Add dataset card with usage instructions

README.md

@@ -0,0 +1,59 @@
+---
+license: mit
+tags:
+  - beasty-bar
+  - game-ai
+  - tablebase
+  - endgame
+---
+
+# Beasty Bar Tablebase
+
+Endgame tablebase for [Beasty Bar](https://www.zoch-verlag.com/zoch_en/brands/beasty-bar/), providing optimal play for positions with 4 or fewer cards remaining per player.
+
+**GitHub**: [diegooprime/beastybar](https://github.com/diegooprime/beastybar)
+
+## Files
+
+| File | Description | Size |
+|------|-------------|------|
+| `endgame_4card_final.tb` | Complete 4-card endgame tablebase | ~1.2GB |
+
+## What is a Tablebase?
+
+A tablebase pre-computes optimal moves for all possible game positions within a certain scope. For Beasty Bar, this covers all positions where each player has 4 or fewer cards remaining in hand.
+
+When the neural network encounters these endgame positions, it can use the tablebase for perfect play rather than relying on learned approximations.
+
+## Usage
+
+```python
+from huggingface_hub import hf_hub_download
+
+# Download tablebase
+tablebase_path = hf_hub_download(
+    repo_id="shiptoday101/beastybar-tablebase",
+    filename="endgame_4card_final.tb",
+    repo_type="dataset"
+)
+
+# Load with the game repo
+from _02_agents.tablebase import Tablebase
+
+tb = Tablebase.load(tablebase_path)
+optimal_action = tb.lookup(game_state)
+```
+
+## Coverage
+
+- **Positions covered**: All 2-player positions with <= 4 cards per hand
+- **Format**: Binary lookup table with state hashing
+
+## Related
+
+- **Neural Network**: [shiptoday101/beastybar-ppo](https://huggingface.co/shiptoday101/beastybar-ppo) - PPO-trained agent for full game play
+- **Game Rules**: [Beasty Bar PDF](https://tesera.ru/images/items/1525203/BeastyBar_EN-online.pdf)
+
+## License
+
+MIT