chess-schema 0.1.2

Pydantic models for structured chess game data, optimized for LLM integration and analysis pipelines

chess-schema

Pydantic models for structured chess game data.

A Python library for parsing, validating, and serializing chess game data with strict type safety. Designed for LLM integration and analysis pipelines where PGN is too messy and you need reliable, validated structured data.

Features

✅ LLM-Optimized: Field descriptions guide LLM outputs, strict validation catches hallucinations
✅ Dual Notation: Both UCI (machine-readable) and SAN (human-readable) for every move
✅ Rich Metadata: Players, ratings, events, dates, comments, variations
✅ Type-Safe: Full Pydantic v2 validation with helpful error messages
✅ Flexible Casing: Python uses snake_case, JSON/LLMs use camelCase
✅ Analysis Trees: Recursive variations for full game analysis support

Installation

# Basic installation
pip install chess-schema

# With PGN parsing support
pip install chess-schema python-chess

Quick Start

Option 1: Parse from PGN

from chess_schema import Game

# Parse a PGN string directly
pgn = """
[Event "World Championship"]
[Site "https://lichess.org/abc123"]
[Date "2024.01.13"]
[Round "1"]
[White "Carlsen, Magnus"]
[Black "Nepomniachtchi, Ian"]
[Result "1-0"]
[WhiteElo "2830"]
[BlackElo "2795"]
[Termination "Normal"]

1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 4. Ba4 Nf6 5. O-O 1-0
"""

game = Game.from_pgn(pgn)
print(f"White: {game.white.name} ({game.white.rating})")
print(f"First move: {game.moves[0].san} (UCI: {game.moves[0].uci})")

Note: from_pgn() requires the python-chess library:

pip install python-chess

Option 2: Build Manually

from chess_schema import Game, Player, Move, GameResult, Termination, GameMetadata
from datetime import date

# Create a game from scratch
game = Game(
    id="lichess_abc123",
    white=Player(name="Alice", rating=1800, title="NM"),
    black=Player(name="Bob", rating=1750),
    moves=[
        Move(uci="e2e4", san="e4", ply=1),
        Move(uci="e7e5", san="e5", ply=2),
        Move(uci="g1f3", san="Nf3", ply=3),
    ],
    result=GameResult.WHITE_WIN,
    termination=Termination.NORMAL,
    metadata=GameMetadata(
        event="Club Championship 2024",
        site="New York, USA",
        date=date(2024, 1, 13),
        round="3"
    )
)

# Serialize to JSON (camelCase for LLMs)
json_output = game.model_dump_json(indent=2, by_alias=True)
print(json_output)

Output:

{
  "id": "lichess_abc123",
  "metadata": {
    "event": "Club Championship 2024",
    "site": "New York, USA",
    "date": "2024-01-13",
    "round": "3"
  },
  "initialFen": "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1",
  "white": {
    "name": "Alice",
    "rating": 1800,
    "title": "NM"
  },
  "black": {
    "name": "Bob",
    "rating": 1750
  },
  "moves": [
    { "uci": "e2e4", "san": "e4", "ply": 1 },
    { "uci": "e7e5", "san": "e5", "ply": 2 },
    { "uci": "g1f3", "san": "Nf3", "ply": 3 }
  ],
  "result": "1-0",
  "termination": "normal"
}

LLM Integration Example

from chess_schema import Game
import anthropic

# Generate schema for LLM prompt
schema = Game.model_json_schema()

client = anthropic.Anthropic(api_key="your-api-key")
response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=2000,
    messages=[{
        "role": "user",
        "content": f"""Analyze this PGN game and return structured JSON matching this schema:

{schema}

PGN:
1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 4. Ba4 Nf6 5. O-O Be7 1-0

Return ONLY valid JSON, no markdown."""
    }]
)

# Parse LLM output
game = Game.model_validate_json(response.content[0].text)
print(f"Game ID: {game.id}")
print(f"Winner: {game.result}")

Or use the built-in parser:

# If you already have PGN, just parse it directly
pgn = "1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 4. Ba4 Nf6 5. O-O Be7 1-0"
game = Game.from_pgn(pgn)

# Then use LLM for analysis/annotation
response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1000,
    messages=[{
        "role": "user",
        "content": f"Analyze this chess game and identify key moments: {game.model_dump_json()}"
    }]
)

Adding Comments and Variations

from chess_schema import Move, Comment

# Move with engine analysis
move = Move(
    uci="e2e4",
    san="e4",
    ply=1,
    comments=[
        Comment(text="Best by test", source="Stockfish 16"),
        Comment(text="King's Pawn Opening", source="ECO"),
    ]
)

# Add a variation
alt_move = Move(uci="d2d4", san="d4", ply=1)
move.variations.append([alt_move])  # 1.d4 as alternative to 1.e4

# Or parse from PGN with comments/variations
pgn_with_analysis = """
1. e4 { Best by test } e5 (1... c5 { Sicilian Defense }) 2. Nf3
"""
game = Game.from_pgn(pgn_with_analysis)
print(game.moves[0].comments[0].text)  # "Best by test"
print(game.moves[0].variations[0][0].san)  # "c5"

Validation Features

chess-schema validates:

• ✅ UCI format (4-5 chars, valid squares, promotion pieces)
• ✅ FEN structure (6 fields, 8 ranks)
• ✅ Sequential ply numbers
• ✅ Rating ranges (0-4000)
• ✅ No extra fields (catches LLM hallucinations)
• ✅ URL formats for source links

from pydantic import ValidationError

try:
    # Invalid UCI (wrong format)
    Move(uci="e2-e4", san="e4")  # ❌ Raises ValidationError
except ValidationError as e:
    print(e)

try:
    # Extra fields rejected
    Game(
        id="test",
        secret_field="hack",  # ❌ Raises ValidationError
        # ... other required fields
    )
except ValidationError as e:
    print("Extra field rejected:", e)

API Reference

Core Models

• Game: Complete game with metadata, players, moves, and result
• Move: Single move with UCI, SAN, comments, and variations
• Player: Player info (name, rating, title)
• GameMetadata: Event, site, date, round, tags
• Comment: Move annotation with source attribution

Enums

• GameResult: WHITE_WIN ("1-0"), BLACK_WIN ("0-1"), DRAW ("1/2-1/2"), UNTERMINATED ("*")
• Termination: NORMAL, TIME_FORFEIT, RULES_INFRACTION, ABANDONED, etc.
• Color: WHITE, BLACK

Base Class

• ChessBaseModel: Inherit for custom models with same validation behavior

Why chess-schema?

Problem: PGN is great for humans but messy for code. It's ambiguous, inconsistent, and hard to parse reliably—especially when dealing with LLM outputs.

Solution: chess-schema provides:

1. Strict validation that catches errors early
2. Clear structure that LLMs can reliably produce
3. Type safety for confident pipeline development
4. Rich metadata for analysis and filtering

Project Structure

chess-schema/
├── chess_schema/
│   ├── __init__.py      # Public API exports
│   ├── base.py          # ChessBaseModel configuration
│   ├── enums.py         # GameResult, Termination, Color
│   ├── move.py          # Move and Comment models
│   └── game.py          # Game, Player, GameMetadata models
├── tests/               # Unit tests
├── README.md
└── pyproject.toml

Contributing

Contributions welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Add tests for new functionality
4. Submit a pull request

License

MIT License - see LICENSE file for details

Credits

Built with Pydantic v2.

---

Questions? Open an issue on GitHub or check the documentation.