pollywog

Added

• Decompilation Module (pollywog/decomp.py) - Convert .lfcalc files back to Python code
  • decompile() - Read .lfcalc file and generate Python code
  • decompile_to_string() - Convert CalcSet to Python code string
  • Phase 1: Direct conversion (exact structure recreation)
  • Phase 2: Pattern detection (automatic helper recognition)
• Pattern Detection (pollywog/patterns.py) - Smart recognition of common calculation patterns
  • WeightedAverage pattern detection
  • Sum pattern detection
  • Product pattern detection
  • Average pattern detection
  • Confidence scoring (>90% threshold)
  • Graceful fallback to direct code
• Jupyter Magic Command - %pw.load for interactive .lfcalc loading
  • Behaves like built-in %load command
  • First run: decompiles and replaces cell content
  • Second run: executes loaded Python code
  • Integrated with pattern detection
• Examples and documentation for decompilation workflows
  • examples/decompile_example.py - Comprehensive usage examples
  • examples/pw_load_magic_demo.ipynb - Jupyter magic demonstration

Changed

• Decompilation now automatically detects and uses helper functions (WeightedAverage, Sum, Product, Average)
• Generated Python code imports helpers when patterns are detected
• Enhanced CLAUDE.md with decompilation architecture details

Testing

• Added 28 new tests (21 pattern detection + 7 integration tests)
• All 146 tests passing
• Roundtrip validation: .lfcalc → Python → exec → CalcSet → .lfcalc