v0.5.2 — Increase Maze Size

[0.5.2] increase maze size

Added

• Enhanced Maze Size Support: Increased maximum maze size from 32×32 to 64×64
• Updated validation constants in strategic_evaluator.py
• Modified prompt.md to reflect new size constraints
• Updated README.md documentation
• Multiple Run Support: Leaderboard now supports multiple runs per model
• Migrated from single-result dict format to list-based storage
• Enhanced get_rankings() to include all runs for each model
• Added migration logic for backward compatibility
• Updated ingest_manual_output to generate timestamped output files
• Improved Rescoring: Enhanced --rescore functionality
• Searches for multiple benchmark output files per model
• Preserves existing metadata (timing, token usage) during rescore
• Better error handling and file discovery

Changed

• Leaderboard Storage: Migrated from single-result to multiple-run format
• Supports historical comparison and progress tracking
• Maintains backward compatibility with existing data
• Input Processing: Enhanced pattern matching and file handling
• Updated regex patterns for better model detection
• Added timestamp support for multiple manual runs

Fixed

• Regex Pattern Issues: Improved model header detection in manual output ingestion
• File Overwriting: Prevented multiple manual runs from overwriting previous results
• Metadata Preservation: Fixed token usage and timing data loss during rescoring

Improved

• Scalability: Support for larger, more complex mazes (up to 64×64)
• Performance: Better file handling for multiple benchmark outputs
• User Experience: Enhanced error messages and file management