ArGO v1.0.0

Major Release Highlights

This first major release of the ArGO chess engine marks a significant milestone with substantial improvements in playing strength, stability, and performance. The engine now incorporates a fully-functional NNUE evaluation system and sophisticated search techniques that bring it to a competitive level.

New Features

• NNUE Evaluation System: Implemented an Efficiently Updated Neural Network for position evaluation, ported from Koivisto
• Advanced Time Management: Dynamic time allocation based on position complexity and search stability
• Improved Search Algorithms:
  • Principal Variation Search (PVS) for more efficient alpha-beta pruning
  • Enhanced Late Move Reduction with dynamic depth adjustments
  • Null move pruning for quick detection of strong positions
• Transposition Table Enhancements:
  • Age-based replacement strategy
  • Improved hash entry utilization
• Magic Bitboards: Optimized move generation for sliding pieces with perfect hashing
• Check Extensions: Automatic search depth extension in check positions
• Mate Distance Pruning: Early termination for detected forced mates

Performance Improvements

• Assembly Optimizations: Critical NNUE calculations using hand-tuned assembly for AMD64
• Move Ordering Refinements:
  • Fine-tuned MVV-LVA capture scoring
  • Improved history heuristic scaling
  • Better killer move selection
• Reduced Memory Footprint: More efficient data structures for core components
• Cache-Friendly Data Layouts: Restructured critical arrays for better cache utilization

UCI Compliance

• Full UCI Protocol Support: Compatible with standard chess GUIs
• Configurable Options: Hash size, threads, and other engine parameters
• Detailed Search Information: Proper reporting of depth, score, nodes, and principal variation

Technical Documentation

• Added comprehensive documentation for all major engine components
• Detailed explanations of core algorithms and data structures
• Architecture overview and design principles

Bug Fixes

• Fixed evaluation inconsistencies in endgame positions
• Corrected castling detection in certain edge cases
• Resolved issues with en passant capture generation
• Fixed time management in tournament time controls
• Eliminated memory leaks in long-running searches

Known Limitations

• NNUE assembly optimizations only available on Linux AMD64 platforms
• No multi-threaded search implementation yet
• No endgame tablebase support

Future Plans

• Three-fold repetition check
• Move Generation through iterators
• Aspiration windows
• Updated Transposition Table and Reduction Table
• Refactoring the board implementation for easier move making
• Multi-threaded search
• Syzygy tablebase integration
• NNUE training infrastructure
• More configurable search and evaluation parameters