GambitFlow/Endgame-Tablebase

🔮 GambitFlow: Endgame Oracle (Perfect Knowledge)

📊 Dataset Overview

This dataset provides Perfect Information (God Mode) for chess endgames. It was generated using Syzygy Tablebases, meaning there is no estimation—every evaluation is mathematically proven.

It covers 3, 4, and 5-piece endgames (e.g., King+Pawn vs King, Rook vs Knight). This data cures the "Horizon Effect" often seen in neural networks during the endgame phase.

• Source: Generated via Syzygy WDL (Win/Draw/Loss) Probing
• Content: 3, 4, and 5 Piece Configurations
• Labeling: Perfect WDL Scores (1.0 = Win, 0.0 = Draw, -1.0 = Loss)
• Format: SQLite Database (.db)

📂 Dataset Structure

The data is stored in an endgame table. Note that multiple database files (e.g., endgame_oracle_w1.db) may exist in the repository, generated by distributed workers.

Column	Type	Description
fen	TEXT (PK)	The endgame position in FEN format.
wdl	INTEGER	The ground truth result: 1 (Win), 0 (Draw), -1 (Loss).
dtz	INTEGER	Distance to Zero (kept as 0 for this version).

🛠️ Usage

import sqlite3
from huggingface_hub import hf_hub_download

# Download the database
db_path = hf_hub_download(
    repo_id="GambitFlow/Synapse-Endgame-tablebase",
    filename="endgame/endgame_oracle_w1.db",
    repo_type="dataset"
)

# Connect
conn = sqlite3.connect(db_path)
cursor = conn.cursor()

# Check a King + Pawn vs King position
cursor.execute("SELECT fen, wdl FROM endgame LIMIT 5")
results = cursor.fetchall()

for fen, wdl in results:
    result_text = "Win" if wdl == 1 else ("Loss" if wdl == -1 else "Draw")
    print(f"Position: {fen} | Result: {result_text}")

⚖️ Credits & License

• Tooling: Syzygy Tablebases by Ronald de Man.
• Tablebase Source: Lichess & Sesse Mirrors.
• Processed By: GambitFlow Team
• License: CC0 1.0 Universal (Public Domain)

This dataset contains derived data from public domain tablebases.