Add requirements specification v2.1 - production-ready version
This v2.1 represents comprehensive engineering improvements based on three expert reviews and technical feasibility research (2026 latest tech stack). Major improvements over v2.0: **Data Consistency** ⭐ - wrong_count completely driven by attempts (no manual updates) - Enforced rule: any "correct/wrong/unknown" action must create an attempt - Provides rebuild_wrong_counts repair tool **Offline Architecture Clarification** ⭐ - Online: PostgreSQL FTS + pg_jieba - Offline: SQLite FTS5 + trigram tokenization - Event Sourcing sync strategy (incremental action logs) - Outbox queue + idempotency + conflict resolution **FSRS Replaces SM-2** ⭐ - Using dart-fsrs v2.0.0 (official package) - More scientific than SM-2 (adopted by Anki) - Boundary conditions: ease_factor[1.3,2.8], max_interval=90 days - Anti-gaming mechanisms (prevent same-day double attempts) **Multi-question Segmentation Strategy** ⭐ - P0 (MVP): Manual box selection + intelligent hints (stable) - P1 (Week 5-8): Auto-segmentation via PP-DocLayoutV3 - "Puzzle-style adjustment": drag boundaries with auto-snap - Significantly reduces MVP risk (4-week delivery) **Strict KPI Metrics** ⭐ - Cohort definition: Week 1 questions tracked to Week 4 - "Second review" definition: ≥24 hours interval, review_mode='daily_review' - "Heavy wrong" definition: attempts wrong ≥3 on different dates - All metrics aggregatable from attempts/review_sessions tables **Latest Tech Stack (2026)** ⭐ - PP-OCRv5 (May 2025, +13% accuracy) - PP-FormulaNet_plus-M (Chinese formula BLEU 89.76%) - PP-DocLayoutV3 (Jan 2026, non-planar document support) - PostgreSQL + pg_jieba (online FTS) - SQLite FTS5 (offline FTS) - FSRS via dart-fsrs v2.0.0 - pHash image fingerprinting + TF-IDF text similarity **UX Optimizations** ⭐ - Anti-cheating: 3-second delay or key value input for "Got it" button - Anti-frustration: Show hints/tips before full answer on "Don't know" - Puzzle-style adjustment for manual question segmentation - pHash first-pass deduplication (instant, OCR-error proof) **Gamification Prep** ⭐ - Database schema reserves: points, current_streak, longest_streak - badges and user_badges tables - P2 implementation ready **MVP Degradation Paths** ⭐ - Every P0 feature has minimum viable standard defined - Clear fallback options if auto-features fail - 4-week delivery risk significantly reduced Addresses all critical feedback from: - Expert Review 1 (S-grade): UX details, sync strategy, gamification - Expert Review 2 (Tech feasibility): Data consistency, offline arch, MVP scope - Expert Review 3 (Tech research): Latest 2026 tech stack Co-Authored-By: Claude Sonnet 4.5 <[email protected]>
This commit is contained in:
Claude Sonnet 4.5