Add history heuristic for quiet move ordering

[luccabb]

Summary

Track which quiet moves (from_square → to_square) cause beta cutoffs across the search tree using a 64×64 history table. Quiet moves with higher history scores are searched first, dramatically improving alpha-beta cutoff rates.

• History table: 64×64 integer array (from_square, to_square) initialized to zero at search start
• Update rule: On beta cutoff for a quiet move, increment by depth² (deeper cutoffs weighted more)
• Move ordering: Quiet moves sorted by history score descending (replaces random shuffle)
• Backward compatible: organize_moves() accepts optional history parameter; without it, behavior is unchanged

Benchmark results (depth 4)

Metric	Before	After	Change
Total nodes	4,760,507	3,124,410	−34.4%
NPS	22,634	24,696	+9.1%
Total time	210.32s	126.51s	−39.8%

Notable per-position improvements:

• Position 46: 986,591 → 20,495 (−97.9%)
• Position 1: 16,405 → 6,060 (−63.1%)
• Position 8: 134,526 → 51,062 (−62.1%)

Local Stockfish Benchmark

Settings: 20 games, Stockfish skill 3, 10s/move, no opening book.

	W	L	D	Win Rate
Master (baseline)	19	1	0	95%
This PR	20	0	0	100%

Use /run-stockfish-benchmark for CI validation with opening book and longer time control.

Test plan

• All alpha_beta unit tests pass (16/16)
• /run-nps-benchmark
• /run-stockfish-benchmark

[github-actions]

Benchmarks

The following benchmarks are available for this PR:

Command	Description
/run-nps-benchmark	NPS speed benchmark (depth 5, 48 positions)
/run-stockfish-benchmark	Stockfish strength benchmark (300 games)

Post a comment with the command to trigger a benchmark run.

[greptile-apps]

Greptile Summary

Implements history heuristic for quiet move ordering using a 64×64 table indexed by (from_square, to_square). Quiet moves that cause beta cutoffs increment their history score by depth², and subsequent searches prioritize high-scoring moves. This dramatically improves alpha-beta pruning efficiency (−34.4% nodes, −39.8% search time at depth 4).

Key changes:

• Added new_history_table() function and HISTORY_TYPE alias in alpha_beta.py
• History table created per search in search_move() and passed through all negamax() calls
• Beta cutoffs for quiet moves increment history[from_square][to_square] += depth * depth
• organize_moves() now sorts quiet moves by history score descending (when history provided)
• Backward compatible: parallel engines (LazySMP, layer-based) will use history=None and maintain existing behavior

Architecture:

• History table is per-search (not shared across parallel processes), avoiding race conditions
• Only quiet moves tracked (captures excluded via board.is_capture() check)
• Depth-squared weighting ensures deeper cutoffs have more influence

Confidence Score: 5/5

• This PR is safe to merge - well-tested optimization with clean implementation
• Score reflects a clean, focused implementation of a standard chess optimization technique. The history heuristic is correctly implemented (depth² weighting, quiet-only, beta-cutoff triggered), properly threaded through recursion, and backward compatible. No race conditions (per-search table), no logic errors, and impressive benchmark results validate the approach. All existing tests pass.
• No files require special attention

Important Files Changed

Filename	Overview
moonfish/engines/alpha_beta.py	Added history heuristic with 64×64 table tracking quiet moves causing beta cutoffs, properly threaded through negamax recursion
moonfish/move_ordering.py	Updated organize_moves() to sort quiet moves by history score when provided, maintains backward compatibility

Flowchart

flowchart TD
    A[search_move: Initialize history table 64×64] --> B[Call negamax with history]
    B --> C{Depth > 0?}
    C -->|No| D[Quiescence search]
    C -->|Yes| E[organize_moves with history]
    E --> F[Sort: captures first, then quiet moves by history score]
    F --> G[Loop through moves]
    G --> H[Recursive negamax with history]
    H --> I{Score >= beta?}
    I -->|Yes - Beta cutoff| J{Is quiet move?}
    J -->|Yes| K[history from_sq to_sq += depth²]
    J -->|No - Capture| L[Skip history update]
    K --> M[Return beta cutoff]
    L --> M
    I -->|No| N[Update alpha, continue loop]
    N --> G
    D --> O[Return evaluation]
    M --> O

Loading

_{Last reviewed commit: b85883e}

[greptile-apps]

_{2 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}