Frequently Asked Questions

Each question tests multiple knowledge points at once (weights are split), errors are attributed to confusion / carelessness / knowledge gaps, and the training mix adjusts dynamically with frustration — avoiding 'the more I practice, the more it hurts'.

The review queue is driven by a simplified forgetting model, retention = exp(-t / stability); the more overdue an item is, the higher its priority.

AI Tutor / conversation / characters are in Phase 5+. Your learning state (what you know) is stored independently in the database and doesn't rely on LLM memory.