v1.5.0: Simplified 2-Tier Format Support

🎉 Release v1.5.0

🎯 Simplified Format Support System

This release simplifies the format support system from 3 tiers to 2 tiers, making benchmarking choices clearer and more focused on document extraction use cases.

Format Tiers (Simplified)

Tier 1: Universal Support (5/5 frameworks) ✅

• 7 formats: PDF, PPTX, XLSX, PNG, BMP, HTML, CSV
• 100% framework support for fair performance comparison
• Use: --format-tier universal

Tier 2: Common Support (4/5 frameworks) 🌟 RECOMMENDED

• 11 formats: Tier 1 + XLS, MD, JPEG, TXT
• 80%+ framework support for real-world scenarios
• Use: --format-tier common (default)

All Formats 📊

• 18 formats: Complete capability assessment
• Use: --format-tier all

🔄 What Changed

Removed Tier 3 (Partial Support):

• Eliminated formats with only 3/5 framework support
• Removed: JPG variant, EML, MSG, JSON, YAML
• These specialized formats (email, data) don't align with document extraction benchmarking

💡 Why This Change?

1. Simpler Choice: Only decide between fair comparison vs real-world testing
2. Better Focus: Concentrates on actual document formats
3. Clearer Purpose: Document extraction benchmarks shouldn't include email/data formats
4. User-Friendly: Reduces confusion about which tier to choose

📊 Expected Success Rates

• Tier 1: All frameworks achieve ~100% success
• Tier 2: Most frameworks achieve 85-95% success
• All: Shows full capabilities including edge cases

🚀 Usage

# Fair performance comparison
uv run python -m src.cli benchmark --framework all --format-tier universal

# Real-world benchmarking (default and recommended)
uv run python -m src.cli benchmark --framework all --format-tier common

# Complete framework evaluation
uv run python -m src.cli benchmark --framework all --format-tier all

🔧 Technical Details

• Updated config.py to remove PARTIAL_FORMATS
• CLI now accepts only: universal, common, all
• CI workflow updated to reflect 2-tier system
• Documentation simplified throughout

This release makes the benchmarking suite more intuitive while maintaining its effectiveness for comparing text extraction frameworks on relevant document formats.