- Added structured warning system with `AdminAiWarning` type in backend and Flutter
- Implemented detailed reason descriptors with `FlyerReasonDescriptor` for parse and match operations
- Added `legacyWarnings` field to maintain backward compatibility
- Enhanced AI trace service to collect and format warnings with item-level context
- Updated flyer import services to include detailed reason descriptions in responses
- Added Swedish diacritic preservation for cheese variants (Prästost, Herrgårdsost, Grevéost)
- Implemented UTF-8 content validation for AI responses
- Added new reason code definitions in `reason-codes.ts`
- Updated Flutter UI to display structured warnings with severity indicators
- Added error report generation and copy functionality in admin panel
- Added comprehensive test coverage for new warning system and cheese normalization
BREAKING CHANGE: AI trace warnings are now structured objects instead of simple strings
- Add FlyerTraceSupplement type for AI trace metadata
- Implement getFlyerTraceSupplements method to fetch trace supplements
- Update AiTraceService to include prompt/rawOutput and counters in flyer traces
- Add persistFlyerTrace method to FlyerImportService for trace persistence
- Enhance AiFlyerParserService to return structured trace data with prompts and retries
- Update FlyerNormalizerService with OCR typo fixes for cheese variants and spröd bakad firre
- Improve Flutter admin panel with selectable text, warnings display, and tooltips
- Add comprehensive tests for AI trace supplements and normalization rules
- Add logic to split Swedish cheese variants (Präst, Herrgård, Grevé) into separate products
- Implement brand normalization for "Arla Ko" and category assignment to "Hårdost"
- Update flyer parser with detailed rules for bundle/group announcements
- Add unit tests for variant splitting and brand/category normalization
- Replace single-item return with flatMap for expanded product lists
- Add bundle support with isBundle, bundleWeight, and bundleItems fields
- Add brand, weight, and comparisonUnit fields to FlyerItem model
- Update AI flyer parser to extract bundle information
- Add sanitization for bundle items in FlyerNormalizerService
- Update DTOs and interfaces to include new fields
- Add migration for new database fields
- Update tests to cover bundle item handling
- Replace fixed text window retry strategy with dynamic chunk-based approach
- Add configurable chunk size, overlap, and maximum chunks via environment variables
- Implement chunk splitting with overlap handling for context preservation
- Add chunk processing with retry logic per chunk
- Include deduplication of items across chunks
- Update logging to reflect chunk-based processing
BREAKING CHANGE: Changes the retry strategy from fixed text windows to dynamic chunk-based processing. Existing configurations may need adjustment for FLYER_AI_CHUNK_SIZE_CHARS, FLYER_AI_CHUNK_OVERLAP_CHARS, and FLYER_AI_MAX_CHUNKS.
- Add FLYER_AI_TIMEOUT_MS and FLYER_AI_RETRIES environment variables
- Configure timeout and retry settings in compose.yml
- Update AiFlyerParserService with configurable timeout and retry logic
- Add text window reduction strategy for retry attempts
- Update documentation in TEKNISK_BESKRIVNING.md
- Fix ESLint configuration in app.security.spec.ts
- Add support for PNG, JPEG, and WebP image formats in flyer import
- Replace external importer service with internal AI-based parsing pipeline
- Add new services: TextExtractorService, AiFlyerParserService, FlyerNormalizerService
- Integrate Mistral AI, pdf-parse, and tesseract.js dependencies
- Add quality confidence indicators and warning panels in Flutter UI
- Update package.json with new dependencies and transform ignore patterns
- Add documentation for flyer importer system
- Add Kilo AI planning file for Happy Island project
BREAKING CHANGE: Flyer import now uses internal AI parsing instead of external importer service
Nils-Johan Gynther