scenario`: The game situation
- `role`: Player's role (analyst, doctor, jester, mafia, sheriff, silencer, vigilante)
- `situation_type`: Type of scenario (strategy, bluffing, psychology, etc.)
- `response`: Optimal strategic response
- `reasoning`: Why this response is optimal
- `common_mistake`: What bad players do in this situation
- `id`: Unique identifier

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("Srivatsormylord/mafia-training-scenarios-claude")
print(dataset['train'][0])
```

## Example

```json
{
"scenario": "You are the Silencer. Should you silence the Jester to prevent them from generating suspicious behavior that might overshadow Mafia's actual work?",
"role": "silencer",
"situation_type": "strategy",
"response": "Generally no. The Jester draws attention away from Mafia, which is beneficial. Only silence the Jester if they're about to accidentally expose a Mafia member through their chaotic accusations.",
"reasoning": "Jester chaos is usually Mafia's ally. Silencing the Jester removes a useful distraction. Preserve your action for information-critical targets like the Sheriff.",
"common_mistake": "Silencing the Jester because they're annoying to deal with, losing your action on a target that was helping Mafia's narrative confusion.",
"id": 1
}
```

## Statistics

- **Total Examples:** 9,950
- **Unique Roles:** 7
- **Most Common Role:** Jester (2,039 scenarios)
- **Least Common Role:** Analyst (875 scenarios)

## Training

This dataset is designed for training language models (10-100M parameters) to play Mafia with:
- Strategic reasoning
- Social deduction
- Bluffing and psychology
- Role-specific tactics

Recommended for:
- Fine-tuning small language models
- Training from scratch
- Game AI research
- Social deduction strategy analysis

## License

MIT License - Free to use for any purpose.