LECTOR: Complete semantic spaced review to MVP

[zijinz456]

Current State (30-40% complete)

LECTOR is the semantic spaced review system that goes beyond basic FSRS by considering concept interference, confusion pairs, and prerequisite dependencies.

What works:

• Review session endpoint with priority scoring
• FSRS integration for scheduling (stability, retrievability, due dates)
• Priority factors: low_mastery (0.5), never_practiced (0.3), time_decay (0.3), prerequisite (0.2), confusion (0.1)
• Review rating submission with mastery updates

What's needed to reach MVP:

• Interference matrix integration (conceptual similarity threshold: 0.6)
• Proactive review scheduling (background job)
• Session clustering — group related concepts in review sessions
• Consolidation bonuses when prerequisite groups are mastered
• End-to-end test: practice → build mastery → verify review ordering

Key Files

• apps/api/services/lector.py — Core LECTOR service
• apps/api/routers/progress_knowledge.py — Review session endpoints (gated behind ENABLE_EXPERIMENTAL_LECTOR)
• apps/api/config.py — LECTOR factor weights (lines 148-157)

How to Enable

ENABLE_EXPERIMENTAL_LECTOR=true

Tests

• tests/test_lector.py