Release 1.2.0

Added

• File Chunking System - Advanced chunking for large files with 40-line chunks and 5-line overlap preservation
• False Positive Rate Tracking - New metric showing percentage of clean code incorrectly flagged as vulnerable
• Multi-Directory Support - Changed sourceDirectory to sourceDirectories list for scanning multiple directories
• Issue Deduplication - Sophisticated deduplication system for chunk-based analysis preventing duplicate issues
• Enhanced Evaluation Metrics - Added DetectionMetrics and DetectionResults models for comprehensive performance tracking
• Negative Test Cases - Comprehensive 238-line Java file with false positive test scenarios
• Package Reorganization - Moved classes to domain-specific packages: files/, llm/, service/, util/
• New Service Components - Added PromptGenerator, IssueDeduplicator, ScannerDefaults for better code organization

Changed

• Detection Rate Calculation - Fixed calculation to prevent values >100% by tracking unique matched expected issues
• False Positive Rate Logic - Returns 100% when files cannot be analyzed (timeout/error scenarios)
• Code Structure - Refactored CodeAnalyzer following Single Responsibility Principle
• Class Naming - Renamed FileScanner to FileFinder, AnalysisResultMapper to IssueParser
• Configuration Management - Extracted constants to ScannerDefaults object for better maintainability
• Evaluation Directory Structure - Reorganized to test-cases/positive/ and test-cases/negative/
• Model Performance - Updated with latest evaluation results showing significant improvements:
  • ai/phi4:latest now achieves 93.8% detection rate (up from 76.7%)
  • Added zero false positive models: ai/deepcoder-preview:latest, ai/mistral-nemo:latest

Fixed

• Duplicate Scanner Icons - Removed duplicate 🔍 emoji from evaluation output
• Expected Files Exclusion - Added **/expected/** pattern to exclude JSON expected results from scanning
• Detection Rate Accuracy - Fixed double-counting of issues in chunk-based analysis